This document summarizes a report on short term load forecasting focusing on older power generation companies and power trading companies, using the case of NTPC Dadri power plant. The report aims to forecast load 4 hours ahead to help older plants better plan generation and utilize excess capacity by selling power on the short term market. An artificial neural network model was developed in Matlab to provide adaptive short term load forecasts to aid in operations and cost optimization. Accurate short term forecasts can help older plants schedule maintenance, fuel procurement and manage costs more efficiently.
This document summarizes a report on short term load forecasting focusing on older power generation companies and power trading companies, using the case of NTPC Dadri power plant. The report aims to forecast load 4 hours ahead to help older plants better plan generation and utilize excess capacity by selling power on the short term market. An artificial neural network model was developed in Matlab to provide adaptive short term load forecasts to aid in operations and cost optimization. Accurate short term forecasts can help older plants schedule maintenance, fuel procurement and manage costs more efficiently.
This document summarizes a report on short term load forecasting focusing on older power generation companies and power trading companies, using the case of NTPC Dadri power plant. The report aims to forecast load 4 hours ahead to help older plants better plan generation and utilize excess capacity by selling power on the short term market. An artificial neural network model was developed in Matlab to provide adaptive short term load forecasts to aid in operations and cost optimization. Accurate short term forecasts can help older plants schedule maintenance, fuel procurement and manage costs more efficiently.
This document summarizes a report on short term load forecasting focusing on older power generation companies and power trading companies, using the case of NTPC Dadri power plant. The report aims to forecast load 4 hours ahead to help older plants better plan generation and utilize excess capacity by selling power on the short term market. An artificial neural network model was developed in Matlab to provide adaptive short term load forecasts to aid in operations and cost optimization. Accurate short term forecasts can help older plants schedule maintenance, fuel procurement and manage costs more efficiently.
Download as DOCX, PDF, TXT or read online from Scribd
Download as docx, pdf, or txt
You are on page 1of 41
1
Short term Load Forecasting Focusing on Power Utilities
through Power Trading Company: Case Analysis NTPC Dadri
Summer Training Report at NTPC Vidyut Vyapar Nigam Submitted towards partial fulfilment of criteria for the award of PGPM (Energy) by Great Lakes IEMR
Submitted By Axit Jain P121013 PGPM (Energy) 2011-2013
Under the Guidance of
External Guide Internal guide Mr. Rakesh Pandey Dr Meenu Mishra NTPC Badarpur Plant Email: mmishra35@rediffmail.com Email: pandeyrakeshntpc@gmail.com
2 CERTIFICATE
It is certified that Mr. Axit Jain, a PG student at Great Lakes Institute of Energy Management and Research, Gurgaon has undergone Summer Internship at NTPC Limited Badarpur Thermal Power Station, New Delhi from 30 April To 29 June 2013.
His project titled Short term Load Forecasting of power focusing on Power Trading Companies has been completed to our satisfaction and the present report is a bona fide account of the same.
External Guide Internal guide Mr. Rakesh Pandey Dr Meenu Mishra NTPC Badarpur Plant Email: mmishra35@rediffmail.com Email: pandeyrakeshntpc@gmail.com
3 DEDICATION
I dedicate this project to my Mentor, Mr. Umesh Pareek, Additional General Manager (Business Excellence) at NVVN Ltd., who helped me us throughout this project of Demand Forecasting. He gave us the idea to work on this project from a new direction, i.e using artificial neural networks. I had no idea from where to begin and how would I achieve the results expected from me, but just because of his constant support and guidance I could saw my way through it. He did everything in his capacity to enable me to understand the technicalities of the subject and imbibed in me the confidence needed to complete the task. His mentorship helped me in becoming a person with a greater character and helped me learn a lot about the corporate life and also taught me how to get the work done under crunch situations. It is truly my privilege to have worked with him.
I would also to dedicate this project to Mr. P.K Jena, Additional General Manager (Business Development) gave me this interesting project of Demand Forecasting to work on. He always helped me in keeping a clear line of sight to achieve my goal. Whenever I got loaded up with problems and was about to give up, his motivating words always helped me even try harder to achieve the task assigned to me.
I would also like to dedicate this project to the people who are working in power exchanges towards improving the current power situation in India. This project can help the power exchanges in better planning and management of the power generated thus helping India in becoming a power surplus country in the near future.
4 ACKNOWLEDGEMENT
I wholeheartedly thank Mr. Rakesh Pandey, my mentor at NTPC, for providing me with the opportunity to work at NVVN and facilitating the process at a very short notice. No formal word of acknowledgement is enough to express my thankfulness to him. I am greatly indebted to my internal guide, Dr. Meenu Mishra. Her assistance contributed immensely to the project. She was instrumental in enabling me to perform efficiently and was always available to offer her guidance. I acknowledge the contribution of all members of NVVN Ltd and NTPC Ltd to the research. I especially appreciate their cooperation in providing me the data needed to forecast the load. Inputs from Mr. Radesh Kumar Sabharwal, Deputy General Manager (EMG) and Sunil Kumar Maheshawari, Additional General Manager (EEMG) particularly contributed to the progress of the project significantly. I am thankful to the faculty of Great Lakes IEMR for having shaped my mind over the past one year and being ever-ready to make sure that I made this one of the most productive periods in my life. I thank Dr. Himadri Das, Director, Great Lakes IEMR for having been the fatherly figure that he has been and for his guidance and encouragement. I thank Mr Vikas Prakash, faculty at Great Lakes IEMR and Mr. Hari Nair, Placement Coordinator, Great Lakes IEMR, for having enabled my association with NTPC. I also thank the Management of Great Lakes IEMR for having provided me the opportunity to study at such an exceptional institution. In the hindsight, it would also like to thank everyone who has directly or indirectly helped me in innumerable ways and varying proportions as the successful completion of this project would not have been possible without them, even though their names might not have been mentioned here.
5 Certificate of Originality and Authenticity
I, Axit Jain, student of Post Graduate Program in Management 2012-14, Great Lakes Institute of Management, Gurgaon hereby declare that I have completed my summer internship on the topic " Short term Load Forecasting Focusing on Power Utilities through Power Trading Company: Case Analysis NTPC Dadri " at National Thermal Power Corporation (NTPC), New Delhi from 30th April, 2013 to 29th June, 2013. I further declare that the information presented and analysis done in the report is true and original to the best of my knowledge. I also assure that the work that forms the basis of this project "Short term Load Forecasting of Power focusing on Old Generation Companies and Power Trading Companies " and the recommendation provided for Indian Context is the result of the original analysis done by me and has not been submitted anywhere else for the award of any graduation or degree.
Axit Jain Roll. No. P121013 PGPM 2012-14 Great Lakes Institute of Management, Gurgaon
6 Executive Summary
This report on Short term Load Forecasting Focusing on Power Utilities through Power Trading Company: Case Analysis NTPC Dadri is applied to forecast 4 hour ahead load keeping in focus the old generators and power traders. In our Project for short term load forecasting we would be using artificial intelligence methods. There are other methods and tools which could have been used for STLF but we used this tool because this is an adaptive model which handles complexity and non linearity in a better way than the conventional mathematical model and the uncertainties are better covered in this model. In India the power trading market has a huge potential as electricity demand is growing by leaps and bounds and People being aware of their rights demand quality and reliable power 24x7. Earlier in India, only long term forecasting was done, concerning future fuel prices and technical improvements and there was not much competition in the electricity market but the scenario has totally changed in the past decade. Rise of competitive energy markets have forced the companies to forecast with accuracy and also for shorter periods like hour ahead, day ahead, week ahead etc. Short term forecasting of electrical load is important for optimum operation planning of power generation facilities as it affects both system reliability and fuel consumption. Accurate forecasting of electricity and power demand determines the utility to match its generation capabilities to the expected requirements. A short-term forecast is important for unit commitment, economic dispatch, load management, etc. Short Term forecast plays an important role in the day-to-day operations of a utility and is typically used for optimizing system operation and scheduling of hydro units and other peaking plants, such as gas turbines. The objective of the operators is to minimize variable costs without jeopardizing the electric system to power failures. The short-term (one to twenty-four hour) load forecast is of importance in the daily operations of the utility. The life of a generation plant is considered to be 25 years, but because of the innovation in technology new plants are coming up with greater efficiency which require less input in terms of fuel and give more output in terms of electricity generated. Thus the price of power generated is much cheaper than these old plants (in our case NTPC Dadri) thereby making it difficult for these old plants to sell their power. Discoms though having a PPA signed with these plants do not take power from these and as a result their capacities go underutilized. Thus keeping in mind these generation plants, we implemented this project which will them forecast load and have insights about the load flow. If the discoms do not buy power from them they can sell their power in the short term market at a better price thereby recovering the fixed cost invested in the project. It can help them plan their shutdown of generating stations on merit order dispatch basis during periods of low demand thus making them cost effective, scheduling of fuel purchase months ahead thereby saving cost and also help in maintenance scheduling. Forecasting model has been designed on the Matlab platform. The network architecture comprises of 17 input variables, 2 hidden layers in which 6 neurons were placed in first layer
7 and 10 neurons were placed in second layer. The input variables used to forecast the demand are wind speed of each hour, temperature of each hour, relative humidity, demand for previous week, ramp demand (last day), Demand (T-1, T-2, T-3, T-4, T-19, T-20, T-21, T- 22, T-23, T-24, T-25) and average temperature. We tried using various algorithms like Resilient Back propagation, Levenberg-Marquardt, Scaled Conjugate Gradient, Bayesian Regularization etc but the best result was given by Levenberg-Marquardt which is also used by many scholars and researchers and is the fastest algorithm in the whole lot.
MAPE obtained by Levenberg-Marquardt for this dataset was % 2.7%. The regression of the overall model is 97.108. The regression of Training dataset is 98.9%. The regression for validation and testing datasets are 94.02% and 95.012%.
8 Table of Contents CERTIFICATE ............................................................................................................................................ 2 DEDICATION ............................................................................................................................................ 3 ACKNOWLEDGEMENT ............................................................................................................................. 4 Certificate of Originality and Authenticity .............................................................................................. 5 Executive Summary ................................................................................................................................. 6 List of Tables ......................................................................................................................................... 10 List of Figures ........................................................................................................................................ 10 1. Introduction .................................................................................................................................. 11 1.1 Title of the Project ...................................................................................................................... 11 Short term Load Forecasting Focusing on Power Utilities through Power Trading Company: Case Analysis NTPC Dadri ....................................................................................................................... 11 1.2 Necessity of the project .............................................................................................................. 11 1.3 Scope of the project .................................................................................................................... 11 1.4 Objective ..................................................................................................................................... 12 1.5 Research Methodology ............................................................................................................... 12 1.6 Limitations ................................................................................................................................... 12 1. Literature review ........................................................................................................................... 13 3. NTPC VIDYUT VYAPAR NIGAM LTD. ................................................................................................. 15 3.1 COMPANY PROFILE ..................................................................................................................... 15 3.2 MAIN FUNCTIONS ....................................................................................................................... 15 3.3 ORGANISATIONAL STRUCTURE ................................................................................................... 16 3.4 BUSINESS DEVELOPMENT INITIATIVES ....................................................................................... 16 3.5 NVVN - STRENGTHS / COMPETITIVE ADVANTAGES .................................................................... 17 3.6 MARKET DEVELOPMENT INITIATIVES: ........................................................................................ 17 4. Short Term Power Trading ............................................................................................................ 18 4.1 Introduction ................................................................................................................................ 18 4.2 Short-Term Load Forecasting (STLF) ........................................................................................... 18 4.3 STLF in Power Trading ................................................................................................................. 19 5. Load Forecasting ........................................................................................................................... 20 5.1 Introduction ................................................................................................................................ 20 5.2 Types of Loads ............................................................................................................................. 20 5.3 Why do Load Forecasting? .......................................................................................................... 20 5.4 Types of Load Forecasting ........................................................................................................... 21
9 5.5 Factors Affecting Load Forecasting ............................................................................................. 21 5.6 Methods or Tools Used for Load Forecasting ............................................................................. 22 5.7 Advantage of Neural Networks over other techniques .............................................................. 23 6. Electricity Demand Forecasting using Artificial Neural Networks (ANN) ......................................... 25 6.1 Introduction ................................................................................................................................ 25 6.2 Applications of ANN .................................................................................................................... 25 6.3 Construction, Training and Working of ANN .............................................................................. 25 6.4 Network Architectures ................................................................................................................ 28 6.5 Network Configuration ............................................................................................................... 29 7. Case Study Demand Forecasting of UP region ........................................................................... 31 7.1 Case Study ................................................................................................................................... 31 7.2 Simulation Design ....................................................................................................................... 32 7.3 Research Data ............................................................................................................................. 32 7.4 Construction of Network Architecture ....................................................................................... 33 8. Results & Observations ................................................................................................................. 34 8.1 Results ......................................................................................................................................... 34 8.2 Observations ............................................................................................................................... 37 9. Conclusion ..................................................................................................................................... 39 10. Bibliography .................................................................................................................................... 40 11. ANNEXURE ...................................................................................................................................... 41
10 List of Tables
Table 1 Anticipated and actual demand data for Uttar Pradesh LGBR 2013-14 ............................ 31 Table 2 Coal Based Power Station ........................................................................................................ 32 Table 3 Gas Based Power Station ......................................................................................................... 32 List of Figures
Figure 1: Input-Output Schematic for Load Forecasting ...................................................................... 22 Figure 2: Neuron Model ........................................................................................................................ 26 Figure 3: Linear Transfer Function ....................................................................................................... 26 Figure 4: Log-Sigmoid Transfer Function ............................................................................................ 27 Figure 5: Neuron With Vector Input ..................................................................................................... 27 Figure 6: One Layer of Neurons ........................................................................................................... 28 Figure 7: Multiple Layer of Neurons .................................................................................................... 29 Figure 8: Network Architecture ........................................................................................................... 34 Figure 9: Actual Demand Load Curve .................................................................................................. 34 Figure 10: Forecasted Load Curve v/s Demand Load Curve ................................................................ 35 Figure 11: Overall Regression Plot .................................................................................................... 35 Figure 12: Regression plot of Training, Validation and Testing datasets .......................................... 36 Figure 13: Error Histogram ................................................................................................................... 37
11 1. Introduction
1.1 Title of the Project Short term Load Forecasting Focusing on Power Utilities through Power Trading Company: Case Analysis NTPC Dadri 1.2 Necessity of the project NVVN acts as platform where buyers and sellers meet and agree to a set of terms decided mutually. It becomes very difficult for NVVN to identify buyers or sellers as Indian Electricity market is huge and is divided into 5 parts. Weather changes and other situations which like World Cup matches, Festive occasions, Processions, National Holidays, rainfall etc affect the load significantly on a short term basis, creating a difference between demand and supply which leads to fluctuating frequencies, load imbalances, and excessive load on transmission lines, ultimately leading to system failure. These situations also result in losses lead because of the increased variable cost and at times results in increased energy prices by a factor of ten or more for utilities due to peaking load demand.
Old generating stations find it difficult to recover their fixed cost invested in the project because of the coming up of technologically advanced large generating stations(UMPPs projects like Mundra, Sasan) which sell their power at a price as low as Rs 1.8 to Rs 3. It becomes very difficult for the old power stations to stay in the business and always face a risk of incurring losses. Renovation and modernization cost incurred increases the life of these plants but the cost of power is still nowhere near to these new plants. Thus they can the short term power market to sell their power instead of underutilizing their capacity and running the plants at low PLF and sitting idle. 1.3 Scope of the project The principal objective of the project is to in forecasting load on a 4 hour ahead helping old generating station and also power traders to plan their bidding strategies, listing out their prospective buyers and sellers and also achieve operational efficiency. NVVN would have insights about the future load trend and thus they can accordingly advise the generating stations to increase or decrease load. Thus saving or minimizing variable costs and also prevent load peaking. Short-term load forecasting can help to estimate load flows and to make decisions that can prevent overloading. Timely implementations of such decisions lead to the improvement of network reliability and to the reduced occurrences of equipment failures and blackouts.
12 1.4 Objective Short Term load forecasting for old generating stations helping them tackle the frequent shut down due to low power demand from Discoms and provide the way of selling their power on short term power market and realizing better returns. Thereby utilizing their idle resources and recovering the fixed cost invested in the project. Providing Power traders an insight about the future load flow and thus helping them identify their prospective buyers and sellers thereby improving their profitability. 1.5 Research Methodology The research methodology employed during the course of this research was both quantitative in nature. For quantitative research, secondary sources of data were used. Demand data was collected for the state of UP, which is being supplied by NTPC Dadri. Data for the months of October, November, and December 2012, which is recorded there in a log book daily on hour basis. Weather data was collected from Environment Monitoring Group (EMG) of the NTPC Dadri plant. The model was built on MATLAB-2011a platform. Data Compilation of the obtained data was performed using Microsoft Excel. Others parameters which also affect load forecasting were also identified and were built using the collected load data.
1.6 Limitations Some of the variables which influence the electricity demand such as fluctuating fuel prices, regulatory changes etc are not considered as inputs. This variable may affect the forecasting accuracy. Weather data used in the model is not the data for the entire state but it is taken at the site of NTPC Dadri, which may be different from the weather conditions from other regions of the state thus affecting the accuracy of the forecast. Demand Data collected was only for 45 days which make it difficult for us to achieve an netter accuracy and reduce the error margin to 1-2 %.
13 1. Literature review
Economical reliability of a power system depends significantly on the accuracy of forecasted load. Accurate load forecasting helps in improving the operational efficiency of a power system by proper scheduling of fuel and water supply, maintenance of generators and most importantly helps in matching demand and supply. Many researchers have studied forecasting results from different mathematical, statistical, simulation and artificial intelligent tools and have found that only artificial intelligence and some statistical tools are best suited for short term load forecasting as they handle the uncertainty well. They are adaptive in nature and have very high accuracy levels and the margin of error is up to 1-2%.
V.H. Kher and S.K. Joshi from the Maharaja Sayajirao University of Baroda, Vadodara tried short term electric load forecasting for Gujarat Electricity Board. They did not include weather data for their prediction as they thought that in Gujarat, the temperature profile on a particular day will not drastically differ from that of the preceding weekday and it will not affect much when doing a short term forecast. They used Functional Link Networks which is a hybrid neural model and found that it had a significant advantage over conventional artificial neural network models. The hybrid model was much easier to use and gave good performance. The estimation of the network weights in the functional link networks is linear and their use in the hybrid model enables a simple implementation of an adaptive scheme in which weights are re-estimated.
In 2009, scholars A. A. Rasool, A. A. Fttah and I.B.Sadik forecasted Short Term Load based on combining the Wavelet and Neural Network, Wavelet Transform is used to enhance the learning capability of the Artificial Neural Networks. Due to the lack of weather information only the average or maximum and minimum Temperatures were used. They forecasted the load using real data of Erbil in the Iraq Kurdistan Region, for the data of 2006, the MAE% for the last week of August was 2.99 % by using average temperature, while it is 4.56% by using maximum and minimum temperature. In 2013 scholars Amera Ismail Melhum, Lamya abd allateef Omar and Sozan Abdulla Mahmood from Duhok University in Iraq tried load forecasting which takes into account the effect of the load being cut off for a certain period of time. They took data of the city Duhok which has a sort of stability in electricity demand and the amount of cut hours varies between 5-12 hours as in other cities the cut off region vary from 17 20 hrs. They used back propagation algorithm for three layer feed-forward artificial neural network architecture. The load data used was obtained from Duhok ELC. Control Region (Iraq) for 2009 and 2010 years. The data set was divided into two parts of which a part was used to construct the forecasting model and the remaining part was used to evaluate the forecasting process. The mape calculated by them was about 1.9 %.
Accurate load forecasting is very important for electric utilities in a competitive environment created by the electric industry deregulation. Statistical and artificial intelligence techniques are the best tools which can be used for electric load forecasting. Different parameters and
14 variables that affect the forecast and its accuracy are weather data, time factors, customer classes, as well as economic. The basic learning which we get after reading all the published articles is that accurate forecasting can be done by basic research in statistics and artificial intelligence and better understanding of the load dynamics and its statistical properties to implement appropriate models.
15 3. NTPC VIDYUT VYAPAR NIGAM LTD. 3.1 COMPANY PROFILE NTPC Vidyut Vyapar Nigam Ltd. (NVVN) was formed by NTPC Ltd as its wholly owned subsidiary to tap the potential of power trading in the country, thereby promoting optimum capacity utilization of generation and transmission assets in the country and to act as a catalyst in development of a vibrant electricity market in India. NVVN was incorporated on 1st Nov 2002 and received the Certificate for Commencement of Business Activities from the Registrar of Companies on 26th Nov 2002. It has an authorized and fully paid up share capital of Rs. 200 million. NVVN commenced its first trading operations in March 2003 with the supply of surplus power from Eastern Region Stations of NTPC to Meghalaya State Electricity Board and Assam State Electricity Board. The Company was granted Category-E license by the Central Electricity Regulatory Commission (CERC) for inter-state trading of electricity on 23rdJul 2004. The trading license has been upgraded to Category F License on 22nd Mar 2005 by CERC. As per the latest CERC Regulations, 2009 the highest rated Trading Licensees are assigned I Category and NVVN has been given the same. NVVN is the only Government Company in the Power Sector engaged in the business of Power Trading. VISION To be a catalyst in the development of the wholesale power market in India by enabling the trading of surplus power MISSION To provide good value to potential sellers and develop commercial arrangements for their surplus power To enable NTPC to maintain optimal generation levels through mutually beneficial trading transactions To provide viable alternatives to buyers for meeting their demands To plan and establish a Power Exchange at national level using state-of-the-art technology 3.2 MAIN FUNCTIONS Purchase of all forms of power / electricity from Independent Power Producers (IPPs), Captive Power Plants, other Generating Companies, Transmission Companies, State Electricity Boards, State Governments statutory bodies, Licensees, Power utilities and to procure it from other sources (whether in Private, Public or Joint Sector Undertaking) including import from abroad Sale of all forms of electrical power to the State Electricity Boards, Vidyut Boards, Power Utilities, Generating Companies, Transmission Companies, Distribution Companies, State Governments, Licensees, statutory bodies, other organizations and
16 bulk consumers of power, whether in private or public sector or joint sector undertakings in India and abroad 3.3 ORGANISATIONAL STRUCTURE The Chief Executive Officer (CEO) heads the Company and is responsible to the Board of Directors. For the Power Trading Business operations of the Company, Business Development, Systems Operations, Finance and HR Groups support the CEO. The total executive strength of NVVN is 54 at various levels. 3.4 BUSINESS DEVELOPMENT INITIATIVES NVVN has been active in the power trading market since March, 2003. Besides trading of short- term surpluses from various State Power Utilities, it has taken several initiatives towards development of the power market and enhancing utilization of existing capacity. Some of the initiatives taken by it are: I) Utilization of Un-requisitioned Surplus Capacity from NTPC Stations: Transfer of un-requisitioned surplus Capacity on liquid fuel (Naphtha) from NTPC gas stations in NR: NVVN has entered into an arrangement with all the Beneficiaries of NTPC Gas Station in NR (including Railways), to enable utilization of day-to-day. Un-requisitioned surplus (URS) Liquid Fuel power from Anta, Auraiya & Dadri Stations. This URS power has helped the constituents of all the Regions including NR to meet their day to day shortages. The arrangement is in operation since 1st January 2004. Transfer of un-requisitioned surplus capacity from NTPC Coal based Stations: NVVN first attempted transfer of Un-requisitioned Surplus Capacity from NTPC Coal based Stations in September, 2005 from Southern Region as per the provisions of the Tariff Order based on the Generator consent. NVVN was successful in transferring the power to Availing Beneficiaries in Northern, Western & North Eastern Region. The arrangement has now been extended to ER & SR Regions. Transfer of un-requisitioned surplus Capacity on Regasified fuel from NTPC gas stations in NR: NVVN has entered into an arrangement with the Beneficiaries of NTPC Gas Station in NR (including Railways), to enable utilization of Day to Day Un- requisitioned surplus (URS) Regasified Fuel power from Anta, Auraiya & Dadri Stations. This URS power shall help the constituents of the Region to meet their day to day shortages. II) Utilization of Power through Power Swap Arrangements (PSA): NVVN worked out an innovative arrangement called Power Swap Arrangement in April, 2006 with WBSEB and arranged supply of 100 MW Peak Power to WBSEB from PSEB and in return WBSEB provided 100 MW off-peak power to PSEB. Both the transactions were at the same rate. A similar arrangement was also worked out with MeSEB wherein 30 MW Off-peak power was supplied to them from WBSEB and in return, MeSEB provided 110% Peak power which was transferred to PSEB.
17 3.5 NVVN - STRENGTHS / COMPETITIVE ADVANTAGES Strong Promoters: NVVN is promoted by NTPC Limited - the biggest generator having an excellent track record in creating energy infrastructure in India. The promoter provides a strong managerial, technical and financial background to NVVN. Excellent Asset Back-up: NTPC existing capacity is 37014 MW (inclusive of 4364 MW capacity through JVs), and has set a target to have an installed power generating capacity of 128000 MW by the year 2032. The National Electricity Policy (notified in January 2005) envisages part of new generating capacities, say 15% to be sold outside Long - term PPAs. This would provide a substantial boost to NVVN in its business in the future. Formidable Network: Strong networking is one of the most important drivers in the trading industry. Here again, the legacy in the form of strong network of NTPC with the existing utilities would help NVVN to easily establish rapport and credibility with the potential buyers and sellers. Government Support: NVVN being a 100% owned subsidiary of NTPC Ltd, is also a government company under the Companies Act 1956. By virtue of being a government company, NVVN would continue to enjoy strong credibility in the trading market. Professional Manpower: NVVN employees have been drawn from the professional manpower pool of NTPC. Most of the employees are engineers and chartered accountants having extensive experience in dealing various technical and commercial issues in NTPC. 3.6 MARKET DEVELOPMENT INITIATIVES: NVVN has been actively involved in facilitating the development of a wholesale electricity market in India and has developed significant domain knowledge for the development of a Power Exchange. NVVN has been sharing the learning with other stakeholders in the Indian power market through various workshops, etc., thus contributing to capacity building among stakeholders. NVVN has also assisted NTPC Ltd in setting up a national level Power Exchange - jointly promoted by NTPC Ltd, NHPC Ltd, PFC and Tata Consultancy Services Ltd (TCS). Joint Venture Agreement for the same between NTPC, PFC, NHPC and TCS was signed on 3rd September 2008 and the Company under the Companies Act, 1956 with the name National Power Exchange Limited (NPEX) incorporated on 11th December 2008. As per the Memorandum of Articles of Association, the authorized share capital of NPEX is Rs.50 Crore, of which, the initial paid up capital is Rs.5 Crore. NTPC Ltd, NHPC Ltd and PFC have contributed 16.67% each and rest 50% by TCS. Vide its Order dated 1st July 2009; CERC has accorded, in principle, approval to NPEX for start functioning as the third National Level Power Exchange. The NPEX is yet to start its operations.
18 4. Short Term Power Trading 4.1 Introduction The power sector has grown significantly since the enactment of the Electricity Act in 2003. However, it still faces the daunting challenge of providing adequate power to meet the growing needs of the economy. The Central Electricity Regulatory Commission (CERC) has mandated to promote competition, efficiency and economy in the power markets and improve the quality of power supply, which necessitates the development of a healthy short-term power market. A short- term power market can help electricity providers procure unplanned and fluctuating power requirements, and on the sellers' side, enable power producers as well as procurers to sell their surplus power. In India, the short-term power market, which covers contracts of less than a year through bilateral agreements and power exchanges is well developed, constituting approximately 11 percent (close to 95 billion units) of the total electricity market in 2011-12, though this includes power transactions through unscheduled interchange (UI) as well. Level of competition among the trading licensees increased steadily as the no of traders increased from 4 to 17 during a period of 2004-05 to 2011-12. During the year 2004-05 (when trading started) licensees charged 5 paise/kwh or less as the trading margin. However, the weighted average trading margin went up to 10 paise/kwh during April to September 2005. Considering this increase, CERC fixed the trading margin at 4 paise/kwh on 26.1.2006. As a result, the trading margin declined from 9 paise/kwh in 2005-06 to 4 paise/kwh in 2009-10. Again on 11 th January, 2010 CERC has issued new regulations fixing trading margins for inter-state trading in electricity. Trading margins would not exceed 4 paise a unit if the selling price of electricity is less than or equal to Rs3 a unit. The ceiling of the trading margin was capped at 7p a unit if the selling price of electricity exceeded Rs3 a unit. The 4p a unit cap regime was not adequate to cover the operational and market risks borne by trading companies due to strong competitive pressures, especially in the short-term buy and sell agreements. This amendment in the rules helped grow the power trading industry.
4.2 Short-Term Load Forecasting (STLF) Predicting the future load with lead times ranging from one hour to a week ahead is generally defined as short term load forecasting. It has a significant impact on the efficiency of operation of any electrical utility. Load forecasting helps an electric utility to make important decisions including decisions on purchasing and generating electric power, load switching, and infrastructure development. Load forecasts are extremely important for energy suppliers, ISOs, nancial institutions, and other participants in electric energy generation, transmission, distribution, and markets. Short-term forecasts have become increasingly important because of the rise of the competitive electricity markets. Short term forecasting of electrical load is important for optimum operation planning of power generation facilities as it affects both system reliability and fuel consumption. Accurate forecasting of electricity and power demand determines the utility to match its generation capabilities to the expected requirements. A short-term forecast
19 is important for unit commitment, economic dispatch, hydrothermal co-ordination, load management, etc. STLF plays an important role in the day-to-day operations of a utility and can be used for optimizing system operation and scheduling of hydro units and other peaking plants, such as gas turbines. The objective of the operators is to minimize variable costs without jeopardizing the electric system to power failures. STLF has a broader role in utility operations as it is also required for the co-ordination of Load Management programs with conventional system resources since the effectiveness of Load Management programs is sensitive to the system load. Accurate forecasting of power demand is to determine the assessment of the dynamic behavior of the system during disturbances so that the proper preventive action can be taken up. 4.3 STLF in Power Trading Electricity supply was considered a public service almost till the end of last century and any forecasting being done was for long term, concerning future fuel prices and technical improvements. Nowadays, short-term forecasting has become increasingly important since the rise of the competitive electricity markets. In this new competitive framework, short-term forecasting is required by traders to derive their bidding strategies in the electricity market and to list out their prospective buyers and sellers and the price to be offered to them. Accurate forecasting is essential for traders to maximize their profits and avowing profit losses over the misjudgment on the demand side. Traders forecast load to predict the energy requirement of a particular state or city so that they can find out the demand supply gap and accordingly devise their strategy. Forecasting is done for day-ahead market, hour, 2 hour, 4 hour and 6 hours ahead market as per their suitability. Hour ahead forecasting is not much useful because of the operational issues as it takes a times to coordinate between the SLDCs, RLDCs and the TSO for revision of the load.
20 5. Load Forecasting 5.1 Introduction Forecasting is the process of making statements about events whose actual outcomes have not yet been observed. We use the past historic data to determine the future trends. Forecasts systematically reduce uncertainty and thereby reduce the risks and costs associated with forecast errors. Every forecast depends on assumptions and making meaningful, logical, pertinent assumptions is the most important challenge for forecasters. Electric load forecasting is the process used to forecast future electric load, given historical load and weather information and current and forecasted weather information.. Load forecasting has always been an essential and important topic for power systems. However, load forecasting is a difficult task as the load at a given hour is dependent not only on the load at the previous hour but also on the load at the same hour on the previous day and on the load at the same hour on the day with the same denomination in the previous week. The STLF is also difficult to handle due to the nonlinear and random-like behaviors of system load, weather conditions, and variations of social and economic environments. So to improve the forecasting accuracy is still a difficult and critical problem. 5.2 Types of Loads Broadly classifies into five types: Domestic Industrial Commercial Agricultural Other Loads Street Lights, Bulk Supplies, Traction etc. 5.3 Why do Load Forecasting? The need for power varies from season to season, day to day, and even minute to minute. To ensure that an adequate supply of power is available to meet the demand, we must plan far into the future. Load forecasting has always been important for planning and operational decision conducted by utility companies. However, with the deregulation of the energy industries, load forecasting is even more important. With supply and demand uctuating and the changes of weather conditions and energy prices increasing by a factor of ten or more during peak situations, load forecasting is vitally important for utilities. Short-term load forecasting can help to estimate load flow and to make decisions that can prevent overloading. Timely implementations of such decisions lead to the improvement of network reliability and to the reduced occurrences of equipment failures and blackouts.
21 5.4 Types of Load Forecasting Load forecasts can be divided into three categories: short-term forecasts which are usually from one hour to one week, medium forecasts which are usually from a week to a year, and long-term forecasts which are longer than a year. The forecasts for dierent time horizons are important for dierent operations within a utility company. The natures of these forecasts are dierent as well. For example, for a particular region, it is possible to predict the next day load with an accuracy of approximately 1-3%. However, it is impossible to predict the next year peak load with the similar accuracy since accurate long-term weather forecasts are not available. 5.5 Factors Affecting Load Forecasting
Time Factors such as Hours of the Day(day/night) Day of the week(Weekday/weekend) Time of the year(season) Weather Conditions Temperature Humidity Rainfall Cloud Cover Visibility Solar Irradiation Class of customers such as industrial, agricultural, commercial, domestic, etc. Special Events such as TV programmes, public holidays, Bandhs etc. Population Electricity Price Economic Indicators such as per capita income, GDP, GNP, etc.
22 Figure 1: Input-Output Schematic for Load Forecasting
5.6 Methods or Tools Used for Load Forecasting Qualitative vs. quantitative methods Qualitative forecasting techniques are subjective, based on the opinion and judgment of consumers, experts; they are appropriate when past data are not available. They are usually applied to intermediate- or long-range decisions. Examples of qualitative forecasting methods are informed opinion and judgment, the Delphi method, market research, and historical life-cycle analogy. Quantitative forecasting models are used to forecast future data as a function of past data; they are appropriate when past data are available. These methods are usually applied to short- or intermediate-range decisions. Examples of quantitative forecasting methods are[citation needed] last period demand, simple and weighted N-Period moving averages, simple exponential smoothing, and multiplicative seasonal indexes. Time series methods Time series methods use historical data as the basis of estimating future outcomes.
23 Moving average Weighted moving average Exponential smoothing Double Exponential Smoothing Or Brown Method Holts Exponential smoothing Autoregressive moving average (ARMA) Autoregressive integrated moving average (ARIMA) Linear prediction Trend estimation
Causal / econometric forecasting methods These forecasting methods use the assumption that it is possible to identify the underlying factors that might influence the variable that is being forecast. For example, including information about climate patterns might improve the ability of a model to predict umbrella sales. This is a model of seasonality which shows a regular pattern of up and down fluctuations. In addition to climate, seasonality can also be due to holidays and customs; for example, one might predict that sales of college football apparel will be higher during the football season than during the off season. Causal methods include: Regression analysis includes a large group of methods that can be used to predict future values of variable using information about other variables. These methods include both parametric (linear or non-linear) and non-parametric techniques. Autoregressive moving average with exogenous inputs (ARMAX).
Artificial intelligence methods Artificial neural networks Group method of data handling Support vector machines Other methods Simulation Monte Carlo Simulation Prediction market Probabilistic forecasting and Ensemble forecasting 5.7 Advantage of Neural Networks over other techniques In machine learning and computational neuroscience, an artificial neural network (ANN), often called neural network, is a mathematical model inspired by biological neural networks. A neural network consists of an interconnected group of artificial neurons, and it processes information using a connectionist approach to computation. In most cases a neural network is
24 an adaptive system changing its structure during a learning phase. Neural networks are used for modeling complex relationships between inputs and outputs or to find patterns in data. Most of the researches show that ANN based approach is very successful to predict the required output with a fault tolerance in the range of 2-5%. The factors which contribute to the desired output have complex and non-linear relationships with the output. When the number and volume of these factors becomes too large, then the versatile ANN model can tackle the complexity and non-linearity with relative ease than the counterpart conventional mathematical model such as Regression, Smoothing or Moving Averages. Thus, application of ANN in power sector has been witnessing continuous growth in acceptance, making it a preferred choice for building a load forecasting model.
25 6. Electricity Demand Forecasting using Artificial Neural Networks (ANN) 6.1 Introduction Neural networks are named after the cells in the human brain that perform intelligent operations. The brain is made up of billions of neuron cells. Each of these cells is like a tiny computer with extremely limited capabilities; however, connected together, these cells form the most intelligent system known. Neural networks are formed from hundreds or thousands of simulated neurons connected together in much the same way as the brain's neurons. Just like people, neural networks learn from experience, not from programming. Neural networks are good at pattern recognition, generalization, and trend prediction. They are fast, tolerant of imperfect data, and do not need formulas or rules. Neural networks are trained by repeatedly presenting examples to the network. Each example includes both inputs (information used to make a decision) and outputs (the resulting decision, prediction, or response). The network tries to learn each example in turn, calculating its output based on the inputs you provided. If the network output doesn't match the target output, corrections are made in the network by changing its internal connections. This trial-and-error process continues until the network reaches your specified level of accuracy. Once the network is trained and tested, we can feed in the new input information, and it will produce a prediction. Designing a neural network is largely a matter of identifying which data is input, and what we want to predict, assess, classify, or recognize. 6.2 Applications of ANN Artificial Neural Network (ANN) is used in varied applications where statistical methods are traditionally employed. It is a powerful tool used for data analysis. ANN has the following applications Classification - Pattern recognition, feature extraction, image matching Noise Reduction- Recognize patterns in the inputs and produce noiseless outputs Prediction- Extrapolation based on historical data Time Series Applications Predicting stock markets performance 6.3 Construction, Training and Working of ANN The work flow for the neural network design process has seven primary steps: Collect data Create the network Configure the network Initialize the weights and biases Train the network Validate the network
26 Use the network Out of the above 7 steps the first step is carried out separately, rest all is associated with the neural network framework. Neuron Model The fundamental building block for neural networks is the single-input neuron. Figure 2: Neuron Model
There are three distinct functional operations that take place in this example neuron. First, the scalar input p is multiplied by the scalar weight w to form the product wp, again a scalar. Second, the weighted input wp is added to the scalar bias b to form the net input n. Finally, the net input is passed through the transfer function f, which produces the scalar output a. The names given to these three processes are: the weight function, the net input function and the transfer function. Transfer Functions Two of the most commonly used functions are shown below. Figure 3: Linear Transfer Function
27 Figure 4: Log-Sigmoid Transfer Function
Neuron with Vector Input The simple neuron can be extended to handle inputs that are vectors. A neuron with a single R-element input vector is shown below. Here the individual input elements
p 1, p 2, p 3.. p R
are multiplied by weights
w 1,1 ,w 1,2 ,w 1,3 w 1,R
And the weighted values are fed to the summing junction. Their sum is Simply Wp, the dot product of the (single row) matrix W and the vector p.
Figure 5: Neuron With Vector Input
The neuron has a bias b, which is summed with the weighted inputs to form the net input n. The net input n is the argument of the transfer function f. Weights/Synapses are very important for determining the function in the operation of ANN n = w 1 , 1 p1 + w 1,2 p2 ++ w 1 , R p R + b
28 This expression can also be written as n = W*p + b
6.4 Network Architectures Two or more of the neurons can be combined in a layer, and a particular network could contain one or more such layers. First we consider a single layer of neurons.
Figure 6: One Layer of Neurons
In this network, each element of the input vector p is connected to each neuron input through the weight matrix W. The ith neuron has a summer that gathers its weighted inputs and bias to form its own scalar output n(i). The various n(i) taken together form an S-element net input vector n. Finally, the neuron layer outputs form a column vector a. We can create a single (composite) layer of neurons having different transfer functions simply by putting two of the networks shown earlier in parallel. Both networks would have the same inputs, and each network would create some of the outputs. Multiple Layers of Neurons A network can have several layers. Each layer has a weight matrix W, a bias vector b, and an output vector a. To distinguish between the weight matrices, output vectors, etc., for each of these layers in the figures, the number of the layer is appended as a superscript to the variable of interest.
29 Figure 7: Multiple Layer of Neurons
The network shown above has R 1 inputs, S 1 neurons in the first layer, S 2 neurons in the second layer, etc. It is common for different layers to have different numbers of neurons. A constant input 1 is fed to the bias for each neuron.
The layers of a multilayer network play different roles. A layer that produces the network output is called an output layer. All other layers are called hidden layers. The three-layer network shown above has one output layer (layer 3) and two hidden layers (layer 1 and layer 2). After creating the network we have to configure it to give the best result.
6.5 Network Configuration After a neural network has been created, it must be configured. The configuration step consists of examining input and target data, setting the networks input and output sizes to match the data, and choosing settings for processing inputs and outputs that will enable best network performance. The configuration step is normally done automatically, when the training function is called or it can also be done manually using various functions.
Training or Process
Learning is a procedure for modifying the weights and biases of a network. The purpose of learning rule is to train the network to perform some task. They fall into three broad categories:
Supervised learning- The learning rule is provided with a set of training data of proper network behaviour. As the inputs are applied to the network, the network outputs are
30 compared to the targets. The learning rule is then used to adjust the weights and biases of the network in order to move the network outputs closer to the targets. Reinforcement learning- It is similar to supervised learning, except that, instead of being provided with the correct output for each network input, the algorithm is only given a grade. The grade is a measure of the network performance over some sequence of inputs. Unsupervised learning- The weights and biases are modified in response to network inputs only. There are no target outputs available. Most of these algorithms perform some kind of clustering operation. They learn to categorize the input patterns into a finite number of classes.
There are many different kinds of learning rules used by neural networks. They are as follows:
Resilient Back propagation Levenberg-Marquardt Scaled Conjugate Gradient Bayesian Regularization Gradient descent with momentum Back Propagation Batch training BFGS quasi-Newton Back Propagation Conjugate gradient Back Propagation with Powell-Beale restarts One-step secant back propagation Random order incremental training with learning functions Sequential order incremental training with learning
After training the network, we can simulate the model by feeding in the input variables and it will give the predicted result. We can then calculate the error by comparing the predicted value and the actual value.
31 7. Case Study Demand Forecasting of UP region
7.1 Case Study Uttar Pradesh created on 1 April 1937 as the United Provinces with the passing of the States Reorganization Act was earlier known as United Provinces till 1950. On 9 November 2000, a new state was formed out of Uttar Pradesh and was named as Uttarakhand. It is the fifth largest state in India in terms of area. With over 200 million inhabitants as of 2011, it is the most populous state in the country as well as the most populous country subdivision in the whole world. Uttar Pradesh is the second largest Indian state by economy, with a GDP of 708,000 crore (US$120 billion). Agriculture and service industries are the largest parts of the state's economy.
Uttar Pradesh has 5 discoms which have the responsibility of distributing electricity all over the state. They are Poorvanchal, Madhyanchal, Paschimanchal, Dakshinanchal, NPCL, NVVNL and KESCo respectively.
Table 1 Anticipated and actual demand data for Uttar Pradesh LGBR 2013-14
We can see in the above table that UP suffers both from normal and peak load deficit. It is in the best interest of Uttar Pradesh Government to forecast load and to try and manage its power requirement in a more systematic and certain manner so as to provide a better quality of life to the 200 million inhabitants living in the state. To have a load management system we need to prepare for the future load demand. We have created a model which will help UP predict the load and thus can help in managing the states reserves useful and tactfully.
UP
Energy
Peak
Requirement
Availability
Surplus(+)/Deficit (-)
Requirement
Availability
Surplus(+)/Deficit (-) MU MU MU % MU MU MU % Anticipated (13-14) 97785 80203 -17582 -18 14400 11606 -2794 -19.4 Actual (12-13) 91647 76446 -15201 -16.6 13940 12048 -1892 -13.6
32 National Capital Power Station (NCPS) or NTPC Dadri is located at Vidyut Nagar, Uttar Pradesh is the power project to meet the power demand of National capital region. It has a huge coal-fired thermal power plant and a gas-fired plant. BTPS has an installed capacity of 2637 MW altogether built at around 1988-90. It is one of the older plants of NTPC and has a little automation. The performance of the plant is deteriorating due to various reasons including aging, old equipments poor quantity and quality of cooling water etc. It can recover its fixed cost by running the load at full Load and sell the power generated from the underutilized capacity in the short term power market. Table 2 Coal Based Power Station Stage Unit Number Installed Capacity (MW) 1 st 1 210 1 st 2 210 1 st 3 210 1 st 4 210 2nd 5 490 2 nd 6 490 Total Six 1820
Table 3 Gas Based Power Station Stage Unit Number Installed Capacity (MW) 1 st 1 130 1 st 2 130 1 st 3 130 1 st 4 130 2 nd 5 154 2 nd 6 154 Total Six 817
7.2 Simulation Design The simulation design includes 1. Research Data 2. Construction of Network Architecture 3. Selection of input variables 7.3 Research Data The data used in this project is the daily load (unrestricted demand) for the State of Delhi. Raw data have been collected from Environment Monitoring Group (EMG), NTPC Dadri Plant For the period of 18 th November to 1 December 2012. Data for first 25 days have been used for training the model. The data for next 15 days have been used for validation and testing the model.
33 7.4 Construction of Network Architecture The structure of the network depends on the number of neurons, number of hidden layers, and selection of activation function which is sigmoid function in our case. The structure of the architecture affects the precision of the model. The following elements have been used to build the network. 1. Number of Hidden Layers 2 nos. 2. Number of neutron in layer 1 6 nos. 3. Number of neutrons in layer 2 10 nos. 4. Training Function - Levenberg-Marquardt 5. Error Indicator MAPE (Mean Absolute Percentage Error)
Selection of Input Variables There is no such thumb rule for selecting inputs for forecasting model. It depends on judgment, experience and is mostly carried out on trial and error basis. Importance of factors may vary for different models of load forecasting. For a long-term load forecasting many macro-economic factors such as GDP, Inflation, Customer Classes, etc play a pivot role. However, for short-term forecasting model such macro-economic factors are not vital as their impact is not much on the load demand for next 4 hours. Following inputs have been considered to implement this model. They are as below. 1. Wind Speed of the hour (m/s) 2. Temperature of the hour (C) 3. Relative Humidity (%) 4. Demand for Previous Week (W-1) 5. Ramp Demand (last day) 6. Demand (T-1, T-2, T-3, T-4, T-19, T-20, T-21, T-22, T-23, T-24, T-25) 7. Average Temperature
34 8. Results & Observations
The results obtained from testing the trained neural network based on the data collected from NTPC Dadri are shown below.
Levenberg-Marquardt algorithm which is used to rum the model was found to be best among all other algorithms as suggested by many researchers and scholars who have tried their hands in short term load forecasting. The MAPE (Mean Absolute Percentage Error) is found to be 2.7%.
8.1 Results
Following are the screenshots of the load curves and various other results implemented on the Graphic User Interfaces of MATLAB 2011a.
36 Figure 12: Regression plot of Training, Validation and Testing datasets The regression of Training dataset is 98.916%. The regression for validation and testing datasets are 94.02% and 95.012%.
37 Figure 13: Error Histogram
8.2 Observations
Generators Perspective:
In this age of technology it has become very important for generating stations to incorporate new ways to recover the cost invested in the project. A generating station has a useful life of 25 years and the investor of the project tries to recover the money invested in the project within this period of time. As new technologies develop over a period of time, new plants are setup which give higher output (power generated) with lower inputs (fuel consumption), thus selling electricity at a lower price. The old generation companies have a risk of running out of business as the buyers may intend to purchase low cost power from other generation companies and end their Power Purchase Agreement (PPA) or they may also buy less power from these stations. Also running the plant on low PLF reduces the plants efficiency thereby increasing the cost of power.
This problem can be solved tactfully in different ways by using the short term load forecast model and forecasting the future load demand and thereby knowing the capacity which will
38 go underutilized. Thereby they can plan and sell this power on the power market. They can also go for other ways sighted below to recover their fixed cost. Having insights about the future load demand can help generating stations plan the shutdown of the units on merit order thereby providing them a cost benefit. Shutdown and maintenance scheduling can also be planned according by studying the load demand pattern. Short term load forecasting can be used for fuel purchase scheduling beforehand thereby helping them save money. An old generating station can come out of the PPA and try acting as a merchant power plant to sell their power on short term markets thus realizing better rates as it is believed that better rates can be realized on trading platforms. They can also sign PPAs for 50% or 60% of their generating capacity and they can use the rest to sell it on power exchanges. Running the plant at 100% PLF increases their efficiency and reduces the operational cost. Discoms Perspective: In the past few years we have seen private players taking up the role of Discoms and this sector which was once monopolized has now opened up. We can see in states like Mumbai where there are two discoms operating and each of them try and provide better service so as to cannibalize the other operators customers and retain the old ones. The increased competition has forced Discoms to implement new ways to serve the customer better and keep them satisfied so as to gain repeated revenues. This model of Short Term Load forecasting would provide the Discoms with future load demand which can help plan the purchase and scheduling of power from different power stations and also help them estimate the amount of power they would require to meet the power deficit. They can intimate the power traders hours before about the power they would need in the future thus providing them enough time to make some arrangement, this would also result in lesser price as buying power in crunch situations is more costly. Traders Perspective: Traders can earn only a margin of 7 paise/kwh of electricity, thereby forcing the traders to search for new avenues to earn more profits. Traders can use this tool of short term load forecasting to their best interest and improve their earnings drastically. Identification of buyers and sellers of power and devising bidding strategies. Insights about Energy Generation And Purchasing Estimate load flows and to make decisions that can prevent overloading. Timely implementations of such decisions lead to the improvement of network reliability and to the reduced occurrences of equipment failures and blackouts. Improving Operational Efficiency Contract Evaluation
39 9. Conclusion
The result of Artificial Neural network model used for four hour ahead short term load forecast for the NTPC Dadri, has a good performance and reasonable prediction accuracy. Its forecasting reliabilities were evaluated by computing the mean absolute percentage error between the exact and predicted values. We were able to obtain a Mean Absolute Percentage Error (MAPE) of 2.7% which represents a high degree of accuracy. This model can be used by old generation companies, power traders as well as discoms to devise business strategies and can help improve profitability drastically. It can also be used for the scheduling of fuel purchase agreements and also for merit order dispatch of generating units.
The results suggest that ANN model with the developed structure can perform good prediction with least error and finally this neural network could be an important tool for short term load forecasting. This model can also be used for long term and medium term forecast which can be used accordingly to plan addition of generation capacities.
Future studies on this work can incorporate additional information (such as customer class and season of the year) into the network so as to obtain a more representative forecast of future load. Network specialization (i.e. the use of one neural network for the peak periods of the day and another network for the hours of the day) can also be experimented upon.
40 10. Bibliography
[1] Feng PAN, Ming ZONG, A COMPREHENSIVE SYSTEM FOR SHORT-TERM LOAD FORECASTING APPLIED IN SHANGHAI, 21st International Conference on Electricity Distribution, Frankfurt, 6-9 June 2011 [2] V.H. Kher and S.K. Joshi, The Maharaja Sayajirao University of Baroda,Vadodara, INDIA, Short Term Load Forecast Using Artificial Neural Network [3]Pradeepta Kumar Sarangi, Nanhay Singh, R. K. Chauhan and Raghuraj Singh, Short Term Load Forecasting using Artificial Neural Network: A comparison with Genetic Algorithm implementation, ARPN Journal of Engineering and Applied Sciences, VOL. 4, NO. 9, NOVEMBER 2009. [4] Sexton R.S., Dorsey R.E. and Sikander N.A. 2002, Simultaneous Optimization of Neural Network Function and Architecture Algorithm. Decision Support Systems. 1034: 1-13. Hong Chen, Claudio A. Canizares and Ajit Singh, ANN-based Short-Term Load Forecasting in Electricity Markets, Department of Electrical & Computer Engineering Waterloo, ON, Canada N2L 3G1. [5]A. D. Papalexopoulos, S. Hao, and T. M. Peng, An Implementation of a Neural Network Based Load Forecasting Model for the EMS, IEEE Trans. Power Systems, Vol. 9, No. 4, Nov. 1994, pp. 19561962.
Websites:
NTPC Vidyut Vypar Nigam, Ltd. http://www.nvvn.co.in/ Central Electricity Authority, Report on Load Generation Balance Report, 2013-14 http://www.cea.nic.in/reports/yearly/lgbr_report.pdf Neural Networks, Wikipedia, http://en.wikipedia.org/wiki/Neural_network Real Time Data of Demand, Delhi SLDC, http://www.delhisldc.org/Redirect.aspx?Loc=0708
41 11. ANNEXURE
The error calculation sheet is attached with the document.