A load forecasting method and system for a power distribution district

Abstract

The present application relates to the technical field of load prediction, and particularly relates to a load prediction method and system for power distribution areas, which comprises the following steps: collecting population signaling data and performing anonymous preprocessing on the collected data to obtain anonymized regional population data and an anonymized inter-grid population flow matrix; the population signaling data comprises regional population count and a cross-grid moving event sequence set; a mapping relationship between a grid g and an area a of a power distribution network is established in combination with GIS data and grid positions of the power distribution network, and an estimated population flow amount between areas is calculated; a CNN-LSTM-Attention-GNN population prediction model is established, and area population data, weather data, date data and the estimated population flow amount between areas are collected to input the population prediction model to obtain a population prediction value at a future time; and a load estimation model is constructed according to the population prediction value at the future time and a per capita load to realize area-level load prediction. The present application improves the precision of load prediction.

Description

Technical Field

[0001] This invention relates to the field of load forecasting technology, and in particular to a load forecasting method and system for distribution substations. Background Technology

[0002] As a crucial aspect of power system planning and operation, distribution network load forecasting is increasingly important due to the rapid development of smart grids and big data technologies. Improving the accuracy of load forecasting has become a pressing need. For example, some scholars have proposed forecasting models combining meteorological and historical load data, while others have attempted to incorporate socioeconomic indicators into the forecasting framework. However, none of these approaches have considered the impact of population mobility on the accuracy of load forecasting. Current technologies, such as single-layer K-anonymization, only perform simple fuzzing on population signaling data without perturbing spatial coordinates or dynamic grids. This results in insufficient privacy protection in densely populated or highly variable areas, impacts data availability, and introduces biases in the forecasting models based on population characteristics.

[0003] Relying on the initial substation boundaries and base station coverage to construct a fixed mapping relationship cannot reflect the actual population flow and spatial dynamic changes caused by substation boundary adjustments. This leads to a disconnect between the population correlation between substations and the actual population distribution, resulting in a significant decrease in prediction accuracy.

[0004] ARIMA models cannot simultaneously handle multivariate nonlinear relationships and cannot learn the complex interactions between population, weather, and calendar effects. Secondly, and more importantly, each model for a given area operates independently, completely ignoring the spatial network connections between areas formed by population flows.

[0005] Therefore, a load forecasting method and system for distribution radio areas is needed. Summary of the Invention

[0006] To address the problems in existing technologies, this invention provides a load forecasting method and system for distribution transformer areas. The specific technical solution is as follows: A load forecasting method for distribution radio areas includes the following steps: Step S1: Collect population signaling data and perform anonymization preprocessing on the collected data to obtain anonymized regional population data. R(t,g) and anonymized inter-grid population flow matrices; population signaling data includes regional population counts. S(t, g) and a set of cross-grid movement event sequences, where t Indicates time, g Indicates the geographic raster number; Step S2: Combine the GIS data and grid locations of the distribution network to establish the mapping relationship between grid g and transformer area a, and calculate the estimated population flow between transformer areas. Step S3: Establish a CNN-LSTM-Attention-GNN population prediction model. Collect population data, weather data, date data, and estimated population flow between substations and input them into the population prediction model to obtain the population prediction value for future time. Step S4: Construct a load estimation model based on the predicted population and per capita load at future times to achieve load forecasting at the transformer substation level.

[0007] Preferably, in step S1, population signaling data is collected and the collected data is pre-processed for anonymization to obtain anonymized regional population data. R(t,g) The anonymized inter-grid population flow matrix specifically includes the following steps: (1) is a regional population count. S(t,g) By applying differential privacy noise, the region-level population count after noise application is obtained, as follows: ; in, This indicates that the mean is 0 and the variance is 0. Gaussian noise, with noise variance determined by a preset privacy budget. And sensitivity determination; (2) Apply differential privacy noise to the set of cross-grid movement event sequences, as follows: Statistical generation of the original inter-grid flow counting matrix M t Elements of the inter-grid flow counting matrix M t (i,j) represents the raster. i Move to grid j The number of events for each element M t Add noise to (i,j) to obtain the elements of the inter-grid flow counting matrix after adding noise, as follows: ; (3) Apply K-anonymization to each grid cell by setting a population threshold. K pop and population mobility event threshold K flow The elements of the noise-added regional population count and the noise-added inter-grid flow count matrix are filtered to obtain the filtered regional population count and inter-grid flow count matrix, as follows: ; ; (4) Introduce spatial dithering to change the original x-coordinate of the raster. y-axis The x-coordinates of the perturbed raster are generated by Gaussian perturbation. y-axis The details are as follows: ; in, For spatial disturbance; (5) Based on the filtered regional population count and inter-grid flow count matrix, and the corresponding perturbed grids, anonymized regional population data is generated. R(t,g) and anonymized inter-grid population flow matrix .

[0008] Preferably, step S2 specifically includes the following steps: (1) The population data of the raster and the inter-raster population flow matrix are allocated to the station area by a weighted average method, as follows: ; Where G is the set of processed geographic rasters, This indicates that the grid g represents the station area. a The weights; (2) Perform double-weighted aggregation on the anonymized inter-grid population flow matrix to calculate the estimated population flow between grids. F t (a, b) represents the number of people moving from district a to district b, as detailed below: ; in, g i and g j Representing grids i and grid j .

[0009] Preferably, step S3 specifically includes the following steps: Step S31: Collect population data, weather data, and date data for the transformer area. The weather data includes temperature and humidity, and the date data includes weekend, summer, and holiday labels. Construct the input tensor data using a sliding window. , N For the number of districts, T The length of the time series. D The characteristic number; Step S32: Establish a convolutional neural network module to process the multi-feature time series matrix of each transformer area. A convolutional neural network module is used to capture the local correlations between features, resulting in the input features processed by the convolutional neural network module. Step S33: Establish an LSTM network by inputting the input features processed by the convolutional neural network module into the LSTM network to establish long-term dependencies, and obtain the output data of the LSTM network at time t. Step S34: Further enhance the model's ability to capture key information through an attention mechanism, calculate the attention coefficients of the LSTM network output data at time t, and the corresponding time... t Attention weights; Step S35, according to the corresponding time t The attention weights are weighted and summed on the hidden states of the LSTM network to obtain the context vector of the corresponding station area; Step S36: Establish a graph neural network module, using the context vector of the corresponding transformer area as the initial node feature of the graph neural network. The connection relationship between transformer areas is defined by the estimated population flow between transformer areas, and finally obtain the population prediction value at future time.

[0010] Preferably, the convolutional neural network module in step S32 includes two layers of convolutional operations and pooling operations, wherein the convolutional layer uses 3... 3 convolutional kernels, pooling layers using 2 A convolution kernel of 2.

[0011] Preferably, the graph neural network module in step S36 includes two graph convolutional layers and a fully connected layer.

[0012] Preferably, step S4 specifically includes the following steps: Step S41: Statistically analyze historical station data and calculate time. t per capita load The details are as follows: ; in L a (t) For calculating time t The load of transformer area a R a (t) For calculating time t Population count for district a; Step S42: Smooth the per capita load series using a time-weighted sliding window to form a steady-state per capita load. Its definition is as follows: ; Where K represents the smoothing window length, Indicates the time decay weight; Step S43, Construct the predicted time. The temperature correction factor is as follows: in For the predicted time temperature For reference temperature, This refers to the temperature sensitivity coefficient. Step S44: Based on the population projections for future times, obtain the district-level revised projected load, as follows: ; in Indicates the predicted time The predicted load value for transformer area a, Indicates the predicted time Population projections for district A.

[0013] A load forecasting system for distribution transformer areas, employing the method described above, includes: The data acquisition and processing module is used to collect population signaling data and perform anonymization preprocessing on the collected data to obtain anonymized regional population data. R(t,g) and anonymized inter-grid population flow matrices; population signaling data includes regional population counts. S(t,g) and a set of cross-grid movement event sequences, where t Indicates time, g Indicates the geographic raster number; The population flow metering and calculation module is used to combine the GIS data and grid locations of the distribution network to establish the mapping relationship from grid g to transformer area a, and calculate the estimated population flow between transformer areas. The population prediction module is used to build a CNN-LSTM-Attention-GNN population prediction model. It collects population data, weather data, date data, and estimated population flow between substations and inputs them into the population prediction model to obtain the population prediction value at future time. The load forecasting module is used to build a load estimation model based on the population forecast and per capita load at future times, so as to realize the load forecast at the transformer substation level.

[0014] A computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device where the computer-readable storage medium is located to perform the load forecasting method for distribution radio areas.

[0015] A processor for running a program, wherein the program executes the load forecasting method for distribution radio areas.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention achieves raster generation through multi-layer anonymization and realizes privacy protection and effective utilization of regional signaling data through joint processing of differential privacy, K-anonymity, and spatial jitter. Based on a GIS-based transformer substation mapping mechanism, when mapping raster signaling to substations, weighted averaging and version control are introduced to ensure the traceability of the mapping and the consistency of historical data. By establishing a CNN-LSTM-attention-GNN population prediction model, CNN captures local temporal dependencies, LSTM models long-term sequence features, and the attention mechanism strengthens the capture of key spatiotemporal features. Simultaneously, a graph neural network (GNN) is used to model the dynamic population flow relationship between substations, achieving load forecasting and improving the accuracy of load forecasting. Attached Figure Description

[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0018] Figure 1 This is a flowchart of the method of the present invention.

[0019] Figure 2 This is a system schematic diagram of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0022] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0023] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0024] Example 1: like Figure 1 As shown, this embodiment provides a load forecasting method for distribution radio areas, including the following steps: Step S1: Collect population signaling data through a dedicated APN line and perform anonymization preprocessing on the collected data to obtain anonymized regional population data. R(t,g) and anonymized inter-grid population flow matrices; population signaling data includes regional population counts. S(t,g) and a set of cross-grid movement event sequences, where t Indicates time, g This represents the geographic raster number. To protect user privacy, the raw population data undergoes multi-layer anonymization at the edge. First, differential privacy noise is applied to both types of data to ensure that privacy is satisfied. Differential privacy constraints.

[0025] In step S1, population signaling data is collected and the collected data is anonymized and preprocessed to obtain anonymized regional population data. R(t,g) The anonymized inter-grid population flow matrix specifically includes the following steps: (1) is a regional population count. S(t,g) By applying differential privacy noise, the region-level population count after noise application is obtained, as follows: ; in, This indicates that the mean is 0 and the variance is 0. Gaussian noise, with noise variance determined by a preset privacy budget. And the sensitivity (set to 1) is determined; (2) Apply differential privacy noise to the set of cross-grid movement event sequences, as follows: Statistical generation of the original inter-grid flow counting matrix M t Elements of the inter-grid flow counting matrix M t (i,j) represents the raster. i Move to grid j The number of events for each element M t Add noise to (i,j) to obtain the elements of the inter-grid flow counting matrix after adding noise, as follows: ; (3) Apply K-anonymization to each grid cell by setting a population threshold. K pop and population mobility event threshold K flow The elements of the noise-added regional population count and the noise-added inter-grid flow count matrix are filtered to obtain the filtered regional population count and inter-grid flow count matrix, as follows: ; ; (4) In order to further obscure the spatial information, spatial jitter is introduced, and the original horizontal coordinate of the raster is changed. y-axis The x-coordinates of the perturbed raster are generated by Gaussian perturbation. y-axis The details are as follows: ; in, For spatial disturbance; (5) Based on the filtered regional population count and inter-grid flow count matrix, and the corresponding perturbed grids, anonymized regional population data is generated. R(t,g) and anonymized inter-grid population flow matrix .

[0026] This invention enhances privacy protection through multi-layer anonymization, spatial perturbation, and a dynamic adaptive grid mechanism, while ensuring the availability of population signaling data in high-density, high-variability scenarios. By employing various anonymization processes, it achieves population data that balances privacy protection and high availability, thereby improving the quality of prediction inputs and overall accuracy.

[0027] Step S2 involves establishing a mapping relationship between grid g and transformer substation a, based on the GIS data and grid locations of the distribution network, and calculating the estimated population flow between substations. This includes the following steps: (1) The population data of the raster and the inter-raster population flow matrix are allocated to the station area by a weighted average method, as follows: ; Where G is the set of processed geographic rasters, This indicates that the grid g represents the station area. a The weights; (3) Perform double-weighted aggregation on the anonymized inter-grid population flow matrix to calculate the estimated population flow between grids. F t (a, b) represents the number of people moving from district a to district b, as detailed below: ; in, g i and g j Representing grids i and grid j .

[0028] To address changes in the location of distribution transformers or grid grids, the system introduces a caliber version management system. v t This means that each update to a transformer area or grid location generates a new version number, ensuring the traceability of the mapping relationship and the consistency of historical data. This invention establishes a traceable spatial mapping management mechanism to solve the problem of data inconsistencies with historical standards caused by changes in communication grids and distribution network transformer areas. The system generates and maintains a unique version identifier for each change in spatial mapping relationship, ensuring that data processing at any time period can call the correct mapping rules within its effective period, thereby achieving long-term data accuracy.

[0029] This invention proposes a weighted average mapping based on weights and establishes a caliber version management mechanism to accurately track and adapt to changes in the geographical location of grids or transformer areas, ensuring the availability and accuracy of data.

[0030] Step S3: Establish a CNN-LSTM-Attention-GNN population prediction model. Collect population data, weather data, date data, and estimated population flow between substations and input them into the population prediction model to obtain the population prediction value for future time periods.

[0031] After obtaining the mapped signaling data for the transformer substations and cross-grid population flow data, this invention constructs spatiotemporal samples using a sliding time window and inputs them into a deep graph spatiotemporal regression model combining a convolutional neural network (CNN), a long short-term memory network (LSTM), an attention mechanism, and a graph neural network (GNN) to achieve population prediction at the transformer substation level. The CNN module excels at extracting local key patterns from multi-feature time series; the LSTM module captures the long-term temporal dependence of population dynamics; the attention mechanism dynamically strengthens the model's perception weights for key time nodes, while the GNN captures the correlation between different transformer substations. Specifically, the following steps are included: Step S31: Collect population data, weather data, and date data for the transformer area. The weather data includes temperature and humidity, and the date data includes weekend, summer, and holiday labels. Construct the input tensor data using a sliding window. , N For the number of districts, T The length of the time series. D It is the characteristic number.

[0032] Step S32: Establish a convolutional neural network (CNN) module to process the multi-feature time series matrix of each station area. A convolutional neural network (CNN) module is used to capture local correlations between features, resulting in the processed input features. The CNN module includes two convolutional layers and pooling operations, with the convolutional layers using 3x3 convolutional layers. 3 convolutional kernels, pooling layers using 2 A 2x2 convolution kernel. The convolution formula for a CNN is as follows: ; in, W For convolution kernel, σ For activation function, X This represents the input to the convolution. b Indicates bias. H This represents the output of the convolution.

[0033] Step S33: Establish an LSTM network. Input the input features processed by the convolutional neural network module into the LSTM network to establish long-term dependencies, and obtain the output data of the LSTM network at time t. The LSTM network contains 3 layers. The first layer uses 128 hidden units, and the second and third layers each contain 64 hidden units.

[0034] Below is the specific formula for the LSTM neural network: ; ; ; ; ; ; in, x t , h t-1 They are t Input time and t The hidden state at time -1 f t ，i t ，o t ，h t , and c t They represent t The states at each time step are: forget gate state, input gate state, output gate state, hidden state, candidate cell state, and storage cell state. σ represents the activation function, and ⊙ represents element-wise multiplication.W and b These are the weights and biases of the corresponding state matrix.

[0035] Step S34: Further enhance the model's ability to capture key information through an attention mechanism, calculate the attention coefficients of the LSTM network output data at time t, and the corresponding time... t Attention weights. The specific steps are as follows: (1) Further enhance the model's ability to capture key information through an attention mechanism. For each time step, the attention coefficient... e t The calculation method is as follows: ; in, h t It is a moment t The output data of the LSTM network, W It's weight. b It is a bias.

[0036] (2) Next, the attention coefficients are normalized using the softmax function to obtain the attention weights.

[0037] ; in, Indicates time t Attention weights e t Indicates time t Attention coefficient T This represents the length of the time series.

[0038] Step S35, according to the corresponding time t The attention weights are weighted and summed over the hidden states of the LSTM network to obtain the context vector c for the corresponding station area. Specifically: ; This attention mechanism helps the model automatically focus on the factors that have the greatest impact on population change from a large amount of data.

[0039] Step S36: Establish a graph neural network module. To model the spatial relationships between transformer substations, a graph neural network (GNN) is introduced. Each transformer substation is considered a node in a graph, and the context vector of the corresponding substation is used as the initial node feature of the graph neural network. The connection relationship between substations is defined by the estimated population flow between them, ultimately obtaining the predicted population value for future time periods. The graph neural network module includes two layers of graph convolution and fully connected layers.

[0040] The graph neural network module contains two graph convolutional layers, each with 64 hidden units. The graph convolutional operation in each layer follows the formula: ; in, A To incorporate the adjacency matrix (which is an identity matrix) of self-joins, ensure that each node retains its own characteristics during updates. D The degree matrix (diagonal matrix) has elements that are the first degree matrix. l Layer weight matrix. () is a non-linear activation function. For the first l The node feature matrix of the layer That is, the initial context feature matrix. It is the first l The output node feature matrix of the two-layer graph convolution. The final representation of each transformer area after two layers of graph convolution. h i It is a spatiotemporal vector that integrates its own temporal characteristics with its contemporaneous spatial correlation. The output of the graph neural network module is passed through a fully connected layer to obtain the final prediction: ; in, This represents the weight, and b is the bias. This represents the population at the target future time.

[0041] This invention predicts future population figures based on a CNN-LSTM-Attention-GNN model, overcoming the limitations of traditional models like ARIMA, which can only handle single linear patterns. The CNN module extracts local key patterns from the multi-feature time series of each transformer substation, the LSTM module captures long-term temporal changes in the substation population, and the attention mechanism strengthens the focus on key time points, improving the model's sensitivity to special events. The GNN module models the spatial network influence between transformer substations.

[0042] Step S4: Construct a load estimation model based on future population projections and per capita load to achieve district-level load forecasting. This includes the following steps: Step S41: Statistically analyze historical station data and calculate time. t per capita load The details are as follows: ; in L a (t) For calculating time t The load of transformer area a R a (t) For calculating time t Population count for district a; Step S42: To obtain more stable load characteristics, the per capita load sequence is smoothed by time-weighted sliding window to form a steady-state per capita load. Its definition is as follows: ; Where K represents the smoothing window length, Indicates the time decay weight; Step S43: To further improve the model's adaptability under extreme weather conditions, a temperature correction coefficient is introduced. Considering the sensitivity of loads such as air conditioning and heating to temperature changes, a prediction time is constructed. The temperature correction factor is as follows: in For the predicted time temperature For reference temperature, This refers to the temperature sensitivity coefficient. Step S44: Based on the population projections for future times, obtain the district-level revised projected load, as follows: ; in Indicates the predicted time The predicted load value for transformer area a, Indicates the predicted time Population projections for district A.

[0043] To ensure the auditability and compliance of the entire process, the system automatically generates logs containing key parameters: differential privacy parameters. K-anonymous parameters K Spatial disturbance caliber version v t Model and threshold version numbers. Logs are protected by timestamps and digital signatures to ensure traceability throughout the entire process of data processing, modeling, prediction, and early warning. .

[0044] Example 2: Based on the same inventive concept as Embodiment 1, this embodiment provides a load forecasting system for distribution substations, and the method described includes: The data acquisition and processing module is used to collect population signaling data and perform anonymization preprocessing on the collected data to obtain anonymized regional population data. R(t,g) and anonymized inter-grid population flow matrices; population signaling data includes regional population counts. S(t,g) and a set of cross-grid movement event sequences, where tIndicates time, g Indicates the geographic raster number; The population flow metering and calculation module is used to combine the GIS data and grid locations of the distribution network to establish the mapping relationship from grid g to transformer area a, and calculate the estimated population flow between transformer areas. The population prediction module is used to build a CNN-LSTM-Attention-GNN population prediction model. It collects population data, weather data, date data, and estimated population flow between substations and inputs them into the population prediction model to obtain the population prediction value at future time. The load forecasting module is used to build a load estimation model based on the population forecast and per capita load at future times, so as to realize the load forecast at the transformer substation level.

[0045] Example 3: Based on the same inventive concept as Embodiment 1, this embodiment provides a computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device where the computer-readable storage medium is located to execute the aforementioned load forecasting method for distribution radio areas.

[0046] Example 4: Based on the same inventive concept as Embodiment 1, this embodiment provides a processor for running a program, wherein the program executes the load forecasting method for distribution radio areas.

[0047] Those skilled in the art will recognize that the modules of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components of the examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the invention.

[0048] In the embodiments provided by this invention, it should be understood that the division of modules is only a logical functional division. In actual implementation, there may be other division methods, such as multiple modules can be combined into one module, one module can be split into multiple modules, or some features can be ignored.

[0049] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0050] If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0051] 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 foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A load forecasting method for distribution radio areas, characterized in that, Includes the following steps: Step S1: Collect population signaling data and perform anonymization preprocessing on the collected data to obtain anonymized regional population data. R(t,g) and anonymized inter-grid population flow matrices; population signaling data includes regional population counts. S(t,g) and a set of cross-grid movement event sequences, where t Indicates time, g Indicates the geographic raster number; Step S2: Combine the GIS data and grid locations of the distribution network to establish the mapping relationship between grid g and transformer area a, and calculate the estimated population flow between transformer areas. Step S3: Establish a CNN-LSTM-Attention-GNN population prediction model. Collect population data, weather data, date data, and estimated population flow between substations and input them into the population prediction model to obtain the population prediction value for future time. Step S4: Construct a load estimation model based on the predicted population and per capita load at future times to achieve load forecasting at the transformer substation level.

2. The load forecasting method for distribution transformer areas according to claim 1, characterized in that, In step S1, population signaling data is collected and the collected data is anonymized and preprocessed to obtain anonymized regional population data. R(t,g) The anonymized inter-grid population flow matrix specifically includes the following steps: (1) is a regional population count. S(t,g) By applying differential privacy noise, the region-level population count after noise application is obtained, as follows: ； in, This indicates that the mean is 0 and the variance is 0. Gaussian noise, with noise variance determined by a preset privacy budget. And sensitivity determination; (2) Apply differential privacy noise to the set of cross-grid movement event sequences, as follows: Statistical generation of the original inter-grid flow counting matrix M t Elements of the inter-grid flow counting matrix M t (i,j) represents the raster. i Move to grid j The number of events for each element M t Add noise to (i,j) to obtain the elements of the inter-grid flow counting matrix after adding noise, as follows: ； (3) Apply K-anonymization to each grid cell by setting a population threshold. K pop and population mobility event threshold K flow The elements of the noise-added regional population count and the noise-added inter-grid flow count matrix are filtered to obtain the filtered regional population count and inter-grid flow count matrix, as follows: ； ； (4) Introduce spatial dithering to change the original x-coordinate of the raster. y-axis The x-coordinates of the perturbed raster are generated by Gaussian perturbation. y-axis The details are as follows: ； in, For spatial disturbance; (5) Based on the filtered regional population count and inter-grid flow count matrix, and the corresponding perturbed grids, anonymized regional population data is generated. R(t,g) and anonymized inter-grid population flow matrix .

3. The load forecasting method for distribution transformer areas according to claim 1, characterized in that, Step S2 specifically includes the following steps: (1) The population data of the raster and the inter-raster population flow matrix are allocated to the station area by a weighted average method, as follows: ； Where G is the set of processed geographic rasters, This indicates that the grid g represents the station area. a The weights; (2) Perform double-weighted aggregation on the anonymized inter-grid population flow matrix to calculate the estimated population flow between grids. F t (a, b) represents the number of people moving from district a to district b, as detailed below: ； in, g i and g j Representing grids i and grid j .

4. The load forecasting method for distribution transformer areas according to claim 1, characterized in that, Step S3 specifically includes the following steps: Step S31: Collect population data, weather data, and date data for the transformer area. The weather data includes temperature and humidity, and the date data includes weekend, summer, and holiday labels. Construct the input tensor data using a sliding window. , N For the number of districts, T The length of the time series. D The characteristic number; Step S32: Establish a convolutional neural network module to process the multi-feature time series matrix of each transformer area. A convolutional neural network module is used to capture the local correlations between features, resulting in the input features processed by the convolutional neural network module. Step S33: Establish an LSTM network by inputting the input features processed by the convolutional neural network module into the LSTM network to establish long-term dependencies, and obtain the output data of the LSTM network at time t. Step S34: Further enhance the model's ability to capture key information through an attention mechanism, calculate the attention coefficients of the LSTM network output data at time t, and the corresponding time... t Attention weights; Step S35, according to the corresponding time t The attention weights are weighted and summed on the hidden states of the LSTM network to obtain the context vector of the corresponding station area; Step S36: Establish a graph neural network module, using the context vector of the corresponding transformer area as the initial node feature of the graph neural network. The connection relationship between transformer areas is defined by the estimated population flow between transformer areas, and finally obtain the population prediction value at future time.

5. A load forecasting method for distribution transformer areas according to claim 4, characterized in that, The convolutional neural network module in step S32 includes two layers of convolutional operations and pooling operations, wherein the convolutional layer uses 3... 3 convolutional kernels, pooling layers using 2 A convolution kernel of 2.

6. The load forecasting method for distribution transformer areas according to claim 1, characterized in that, The graph neural network module in step S36 includes two graph convolutional layers and a fully connected layer.

7. The load forecasting method for distribution transformer areas according to claim 1, characterized in that, Step S4 specifically includes the following steps: Step S41: Statistically analyze historical station data and calculate time. t per capita load The details are as follows: ； in L a (t) For calculating time t The load of transformer area a R a (t) For calculating time t Population count corresponding to area a; Step S42: Smooth the per capita load series using a time-weighted sliding window to form a steady-state per capita load. Its definition is as follows: ； Where K represents the smoothing window length, Indicates the time decay weight; Step S43, Construct the predicted time. The temperature correction factor is as follows: in For the predicted time temperature For reference temperature, This refers to the temperature sensitivity coefficient. Step S44: Based on the population projections for future times, obtain the district-level revised projected load, as follows: ； in Indicates the predicted time The predicted load value for transformer area a, Indicates the predicted time Population projections for district A.

8. A load forecasting system for distribution transformer areas, characterized in that, The method described by any one of claims 1 to 7 includes: The data acquisition and processing module is used to collect population signaling data and perform anonymization preprocessing on the collected data to obtain anonymized regional population data. R(t,g) and anonymized inter-grid population flow matrices; population signaling data includes regional population counts. S(t,g) and a set of cross-grid movement event sequences, where t Indicates time, g Indicates the geographic raster number; The population flow metering and calculation module is used to combine the GIS data and grid locations of the distribution network to establish the mapping relationship from grid g to transformer area a, and calculate the estimated population flow between transformer areas. The population prediction module is used to build a CNN-LSTM-Attention-GNN population prediction model. It collects population data, weather data, date data, and estimated population flow between substations and inputs them into the population prediction model to obtain the population prediction value at future time. The load forecasting module is used to build a load estimation model based on the population forecast and per capita load at future times, so as to realize the load forecast at the transformer substation level.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform a load forecasting method for a distribution substation as described in any one of claims 1 to 7.

10. A processor, characterized in that, The processor is used to run a program, wherein the program executes a load forecasting method for distribution radio areas as described in any one of claims 1 to 7.

Images