A method for electric heating load forecasting based on probability modeling and error correction

Abstract

The present invention discloses an electric heating load prediction method based on probability modeling and error correction, comprising the following steps: step 1: establishing an association probability model of the load change at the same time of two consecutive days and the load data of the previous day based on the historical data of the electric heating load; step 2: using a histogram model to characterize the actual distribution of the load change under different load fluctuation intervals; step 3: using a normal distribution to fit the histogram model of the load change in each fluctuation interval, obtaining a probability model of the load change under different load intervals to achieve a preliminary prediction of the electric heating load; step 4: establishing a MEMD-BiLSTM error correction model to correct the prediction result according to the error between the actual value and the preliminary prediction result, and obtaining the final prediction result of the electric heating load. The prediction method can fully mine the internal correlation and prediction error information of the electric heating load data, and effectively improve the prediction accuracy of the electric heating load.

Description

Technical Field

[0001] The invention relates to the technical field of smart grids, and in particular to an electric heating load prediction method based on probability modeling and error correction. Background Art

[0002] In recent years, with the rapid economic development in my country, smog weather has also spread widely, reducing the quality of the environment. In order to cope with the depletion of fossil energy and environmental pollution, the national energy department has introduced a series of energy policies. For example, in the "Guiding Opinions on Electricity Substitution" issued by the National Energy Administration in 2016, it was pointed out that the development strategy of replacing coal with electricity for winter heating with electricity substitution as the core project should be implemented, advocating the gradual promotion of the "coal to electricity" project in the field of residential heating, promoting electric heating to replace coal heating, promoting the gradual replacement of traditional energy with clean energy, reducing the emission of carbides, sulfides and nitrides, and improving air quality from the source.

[0003] With the large-scale access of electric heating equipment to the power grid, the peak-to-valley difference of winter power load has further increased, and the daily load characteristics and daily load curves in winter have obvious differences, which has brought new challenges to the power supply quality and the safe and stable operation of the distribution network. Unlike traditional power loads, electric heating is significantly affected by temperature changes and is more similar to the air conditioning and refrigeration load in summer. It not only has obvious seasonal climate characteristics, but also has stronger time series characteristics and nonlinear characteristics. Accurate electric heating load forecasting can not only ensure the economic and stable operation of the power system, but also play an important role in the peak and frequency regulation functions of the power system. This paper combines the probability modeling method with the neural network model, and uses the probability modeling method to realize the preliminary forecast of the electric heating load in the next day. At the same time, considering the large error of the forecast result, an error correction model of multivariate empirical mode decomposition-long short-term memory neural network is established to correct the forecast result and improve the accuracy of electric heating load forecasting.

[0004] Defects and shortcomings of the existing technology:

[0005] 1. There are relatively few studies on short-term electric heating load forecasting methods, and most of them directly use statistical methods or neural networks to establish the mapping relationship between the main influencing factors and the electric heating load to complete the forecast. However, most of these methods ignore the changing rules between the electric heating load data at the same time on two consecutive days, and fail to fully explore the internal correlation of the electric heating load data.

[0006] 2. Electric heating load is affected by many factors such as weather, temperature, and environment, which means that any accurate electric heating load prediction method will have certain errors in its prediction results. The error information is reflected in the difference between the actual value and the predicted value of the electric heating load. Most existing methods ignore this error information, so the prediction accuracy needs to be further improved. Summary of the invention

[0007] The purpose of the present invention is to overcome the shortcomings of the prior art and provide an electric heating load prediction method based on probability modeling and error correction, which can fully explore the internal correlation of electric heating load data and prediction error information, and effectively improve the accuracy of electric heating load prediction.

[0008] A method for predicting electric heating load based on probability modeling and error correction includes the following steps:

[0009] Step 1: Based on the historical data of electric heating load, establish a correlation probability model between the load change at the same time in two consecutive days and the load data of the previous day;

[0010] Step 2: Use the histogram model to characterize the actual distribution of load changes under different load fluctuation ranges;

[0011] Step 3: Use normal distribution to fit the histogram model of load variation in each fluctuation range, and obtain the probability model of load variation in different load ranges to achieve preliminary prediction of electric heating load;

[0012] Step 4: According to the error between the actual value and the preliminary prediction result, a MEMD-BiLSTM error correction model is established to correct the prediction result and obtain the final prediction result of the electric heating load.

[0013] Preferably, step 1 needs to establish a correlation probability model between the load change at the same time on two consecutive days and the load on the previous day based on a normal distribution function, so as to calculate the load change at the same time on two consecutive days in the electric heating load data;

[0014] ΔP i,t =P i,t -P i+1,t

[0015] Where P i,t is the electric heating load value at the tth moment on the i-th day; P i+1,t is the electric heating load value at time t on the i+1th day; ΔP i,t is the change in electric heating load at time t on the i-th day.

[0016] Preferably, in step 2, the maximum and minimum values ​​of the historical load data need to be counted to obtain the fluctuation range of the electric heating load, and the fluctuation range is divided into n small fluctuation ranges, and the histogram model is used to establish the actual distribution of the load change at the same time on two consecutive days in different load fluctuation ranges.

[0017] Preferably, in step 3, the load changes between each fluctuation zone need to be fitted using a normal distribution function based on a histogram model to obtain a normal distribution curve and a probability density function of the load changes in each interval; and a preliminary prediction of the electric heating load is achieved using the actual load data of the day before the test date and an associated probability model of the load changes.

[0018] Preferably, when step 4 establishes the error correction model based on the MEMD-BiLSTM network, it specifically includes the following sub-steps:

[0019] Step 4.1: Calculate the error between the actual data of the training centralized electric heating load and the preliminary predicted value, which is used as the training data for the subsequent error correction model;

[0020] P error =P real -P pred1

[0021] Where P real is the actual value of the electric heating load for training; P pred1 is the initial prediction value of the training set; P error is the error data of the training set;

[0022] Step 4.2: Use multivariate empirical mode decomposition to decompose the preliminary prediction value and the error at the same time, and obtain two groups of load components with high to low frequencies, so as to reduce the unsteady characteristics of the preliminary prediction value and error series and improve the predictability of the model;

[0023] Step 4.3: Reconstruct the two sets of load components into a high-frequency component and a low-frequency component respectively to reduce the model training time and the cumulative error;

[0024] Step 4.4: Use a bidirectional long short-term memory neural network to establish prediction models for high-frequency components and low-frequency components respectively;

[0025] Step 4.5: superimpose the prediction results of the high-frequency component and the low-frequency component to achieve error prediction; use the error prediction result to correct the initial prediction value to obtain the final prediction result of the electric heating load;

[0026] P′ pred =P′ error +P pred

[0027] In the formula, P′ error is the error prediction value; P′ pred It is the final predicted value of electric heating load.

[0028] The advantages and technical effects of the present invention are:

[0029] The present invention discloses an electric heating load prediction method based on probability modeling and error correction. By mining the internal relationship between electric heating load data, a normal distribution model is used to establish a correlation probability model between the load change at the same time on two consecutive days and the load of the previous day, so as to characterize the quantitative correlation between the electric heating load data and the load change. In view of the deviation between the prediction result of the probability model and the actual value, the present invention establishes an error correction model based on MEMD-BiLSTM, extracts the effective information of the error sequence, and corrects the initial prediction value with the error prediction result to obtain the final prediction result, thereby improving the short-term prediction accuracy of the electric heating load. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is the load variation sequence and the corresponding electric heating load sequence at the same time in two consecutive days in the present invention;

[0031] Figure 2 It is the histogram model of load variation under different fluctuation ranges in the present invention (intervals one to four);

[0032] Figure 3 It is the initial prediction value and error sequence diagram of the electric heating load in the present invention;

[0033] Figure 4 It is the MEMD decomposition result diagram of the initial prediction value and error sequence in the present invention;

[0034] Figure 5 This is a comparison chart of the actual value and the predicted value of the embodiment of the present invention on February 23, 2021;

[0035] Figure 6 It is the prediction flow chart of the present invention. DETAILED DESCRIPTION

[0036] In order to further understand the content, features and effects of the present invention, the following embodiments are given as examples and described in detail with reference to the accompanying drawings. It should be noted that the embodiments are illustrative, not restrictive, and the protection scope of the present invention cannot be limited thereby.

[0037] The electric heating load prediction method based on probability modeling and error correction of the present invention establishes a correlation probability model between the load change at the same time on two consecutive days and the load of the previous day based on the normal distribution function:

[0038] Step 1: Obtain the power data of the distribution line with electric heating load in a certain area of ​​Tianjin from December 1, 2020 to February 28, 2021. The historical load data is one point every 15 minutes, with a total of 96 sampling points per day. The electric heating load data from December 1, 2020 to February 22, 2021 is selected as the training set, and the electric heating load data from February 23, 2021 to February 28, 2021 is selected as the test set. On this basis, calculate the load change at the same time on two consecutive days in the training set;

[0039] ΔP i,t =P i,t -P i+1,t (1)

[0040] Where P i,t is the electric heating load value at the tth moment on the i-th day; P i+1,t is the electric heating load value at time t on the i+1th day; ΔP i,t is the change in electric heating load at time t on the i-th day.

[0041] Step 2: Obtain the load variation sequence ΔP according to step 1 i,t and the corresponding load sequence P i,t A two-dimensional array is formed, in which the first column is the load change and the second column is the corresponding electric heating load. The maximum value P of the electric heating load data in the second column is counted. max and the minimum value P min Thus, the fluctuation range of the electric heating load is obtained as [0.8, 3.2]. Then, the two-dimensional array is divided into four small intervals according to the fluctuation range, namely [0.8, 1.4], [1.4, 2.0], [2.0, 2.6] and [2.6, 3.2].

[0042] Step 3: Use the histogram to establish the actual distribution of load changes within each fluctuation interval. When establishing a histogram model, the first thing to do is to select an appropriate group N. Too few groups will distort the distribution law of the histogram, and too many groups will affect the obviousness of the data distribution law. After that, determine the frequency and frequency of each group. Divide the power change sequence into the corresponding histogram groups according to size, and count the number of data in each group, which is the frequency; the ratio of the frequency of each group to its total number is the frequency of each group.

[0043] Step 4: According to the histogram model, the probability distribution of the load change in each interval is fitted using the normal distribution, so as to obtain the normal distribution curve and probability density function of the load change in each interval;

[0044] Step 5: Use the 96-point actual load data of the day before the test date and the associated probability model of the load change in each interval obtained in step 4 to achieve a preliminary prediction of the electric heating load.

[0045] Establish an error correction model based on the MEMD-BiLSTM network:

[0046] Step 6: Calculate the preliminary prediction results P for the training set and test set based on the probability model pred1 and P pred2 On this basis, the error between the actual data of the electric heating load in the training set and the preliminary predicted value is calculated as the training data for the subsequent error correction model;

[0047] P error =P real -P pred1 (2)

[0048] Where P real is the actual value of the electric heating load for training; P pred1 is the initial prediction value of the training set; P error is the error data of the training set.

[0049] Step 7: In view of the strong randomness and non-stationary characteristics of the error sequence, the preliminary prediction value and the error sequence of the training set are decomposed simultaneously using multivariate empirical mode decomposition (MEMD) to obtain the same number of intrinsic mode components and residual components with high to low frequencies, thereby reducing the non-stationary characteristics of the sequence and improving the predictability of the model;

[0050] Step 8: According to the high and low frequencies, the load component and error component of the preliminary prediction value are reconstructed into a high-frequency component and a low-frequency component respectively, so as to reduce the model training time and the accumulated error;

[0051] Step 9: Use the bidirectional long short-term memory neural network (BiLSTM) to establish prediction models for high-frequency components and low-frequency components respectively. The error component of the previous day's 96 points and the preliminary prediction value component of the second day in the training set are used as the input of the BiLSTM network to complete parameter initialization, and the error prediction value of the second day is used as the output of the BiLSTM network. At the same time, the mean square error function is selected as the network loss function, as shown in formula (3):

[0052]

[0053] Where P i is the actual value, is the predicted value.

[0054] The parameters of the BiLSTM network are adjusted in reverse according to the value of the loss function. If the value of the loss function obtained by the current network parameters is less than the preset value or reaches the number of iterations, the network parameter training is completed;

[0055] Input the error component of 96 points on the day before the day to be predicted and the preliminary prediction value component of the day to be predicted into the BiLSTM network that has completed parameter training to obtain the error prediction value of the day to be predicted;

[0056] Step 10: Superimpose the prediction results of the high-frequency component and the low-frequency component of the day to be predicted to achieve error prediction. The error prediction result is superimposed on the initial prediction value to obtain the final prediction result of the electric heating load.

[0057] P′ pred =P′ error +P pred2 (3)

[0058] In the formula, P′ error is the error prediction value; P′ pred It is the final predicted value of electric heating load.

[0059] Finally, all the parts not described in the present invention adopt mature technical means in the prior art.

[0060] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all these improvements and changes should fall within the scope of protection of the appended claims of the present invention.

Claims

1. A method for predicting electric heating load based on probability modeling and error correction, characterized in that: The following steps are involved: Step 1: Based on the historical data of electric heating load, establish a correlation probability model between the load change at the same time in two consecutive days and the load data of the previous day; Step 2: Use the histogram model to characterize the actual distribution of load changes under different load fluctuation ranges; Step 3: Use normal distribution to fit the histogram model of load variation in each fluctuation range, and obtain the probability model of load variation in different load ranges to achieve preliminary prediction of electric heating load; Step 4: According to the error between the actual value and the preliminary prediction result, a MEMD-BiLSTM error correction model is established to correct the prediction result and obtain the final prediction result of the electric heating load; When the error correction model is established based on the MEMD-BiLSTM network, step 4 specifically includes the following sub-steps: Step 4.1: Calculate the error between the actual data of the training centralized electric heating load and the preliminary predicted value, which is used as the training data for the subsequent error correction model; P error =P real -P pred1 Where P real is the actual value of the electric heating load for training; P pred1 is the initial prediction value of the training set; P error is the error data of the training set; Step 4.2: Use multivariate empirical mode decomposition to decompose the preliminary prediction value and the error at the same time, and obtain two groups of load components with high to low frequencies, so as to reduce the unsteady characteristics of the preliminary prediction value and error series and improve the predictability of the model; Step 4.3: Reconstruct the two sets of load components into a high-frequency component and a low-frequency component respectively to reduce the model training time and the cumulative error; Step 4.4: Use a bidirectional long short-term memory neural network to establish prediction models for high-frequency components and low-frequency components respectively; Step 4.5: Superimpose the prediction results of the high-frequency component and the low-frequency component to achieve error prediction; The initial prediction value is corrected with the error prediction result to obtain the final prediction result of the electric heating load; P p ′ red =P e ′ rror +P pred Where P e ' rror is the error prediction value; P p ' red It is the final predicted value of electric heating load.

2. The electric heating load prediction method based on probability modeling and error correction according to claim 1 is characterized in that: The step 1 needs to establish a correlation probability model between the load change at the same time on two consecutive days and the load on the previous day based on the normal distribution function, so as to calculate the load change at the same time on two consecutive days in the electric heating load data; ΔP i,t =P i,t -P i+1,t Where P i,t is the electric heating load value at the tth moment on the i-th day; P i+1,t is the electric heating load value at time t on the i+1th day; ΔP i,t is the change in electric heating load at time t on the i-th day.

3. The electric heating load prediction method based on probability modeling and error correction according to claim 1 is characterized in that: In step 2, the maximum and minimum values ​​of the historical load data need to be counted to obtain the fluctuation range of the electric heating load, and the fluctuation range is divided into n small fluctuation ranges, and the actual distribution of the load change at the same time on two consecutive days in different load fluctuation ranges is established using a histogram model.

4. The electric heating load prediction method based on probability modeling and error correction according to claim 1 is characterized in that: In step 3, the load variation between each fluctuation area needs to be fitted using the normal distribution function according to the histogram model to obtain the normal distribution curve and probability density function of the load variation in each interval; and the actual load data of the day before the test date and the associated probability model of the load variation are used to achieve a preliminary prediction of the electric heating load.

Images