Source network load interaction type user energy consumption space-time characteristic analysis method adaptive to differentiated scene
By constructing a quantitative system of multiple influencing factors and a CNN-LSTM fusion model, the prediction error and adaptability problems in the analysis of the spatiotemporal characteristics of energy consumption of users in medium and low voltage distribution networks were solved, achieving high-precision prediction of user energy consumption trends and improving the stability and planning accuracy of the distribution network.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ELECTRIC POWER RES INST STATE GRID SHANXI ELECTRIC POWER
- Filing Date
- 2025-12-31
- Publication Date
- 2026-05-12
AI Technical Summary
Existing methods for analyzing the spatiotemporal characteristics of user energy consumption have problems such as large prediction errors of distributed generation (DG) output, insufficient quantification of multiple influencing factors, and low prediction accuracy in medium and low voltage distribution networks, making it difficult to adapt to the needs of differentiated scenarios.
By employing multi-source data acquisition technology, combined with principal component analysis, Spearman correlation coefficient, and binary Gauss Copula function, a quantitative system for multiple influencing factors is constructed. Based on a fusion model of convolutional neural network and long short-term memory network, user energy consumption trend is predicted.
It enables precise analysis of the spatiotemporal characteristics of energy consumption by users in medium and low voltage distribution networks, improves prediction accuracy and scenario adaptability, provides high-precision data support for distribution network planning, and enhances operational stability.
Smart Images
Figure CN122026313A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of power system and its automation technology, and relates to a method for analyzing the spatiotemporal characteristics of energy consumption of users with source-grid-load interaction, and more particularly to a method for analyzing the spatiotemporal characteristics of energy consumption of users with source-grid-load interaction that is adapted to different scenarios. Background Technology
[0002] With the advancement of "dual carbon" targets and the construction of new power systems, medium- and low-voltage distribution networks are characterized by blurred boundaries between power sources, grids, and loads, diversified user demands, and a rapid increase in the penetration rate of distributed energy. On the one hand, the output of distributed photovoltaic and wind power (DG) exhibits strong randomness and volatility, while the charging and discharging loads of electric vehicles have spatiotemporal uncertainties, leading to a widening of the peak-valley load difference in the distribution network and challenges to operational stability. On the other hand, the influence mechanisms of multiple factors such as regional geographical characteristics, economic level, population density, and electricity policies on the distribution of user electricity loads are complex, and existing technologies lack systematic quantitative analysis methods.
[0003] Existing methods for analyzing the spatiotemporal characteristics of user energy consumption have the following shortcomings: First, the time sequence and distribution of DG output data are not adequately preserved, resulting in large DG output prediction errors, which makes it difficult to support dynamic dispatching of the distribution network. Secondly, the influence of regional characteristics, social factors and other multi-dimensional factors has not been fully quantified, resulting in poor adaptability between user energy consumption trend predictions and actual scenarios. Third, most prediction models use a single neural network structure, which cannot simultaneously capture the spatial correlation and temporal dependency of data, making it difficult to meet the prediction accuracy requirements of different scenarios.
[0004] Therefore, there is an urgent need to propose a spatiotemporal characteristic analysis method for user energy consumption that can integrate multi-source data, quantify multiple influencing factors, and adapt to differentiated scenarios. This method would provide accurate data support for the optimized configuration and extension planning of user access in medium and low voltage distribution networks, thereby improving the stability of distribution network operation.
[0005] A search revealed no publicly available literature of the same or similar prior art as this invention. Summary of the Invention
[0006] To address the shortcomings of existing technologies, this invention proposes a source-grid-load interaction-based spatiotemporal characteristic analysis method for user energy consumption that is adapted to different scenarios. This method can solve the technical problems of insufficient adaptability, inadequate quantification of multiple influencing factors, and low prediction accuracy in the spatiotemporal characteristic analysis of user energy consumption in medium and low voltage distribution networks.
[0007] The above-mentioned objective of this invention is achieved through the following technical solution: A method for analyzing the spatiotemporal characteristics of energy consumption by users in source-grid-load interaction scenarios, adaptable to different scenarios, includes the following steps: Step S100: Use data acquisition technology to acquire multi-source basic data of medium and low voltage distribution networks; Step S200: Based on the multi-source basic data collected in step S100, and combined with regional characteristics, social factors and population density distribution, construct a quantitative system of multiple influencing factors. Step S300: Based on convolutional neural networks and long short-term memory networks, establish a user energy consumption trend prediction model that integrates multi-source basic data and quantitative system data; Step S400: Train the user energy consumption trend prediction model established in step S300 using the training dataset, optimize the model parameters through iterative optimization of the loss function, determine the optimal structure of the model, and obtain the trained user energy consumption trend prediction model. Step S500: Using the user energy consumption trend prediction model trained in step S400, reason about the source-grid-load interaction data under different regions, different DG penetration rates, different load types and different time scales to generate a spatiotemporal distribution feature map of user energy consumption, and complete the accurate analysis of the spatiotemporal characteristics of source-grid-load interaction type user energy consumption under differentiated scenarios.
[0008] Furthermore, the specific steps of step S100 include: Historical meteorological data, including irradiance, wind speed, temperature and humidity, are collected over a period of no less than one year, with a sampling frequency of 5 minutes per time. Collect output data of distributed power sources, including distributed photovoltaic, wind power and micro gas turbines, with a collection period of not less than 1 year and a sampling frequency of 5 minutes / time. Collect charging and discharging data of electric vehicles, including charging power, discharging power and charging and discharging time, at a frequency of 5 minutes / time; Collect user electricity load data, categorized by industrial users, commercial users, and residential users, with a collection frequency of 5 minutes per time.
[0009] Furthermore, the specific steps of step S200 include: (1) Principal component analysis was used to standardize historical meteorological data and distributed power output data, and to extract spatiotemporal core features; The standardized formula is: Where, x i Here are the original data, and μ is the data mean. Let x be the standard deviation of the data. i 'This is standardized data;' PCA solves for the eigenvalues λ using the covariance matrix. j With eigenvector e j Select the cumulative variance contribution rate The first k principal components are used as the core spatiotemporal features.
[0010] (2) The Spearman correlation coefficient was used to calculate the correlation between regional characteristics, social factors, population density distribution, electric vehicle charging and discharging data and user electricity load type distribution. The formula for Spearman's correlation coefficient is: Where di is the rank difference of the i-th sample on the two variables, and n is the number of samples.
[0011] Based on the spatiotemporal core features and correlation obtained from steps (1) and (2), the weights of each influencing factor on the user's electricity load are quantified by the binary Gauss Copula function, and a multi-influencing factor quantification system is constructed. The Gaussian Copula density function is: Where ρ is the Pearson correlation coefficient between the two variables. Let be the quantile function of the standard normal distribution, and u,v∈(0,1) be the cumulative probability of the marginal distribution.
[0012] Weights of each factor W j The result is obtained by normalizing the joint density fitted by Copula.
[0013] Moreover, the factors in the multi-influencing factor quantification system of step S200 include: irradiance, distributed power output fluctuation, population density, GDP per capita, time-of-use electricity price and smart meter coverage. Furthermore, the specific steps of step S300 include: By constructing multiple convolutional and pooling layers through a convolutional neural network, spatial correlation features of multi-source basic data and multi-influencing factor quantification system data are extracted, and spatial feature vectors are output. The convolution operation formula for the convolutional neural network is as follows: in, For the first Layer convolution kernel weights, For bias terms, is the ReLU activation function, and k is the kernel size.
[0014] By constructing multiple hidden layers through a long short-term memory network, the long and short-term temporal dependencies between multi-source basic data and quantized system data are captured, and a temporal feature vector is output. The update formula for the Long Short-Term Memory (LSTM) network is as follows: Among them, f t For the output of the forget gate, i t For input gate output, o t For output gate output, For candidate cell states, C t In cellular state, h t For the output of the hidden layer, W f W i W C W o Let b be the weight matrix. f b i b C b o As the bias term, x t Input for the current time, h t-1 This is the output of the hidden layer from the previous time step.
[0015] The spatial feature vector and the temporal feature vector are concatenated and input into a fully connected layer. An activation function is then used to construct the complete architecture of the user energy consumption trend prediction model. This completes the construction of the user energy consumption trend prediction model, which integrates multi-source basic data and quantitative system data, and outputs three types of predicted values: distributed power output. Net load of electric vehicles Total energy consumption by users .
[0016] Furthermore, the specific steps of step S400 include: The training dataset is divided into a training set, a validation set, and a test set in a 7:2:1 ratio. The mean absolute percentage error is used as the loss function, and the model parameters are iteratively updated through the backpropagation algorithm. The MAPE formula is: in, This is the actual value. is the predicted value, and n is the number of samples.
[0017] Set an initial learning rate and use an adaptive moment estimation optimizer to dynamically adjust the learning rate. When the loss function value converges to a set threshold or the number of iterations reaches a set maximum value, training is stopped, the optimal structure of the model is determined, and a well-trained user energy consumption trend prediction model is obtained. Furthermore, the initial learning rate is set to 0.001, the threshold for the loss function is set to 0.05, and the maximum number of iterations is set to 1000.
[0018] Furthermore, the specific steps of step S500 include: Using the user energy consumption trend prediction model trained in step S400, inference is performed on the source-grid-load interaction data under different regions, different DG penetration rates, different load types and different time scales, and the output of distributed power generation output prediction results, electric vehicle charging and discharging load prediction results and user energy consumption trend prediction results are output.
[0019] The advantages and beneficial effects of this invention are as follows: 1. This invention proposes a spatiotemporal characteristic analysis method for user energy consumption based on source-grid-load interaction, adaptable to differentiated scenarios. It comprehensively acquires relevant basic data of the distribution network using multi-source data acquisition technology; constructs a quantitative system of multiple influencing factors by combining principal component analysis, Spearman correlation coefficient, and binary Gaussian Copula function; and accurately captures spatial correlation and temporal dependency relationships of data based on a convolutional neural network-long short-term memory network fusion model. This ensures that the prediction accuracy of distributed power generation output error, electric vehicle charging and discharging load error, and user energy consumption trend prediction meets preset requirements, adapting to differentiated scenario needs. It can provide accurate data support for distribution network planning and has better scenario adaptability and prediction effect.
[0020] 2. This invention is based on a multi-source high-frequency data acquisition system. By synchronously acquiring 5-minute time-series data of meteorology, distributed power output, electric vehicle charging and discharging, and classified user loads, it ultimately constructs a spatiotemporal analysis input foundation covering all elements of "source-grid-load". This provides highly complete and timely data support for subsequent quantification of multiple influencing factors and model training, significantly improving the physical authenticity and scenario representativeness of user energy consumption characteristic analysis.
[0021] 3. This invention is based on a multi-influencing factor quantification system constructed using principal component analysis, Spearman correlation analysis, and Gauss Copula function. By reducing dimensionality to retain core spatiotemporal features, quantifying the correlation between regional and social factors, and modeling nonlinear joint dependencies, it ultimately generates a multi-dimensional weighted input with physical interpretability. This effectively integrates unstructured external factors and power operation data, enhancing the analytical model's adaptability to differentiated regional scenarios.
[0022] 4. This invention is based on a CNN-LSTM fusion prediction model. It extracts spatial correlation features through convolutional layers, captures long-term and short-term temporal dependencies through LSTM layers, and fuses spatiotemporal information through fully connected layers. It also combines a 7:2:1 data partitioning, the MAPE loss function, and the Adam optimizer for iterative parameter optimization. Finally, it outputs spatiotemporal characteristic analysis results of user energy consumption that are adapted to different distributed power penetration scenarios. This provides high-precision and interpretable data support for the optimized configuration and extension planning of medium and low voltage distribution network users, thereby improving the stability of distribution network operation. Attached Figure Description
[0023] Figure 1 This is an overall flowchart of the method of the present invention; Figure 2 This is a diagram of the multi-source basic data acquisition architecture for medium and low voltage power distribution networks; Figure 3 Flowchart of multi-source data acquisition for medium and low voltage power distribution networks; Figure 4 This is a flowchart for constructing a quantitative system for multiple influencing factors; Figure 5 This is a diagram of the CNN-LSTM fusion model architecture; Figure 6 This is a comparison chart of the predicted output of distributed power sources and the actual values. Figure 7 This is a comparison chart of the predicted and actual energy consumption trends of users; Detailed Implementation
[0024] The structure of the present invention will be further described below with reference to the accompanying drawings and embodiments. It should be noted that these embodiments are descriptive and not limiting.
[0025] A method for analyzing the spatiotemporal characteristics of user energy consumption in source-grid-load interaction scenarios that is adaptable to different scenarios, such as Figure 1 As shown, it includes the following steps: Step S100: Use data acquisition technology to acquire multi-source basic data of medium and low voltage distribution networks; Step S200: Based on the multi-source basic data collected in step S100, and combined with regional characteristics, social factors and population density distribution, construct a quantitative system of multiple influencing factors. Step S300: Based on convolutional neural networks and long short-term memory networks, establish a user energy consumption trend prediction model that integrates multi-source basic data and quantitative system data; Step S400: Train the user energy consumption trend prediction model established in step S300 using the training dataset, optimize the model parameters through iterative optimization of the loss function, determine the optimal structure of the model, and obtain the trained user energy consumption trend prediction model. Step S500: Using the user energy consumption trend prediction model trained in step S400, reasoning is performed on the source-grid-load interaction data under different regions, different DG penetration rates, different load types, and different time scales to generate a spatiotemporal distribution feature map of user energy consumption. This completes the accurate analysis of the spatiotemporal characteristics of source-grid-load interaction type user energy consumption under differentiated scenarios, providing scenario-adaptive data support for the planning, scheduling, and operation of medium and low voltage distribution networks.
[0026] In this embodiment, the multi-source basic data includes historical meteorological data, distributed power generation output data, electric vehicle charging and discharging data, and user electricity load data. Distributed power sources include distributed photovoltaic, wind power, and micro gas turbines. User types are categorized as industrial, commercial, and residential users. The specific data acquisition equipment is not limited; in this embodiment, a meteorological monitoring station, inverter control cabinet, charging pile management system, and smart meters are used to collect the corresponding data. The data acquisition architecture is as follows: Figure 2 As shown, from bottom to top, there are three layers: data acquisition layer, data transmission layer, and data storage layer. The data acquisition layer collects multi-source data in real time through various monitoring devices, and then transmits the data to the data storage layer through the data transmission layer to achieve centralized management and retrieval of the data, providing data support for the subsequent construction of the quantitative system and model training.
[0027] In this embodiment, in step S100, multi-source basic data of the medium and low voltage distribution network is acquired using data acquisition technology, specifically as follows: Historical meteorological data, including irradiance, wind speed, temperature and humidity, are collected over a period of no less than one year, with a sampling frequency of 5 minutes per time. Data on the output of distributed power sources, including distributed photovoltaic, wind power and micro gas turbines, are collected at a frequency consistent with historical meteorological data. Collect charging and discharging data of electric vehicles, including charging power, discharging power and charging and discharging time, at a frequency of 5 minutes / time; Collect user electricity load data, categorized by industrial users, commercial users, and residential users, with a collection frequency of 5 minutes per time.
[0028] like Figure 3 As shown in the data acquisition flowchart, each acquisition device collects corresponding data at a set frequency. After data preprocessing, the data is stored on the cloud server. Outlier removal uses the 3σ criterion, and missing value imputation uses linear interpolation to ensure data quality meets the requirements of subsequent analysis.
[0029] In this embodiment, step S200 involves constructing a multi-influencing factor quantification system based on the multi-source basic data, combined with regional characteristics, social factors, and population density distribution. This specifically includes the following steps: S201. Principal component analysis is used to reduce the dimensionality of historical meteorological data and distributed power generation data, extracting core spatiotemporal features. The proportion of original data information retained after dimensionality reduction is no less than 85%. S202. The Spearman correlation coefficient was used to calculate the correlation between regional characteristics, social factors, population density distribution, electric vehicle charging and discharging data and user electricity load type distribution. S203. Based on the aforementioned correlation and spatiotemporal core characteristics, the weights of each influencing factor on the user's electricity load are quantified using a binary Gaussian Copula function, thereby constructing a multi-influencing factor quantification system.
[0030] The regional characteristics include geographical characteristics, economic characteristics, and industrial and commercial structure characteristics; the social factors include population structure, electricity policies, and the completeness of infrastructure.
[0031] Specifically, such as Figure 4 As shown, the process of constructing a quantitative system for multiple influencing factors first involves standardizing historical meteorological data and distributed power generation data, and then extracting spatiotemporal core features through principal component analysis. Taking historical meteorological data as an example, the standardization formula is: Where, x i Here are the original data, and μ is the data mean. Let x be the standard deviation of the data. i 'This is the standardized data.'
[0032] Principal component analysis (PCA) is used to calculate the covariance matrix, solve for eigenvalues and eigenvectors, and select principal components with eigenvalues greater than 1 to ensure that the cumulative variance contribution rate is not less than 85%. In this embodiment, three principal components are extracted from historical meteorological data, with a cumulative variance contribution rate of 89.7%; two principal components are extracted from distributed power generation output data, with a cumulative variance contribution rate of 87.2%.
[0033] Principal component analysis (PCA) is used to calculate the covariance matrix, solve for eigenvalues and eigenvectors, and select principal components with eigenvalues greater than 1 to ensure that the cumulative variance contribution rate is not less than 85%. In this embodiment, three principal components are extracted from historical meteorological data, with a cumulative variance contribution rate of 89.7%; two principal components are extracted from distributed power generation output data, with a cumulative variance contribution rate of 87.2%.
[0034] The Spearman correlation coefficient was used to calculate the correlation between each influencing factor and the distribution of user electricity load types. The formula for the Spearman correlation coefficient is as follows: Where di is the rank difference of the i-th sample on the two variables, and n is the number of samples. Spearman correlation coefficient.
[0035] In this embodiment, the correlation between each influencing factor and the industrial user load is calculated as follows: irradiance 0.82, GDP per capita 0.67, population density 0.73, time-of-use electricity price 0.58, and smart meter coverage rate 0.62.
[0036] The weights of each influencing factor are quantified using a binary Gaussian Copula function. The Gaussian Copula density function is as follows: Where ρ is the Pearson correlation coefficient between the two variables. Let be the quantile function of the standard normal distribution, and u,v∈(0,1) be the cumulative probability of the marginal distribution.
[0037] In this embodiment, the weights of each influencing factor are quantified as follows: irradiance 0.23, distributed power output fluctuation 0.18, population density 0.15, GDP per capita 0.12, time-of-use electricity price 0.09, smart meter coverage 0.08, and other factors 0.15, thus completing the quantitative system of multiple influencing factors.
[0038] In this embodiment, step S300 involves establishing a user energy consumption trend prediction model based on convolutional neural networks and long short-term memory networks, which integrates multi-source basic data and quantitative system data. This specifically includes the following steps: S301. Construct multiple convolutional and pooling layers through a convolutional neural network to extract the spatial correlation features between multi-source basic data and quantized system data, and output spatial feature vectors. S302. Construct a multi-layer hidden layer through a long short-term memory network to capture the long and short-term temporal dependencies between multi-source basic data and quantized system data, and output a temporal feature vector. S303. The spatial feature vector and the temporal feature vector are concatenated and then input into a fully connected layer. The complete architecture of the user energy consumption trend prediction model is constructed by combining the activation function.
[0039] like Figure 5 As shown in the CNN-LSTM fusion model architecture diagram, the input layer dimensions are: None, 24, 12 (time step 24, feature dimension 12), the convolutional layers use 3×3 convolutional kernels with 2, 64, and 128 kernels respectively, and the pooling layers use 2×2 max pooling; the LSTM module constructs 4 hidden layers, each with 256 neurons, the activation function is tanh, and the dropout coefficient is set to 0.2; the fully connected layers include 2 fully connected layers (with 128 and 64 neurons respectively) and the output layer (3 output dimensions, corresponding to distributed power output, electric vehicle charging and discharging load, and total user energy consumption respectively), and the output layer activation function is the Linear function.
[0040] The formula for feature extraction from the convolutional layer is as follows: in, For the first Layer convolution kernel weights, For bias terms, is the ReLU activation function, and k is the kernel size.
[0041] The temporal feature capture process of the LSTM module is implemented through a gating mechanism. The formulas for the forget gate, input gate, output gate, and cell state update are as follows: Among them, f t For the output of the forget gate, i t For input gate output, o t For output gate output, For candidate cell states, C t In cellular state, h t For the output of the hidden layer, W f W i W C W o Let b be the weight matrix. f b i b C b o As the bias term, x t Input for the current time, h t-1 This is the output of the hidden layer from the previous time step.
[0042] In this embodiment, step S400 involves training the user energy consumption trend prediction model using a training dataset, iteratively optimizing the model parameters through a loss function, and determining the optimal model structure. This specifically includes the following steps: S401. Divide the training dataset into a training set, a validation set, and a test set in a ratio of 7:2:1. S402. The mean absolute percentage error is used as the loss function, and the model parameters are iteratively updated through the backpropagation algorithm. S403. Set the initial learning rate and use the adaptive moment estimation optimizer to dynamically adjust the learning rate. S404. When the loss function value converges to the set threshold or the number of iterations reaches the set maximum value, stop training and determine the optimal model structure.
[0043] In this embodiment, the initial learning rate is set to 0.001, the threshold for the loss function is set to 0.05, and the maximum number of iterations is set to 1000. The formula for Mean Absolute Percentage Error (MAPE) is: Among them, y i This is the actual value. is the predicted value, and n is the number of samples.
[0044] The parameter update formula for the Adaptive Moment Estimation (Adam) optimizer is as follows: Where, m t For the first-order moment estimate, v t For second-order moment estimation, This is the first-order moment estimate after bias correction. For the second-order moment estimate after bias correction, g t For the gradient, β1=0.9, β2=0.999, =10 -8 α is the learning rate, θ t θ is the parameter at the current time. t-1 These are the parameters from the previous time step.
[0045] During model training, the loss function values for the training and validation sets are recorded every 100 iterations. The model is considered successful if the change in the loss function value is less than 10 over 50 consecutive iterations. -4 When the model converges, it is determined that the model has converged. In this embodiment, the model converges after 800 iterations, with a training set loss function value of 0.038 and a validation set loss function value of 0.042, thus determining the optimal model structure.
[0046] In this embodiment, in step S500, the differentiated scenarios include basic power supply scenarios with a distributed power penetration rate of <15%, collaborative power supply scenarios with a distributed power penetration rate of 15%-30%, and high-proportion distributed power access scenarios with a distributed power penetration rate of >30%.
[0047] The user energy consumption trend prediction model is used to output multi-dimensional prediction results for differentiated scenarios, specifically: The output prediction results of distributed power sources are provided, with a prediction error of ≤8%. Output electric vehicle charging and discharging load prediction results, with a prediction error ≤10%; Output the user energy consumption trend prediction results with a prediction accuracy of ≥92%.
[0048] Figure 6 This graph compares the predicted output of distributed power sources with the actual output in a collaborative power supply scenario. The horizontal axis represents time, and the vertical axis represents output power. The solid line represents the actual value, and the dashed line represents the predicted value. The graph shows that the predicted results have a high degree of fit with the actual values, with a daily average prediction error of 7.3%, meeting the requirement of ≤8%.
[0049] Figure 7 This chart compares the predicted and actual energy consumption trends of users in a collaborative power supply scenario. The horizontal axis represents time, and the vertical axis represents energy consumption. The solid line represents the actual value, and the dashed line represents the predicted value. The hourly prediction accuracy is 93.5%, and the daily prediction accuracy is 95.2%, both meeting the requirement of ≥92%.
[0050] Although embodiments and drawings of the present invention have been disclosed for illustrative purposes, those skilled in the art will understand that various substitutions, variations and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the scope of the present invention is not limited to the contents disclosed in the embodiments and drawings.
Claims
1. A method for analyzing the spatiotemporal characteristics of user energy consumption in a source-grid-load interactive model adapted to differentiated scenarios, characterized in that: Includes the following steps: Step S100: Use data acquisition technology to acquire multi-source basic data of medium and low voltage distribution networks; Step S200: Based on the multi-source basic data collected in step S100, and combined with regional characteristics, social factors and population density distribution, construct a quantitative system of multiple influencing factors. Step S300: Based on convolutional neural networks and long short-term memory networks, establish a user energy consumption trend prediction model that integrates multi-source basic data and quantitative system data; Step S400: Train the user energy consumption trend prediction model established in step S300 using the training dataset, optimize the model parameters through iterative optimization of the loss function, determine the optimal structure of the model, and obtain the trained user energy consumption trend prediction model. Step S500: Using the user energy consumption trend prediction model trained in step S400, reason about the source-grid-load interaction data under different regions, different DG penetration rates, different load types and different time scales to generate a spatiotemporal distribution feature map of user energy consumption, and complete the accurate analysis of the spatiotemporal characteristics of source-grid-load interaction type user energy consumption under differentiated scenarios.
2. The method for analyzing the spatiotemporal characteristics of user energy consumption in a source-grid-load interactive manner adapted to differentiated scenarios, as described in claim 1, is characterized in that: The specific steps of step S100 include: Historical meteorological data, including irradiance, wind speed, temperature and humidity, are collected over a period of no less than one year, with a sampling frequency of 5 minutes per time. Collect output data of distributed power sources, including distributed photovoltaic, wind power and micro gas turbines, with a collection period of not less than 1 year and a sampling frequency of 5 minutes / time. Collect charging and discharging data of electric vehicles, including charging power, discharging power and charging and discharging time, at a frequency of 5 minutes / time; Collect user electricity load data, categorized by industrial users, commercial users, and residential users, with a collection frequency of 5 minutes per time.
3. The method for analyzing the spatiotemporal characteristics of user energy consumption in a source-grid-load interactive manner adapted to differentiated scenarios, as described in claim 1, is characterized in that: The specific steps of step S200 include: (1) Principal component analysis was used to standardize historical meteorological data and distributed power output data, and to extract spatiotemporal core features; The standardized formula is: ； Where, x i Here are the original data, and μ is the data mean. Let x be the standard deviation of the data. i 'This is standardized data;' PCA solves for the eigenvalues λ using the covariance matrix. j With eigenvector e j Select the cumulative variance contribution rate The first k principal components are used as the core spatiotemporal features; (2) The Spearman correlation coefficient was used to calculate the correlation between regional characteristics, social factors, population density distribution, electric vehicle charging and discharging data and user electricity load type distribution. The formula for Spearman's correlation coefficient is: ； Where di is the rank difference of the i-th sample on the two variables, and n is the number of samples; Based on the spatiotemporal core features and correlation obtained from steps (1) and (2), the weights of each influencing factor on the user's electricity load are quantified by the binary Gauss Copula function, and a multi-influencing factor quantification system is constructed. The Gaussian Copula density function is: ； Where ρ is the Pearson correlation coefficient between the two variables. Let be the quantile function of the standard normal distribution, and u,v∈(0,1) be the cumulative probability of the marginal distribution; Weights of each factor W j The result is obtained by normalizing the joint density fitted by Copula.
4. The method for analyzing the spatiotemporal characteristics of user energy consumption in a source-grid-load interactive manner adapted to differentiated scenarios, as described in claim 1, is characterized in that: The factors in the multi-influencing factor quantification system of step S200 include: irradiance, distributed power output fluctuation, population density, GDP per capita, time-of-use electricity price, and smart meter coverage.
5. The method for analyzing the spatiotemporal characteristics of user energy consumption in a source-grid-load interactive manner adapted to differentiated scenarios, as described in claim 1, is characterized in that: The specific steps of step S300 include: By constructing multiple convolutional and pooling layers through a convolutional neural network, spatial correlation features of multi-source basic data and multi-influencing factor quantification system data are extracted, and spatial feature vectors are output. The convolution operation formula for the convolutional neural network is as follows: ； in, For the first Layer convolution kernel weights, For bias terms, Here, k is the ReLU activation function, and k is the kernel size. By constructing multiple hidden layers through a long short-term memory network, the long and short-term temporal dependencies between multi-source basic data and quantized system data are captured, and a temporal feature vector is output. The update formula for the Long Short-Term Memory (LSTM) network is as follows: ； ； ； ； ； ； Among them, f t For the output of the forget gate, i t For input gate output, o t For output gate output, For candidate cell states, C t In cellular state, h t For the output of the hidden layer, W f W i W C W o Let b be the weight matrix. f b i b C b o As the bias term, x t Input for the current time, h t-1 This is the output of the hidden layer from the previous time step; The spatial feature vector and the temporal feature vector are concatenated and input into a fully connected layer. An activation function is then used to construct the complete architecture of the user energy consumption trend prediction model. This completes the construction of the user energy consumption trend prediction model, which integrates multi-source basic data and quantitative system data, and outputs three types of predicted values: distributed power output. Net load of electric vehicles Total energy consumption by users .
6. The method for analyzing the spatiotemporal characteristics of user energy consumption in a source-grid-load interactive manner adapted to differentiated scenarios, as described in claim 1, is characterized in that: The specific steps of step S400 include: The training dataset is divided into a training set, a validation set, and a test set in a 7:2:1 ratio. The mean absolute percentage error is used as the loss function, and the model parameters are iteratively updated through the backpropagation algorithm. The MAPE formula is: ； in, This is the actual value. The value is the predicted value, and n is the number of samples. Set an initial learning rate and use an adaptive moment estimation optimizer to dynamically adjust the learning rate. When the loss function value converges to a set threshold or the number of iterations reaches a set maximum value, training stops, the optimal model structure is determined, and a well-trained user energy consumption trend prediction model is obtained.
7. The method for analyzing the spatiotemporal characteristics of user energy consumption in a source-grid-load interactive manner adapted to differentiated scenarios, as described in claim 6, is characterized in that: The initial learning rate is set to 0.001, the threshold of the loss function is set to 0.05, and the maximum number of iterations is set to 1000.
8. The method for analyzing the spatiotemporal characteristics of user energy consumption in a source-grid-load interactive manner adapted to differentiated scenarios, as described in claim 1, is characterized in that: Furthermore, the specific steps of step S500 include: Using the user energy consumption trend prediction model trained in step S400, inference is performed on the source-grid-load interaction data under different regions, different DG penetration rates, different load types and different time scales, and the output of distributed power generation output prediction results, electric vehicle charging and discharging load prediction results and user energy consumption trend prediction results are output.