WO2021213192A1

WO2021213192A1 - Load prediction method and load prediction system employing general distribution

Info

Publication number: WO2021213192A1
Application number: PCT/CN2021/086342
Authority: WO
Inventors: 陶叶炜; 徐箭; 董甜; 华夏; 麦锦雯; 柏筱飞; 项敏; 周力; 童充; 钱艺琳; 李荷婷; 王丹; 陆峰; 廖思阳; 吴迪; 杨昊; 沈韵
Original assignee: 国网江苏省电力有限公司苏州供电分公司; 武汉大学
Priority date: 2020-04-22
Filing date: 2021-04-11
Publication date: 2021-10-28
Also published as: CN111552923B; CN111552923A

Abstract

Provided are a load prediction method and a load prediction system employing general distribution and capable of predicting short-term load power of an electrical power system. In the method, historical weather information data and electricity-related data are combined, and corresponding statistical analysis models are respectively constructed for different temperature increment levels. First, a histogram model is used to approximately characterize actual distribution, and then an appropriate model is selected to perform fitting for the actual distribution. The present invention employs a general distribution function model to perform fitting, and obtains a confidence interval for predicted power at a certain confidence level according to a closed-form analytical expression of an inverse function of a general distribution CDF.

Description

A load forecasting method and load forecasting system based on universal distribution

Technical field:

The invention belongs to the technical field of power load forecasting, and specifically relates to a load forecasting method and a forecasting system based on the influence of universally distributed meteorological factors on load power consumption.

Background technique:

With the improvement of the national economy, the demand for electricity is also increasing. There are many factors that need to be considered for the scientific and reasonable dispatch of electricity, of which temperature is an important factor. Changes in temperature will cause significant changes in regional load power on both civil and industrial electricity. On the one hand, the use of civil high-power power-consuming equipment (such as electrical equipment for cooling and heating) is closely related to meteorological factors such as temperature; on the other hand, temperature changes will also affect the production plan of enterprises and cause changes in industrial load power. Therefore, it is extremely necessary to analyze the quantitative relationship between load power and temperature and apply it to the load forecasting process.

At present, the existing research methods of electricity consumption forecasting can be divided into two categories, one is the indirect calculation method based on probability statistics, and the more common is the multiple regression model. In regression analysis, the dependent variable of the regression equation is often power load, which can be short-term load or long-term load. The independent variables are various factors that affect power load, such as economy, policy, electricity price, and climate. The disadvantage of regression analysis is that its realization requires a certain number of samples, and the samples must have a good distribution law and a definite development trend, and the computational workload is relatively large. The other is the indirect method based on historical electricity consumption data. Such as conventional mathematical models, such as gray prediction models and neural network models. For the power system, the affected power supply units, grid capacity, user conditions, and other information are known. However, many other factors that affect the load, such as meteorological conditions, changes in administrative and management policies, and regional economic activities, are difficult. Know exactly, therefore, the electrical load is a gray system. Grey theory accumulates the irregular historical data series into a series of ascending shapes with exponential growth law. Since the formation of the first-order differential equation is an exponential growth form, a differential equation can be established for the generated series. Model. So the gray model is actually a model built by generating a sequence of numbers. The data obtained from the GM model must be inversely generated, that is, the cumulative subtractive generation and reduction can be applied. The disadvantage is that the forecasting process does not consider the impact of the load by economy, climate, and policies, but directly processes and analyzes the time series data of electricity consumption. In addition, the greater the degree of dispersion of the original data, the greater the gray level, and the lower the prediction accuracy. . The neural network theory uses the learning function of the neural network to allow the computer to learn the mapping relationship contained in the historical load data, and then use this mapping relationship to predict the future load. BP neural network can simulate the nonlinear mapping relationship between any input and output. A three-layer BP network is composed of input layer, hidden layer and output layer. The training method is to calculate the actual output through a BP neural network for a set of input samples. The error between the actual output of the BP network and the output sample is used to correct the connection weight of the network until the error between the two reaches the set value. Generally, the input samples include load, temperature, weather conditions, and date types. The target sample during training is load by time period. The model application is to predict the load value of the next day from the load data of the previous one. The disadvantage is that there are certain difficulties in processing non-textual historical data information and the existing experience and knowledge of dispatchers; it has high requirements for input data, and when the number of samples is relatively small, the prediction accuracy is reduced; for those that are far away from the training samples Data and forecast results are not ideal.

In the prior art, the wind power probability prediction method based on the numerical weather forecast ensemble forecast results, the specific method steps are: 1. Establish a short-term wind power prediction model for each member of the numerical weather forecast ensemble forecast, and input the numerical weather forecast results separately The wind power prediction result of each member is obtained, and the wind power prediction result of the ensemble forecast of the weather forecast is obtained. 2. Identify the error type of each set according to the wind power prediction result of each member in the set. 3. Divide the power levels of each set according to the wind power prediction result of the set. 4. Calculate the set of relative errors for sets of different error types and power levels. 5. Use the kernel estimation method to obtain the probability density function of each sample in the set. 6. Use non-parametric fitting methods to fit the probability density distribution and fit the regression function. 7. Perform regression verification on the fitted regression results. 8. Calculate the upper limit of error and the lower limit of error under a certain confidence level. 9. Calculate the estimated interval of power prediction under a certain confidence level according to the upper and lower limits of the error.

In the existing research methods of electricity consumption forecasting, the multiple regression model has a simple model and perfect theory, which can fully consider various influencing factors, but it is easy to cause poor forecast accuracy due to improper selection of factors. Conventional mathematical models have good short-term forecasting effects, but they rely too much on mathematical and physical mechanisms, and have greater limitations in long-term forecasts and practical applications.

Summary of the invention:

In view of the above problems in the prior art, the present invention discloses a load forecasting method and a forecasting system based on universal distribution, and is applied to load power forecasting.

The above-mentioned technical problems of the present invention are mainly solved by the following technical solutions:

A load forecasting method based on general distribution, including the following steps:

Step 1. Collect historical data of load power and temperature at each moment of the load to be predicted within a set time period:

Step 2: Process the load power and temperature history data collected in Step 1, and calculate an array containing the temperature increment and the power ratio, where the temperature increment refers to the day after the same time minus the previous day at each time point The temperature increment of, the power ratio refers to the ratio of the power of the day after the same moment at each time point to the power of the previous day;

Step 3. Perform binning processing on the arrays containing temperature increments and power ratios obtained in step 2, and all the arrays are stored in their corresponding temperature increment grade data boxes:

Step 4. Establish the actual distribution model of the load power in each level data box based on the histogram:

Step 5: Use the general distribution function model to fit the data in each level data box to obtain the general distribution function corresponding to each level data box;

Step 6. According to the current power value of the load to be predicted and the temperature of the forecast day, based on the general distribution function fitted in step 5, determine the confidence interval of the power ratio of the load to be predicted under the forecast daily temperature condition to achieve the given weather Power prediction.

The present invention further includes the following preferred solutions.

In step 1, collect the load power of the load to be predicted in the last three months and the temperature data of the corresponding time at 24 hours and minutes; among them, in the historical data, the sampling period of the load power is 1 min, which corresponds to every Each load power data point is the active power at 1 min; the sampling period of the temperature data is 5 min, that is, the temperature corresponding to each temperature data point at 5 min.

In step 2, further include the following:

Step 2.1, for the historical data collected in step 1, establish a load power time series and a temperature time series respectively, where the load power time series is a time series every 1 minute; the temperature time series is a time series every 5 minutes;

Step 2.2, perform frequency reduction processing on the load power time series, take a point every 5 minutes, and unify the power and temperature into a time period of 5 minutes;

Step 2.3: Calculate the temperature increment of the next day minus the previous day at the same time at each time point and the ratio of the corresponding power of the next day to the power of the previous day; thus obtain the temperature increment sequence and the power ratio sequence, which correspond to each other Form an array, the first column of the array is the temperature increment, and the second column is the power ratio.

In step 3, specifically include the following:

Step 3.1, classify the temperature increment, divide the temperature increment into R levels, that is, R data boxes, the R levels are divided into R levels evenly between the maximum and minimum temperature increments;

Step 3.2, according to the size of the "temperature increment" in each historical data array, store all the arrays calculated in step 2 into their corresponding temperature increment level data boxes to complete the binning process of the arrays.

Preferably, the R=7.

In step 4, draw the probability density histogram of the power ratio for each data box according to the "temperature increment" in the data box of each level.

In step 4, specifically include the following:

Step 4.1, determine the total width of the histogram; obtain the range from the difference between the maximum value and the minimum value of the power ratio in the data box sample sets of different temperature increment levels. Product) to take the number greater than the range;

Step 4.2, determine the group number, group distance, and each group limit of the histogram, assuming that the group number of the histogram is N; the width of each group is 1/N, which is the group distance;

Step 4.3, determine the frequency of each group, divide the power ratio data in the sample set into each group of the histogram according to the size, and count the number of data in each group, which is the frequency;

Step 4.4, draw a probability density histogram, the abscissa is the power ratio, the ordinate is the probability density, each group of the histogram corresponds to a rectangle, the width of the rectangle is the group distance, the height is the probability density of each group, the conversion of probability density and frequency Relationship: probability density = frequency/total number of samples * number of groups.

Preferably N=50.

In step 5, the general distribution is defined as follows:

If the continuous random variable X obeys a general distribution with shape parameters α, β and γ, it is recorded as:

X～V(α,β,γ)

Among them, the shape parameters α, β and γ respectively satisfy:

α＞0, β＞0, -∞＜γ＜+∞

The probability density function (Probability Density Function, PDF) f(x) expression of the general distribution is:

The cumulative distribution function (Cumulative Distribution Function, CDF) F(x) expression of the general distribution is:

F(x)=(1+e ^-α(x-γ) ) ^-β

The inverse function of CDF F ^-1 (c) is expressed as:

Among them, f(x) is the probability density of the general distribution function, and F(x) is the cumulative distribution value of the general distribution function.

In step 6, the temperature forecast value at each time point of the day to be predicted and the temperature value at the corresponding time point of the current day are compared to obtain the temperature increment sequence. According to the corresponding different temperature increment data boxes, use step 5 Determine the general distribution function shape parameter value fitted by the different temperature increment data boxes, calculate the expression of the general distribution CDF inverse function F ^-1 (c), and obtain the confidence interval of the load forecast power under the set confidence level :

Among them, w _l,up and w _l,low are the upper and lower bounds of the load power confidence interval, respectively.

The preferred confidence level conf is 0.95.

This application also discloses a load forecasting system based on the aforementioned load forecasting method, which includes a historical database, a data collection and processing unit, a data binning processing unit, a histogram unit, a fitting calculation unit, and a load power prediction calculation unit; its features Lies in:

The data collection and processing unit collects time-by-time load power and temperature in the historical data of the load to be predicted, unifies the load power and temperature sampling values into the same sampling frequency, and establishes an array containing a temperature increment sequence and a power ratio sequence;

The data binning processing unit performs binning processing on the array established by the data acquisition and processing unit based on the temperature increment;

The histogram unit draws a power ratio probability density histogram for each bin data, and calculates the power ratio probability density in each group in the histogram;

The fitting calculation unit uses a general distribution function model to fit the data in each level data box to obtain a general distribution function corresponding to each level data box;

The load power prediction calculation unit calculates the temperature of the load to be predicted on the predicted day based on the general distribution function of each corresponding level data box fitted by the fitting calculation unit according to the input current power value and the weather forecast temperature value of the predicted day The confidence interval of the power ratio under the conditions can realize the power prediction under the given weather.

Compared with the prior art, this application has the following beneficial technical effects.

Through the method of probabilistic modeling, the associated probability model of the load change and the change of meteorological information is established. The histogram model is used to approximate the actual distribution of the load power change under different temperature increments. The general distribution model is selected to compare the actual distribution histogram The graph is fitted, and the model can effectively characterize the peak characteristics and off-axis characteristics of the distribution law with high fitting accuracy. Using the reversible mathematical properties of the cumulative probability density function of the distribution function, the incremental interval of the load power under the known weather forecast results can be analytically calculated under a certain confidence level, so as to realize the influence of weather on the power system load power forecast results. Dynamic correction after the factor. Based on how many degrees the temperature has increased at the same time on different dates and how many times the load power will increase accordingly, exploring the quantitative relationship between the temperature increase and the power ratio is innovative and feasible, and the established load changes The probabilistic model of the correlation between the temperature change and the temperature change can provide a quantitative data analysis method for analyzing the influence of temperature change on the load power, and this method is also suitable for analyzing the quantitative influence of other external factors such as humidity and rainfall on the load power change. relation.

Description of the drawings:

Figure 1 is a schematic flow chart of a general distributed load forecasting method according to this application;

Fig. 2 is a probability distribution histogram of the fourth data box in the embodiment of the application;

FIG. 3 is an effect diagram after fitting the "power ratio" in the fourth box according to the embodiment of the application;

Figure 4 is a structural block diagram of a general distributed load forecasting system of this application.

Detailed ways:

The technical solution of the present invention will be further described in detail below in conjunction with the drawings and specific embodiments of the specification.

As shown in FIG. 1, it is a schematic flowchart of a general distributed load forecasting method disclosed in this application. It includes the following steps:

This application preferably collects the load power of the load to be predicted in the last three months, 24 hours a day, and the corresponding temperature data; among them, in the historical data, the sampling period of the load power is 1 min, which corresponds to each load The power data point is the active power at 1 min; the sampling period of the temperature data is 5 min, which means that each temperature data point corresponds to the temperature at 5 min.

Step 3: Perform binning processing on the arrays containing temperature increments and power ratios obtained in step 2, and all the arrays are stored in their corresponding temperature increment level data boxes;

Step 3 specifically includes the following:

Among them, in the preferred solution of the present application, R=7 is preferred.

Step 4. Establish the actual distribution model of the load power in each level data box based on the histogram;

In step 4, specifically include the following:

Step 4.1, determine the total width of the histogram; subtract the minimum value from the maximum value in the temperature increment sequence to obtain the difference, and the total width of the histogram shall be the smallest integer greater than the difference;

Based on a large number of statistical results and calculation analysis, this application selects N=50.

The general distribution is defined as follows:

X～V(α,β,γ)

Among them, the shape parameters α, β and γ respectively satisfy:

α＞0, β＞0, -∞＜γ＜+∞

F(x)=(1+e ^-α(x-γ) ) ^-β

The inverse function of CDF F ^-1 (c) is expressed as:

The preferred confidence level conf is 0.95.

As shown in FIG. 4, this application also discloses a load forecasting system based on the aforementioned load forecasting method, including a historical database, a data collection and processing unit, a data binning processing unit, a histogram unit, a fitting calculation unit, and a load forecasting system. Power prediction calculation unit.

The data collection and processing unit collects time-by-time load power and temperature in the historical data of the load to be predicted, unifies the load power and temperature sampling values into the same sampling frequency, and establishes an array containing a temperature increment sequence and a power ratio sequence; The data binning processing unit performs binning processing on the array established by the data acquisition and processing unit based on the temperature increment; the histogram unit draws a power ratio probability density histogram for each binning data, and calculates the histogram The power ratio probability density in each group; the fitting calculation unit uses a general distribution function model to fit the data in each level data box to obtain the general distribution function corresponding to each level data box; the load power prediction calculation Based on the input current power value and the weather forecast temperature value of the forecast day, the unit calculates the power ratio confidence of the load to be predicted under the forecast daily temperature based on the general distribution function of each corresponding level data box fitted by the fitting calculation unit Interval, to realize the power prediction under a given weather.

The following takes an industrial park as an example to introduce:

Step 1. Collect historical data of load power and temperature at each moment of the load to be predicted within a set time period. In this embodiment, collect 2016/11/1-2016/11/30; 2017/7/1-2017/8 /31 A total of three months of the load power at 24 hours a day and the temperature at the corresponding time. Among them, the sampling period of load power is 1min, that is, the active power corresponding to each load power data point is 1min; the sampling period of temperature data is 5min, that is, the temperature corresponding to each temperature data point is 5min.

Step 2. Data processing:

Step 2.1. In the collected data, the power is a time series of one period every 1min, corresponding to each data point is the active power at 1min; the temperature is a time series of every 5min, corresponding to each data point is 5min The temperature; every 24 hours a day, there are a total of 144 1min time periods and 288 5min time periods, so each power time series P has a total of 144 data, and each temperature time series has a total of 288 data.

Step 2.2, perform down-sampling on the power time series, take a point every 5 minutes, and unify the power and temperature into a time period of 5 minutes;

Step 3: Binning processing based on the temperature increment prediction level:

Step 3.1, classify the temperature increment:

According to the temperature increment, it is divided into 7 levels, that is, 7 data boxes. Store the corresponding arrays into the temperature increment level data box respectively. In this embodiment, the value range of the temperature increment is within [-15.5°C, 18.5°C]. Divide it into 7 boxes, and the temperature increment in each box ranges from [-16,-11], [-11,-6], [-6,-1], [-1,4] , [4,9], [9,14], [14,19].

Step 3.2, after the data processing of step 2, the historical data has been divided into arrays with historical moments as the unit; according to the size of the "temperature increment" in each historical data array, all the arrays in the database are stored in it. In the corresponding temperature increment level data box, the binning process of the array is completed.

Step 4. The actual distribution model based on the histogram:

According to the "temperature increment" samples in the data boxes of each level, draw the probability density histograms of the power ratios for each data box;

Step 4.2, determine the group number, group distance and each group limit of the histogram, assuming that the group number of the histogram is N=50; the width of each group is 1/N, that is, the group distance;

Step 4.3, determine the frequency of each group, divide the power ratio data in the sample set into the histogram groups according to the size, and count the number of data in each group, which is the frequency; step 4.4, draw the probability density histogram, the abscissa is Power ratio, the ordinate is the probability density, each group of the histogram corresponds to a rectangle, the width of the rectangle is the group distance, and the height is the probability density of each group. The conversion relationship between probability density and frequency: probability density = frequency/total number of samples * group number.

Step 5. Fit the data in the box:

Use general distribution function model to fit the data in the box;

The general distribution is defined as follows:

X～V(α, β, γ) Among them, the shape parameters α, β and γ respectively satisfy:

α＞0, β＞0, -∞＜γ＜+∞

F(x)=(1+e ^-α(x-γ) ) ^-β

The inverse function of CDF F ^-1 (c) is expressed as:

Fitting is performed for different temperature increment levels respectively, and the α, β and γ values obtained by the fitting are shown in Table 1 below: Table 1

The histogram is the actual distribution, and the continuous curve is the general distribution model effect. Fig. 2 shows the probability distribution histogram of the fourth data box according to the embodiment of the application, and Fig. 3 shows the effect diagram after fitting the "power ratio" in the fourth box according to the embodiment of the application.

Obtain the maximum and minimum power ratio at a certain confidence level (take 95% as an example), and the power prediction under a given weather. Assuming that the current power value is known to be 10, the minimum and maximum power of tomorrow at the 95% confidence level can be calculated as shown in Table 2 below:

Table 2

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: Modifications or equivalent replacements are made to the specific embodiments of, and any modification or equivalent replacement that does not depart from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

Claims

A load forecasting method based on general distribution, which is characterized by combining historical weather information data and electrical data, and constructing corresponding statistical analysis models for different temperature increment levels, so as to calculate the predicted power at a set confidence level Confidence interval.
A load forecasting method based on universal distribution, characterized in that the load forecasting method includes the following steps:

Step 1. Collect historical data of load power and temperature at each time of the load to be predicted within a set time period;

Step 2. Process the load power and temperature history data collected in Step 1, and calculate an array containing the temperature increment and the power ratio, where the temperature increment refers to the temperature of the day after the same time minus the previous day at each time point The temperature increment obtained by the temperature, the power ratio refers to the ratio of the power of the day after the same time at each time point to the power of the previous day;

Step 3: Perform binning processing on the arrays containing temperature increments and power ratios obtained in step 2, that is, storing all the arrays in their corresponding temperature increment level data boxes;

Step 4. Establish the actual distribution model of the load power in each level data box based on the histogram;

Step 5: Use the general distribution function model to fit the data in each level data box to obtain the general distribution function corresponding to each level data box;

Step 6. According to the current power value of the load to be predicted and the temperature of the forecast day, based on the general distribution function fitted in step 5, determine the confidence interval of the power ratio of the load to be predicted under the forecast daily temperature condition to achieve the given weather Power prediction.
The load forecasting method according to claim 2, characterized in that:

In step 1, collect the load power of the load to be predicted in the last three months and the temperature data of the corresponding time at 24 hours and minutes; among them, in the historical data, the sampling period of the load power is 1 min, which corresponds to every Each load power data point is the active power at 1 min; the sampling period of the temperature data is 5 min, that is, the temperature corresponding to each temperature data point at 5 min.
The load forecasting method according to claim 2, characterized in that:

In step 2, further include the following:

Step 2.1, for the historical data collected in step 1, establish a load power time series and a temperature time series respectively, where the load power time series is a time series every 1 minute; the temperature time series is a time series every 5 minutes;

Step 2.2, perform frequency reduction processing on the load power time series, take a point every 5 minutes, and unify the power and temperature time series into a period of 5 minutes;

Step 2.3, find the temperature increment obtained by subtracting the temperature of the previous day from the temperature of the next day at the same time at each time point and the power ratio of the power of the next day to the power of the previous day; thereby obtaining the temperature increment sequence and the power ratio sequence , The two correspond to form an array, the first column of the array is the temperature increment, and the second column is the power ratio.
The load forecasting method according to claim 2, characterized in that:

In step 3, specifically include the following:

Step 3.1, the temperature increment is divided into R levels according to the temperature increment, that is, R data boxes. The R levels refer to the R levels divided equally between the maximum and minimum temperature increments. grade;

Step 3.2, according to the size of the temperature increment in each historical data array, store all the arrays calculated in step 2 into their corresponding temperature increment level data boxes respectively to complete the binning process of the arrays.
The load forecasting method according to claim 5, characterized in that:

The R=7.
The load forecasting method according to claim 2, characterized in that:

In step 4, draw the probability density histogram of the power ratio for each data box according to the temperature increment in each level data box.
The load forecasting method according to claim 2, characterized in that:

In step 4, specifically include the following:

Step 4.1, determine the total width of the histogram; subtract the minimum value from the maximum value in the temperature increment sequence to get the difference. The total width of the histogram (ie the product of the group distance and the number of groups) is the smallest value greater than this difference Integer

Step 4.2, determine the group number, group distance, and each group limit of the histogram, assuming that the group number of the histogram is N; the width of each group is 1/N, which is the group distance;

Step 4.3, determine the frequency of each group, divide the sample set, that is, the power ratio data in all the arrays processed by binning, into each group of the histogram according to the size, and count the number of data in each group, which is the frequency;

Step 4.4, draw a probability density histogram, the abscissa is the power ratio, the ordinate is the probability density, each group of the histogram corresponds to a rectangle, the width of the rectangle is the group distance, the height is the probability density of each group, the conversion of probability density and frequency Relationship: probability density = frequency/total number of samples * number of groups.
The load forecasting method according to claim 8, characterized in that:

The number of groups of the histogram N=50.
The load forecasting method according to claim 2, characterized in that:

In step 5, the general distribution is defined as follows:

If the continuous random variable X obeys a general distribution with shape parameters α, β and γ, it is recorded as:

X～V(α,β,γ)

Among them, the shape parameters α, β and γ respectively satisfy:

α＞0, β＞0, -∞＜γ＜+∞

The probability density function (Probability Density Function, PDF) f(x) expression of the general distribution is:

The cumulative distribution function (Cumulative Distribution Function, CDF) F(x) expression of the general distribution is:

F(x)=(1+e -α(x-γ) ) -β

The inverse function of CDF F -1 (c) is expressed as:

Among them, f(x) is the probability density of the general distribution function, and F(x) is the cumulative distribution value of the general distribution function.
The load forecasting method according to claim 2, characterized in that:

In step 6, the temperature forecast value at each time point of the day to be predicted is the difference from the temperature value at the corresponding time point of the current day to obtain the temperature increment sequence, and the temperature increment data box corresponding to the corresponding different temperature increment data box is used to determine the temperature in step 5. The general distribution function shape parameter values obtained by fitting the different temperature increment data boxes of, calculate the expression of the general distribution CDF inverse function F -1 (c) to obtain the confidence interval of the load forecast power under the set confidence level:

Among them, w l,up and w l,low are the upper and lower bounds of the load power confidence interval, respectively.
The load forecasting method according to claim 11, characterized in that:

The preferred confidence level conf is 0.95.
A load forecasting system based on the load forecasting method of any one of the preceding claims 1-12, comprising a historical database, a data collection and processing unit, a data binning processing unit, a histogram unit, a fitting calculation unit, and a load Power prediction calculation unit; characterized by:

The data collection and processing unit collects time-by-time load power and temperature in the historical data of the load to be predicted, unifies the load power and temperature sampling values into the same sampling frequency, and establishes an array containing a temperature increment sequence and a power ratio sequence;

The data binning processing unit performs binning processing on the array established by the data acquisition and processing unit based on the temperature increment;

The histogram unit draws a power ratio probability density histogram for each bin data, and calculates the power ratio probability density in each group in the histogram;

The fitting calculation unit uses a general distribution function model to fit the data in each level data box to obtain a general distribution function corresponding to each level data box;

The load power prediction calculation unit calculates the temperature of the load to be predicted on the predicted day based on the general distribution function of each corresponding level data box fitted by the fitting calculation unit according to the input current power value and the weather forecast temperature value of the predicted day The confidence interval of the power ratio under the conditions can realize the power prediction under the given weather.