CN109165770A - Novel photovoltaic power forecasting method based on AFSA-Elman - Google Patents
Novel photovoltaic power forecasting method based on AFSA-Elman Download PDFInfo
- Publication number
- CN109165770A CN109165770A CN201810745300.0A CN201810745300A CN109165770A CN 109165770 A CN109165770 A CN 109165770A CN 201810745300 A CN201810745300 A CN 201810745300A CN 109165770 A CN109165770 A CN 109165770A
- Authority
- CN
- China
- Prior art keywords
- elman
- power
- value
- afsa
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000013277 forecasting method Methods 0.000 title claims abstract description 14
- 238000013528 artificial neural network Methods 0.000 claims abstract description 39
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 18
- 230000006399 behavior Effects 0.000 claims description 26
- 241000251468 Actinopterygii Species 0.000 claims description 14
- 238000000034 method Methods 0.000 claims description 11
- 238000005286 illumination Methods 0.000 claims description 10
- 238000012549 training Methods 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 8
- 230000019637 foraging behavior Effects 0.000 claims description 7
- 210000005036 nerve Anatomy 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 235000013305 food Nutrition 0.000 claims description 5
- 239000013598 vector Substances 0.000 claims description 4
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 abstract 2
- 235000019688 fish Nutrition 0.000 description 35
- 230000000007 visual effect Effects 0.000 description 6
- 241000196324 Embryophyta Species 0.000 description 3
- 230000005856 abnormality Effects 0.000 description 3
- 238000010248 power generation Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000013178 mathematical model Methods 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 206010033799 Paralysis Diseases 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000000428 dust Substances 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- BULVZWIRKLYCBC-UHFFFAOYSA-N phorate Chemical compound CCOP(=S)(OCC)SCSCC BULVZWIRKLYCBC-UHFFFAOYSA-N 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000005295 random walk Methods 0.000 description 1
- 238000013517 stratification Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Economics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Operations Research (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Quality & Reliability (AREA)
- Evolutionary Computation (AREA)
- Entrepreneurship & Innovation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses the novel photovoltaic power forecasting methods based on AFSA-Elman, the original power of acquisition is subjected to wavelet decomposition first, obtain low frequency trend component and high frequency detail component, input in conjunction with collected weather data as model is predicted respectively, finally corresponding predicted value is reconstructed, obtains final power prediction value；Using artificial fish-swarm algorithm to the weight and threshold optimization of Elman neural network, then by the Elman after optimization applied in the short-term forecast of photovoltaic output power.The present invention can be according to the weather history and output power at photovoltaic station, and the weather condition of combination in a short time predicts the output power of subsequent time, prediction model can be using autocorrelation present in historical data, without being modeled with the cumbersome physical equation of numerous complicated.
Description
Technical field
The invention belongs to technical field of photovoltaic power generation, more particularly to the novel photovoltaic power prediction side based on AFSA-Elman
Method.
Background technique
Solar energy belongs to intermittent energy source, so that the generated output of photovoltaic plant shows fluctuation and Intermittent Features.
When grid-connected photovoltaic power station installed capacity is larger, safety and the stability of electric system will be affected.Studies have shown that working as light
When proportion is more than 15% in the power system, fluctuation may lead electric system and paralyse, band for overhead utility installed capacity
Carry out immeasurable loss.It is reasonable to cooperate power department to carry out that therefore, it is necessary to the generated outputs of Accurate Prediction photovoltaic plant
Plan and scheduling.
Accurate Prediction generated output, cooperation electric dispatching department rational management electricity consumption arrange spare capacity, can not only mention
Power quality, enhancing grid stability are risen, and the operation expense of electric system can be greatly reduced.In this regard, national
The grid-connected photovoltaic power station that power grid is more than 10MW to installed capacity in photovoltaic electric station grid connection acceptance specification proposes specific power
Prediction requires.Grid-connected power station is needed using 15min as prediction step, provides the generated power forecasting of future 4h to electric dispatching department
Value, it is also required to provide the power prediction of next day for 24 hours daily.
Artificial fish-swarm algorithm (Artificial Fish Swarm Algorithm, AFSA) is as a kind of novel intelligence
There is algorithm the of less demanding, global convergence to initial value the advantages such as to require, is of less demanding to parameter setting be used in industry
Control field.
Summary of the invention
The present invention makes full use of wavelet decomposition to excavate inner link existing for photovoltaic output power and ambient weather factor,
The Elman neural network by the accurate optimizing of AFSA and part with global optimizing combines simultaneously, overcomes Elman nerve
The randomness of network initial weight and threshold value, easy the shortcomings that falling into local optimum, under the premise of prediction model accelerates convergent, into
One step improves precision of prediction.
The technical solution of the method for the present invention is as follows:
Novel photovoltaic power forecasting method based on AFSA-Elman, comprising the following steps:
Step 1: data collect photovoltaic station history output power and corresponding weather condition, to collected number
According to pre-processing, the high frequency detail component and low frequency trend component of power are obtained to Power Decomposition, establish sample set；It is collected
Data include power P, intensity of illumination Q, temperature T, humidity H, intensity of illumination G, wind speed S, pretreated detailed process are as follows: reject function
Rate is less than or equal to zero power points, reject with the very big abnormal point of proximity data deviation, will treated that data normalization is handled,
History generated energy sequence after normalization is subjected to wavelet decomposition, is high frequency detail signal d by signal decompositioniWith low frequency trend point
Measure an, i=1,2,3 ... n, n are maximum decomposition level number.
Step 2: determining nerve network input parameter, Elman is established to high frequency detail component and low frequency trend component respectively
Neural network model predicts the high and low frequency component of subsequent time；
It determines nerve network input parameter: is weighed between different variables using Pearson correlation coefficient and linearly depend on journey
Degree, calculates the pertinency factor ρ between each component and weather condition:
N is the number of samples of training set, x in formulaiFor the input parameter of certain prediction model to be selected,For sample parameter
Average value, yiIt is model output,For the average value for exporting sample.
Step 3: Elman neural network parameter being optimized by artificial fish-swarm algorithm AFSA
Step 3.1: with the connection weight ω 1, ω 2, ω 3 of Elman neural network, threshold value b1, b2 as parameter to be optimized,
Five dimensional vectors are constructed as an Artificial Fish individual, initialize the shoal of fish；
Step 3.2: using the difference e between the predicted value of Elman neural network and desired output as the suitable of Artificial Fish
Response objective function, and be recorded as target function value is the smallest optimum individual and be recorded in bulletin board；
Step 3.3: the current shoal of fish executes foraging behavior, clustering behavior, behavior of knocking into the back respectively；Execute foraging behavior, clustering
Before behavior, behavior of knocking into the back, Artificial Fish explores the environment being presently in, and reattempts two kinds of behaviors of bunching and knock into the back, and calculates target letter
Whether number fitness value improves, and actual selection executes the lesser behavior of fitness value, and default behavior is to look for food；
Step 3.4: each round finishes, and calculates the fitness target function value i.e. e of each Artificial Fish, and in contrast to bulletin board
Optimal value is stored in bulletin board；
Step 3.5: judging whether to meet maximum number of iterations termination condition, export optimized parameter if meeting, otherwise, repeatedly
Generation number k=k+1, return step 3.3；After each round iteration optimizing all can calculating target function value e, it is minimum to find e value
Individual, compared with advancing with the optimum individual that is saved in last round of bulletin board, by target function value e it is small be denoted as optimum individual
It is stored in bulletin board, and continues next round iteration, avoids falling into local optimum in searching process, leans on Artificial Fish to global optimum
Closely；
Step 3.6: after optimizing, using the best initial weights of output and threshold value as the optimized parameter of Elman neural network,
And establish the corresponding Elman neural network prediction model of high and low frequency component；
Step 3.7: test analysis is carried out to the model optimized, reconstruct obtains the predicted value of subsequent time power, if
Reach error requirements, then saving this group of weight and threshold value, while exporting the prediction result of subsequent time power, otherwise go to step
3.1, restart to optimize；
The predicted value of subsequent time power predicts the high frequency division of subsequent time by the Elman neural network optimized respectively
Measure Di, i=1,2,3 ... n, n are maximum decomposition level number, low frequency component An, the predicted value of high and low frequency component is reconstructed, is obtained
To the predicted value P of subsequent time powert+1:
Compared with prior art, the beneficial effects of the present invention are:
It is easily realized 1. algorithm is concise.It does not need to establish accurate photovoltaic power generation number with the cumbersome physical equation of numerous complicated
Model is learned, the relevant parameter for determining mathematical model is less required to find in historical data under the conditions of only given original state
Existing autocorrelation (factors such as battery plate suqare, service wear, dust stratification can embody in history generated output).
2. use wavelet decomposition method, by historical data details coefficients and trend component analyzed respectively, further
Characteristic information present in data is excavated, precision of prediction is improved.
3. traditional Elman neural network gradient descent method solves, it is slow etc. easily to fall into local optimum, late convergence
Defect.Comprising the operator that knocks into the back in artificial fish-swarm algorithm, accelerates Artificial Fish and move about to more preferably position, make the people for falling into local optimum
Work fish is mobile to the Artificial Fish direction of global optimum and flees from local optimum, improves the convergence rate of network, reduces Elman and fall into
The risk of local optimum.
Detailed description of the invention
Fig. 1 is the flow chart of the novel photovoltaic power forecasting method based on AFSA-Elman；
Fig. 2 is wavelet decomposition schematic diagram；
Fig. 3 is the mapping graph of Elman neural network；
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that described herein, the specific embodiments are only for explaining the present invention, not
For limiting the present invention.
Invention is further described in detail for 1 flow chart and specific embodiment with reference to the accompanying drawing, but of the invention
Protection scope is not limited to this.
Below in conjunction with shown in attached drawing, the embodiment of the present invention is described in detail:
As shown in Figure 1, the novel photovoltaic power forecasting method based on AFSA-Elman, includes the following steps:
Step 1: the SCADA (Supervisory Control And Data Acquisition) installed by photovoltaic station
System, collect photovoltaic station history output power and corresponding weather condition, comprising: intensity of illumination Q, power P, temperature T,
Humidity H, wind speed S；SCADA system is to acquire initial data by various kinds of sensors and be transferred to monitoring client, due to real data
In collection process, sensor abnormality, maloperation, transmission error etc. will lead to the appearance of Outliers, and therefore, it is necessary to make to data
Further pretreatment, rejecting abnormalities data；Then 3 grades of wavelet decompositions are made to power, obtains 3 high frequency detail components of power
With 1 low frequency trend component, sample set is established.
The initial data of acquisition is pre-processed, specifically includes the following steps:
Step 1.1, the power points that power is less than or equal to zero is rejected, is rejected and the very big abnormal point of proximity data deviation, general
Data normalization processing that treated；
Each attribute input parameter there are specific physical significance, there are different unit dimensions, between different attribute
There may be the gaps of the order of magnitude.The input parameter of each attribute is limited in a certain range by normalized, is promoted
While training speed, and reduce the probability that neural network falls into local optimum.The selection for limiting section also influences whether mind
Service performance through network is obtained reaching optimum prediction effect when mode input is limited between [0,1], be limited by test
Determine formula to be given by:
In formula, ziFor the value before normalization；zminFor the minimum value in this attribute data；zmaxFor in this attribute data
Maximum value；
Step 1.2, the historical power of treated photovoltaic power generation is subjected to 3 grades of wavelet decompositions, obtains the son of power sequence
The trend component a of frequency sequence low-frequency range3(t) and the random component d of high bandi(t) (i=1,2,3), the power of t moment
It may be expressed as:
By treated, sample data is randomly divided into training set and test set.Training dataset is used to train Elman neural
Network, test data set are not involved in trained Elman neural network, but pre- by the Elman neural network optimized for testing
Survey the accuracy of model.
Step 2: calculating separately the pertinency factor ρ between different weather situation and high and low frequency component, choose pertinency factor value
The biggish input parameter as Elman neural network establishes Elman mind to high frequency detail component and low frequency trend component respectively
Through network model, high frequency, the low frequency component of subsequent time are predicted；
Step 2.1, weighed using Pearson correlation coefficient between different variables and linearly depend on degree, calculate each component with
Pertinency factor ρ between weather condition, so that it is determined that the input parameter of neural network:
N is the number of samples of training set in formula；xiFor the input parameter of certain prediction model to be selected；For sample parameter
Average value；yiIt is model output (low frequency component a3With high fdrequency component di)；For the average value for exporting sample.
Using the weather variable high with power component pertinency factor as the input of model, it is not necessary that whole influence factors is defeated
Enter model, greatly reduce the dimension of mode input, reduce the complexity of algorithm, improves the speed of service.
Step 2.2, build the method for Elman neural network: low frequency component approaches original power, reflects normal illumination feelings
Related coefficient under condition between power Main change trend, with intensity of illumination Q, temperature T is larger, therefore only by t when establishing model
Q, T, a at moment3(t) and the Q ' at t+1 moment, T ' are used as mode input, establish the Elman neural network prediction of low frequency component
Model；High fdrequency component is details coefficients, reflects the randomness of changed power, and influence the part of power prediction precision, but
The part is not interference parameter, has reflected the influence of the factors such as Changes in weather and hardware device.Therefore in prediction high frequency
When component, by the power high frequency component d of t momenti(t) (i=1,2,3), a3(t), intensity of illumination Q, temperature T, humidity H, wind
Input of the fast S as network.Similarly, the Q ' at t+1 moment, temperature T ', humidity H ', wind speed S ' are also as the input of model, training
The Elman neural network prediction model of 3 high fdrequency components out.
The mathematical model of Elman neural network such as following formula:
T is current time in formula, and X (t) is the output valve of hidden layer, and U (t-1) is the output valve of eve network, Xc(t)
For the output for accepting layer, Y (t) is the output valve for predicting network, and ω 1 is the weight between connection undertaking layer and hidden layer, and ω 2 is
Connect the weight of hidden layer and output layer, ω 3 is the connection weight between input layer and hidden layer, b1, b2 be imply, output layer
In threshold value.F (x) is middle layer neuron function, takes Sigmoid function, it may be assumed that
G (x) takes linear function, and structure such as attached drawing 3 is mended in opening up for Elman neural network model.
Step 3: it is excellent that parameter being carried out to the Elman neural network that high and low frequency component is established respectively using artificial fish-swarm algorithm
Change, with the connection weight ω 1, ω 2, ω 3 of Elman neural network, threshold value b1, b2 constructs five dimensional vectors as parameter to be optimized
[ω 1, ω 2, ω 3, b1, b2] is used as an Artificial Fish individual, and ω 1, ω 2, ω 3, b1, b2 ∈ (0,1) initialize the shoal of fish；
Artificial fish-swarm is initialized, realization comprises determining that Artificial Fish number N, step-length Step, visual field Visual, gathers around
It squeezes degree factor delta, sound out number Try-number, maximum number of iterations K, note k is current iteration number (k=0,1,2 ...).
Step 4: by the difference e between the predicted value and desired output of Elman neural network, as the suitable of fish-swarm algorithm
Response objective function, and be recorded as target function value is the smallest optimum individual and be recorded in bulletin board；
Step 5: the current shoal of fish executes foraging behavior, clustering behavior, behavior of knocking into the back, as one wheel respectively；
It executes foraging behavior and specifically includes step:
A state X is randomly choosed in the sensing range of current manual fishj[ω1,ω2,ω3,b1,b2]
Xj=Xi+Visual·Rand()
In formula, Xi[ω 1, ω 2, ω 3, b1, b2] is the current state of Artificial Fish, Rand () be between 0 and 1 with
Machine number, Visual are the field range of every Artificial Fish, judge current tracking error eiWith target following error amount ejSize,
If meeting ej<ei, then it takes a step forward to the direction, even if also:
Step is the every primary step-length of advancing of Artificial Fish.
It executes clustering behavior and specifically includes step:
If the center of artificial fish-swarm: Xcenter[ω1c,ω2c,ω3c,b1c,b2c], wherein
Wherein m is the number of the Artificial Fish in sensing range centered on self-position.
Calculate center XcenterCorresponding error amount ecenter, judge whether to meet condition ecenter<ei, and
ecenterM < δ × ei, wherein δ is crowding factor (0 < δ < 1), eiFor the fitness value of the Artificial Fish, if so, towards artificial
Shoal of fish center takes a step forward, namely:
The execution behavior of knocking into the back specifically includes step:
If Artificial Fish current state Xi[ω 1, ω 2, ω 3, b1, b2] is explored in the partner in current neighborhood and is corresponded to error
The partner X of value e minimum valuej.If ej/ n < δ × ei, then to XjIt takes a step forward, n is the artificial fish-swarm number in sensing range, and δ is to gather around
It squeezes the factor (0 < δ < 1)；
Step 6: each round finishes, and calculates the fitness target function value i.e. e of each Artificial Fish, and compares in bulletin board and protect
Smaller value and its corresponding individual are stored in bulletin board by the optimum individual deposited；
Step 7: judging whether to meet maximum number of iterations termination condition, export optimized parameter, otherwise, iteration if meeting
Number k=k+1, return step 5；
Step 8: after optimizing, using the best initial weights of output and threshold value as the optimized parameter of Elman neural network, and
Establish the corresponding Elman neural network prediction model of high and low frequency component；
Step 9: test analysis being carried out to the 4 groups of Elman neural networks optimized, high and low frequency component predicted value is carried out
Wavelet reconstruction obtains final power prediction value, while compared with the moment practical photovoltaic power output valve, if reaching error
It is required that otherwise going to step 3, restarting to optimize then save this group of weight and threshold value；
Step 9.1: 4 prediction models for being adjusted to optimized parameter being tested, predict the high and low frequency division of subsequent time respectively
Amount, then high frequency detail component Di (i=1,2,3) and low frequency component A that 4 neural network forecasts are gone out3By wavelet reconstruction, acquisition
The output power predicted value P at subsequent time photovoltaic stationt+1:
Step 9.2: prediction result is analyzed using root-mean-square error index, if error is larger, re-starts training,
If error is in allowed limits, Elman neural metwork training is qualified；
σMSEFor root-mean-square error, N is forecast sample number, and y (i) is the true value of sequence,For prediction result.
Embodiment:
Initial data is acquired first, and establish set of data samples: the raw data set that this experiment uses comes from Zhejiang photovoltaic
Power station on March 1st, 2015, the SCADA data that acquisition resolution is 5 seconds was divided into electric data and meteorological data two to August 31st
Class.Electric data is the electric current and voltage value of inverter output, the weather station that meteorological data is installed in photovoltaic plant, packet
Intensity of illumination Q, power P, temperature T, humidity H, wind speed S etc. are contained.Further pretreatment, rejecting abnormalities number are made to data again
According to, then the data handled well are done into normalized by formula (1), mode input value is limited between [0,1]:
The Mallet fast algorithm that orthogonal transformation is used to treated power sequence, it is thin to be decomposed into high frequency for power signal
Save signal di(i=1,2,3 ... n) and low frequency trend component an, n is maximum decomposition level number, uses db3 wavelet basis as shown in Figure 2
Carry out 3 grades of decomposition.The power of t moment in initial data can be expressed as formula (2).
Then the input parameter for determining Elman neural network, establishes prediction model: first calculate each component and weather condition it
Between pertinency factor, using the high variable of pertinency factor as the input of model (formula (3)).
It regard Q, T, a3 (t) of t moment and the Q ' at t+1 moment, T ' as mode input, establishes the Elman of low frequency component
Neural network prediction model；By the power high frequency component d of t momenti(t) (i=1,2,3), a3(t), intensity of illumination Q, temperature T, wet
Input as network of H, wind speed S is spent, similarly, the Q ' at t+1 moment, temperature T ', humidity H ', wind speed S ' are also as the defeated of model
Enter, trains the Elman neural network prediction model of 3 high fdrequency components.
After establishing Elman neural network prediction model, by the connection weight ω 1, ω 2, ω 3 of network, threshold value b1, b2
Five dimensional vectors are constructed as an Artificial Fish individual as parameter to be optimized, initialize the shoal of fish: setting fish Artificial Fish number N=
100, step-length step-length Step1=0.45, Step2=0.4, Step3=0.68, Step4=0.5, Step5=0.56, the visual field
Visual=0.62, crowding factor delta=0.618, number Try-number=40, maximum number of iterations K=100 are soundd out；
Using the difference e between predicted value and desired output as the fitness objective function of whole Artificial Fishs, and by mesh
Offer of tender numerical value is the smallest to be recorded as optimum individual, while minimum value and corresponding individual are charged to bulletin board；
Setting is current to count k=1 repeatly, executes foraging behavior to 100 Artificial Fishs being randomly generated, clustering behavior, knocks into the back
Behavior: Artificial Fish explores current locating environment, attempts to bunch respectively and two kinds of behaviors of knocking into the back, calculating target function fitness
Whether value improves, and selection executes the lesser behavior of fitness value, and default behavior is to look for food.If certain Artificial Fish exploration is bunched and is chased after
After tail behavior, fitness value does not improve, then this Artificial Fish executes foraging behavior, if this Artificial Fish is reaching row of looking for food
For maximum attempts after, fitness value then executes random behavior still without improvement, i.e., this Artificial Fish is around oneself
Random walk is to a new position in environment.
Due to each Artificial Fish all can around it local optimum it is close, all cross comparison all people work fish target letter
Numerical value e obtains the smallest individual of epicycle searching process target function value e, and in contrast to optimum individual last round of in bulletin board, if
The target function value e that epicycle optimizing obtains optimum individual is smaller, then replaces the optimum individual taken turns to be placed in bulletin board, carrying out
Next round optimizing promotes the shoal of fish close to global optimum；
Optimizing judges whether to reach termination condition i.e. maximum number of iterations K for a period of time afterwards, exports optimal ginseng if meeting
Number, otherwise enables the number of iterations add 1, carries out optimizing again by fish-swarm algorithm；
After reaching maximum number of iterations, the best initial weights and threshold value that optimizing is obtained as Elman neural network most
Excellent parameter, and establish the corresponding Elman neural network prediction model of high and low frequency component；
Finally 4 prediction models are analyzed using test set data, predict the high frequency detail point of subsequent time power respectively
Measure Di(i=1,2,3) and low frequency component A3, substitute into formula (5) reconstruct and obtain the power prediction value P of subsequent timet+1。
Using root-mean-square error index σMSEError (formula (6)) are predicted to analyze, if reaching error requirements, are protected
This group of weight and threshold value are deposited, the shoal of fish is otherwise initialized and restarts to optimize.
The above description is only an embodiment of the present invention, is not intended to limit the scope of the invention, all to utilize this hair
Equivalent structure or equivalent flow shift made by bright specification and accompanying drawing content is applied directly or indirectly in other relevant skills
Art field, is included within the scope of the present invention.
Claims (8)
1. the novel photovoltaic power forecasting method based on AFSA-Elman, which comprises the following steps:
Step 1: data collect photovoltaic station history output power and corresponding weather condition, to collected data make
Pretreatment, obtains the high frequency detail component and low frequency trend component of power to Power Decomposition, establishes sample set；
Step 2: determining nerve network input parameter, Elman nerve is established to high frequency detail component and low frequency trend component respectively
Network model predicts the high and low frequency component of subsequent time；
Step 3: Elman neural network parameter being optimized by artificial fish-swarm algorithm AFSA.
2. the novel photovoltaic power forecasting method based on AFSA-Elman as described in claim 1, which is characterized in that the step
Rapid 3 specifically:
Step 3.1: with the connection weight ω 1, ω 2, ω 3 of Elman neural network, threshold value b1, b2 is as parameter to be optimized, building
Five dimensional vectors initialize the shoal of fish as an Artificial Fish individual；
Step 3.2: using the difference e between the predicted value of Elman neural network and desired output as the fitness of Artificial Fish
Objective function, and be recorded as target function value is the smallest optimum individual and be recorded in bulletin board；
Step 3.3: the current shoal of fish executes foraging behavior, clustering behavior, behavior of knocking into the back respectively；
Step 3.4: each round finishes, and calculates the fitness target function value i.e. e of each Artificial Fish, and will most in contrast to bulletin board
The figure of merit is stored in bulletin board；
Step 3.5: judge whether to meet maximum number of iterations termination condition, exports optimized parameter if meeting, otherwise, iteration time
Number k=k+1, return step 3.3；
Step 3.6: after optimizing, using the best initial weights of output and threshold value as the optimized parameter of Elman neural network, and building
The vertical corresponding Elman neural network prediction model of high and low frequency component；
Step 3.7: test analysis being carried out to the model optimized, reconstruct obtains the predicted value of subsequent time power, if reached
Error requirements then saving this group of weight and threshold value, while exporting the prediction result of subsequent time power, otherwise go to step 3.1,
Restart to optimize.
3. the novel photovoltaic power forecasting method based on AFSA-Elman as described in claim 1, which is characterized in that collect
Data include power P, intensity of illumination Q, temperature T, humidity H, intensity of illumination G, wind speed S.
4. the novel photovoltaic power forecasting method based on AFSA-Elman as described in claim 1, which is characterized in that described pre-
The detailed process of processing are as follows: reject the power points that power is less than or equal to zero, reject and the very big abnormal point of proximity data deviation, general
Data normalization processing that treated, carries out wavelet decomposition for the history generated energy sequence after normalization, is height by signal decomposition
Frequency detail signal diWith low frequency trend component an, i=1,2,3 ... n, n are maximum decomposition level number.
5. the novel photovoltaic power forecasting method based on AFSA-Elman as described in claim 1, which is characterized in that described true
Determine nerve network input parameter: is weighed between different variables using Pearson correlation coefficient and linearly depend on degree, calculate each point
Pertinency factor ρ between amount and weather condition:
N is the number of samples of training set, x in formulaiFor the input parameter of certain prediction model to be selected,For being averaged for sample parameter
Value, yiIt is model output,For the average value for exporting sample.
6. the novel photovoltaic power forecasting method based on AFSA-Elman as claimed in claim 2, which is characterized in that execution is looked for
Before food behavior, clustering behavior, behavior of knocking into the back, Artificial Fish explores the environment being presently in, and reattempts two kinds of behaviors of bunching and knock into the back,
Whether calculating target function fitness value improves, and actual selection executes the lesser behavior of fitness value, and default behavior is to look for food.
7. the novel photovoltaic power forecasting method based on AFSA-Elman as claimed in claim 2, which is characterized in that each round
After iteration optimizing all can calculating target function value e, find the smallest individual of e value, and save in last round of bulletin board
Optimum individual traveling is compared, and the small optimum individual that is denoted as of target function value e is stored in bulletin board, and continue next round iteration,
It avoids falling into local optimum in searching process, keeps Artificial Fish close to global optimum.
8. the novel photovoltaic power forecasting method based on AFSA-Elman as claimed in claim 2, which is characterized in that lower a period of time
The predicted value for carving power, predicts subsequent time high fdrequency component D by the Elman neural network optimized respectivelyi, i=1,2,
3 ... n, n are maximum decomposition level number, low frequency component An, the predicted value of high and low frequency component is reconstructed, subsequent time function is obtained
The predicted value P of ratet+1:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810745300.0A CN109165770A (en) | 2018-07-09 | 2018-07-09 | Novel photovoltaic power forecasting method based on AFSA-Elman |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810745300.0A CN109165770A (en) | 2018-07-09 | 2018-07-09 | Novel photovoltaic power forecasting method based on AFSA-Elman |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109165770A true CN109165770A (en) | 2019-01-08 |
Family
ID=64897574
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810745300.0A Pending CN109165770A (en) | 2018-07-09 | 2018-07-09 | Novel photovoltaic power forecasting method based on AFSA-Elman |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109165770A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110414748A (en) * | 2019-08-12 | 2019-11-05 | 合肥阳光新能源科技有限公司 | Photovoltaic power prediction technique |
| CN110738253A (en) * | 2019-10-14 | 2020-01-31 | 南京信息工程大学 | short-term wind power prediction method based on FCM and AFSA-Elman |
| CN113779861A (en) * | 2021-07-23 | 2021-12-10 | 国网河北省电力有限公司电力科学研究院 | Photovoltaic power prediction method and terminal equipment |
| CN114004424A (en) * | 2021-12-21 | 2022-02-01 | 南方电网科学研究院有限责任公司 | Photovoltaic power prediction method, system, equipment and storage medium |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106446494A (en) * | 2016-05-11 | 2017-02-22 | 新疆大学 | Wavelet packet-neural network-based wind/photovoltaic power prediction method |
-
2018
- 2018-07-09 CN CN201810745300.0A patent/CN109165770A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106446494A (en) * | 2016-05-11 | 2017-02-22 | 新疆大学 | Wavelet packet-neural network-based wind/photovoltaic power prediction method |
Non-Patent Citations (4)
| Title |
|---|
| LI WANG等: "Dynamic Stability Analysis of a Hybrid Wave and Photovoltaic Power Generation System Integrated Into a Distribution Power Grid", 《IEEE TRANSACTIONS ON SUSTAINABLE ENERGY》 * |
| 刘洪波等: "基于人工鱼群神经网络的城市时用水量预测方法", 《天津大学学报(自然科学与工程技术版)》 * |
| 韩伟等: "基于FOA-Elman神经网络的光伏电站短期出力预测模型", 《电测与仪表》 * |
| 黎静华等: "考虑气象因素的短期光伏出力预测的奇异谱分析方法", 《电力自动化设备》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110414748A (en) * | 2019-08-12 | 2019-11-05 | 合肥阳光新能源科技有限公司 | Photovoltaic power prediction technique |
| CN110414748B (en) * | 2019-08-12 | 2022-04-08 | 阳光新能源开发股份有限公司 | Photovoltaic power prediction method |
| CN110738253A (en) * | 2019-10-14 | 2020-01-31 | 南京信息工程大学 | short-term wind power prediction method based on FCM and AFSA-Elman |
| CN113779861A (en) * | 2021-07-23 | 2021-12-10 | 国网河北省电力有限公司电力科学研究院 | Photovoltaic power prediction method and terminal equipment |
| CN113779861B (en) * | 2021-07-23 | 2023-08-22 | 国网河北省电力有限公司电力科学研究院 | Photovoltaic Power Prediction Method and Terminal Equipment |
| CN114004424A (en) * | 2021-12-21 | 2022-02-01 | 南方电网科学研究院有限责任公司 | Photovoltaic power prediction method, system, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109546659B (en) | Power distribution network reactive power optimization method based on random matrix and intelligent scene matching | |
| CN108388962B (en) | Wind power prediction system and method | |
| CN109165770A (en) | Novel photovoltaic power forecasting method based on AFSA-Elman | |
| CN108921359A (en) | A kind of distribution gas density prediction technique and device | |
| CN104408562A (en) | Photovoltaic system generating efficiency comprehensive evaluation method based on BP (back propagation) neural network | |
| CN102938093A (en) | Wind power forecasting method | |
| CN110969306B (en) | Power distribution low-voltage transformer area load prediction method and device based on deep learning | |
| CN111628494A (en) | Low-voltage distribution network topology identification method and system based on logistic regression method | |
| CN113837434A (en) | Solar photovoltaic power generation prediction method and device, electronic equipment and storage medium | |
| CN114357670A (en) | Power distribution network power consumption data abnormity early warning method based on BLS and self-encoder | |
| CN107194507A (en) | A kind of short-term wind speed forecasting method of wind farm based on combination SVMs | |
| CN110659672A (en) | Wind turbine generator output step uncertainty prediction method and device | |
| CN113762591B (en) | Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning | |
| CN110796292A (en) | Photovoltaic power short-term prediction method considering haze influence | |
| CN108345996B (en) | System and method for reducing wind power assessment electric quantity | |
| CN117394439A (en) | Method for improving power generation self-utilization rate of energy storage system | |
| CN112348236B (en) | Abnormal daily load demand prediction system and method for intelligent electricity consumption monitoring terminal | |
| CN115545962A (en) | Crop growth period control method and system based on multi-sensor system | |
| CN114372615A (en) | Short-term photovoltaic power prediction method and system considering spatial correlation | |
| Wang et al. | The power of short-term load algorithm based on LSTM | |
| CN118228897B (en) | CEEMDAN-SE-TCN-GRU model-based power load prediction method | |
| Zheng et al. | Short-Term Load Forecast Based on Feature Reconstruction and Bidirectional LSTM | |
| CN111626468A (en) | Photovoltaic interval prediction method based on partial convex loss function | |
| CN114243695B (en) | Power load prediction method based on bidirectional long-short-term memory neural network | |
| Lijuan | Research on Algorithm and Modification of Elman Network Based on Fuzzy Classification and Short-Term Electricity Consumption Prediction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190108 |