Power grid planning-oriented demand side resource response potential prediction method

Abstract

The invention provides a demand side resource response potential prediction method for power grid planning, and the method comprises the steps: firstly obtaining power grid data of a region to be planned for multiple days in recent periods, and carrying out the preprocessing of the obtained power grid data, and obtaining a corresponding preprocessing data set; inputting the preprocessed data set into a hybrid load prediction model trained based on historical power grid data of the established power grid to obtain a load prediction result; wherein the hybrid load prediction model comprises a CaualTransformer unit, a GRU unit, and a VAE-GRU unit, and the VAE-GRU unit is connected with the CaualTransformer unit and the VAE-GRU unit; and finally, according to a load prediction result, calculating the demand side resource response potential of the region to be planned by the power grid. Due to the fact that the hybrid load prediction model is utilized, an accurate and reliable load prediction result can be obtained, accurate and reliable data support is provided for prediction of the demand side resource response potential of the area to be planned for the power grid, and then it is guaranteed that a reliable data basis is provided for planning operation of a power system.

Description

Technical Field

[0001] The present invention relates to the field of low-voltage power distribution, and in particular to a demand-side resource response potential prediction method for power grid planning. Background Art

[0002] According to relevant statistical analysis, the power system produces 15% of the world's carbon emissions through the burning of fossil fuels. Reducing carbon emissions from the power system will provide a huge impetus for green and sustainable development. In order to achieve carbon emission reduction in the power system, the fundamental means is energy substitution - using large-scale renewable energy to replace fossil energy such as coal for power generation to effectively reduce carbon emissions.

[0003] However, renewable energy represented by wind and solar power will reduce the system operation performance due to its randomness, so it needs to be coordinated with the energy storage system. By setting up an energy storage system, the grid-connected rate of new energy can be increased, while optimizing the system flow distribution and supporting the system frequency and peak regulation, which can greatly improve the system operation performance. As the penetration rate of electric vehicles increases rapidly, its impact on the operation of the power grid cannot be ignored. At the same time, electric vehicles have the characteristics of large single-machine capacity, fast response speed, and certain regularity in the time of access to the system. They have great potential for flexible scheduling. Electric vehicles can be used as system dispatchable energy storage equipment to smooth out the negative impact of the randomness and volatility of renewable energy such as wind power and photovoltaic power on the system operation, thereby achieving system operation optimization, improving the utilization rate of wind power and photovoltaic power, and thus reducing the system's carbon emissions.

[0004] However, how to optimize system operation by scheduling the charging and discharging process of electric vehicles while meeting the charging needs of electric vehicles, effectively absorb renewable energy within the system, and reduce carbon emissions from system operation is an urgent problem to be solved. Summary of the invention

[0005] Based on the problems that need to be solved urgently in the prior art, the present invention provides a demand-side resource response potential prediction method for power grid planning. The demand-side resource response potential of the planned power grid (target power grid) is estimated through the rich historical power grid data of the built power grid, providing reliable data support for the planning and operation of the power system.

[0006] The method for predicting the demand-side resource response potential for power grid planning provided by the present invention comprises the following steps:

[0007] S1: Obtain the power grid data of the planned area for several consecutive days in the near future, and preprocess the acquired power grid data to obtain the corresponding preprocessed data set;

[0008] S2: Input the preprocessed data set into a hybrid load forecasting model trained based on historical power grid data of an established power grid to obtain a load forecasting result; wherein the hybrid load forecasting model includes: a Causal Transformer unit, a GRU unit, and a VAE-GRU unit;

[0009] S3: Based on the load forecast results, calculate the demand-side resource response potential of the area to be planned.

[0010] According to a specific implementation, in the demand-side resource response potential prediction method for power grid planning of the present invention, step S1 includes:

[0011] S101: Perform a preliminary screening on the acquired power grid data to obtain a preliminary screening data set; wherein the preliminary screening data set is represented by H = {H eh ,H el ,H ea ,H th ,H tl ,H ta ,H w ,H r ,H h ,H1,H2,H3,H c ,H v}, H eh represents the maximum daily load, H el Represents the minimum daily load, H ea represents the daily average load, H th Represents the maximum daily temperature, H tl Represents the daily minimum temperature, H ta represents the daily average temperature, H w Represents wind speed, H r Represents rainfall, H h represents humidity, H1 represents electricity consumption in the primary industry, H2 represents electricity consumption in the secondary industry, H3 represents electricity consumption in the tertiary industry, and H c Represents the air conditioning load, H v represents the electric vehicle charging load;

[0012] S102: Preprocessing the primary screening data set to obtain the preprocessed data set; wherein the preprocessing includes: calculating the mutual information value between the data elements in the primary screening data set, filling the data elements in the primary screening data set according to the time scale, and standardizing the data elements in the primary screening data set; the preprocessed data set is represented by D={D eh ,D el ,D ea ,D th ,D tl ,D w ,D1,D2,D3,Dc ,D v}, where D eh represents the standardized daily maximum load, D el represents the standardized daily minimum load, D ea represents the standardized daily average load, D th represents the standardized daily maximum temperature, D tl represents the standardized daily minimum temperature, D w represents the standardized wind speed, D1 represents the filled and standardized electricity consumption of the primary industry, D2 represents the filled and standardized electricity consumption of the secondary industry, D3 represents the filled and standardized electricity consumption of the tertiary industry, and D c represents the standardized air conditioning load, D v Represents the normalized electric vehicle charging load.

[0013] Furthermore, in step S102, the mutual information value between each member in the primary screening data set is calculated as follows:

[0014]

[0015] Among them, p(h ei ,h oi ) indicates H e , the joint probability function of Ho; p(h ei ),p(h oi ) indicates H e , the marginal probability function of Ho; H e ∈{H1,H2,H3,H c ,H v}, H o ∈{H eh ,H el ,H ea ,H th ,H tl ,H ta ,H w ,H r ,H h};

[0016] The data elements in the preliminary screening data set are filled in the following manner: the data elements H1, H2, and H3 in the preliminary screening data set H are copied n-1 times according to the number of days n in the month, and the data of the month are filled;

[0017] The method for standardizing the data elements in the primary screening data set is:

[0018]

[0019] Among them, H i∈H, Represents H i The mean of Represents H i The variance of , ε is taken as 1e-6.

[0020] According to a specific implementation, in the demand-side resource response potential prediction method for power grid planning of the present invention, the Causal Transformer unit aims to minimize the mean square error between the load forecast value and the load actual value, and configures the objective function as follows:

[0021]

[0022] Among them, the data set D e ∈{D eh ,D el ,D ea ,D1,D2,D3,D c ,D v}, and determine the data set D according to the mutual information value between each member in the initial screening data set e Contains data elements; D epred is the corresponding load forecast value; B represents the amount of data in a batch used for training; T represents the number of days corresponding to the input data used for prediction; o represents the data offset value. When o=1, it means that the data from the 1st to T days are used to predict the load data from the 2nd to T+1st days;

[0023] The GRU unit aims to minimize the mean square error between the load prediction value and the actual load value, and configures the objective function as:

[0024] L GRU =MSE(D e ,D epred |o=1)+MSE(D e ,D epred |o=k)

[0025] Where k represents the number of days of future load that need to be directly predicted;

[0026] The VAE-GRU unit aims to minimize the mean square error between the load decoding value and the load actual value, and minimize the mean square error between the load prediction value and the load actual value, and configures the objective function as:

[0027] L VAE =KLD(Z,N)+MSE(D,D pred )

[0028] L GRU,VAE =MSE(Z e ,Z epred|o=1)+MSE(Z e ,Z epred |o=k)

[0029] Wherein, Z represents the multivariate Gaussian distribution represented by the code corresponding to each member in the preprocessed data set D after only encoding; N represents the multivariate standard normal distribution; KLD represents the KL divergence calculation symbol; D pred represents the predicted value of the preprocessed data set D; Z e Representative data set D e The encoding set after VAE encoding; Z epred Represents Z e The predicted value of .

[0030] According to a specific implementation, in the demand-side resource response potential prediction method for power grid planning of the present invention, the hybrid load prediction model is trained based on the historical power grid data of the built power grid as follows:

[0031] Using the historical power grid data of the built power grid to obtain the corresponding pre-processed data set;

[0032] Using the obtained preprocessed data set, the CausalTransformer unit, the GRU unit and the VAE-GRU unit are trained respectively; wherein, the data set D belonging to the preprocessed data set is used e The CausalTransformer unit is trained with o=1, and the GRU unit and the VAE-GRU unit are trained with o=1 and o=k respectively.

[0033] Furthermore, the CausalTransformer unit predicts the load within the next k days in sequence in an autoregressive manner according to the input preprocessed data set; the GRU unit predicts the load within the next k days in sequence in an autoregressive manner and a direct prediction manner according to the input preprocessed data set; the VAE-GRU unit predicts the encoding of the load within the next k days in sequence in an autoregressive manner and a direct prediction manner according to the input preprocessed data set, and decodes it through a decoder to obtain the predicted load within the next k days.

[0034] According to a specific implementation, in the demand-side resource response potential prediction method for power grid planning of the present invention, in step S2, the load prediction result obtained by the hybrid load prediction model is expressed as:

[0035] R=αP GPT +β1P GRU,ag +β2P GRU +γP GRU,vae,ag +(1-α-β1-β2-γ)P GRU,vae

[0036] Among them, P GPT represents the load predicted by the Causal Transformer unit; P GRU,ag represents the load predicted by the GRU unit in an autoregressive manner; P GRU represents the load directly predicted by the GRU unit; P GRU,vae,ag represents the load predicted by the VAE-GRU unit after autoregression and decoding; P GRU,vae represents the load predicted directly by the VAE-GRU unit and after decoding; α, β1, β2, γ represent the weights of the prediction results of each unit.

[0037] According to a specific implementation, in the demand-side resource response potential prediction method for power grid planning of the present invention, in step S3, the method of calculating the demand-side resource response potential of the area to be planned is:

[0038] P total =P c +P i +P v

[0039]

[0040]

[0041]

[0042] Among them, P total is the demand-side resource response potential of the target power grid; P c is the demand response potential of air conditioning load; P i is the demand response potential of industrial load; P v is the demand response potential of electric vehicle charging load; d is the responsiveness parameter of demand response, which is (1, 2, 3) from low to high according to the responsiveness requirement; T cr is the peak shaving period set; N r T cr Length; P c1,max,t is the maximum heating load during the heating period; P c1,min,t is the minimum heating load during the heating period; P c2,max,t is the maximum load during the cooling period; P c2,min,t is the minimum load during the cooling period; E i is the total annual power consumption of industrial load; T imax is the annual maximum load utilization hours of industrial load; n v is the total number of private electric vehicles in the target grid area; L is the charging area, L = {residential area, industrial area, commercial area}; r v,l is the charging ratio in the charging area; It is the average charging power of a single vehicle in different charging areas.

[0043] In this way, the demand-side resource response potential prediction method for power grid planning provided by the present invention first obtains the power grid data of the proposed power grid planning area for several consecutive days in the recent period, and preprocesses the obtained power grid data to obtain a corresponding preprocessed data set; then the preprocessed data set is input into a hybrid load prediction model trained based on the historical power grid data of the built power grid to obtain a load forecast result; wherein the hybrid load prediction model includes: Causal Transformer unit, GRU unit, VAE-GRU unit; finally, according to the load forecast result, the demand-side resource response potential of the proposed power grid planning area is calculated. Since the present invention utilizes a hybrid load forecasting model, it can obtain accurate and reliable load forecast results, thereby providing accurate and reliable data support for the prediction of the demand-side resource response potential of the proposed power grid planning area, thereby ensuring that a reliable data basis is provided for the planning and operation of the power system. Description of the drawings:

[0044] Figure 1 It is a schematic diagram of the process of the present invention;

[0045] Figure 2 This is a schematic diagram of the working principle of the hybrid load forecasting model;

[0046] Figure 3 It is a schematic diagram of the working principle of the Causal Transformer unit;

[0047] Figure 4 This is a schematic diagram of the working principle of the GRU unit;

[0048] Figure 5 Schematic diagram of the working principle of the VAE-GRU unit;

[0049] Figure 6 A comparison chart of the actual value and the predicted value of the air conditioning load by the hybrid load prediction model of the present invention;

[0050] Figure 7 A comparison chart of the actual value and the predicted value of the secondary industry load by the hybrid load forecasting model of the present invention;

[0051] Figure 8 This is a comparison chart of the actual value and predicted value of the tertiary industry load by the hybrid load forecasting model of the present invention. DETAILED DESCRIPTION

[0052] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. However, this should not be understood as the scope of the above subject matter of the present invention being limited to the following embodiments, and all technologies realized based on the content of the present invention belong to the scope of the present invention.

[0053] In one embodiment of the present invention, Figure 1 As shown, the demand-side resource response potential prediction method for power grid planning provided by the present invention includes the following steps:

[0054] S1: Obtain the power grid data of the planned area for a number of consecutive days in the recent period, and pre-process the acquired power grid data to obtain a corresponding pre-processed data set; specifically, step S1 includes:

[0055] S101: Perform a preliminary screening on the acquired power grid data to obtain a preliminary screening data set; wherein the preliminary screening data set is represented by H = {H eh ,H el ,H ea ,H th ,H tl ,H ta ,H w ,H r ,H h ,H1,H2,H3,H c ,H v}, H eh represents the maximum daily load, H el Represents the minimum daily load, H ea represents the daily average load, H th Represents the maximum daily temperature, H tl Represents the daily minimum temperature, H ta represents the daily average temperature, H w Represents wind speed, H r Represents rainfall, H h represents humidity, H1 represents electricity consumption in the primary industry, H2 represents electricity consumption in the secondary industry, H3 represents electricity consumption in the tertiary industry, and H c Represents the air conditioning load, H v represents the electric vehicle charging load;

[0056] S102: Preprocessing the primary screening data set to obtain the preprocessed data set; wherein the preprocessing includes: calculating the mutual information value between the data elements in the primary screening data set, filling the data elements in the primary screening data set according to the time scale, and standardizing the data elements in the primary screening data set; the preprocessed data set is represented by D={D eh ,D el ,D ea ,D th ,D tl ,D w ,D1,D2,D3,D c ,D v}, where D eh represents the standardized daily maximum load, D elrepresents the standardized daily minimum load, D ea represents the standardized daily average load, D th represents the standardized daily maximum temperature, D tl represents the standardized daily minimum temperature, D w represents the standardized wind speed, D1 represents the filled and standardized electricity consumption of the primary industry, D2 represents the filled and standardized electricity consumption of the secondary industry, D3 represents the filled and standardized electricity consumption of the tertiary industry, and D c represents the standardized air conditioning load, D v Represents the normalized electric vehicle charging load.

[0057] Furthermore, in step S102, the mutual information value between each member in the primary screening data set is calculated as follows:

[0058]

[0059] Among them, p(h ei ,h oi ) indicates H e , the joint probability function of Ho; p(h ei ),p(h oi ) represents the marginal probability function of He and Ho; H e ∈{H1,H2,H3,H c ,H v}, H o ∈{H eh ,H el ,H ea ,H th ,H tl ,H ta ,H w ,H r ,H h};

[0060] The data elements in the preliminary screening data set are filled in the following manner: the data elements H1, H2, and H3 in the preliminary screening data set H are copied n-1 times according to the number of days n in the month, and the data of the month are filled;

[0061] The method for standardizing the data elements in the primary screening data set is:

[0062]

[0063] Among them, H i ∈H, Represents H i The mean of Represents H i The variance of , ε is taken as 1e-6.

[0064] S2: Input the preprocessed data set into a hybrid load forecasting model trained based on historical power grid data of the built power grid to obtain a load forecasting result; wherein, Figure 2 As shown, the hybrid load forecasting model includes: CausalTransformer unit, GRU unit, VAE-GRU unit; at the same time, the load forecasting result obtained by the hybrid load forecasting model is expressed as:

[0065] R=αP GPT +β1P GRU,ag +β2P GRU +γP GRU,vae,ag +(1-α-β1-β2-γ)P GRU,vae

[0066] Among them, P GPT represents the load predicted by the Causal Transformer unit; P GRU,ag represents the load predicted by the GRU unit in an autoregressive manner; P GRU represents the load directly predicted by the GRU unit; P GRU,vae,ag represents the load predicted by the VAE-GRU unit after autoregression and decoding; P GRU,vae represents the load predicted directly by the VAE-GRU unit and after decoding; α, β1, β2, γ represent the weights of the prediction results of each unit. In implementation, α, β1, β2, γ are 0.4, 0.1, 0.4, 0.05 respectively.

[0067] Specifically, the CausalTransformer unit aims to minimize the mean square error between the load prediction value and the actual load value, and configures the objective function as:

[0068]

[0069] Among them, the data set D e ∈{D eh ,D el ,D ea ,D1,D2,D3,D c ,D v}, and determine the data set D according to the mutual information value between each member in the initial screening data set H e The data elements included, that is, the greater the mutual information value between the members in the preliminary screening data set H, the higher the correlation. In this way, by screening other data with high correlation with the data to be predicted as input data to predict the data to be predicted, the interference of redundant input can be removed, the learning difficulty of the neural network can be reduced, and the prediction accuracy can be improved; D epredis the corresponding load forecast value; B represents the amount of data in a batch used for training; T represents the number of days corresponding to the input data used for prediction; o represents the data offset value. When o=1, it means that the data from the 1st to T days are used to predict the load data from the 2nd to T+1st days;

[0070] The GRU unit aims to minimize the mean square error between the load prediction value and the actual load value, and configures the objective function as:

[0071] L GRU =MSE(D e ,D epred |o=1)+MSE(D e ,D epred |o=k)

[0072] Where k represents the number of days of future load that need to be directly predicted;

[0073] The VAE-GRU unit aims to minimize the mean square error between the load decoding value and the load actual value, and minimize the mean square error between the load prediction value and the load actual value, and configures the objective function as:

[0074] L VAE =KLD(Z,N)+MSE(D,D pred )

[0075] L GRU,VAE =MSE(Z e ,Z epred |o=1)+MSE(Z e ,Z epred |o=k)

[0076] Wherein, Z represents the multivariate Gaussian distribution represented by the code corresponding to each member in the preprocessed data set D after only encoding; N represents the multivariate standard normal distribution; KLD represents the KL divergence calculation symbol; D pred represents the predicted value of the preprocessed data set D; Z e Representative data set D e The encoding set after VAE encoding; Z epred Represents Z e The predicted value of .

[0077] After determining the objective functions of the CausalTransformer unit, the GRU unit, and the VAE-GRU unit, the hybrid load forecasting model is trained based on the historical power grid data of the built power grid. The specific training method is:

[0078] Firstly, the corresponding preprocessed data set is obtained by using the historical grid data of the built grid; among which, the historical grid data of the built grid is preferably obtained from areas with similar economic development and electricity consumption structure to the planned grid planning area.

[0079] Then, the obtained preprocessed data set is used to train the Causal Transformer unit, the GRU unit, and the VAE-GRU unit respectively; wherein, the data set D belonging to the preprocessed data set is used e The CausalTransformer unit is trained with o=1, and the GRU unit and the VAE-GRU unit are trained with o=1 and o=k respectively.

[0080] After the hybrid load forecasting model is trained, when forecasting is performed based on the input preprocessed data set, such as Figure 3 As shown, the Causal Transformer unit predicts the load in the next k days in sequence in an autoregressive manner according to the input preprocessed data set; Figure 4 As shown, the GRU unit predicts the load in the next k days in turn by autoregression and direct prediction according to the input preprocessed data set; Figure 5 As shown, the VAE-GRU unit predicts the encoding of the load within the next k days in turn by autoregression and direct prediction according to the input preprocessed data set, and decodes it through the decoder to obtain the predicted load within the next k days.

[0081] S3: According to the load forecast result, the demand-side resource response potential of the area to be planned is calculated; specifically, in step S3, the method of calculating the demand-side resource response potential of the area to be planned is:

[0082] P total =P c +P i +P v

[0083]

[0084]

[0085]

[0086] Among them, P total is the demand-side resource response potential of the target power grid; P c is the demand response potential of air conditioning load; P i is the demand response potential of industrial load; P vis the demand response potential of electric vehicle charging load; d is the responsiveness parameter of demand response, which is (1, 2, 3) from low to high according to the responsiveness requirement; T cr is the peak shaving period set; N r T cr Length; P c1,max,t is the maximum heating load during the heating period; P c1,min,t is the minimum heating load during the heating period; P c2,max,t is the maximum load during the cooling period; P c2,min,t is the minimum load during the cooling period; E i is the total annual power consumption of industrial load; T imax is the annual maximum load utilization hours of industrial load; n v is the total number of private electric vehicles in the target grid area; L is the charging area, L = {residential area, industrial area, commercial area}; r v,l is the charging ratio in the charging area; It is the average charging power of a single vehicle in different charging areas.

[0087] In order to further verify the effectiveness of the demand-side resource response potential prediction method for power grid planning provided by the present invention, training, prediction and demand response potential evaluation experiments were carried out based on the power load data of the power grid in A from 2019 to 2022 and the local meteorological data; the evaluation experiment was carried out based on Python and Tensorflow 2.0 framework, and the computer configuration used was Inter Core i5-9300H CPU@2.40GHz and 1 NVIDIA GTX 1660Ti GPU.

[0088] Finally, we get Figure 6 , Figure 7 and Figure 8 The prediction results of the hybrid load forecasting model designed by the present invention for part of the power load in the medium-term future are shown respectively. It can be seen that the prediction error is relatively small, and accurate and reliable load forecasting results can be obtained, thereby providing accurate and reliable data support for the prediction of the demand-side resource response potential in the proposed power grid planning area, and further ensuring that a reliable data basis is provided for the planning and operation of the power system.

[0089] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A demand-side resource response potential prediction method for power grid planning, characterized in that: The following steps are involved: S1: Obtain the power grid data of the proposed power grid planning area for several consecutive days in the near future, and preprocess the acquired power grid data to obtain the corresponding preprocessed data set; S2: Input the preprocessed data set into a hybrid load forecasting model trained based on historical power grid data of an established power grid to obtain a load forecasting result; wherein the hybrid load forecasting model includes: a Causal Transformer unit, a GRU unit, and a VAE-GRU unit; S3: Based on the load forecast results, the demand-side resource response potential of the proposed power grid planning area is calculated.

2. The method for predicting demand-side resource response potential for power grid planning according to claim 1, characterized in that: Step S1 includes: S101: Perform a preliminary screening on the acquired power grid data to obtain a preliminary screening data set; wherein the preliminary screening data set is represented by H = {H eh ,H el ,H ea ,H th ,H tl ,H ta ,H w ,H r ,H h ,H1,H2,H3,H c ,H v }, H eh represents the maximum daily load, H el Represents the minimum daily load, H ea represents the daily average load, H th Represents the maximum daily temperature, H tl Represents the daily minimum temperature, H ta represents the daily average temperature, H w Represents wind speed, H r Represents rainfall, H h represents humidity, H1 represents electricity consumption in the primary industry, H2 represents electricity consumption in the secondary industry, H3 represents electricity consumption in the tertiary industry, and H c Represents the air conditioning load, H v represents the electric vehicle charging load; S102: Preprocessing the primary screening data set to obtain the preprocessed data set; wherein the preprocessing includes: calculating the mutual information value between the data elements in the primary screening data set, filling the data elements in the primary screening data set according to the time scale, and standardizing the data elements in the primary screening data set; the preprocessed data set is represented by D={D eh ,D el ,D ea ,D th ,D tl ,D w ,D1,D2,D3,D c ,D v }, where D eh represents the standardized daily maximum load, D el represents the standardized daily minimum load, D ea represents the standardized daily average load, D th represents the standardized daily maximum temperature, D tl represents the standardized daily minimum temperature, D w represents the standardized wind speed, D1 represents the filled and standardized electricity consumption of the primary industry, D2 represents the filled and standardized electricity consumption of the secondary industry, D3 represents the filled and standardized electricity consumption of the tertiary industry, and D c represents the standardized air conditioning load, D v Represents the normalized electric vehicle charging load.

3. The method for predicting demand-side resource response potential for power grid planning according to claim 2, characterized in that: In step S102, the mutual information value between each member in the primary screening data set is calculated as follows: Among them, p(h ei ,h oi ) represents the joint probability function of He and Ho; p(h ei ),p(h oi ) indicates H e , the marginal probability function of Ho; H e ∈{H1,H2,H3,H c ,H v }, H o ∈{H eh ,H el ,H ea ,H th ,H tl ,H ta ,H w ,H r ,H h }; The data elements in the preliminary screening data set are filled in the following manner: the data elements H1, H2, and H3 in the preliminary screening data set H are copied n-1 times according to the number of days n in the month, and the data of the month are filled; The method for standardizing the data elements in the primary screening data set is: Among them, H i ∈H, Represents H i The mean of Represents H i The variance of ε is 10 -6 .

4. The method for predicting demand-side resource response potential for power grid planning according to claim 3, characterized in that: The Causal Transformer unit aims to minimize the mean square error between the load forecast value and the actual load value, and configures the objective function as: Among them, the data set D e ∈{D eh ,D el ,D ea ,D1,D2,D3,D c ,D v }, and determine the data set D according to the mutual information value between each member in the preliminary screening data set e Contains data elements; D epred is the corresponding load forecast value; B represents the amount of data in a batch used for training; T represents the number of days corresponding to the input data used for prediction; o represents the data offset value. When o=1, it means that the data from the 1st to T days are used to predict the load data from the 2nd to T+1st days; The GRU unit aims to minimize the mean square error between the load prediction value and the actual load value, and configures the objective function as: L GRU =MSE(D e ,D epred |o=1)+MSE(D e ,D epred |o=k) Where k represents the number of days of future load that need to be directly predicted; The VAE-GRU unit aims to minimize the mean square error between the load decoding value and the load actual value, and minimize the mean square error between the load prediction value and the load actual value, and configures the objective function as: L VAE =KLD(Z,N)+MSE(D,D pred ) L GRU,VAE =MSE(Z e ,Z epred |o=1)+MSE(Z e ,Z epred |o=k) Wherein, Z represents the multivariate Gaussian distribution represented by the code corresponding to each member in the preprocessed data set D after only encoding; N represents the multivariate standard normal distribution; KLD represents the KL divergence calculation symbol; D pred represents the predicted value of the preprocessed data set D; Z e Representative data set D e The encoding set after VAE encoding; Z epred Represents Z e The predicted value of .

5. The method for predicting demand-side resource response potential for power grid planning according to claim 4, characterized in that: The hybrid load forecasting model is trained based on the historical power grid data of the built power grid as follows: Using the historical power grid data of the built power grid to obtain the corresponding pre-processed data set; Using the obtained preprocessed data set, the CausalTransformer unit, the GRU unit and the VAE-GRU unit are trained respectively; wherein, the data set D belonging to the preprocessed data set is used e The CausalTransformer unit is trained with o=1, and the GRU unit and the VAE-GRU unit are trained with o=1 and o=k respectively.

6. The method for predicting demand-side resource response potential for power grid planning according to claim 5, characterized in that: The CausalTransformer unit predicts the load within the next k days in sequence in an autoregressive manner based on the input preprocessed data set; the GRU unit predicts the load within the next k days in sequence in an autoregressive manner and a direct prediction manner based on the input preprocessed data set; the VAE-GRU unit predicts the encoding of the load within the next k days in sequence in an autoregressive manner and a direct prediction manner based on the input preprocessed data set, and decodes it through a decoder to obtain the predicted load within the next k days.

7. The method for predicting demand-side resource response potential for power grid planning according to claim 6, characterized in that: In step S2, the load forecasting result obtained by the hybrid load forecasting model is expressed as: R=αP GPT +β1P GRU,ag +β2P GRU +γP GRU,vae,ag +(1-α-β1-β2-γ)P GRU,vae Among them, P GPT represents the load predicted by the Causal Transformer unit; P GRU,ag represents the load predicted by the GRU unit in an autoregressive manner; P GRU represents the load directly predicted by the GRU unit; P GRU,vae,ag represents the load predicted by the VAE-GRU unit after autoregression and decoding; P GRU,vae represents the load predicted directly by the VAE-GRU unit and after decoding; α, β1, β2, γ represent the weights of the prediction results of each unit.

8. The method for predicting demand-side resource response potential for power grid planning according to any one of claims 1 to 7, characterized in that: In step S3, the method for calculating the demand-side resource response potential in the proposed power grid planning area is: P total =P c +P i +P v Among them, P total is the demand-side resource response potential of the target power grid; P c is the demand response potential of air conditioning load; P i is the demand response potential of industrial load; P v is the demand response potential of electric vehicle charging load; d is the responsiveness parameter of demand response, which is (1, 2, 3) from low to high according to the responsiveness requirement; T cr is the peak shaving period set; N r T cr Length; P c1,max,t is the maximum heating load during the heating period; P c1,min,t is the minimum heating load during the heating period; P c2,max,t is the maximum load during the cooling period; P c2,min,t is the minimum load during the cooling period; E i is the total annual power consumption of industrial load; T imax is the annual maximum load utilization hours of industrial load; n v is the total number of private electric vehicles in the target grid area; L is the charging area, L = {residential area, industrial area, commercial area}; r v,l is the charging ratio in the charging area; It is the average charging power of a single vehicle in different charging areas.

Images