Power load dynamic distribution control system and method

Abstract

The invention relates to the field of power load dispatching, in particular to a power load dynamic distribution control system and method. According to the method, historical power load data of each region is preprocessed, time sequence characteristics of the power load of each region are extracted, and the future power load is predicted more accurately in combination with spatial distance information between the regions contained in a physical spatial topology matrix, so that the power load is scheduled. Therefore, the optimized power load dynamic distribution control method is adopted to perform scheduling arrangement, including generator set scheduling, energy storage equipment scheduling and power transmission network scheduling, so that the operation cost can be reduced and the flexibility of the system can be improved on the premise of ensuring the power supply safety.

Description

Technical Field

[0001] The present application relates to the field of power load dispatching, and specifically to a power load dynamic distribution control system and method. Background Art

[0002] With the rapid development of the economy and the acceleration of urbanization, the demand for electricity is growing, and the power system is facing increasingly complex load management challenges. Power load forecasting is an important part of the operation and management of the power system. It plays a vital role in ensuring the balance of power supply and demand, improving the operating efficiency of the power system, and reducing operating costs. However, traditional power load forecasting and scheduling methods are usually based on statistical models and manual planning, and have problems such as low prediction accuracy, lack of real-time and dynamic performance, and limited data processing capabilities.

[0003] In response to the above technical problems, Chinese patent CN116646933A discloses a power load scheduling method and system based on big data, which can use big data technology to collect historical power load data from different regions, preprocess and extract features, and then use the extracted features (including time, season, holidays and weather characteristics) to establish a power load prediction model through machine learning. Finally, the model is used to predict the load, and the scheduling of generator sets, energy storage equipment and transmission networks is carried out according to the prediction results and the actual situation of the current power system.

[0004] However, in the above-mentioned power load scheduling method, when predicting power load, the scheme mainly relies on historical power load data, and infers future load trends by analyzing past data, without fully considering the physical spatial relationship between various regions, and ignoring the mutual influence between power loads in different geographical regions, resulting in poor performance of the model when facing complex changes in power demand, which limits the accuracy of the prediction results and thus affects the accuracy of dynamic allocation of power loads.

[0005] Therefore, an optimized power load dynamic distribution control scheme is desired. Summary of the invention

[0006] This application is made in view of the above problems. One object of this application is to provide a power load dynamic distribution control system and method.

[0007] The embodiment of the present application provides a method for controlling dynamic distribution of electric loads, which includes:

[0008] Collect historical power load data for each region;

[0009] Preprocessing the historical power load data of each region to obtain preprocessed historical power load data of each region;

[0010] Extracting the time series characteristics of power load in each region from the pre-processed historical power load data of each region;

[0011] Constructing a physical space topology matrix between the regions, wherein the value of each non-diagonal position in the physical space topology matrix is ​​used to represent the Euclidean distance between the corresponding two regions;

[0012] Inputting the physical space topology matrix and the time series characteristics of the power loads in each region into the trained load forecasting model to obtain a load forecasting result;

[0013] According to the load forecast results and the actual situation of the current power system, an optimization algorithm is used to dispatch the power load, including generator group scheduling, energy storage equipment scheduling and transmission network scheduling.

[0014] For example, according to the power load dynamic distribution control method of an embodiment of the present application, extracting the power load time series characteristics of each region from the pre-processed historical power load data of each region includes:

[0015] Extracting the time series correlation features of the power loads based on the multi-dimensional time scale from the pre-processed historical power load data of each region to obtain a sequence of time series correlation feature vectors of the power loads at the first time scale, a sequence of time series correlation feature vectors of the power loads at the second time scale, and a sequence of time series correlation feature vectors of the power loads at the third time scale;

[0016] Performing contribution feature enhancement aggregation based on cluster analysis on the sequence of the first time scale electric load time series associated feature vectors, the sequence of the second time scale electric load time series associated feature vectors, and the sequence of the third time scale electric load time series associated feature vectors, respectively, to obtain the first time scale electric load time series significantly aggregated feature vectors, the second time scale electric load time series significantly aggregated feature vectors, and the third time scale electric load time series significantly aggregated feature vectors;

[0017] The first time scale power load time series significant aggregation feature vector, the second time scale power load time series significant aggregation feature vector and the third time scale power load time series significant aggregation feature vector are fused to obtain a multi-time scale power load time series significant aggregation feature vector as the power load time series feature of each region.

[0018] For example, according to the power load dynamic distribution control method of an embodiment of the present application, the pre-processed historical power load data of each area is subjected to power load time series correlation feature extraction based on a multi-dimensional time scale to obtain a sequence of first time scale power load time series correlation feature vectors, a sequence of second time scale power load time series correlation feature vectors, and a sequence of third time scale power load time series correlation feature vectors, including:

[0019] Performing one-dimensional convolution coding based on the first convolution layer on the preprocessed historical power load data of each region to obtain a sequence of time series correlation feature vectors of the power load at the first time scale;

[0020] Performing one-dimensional convolutional coding based on a second convolutional layer on the preprocessed historical power load data of each region to obtain a sequence of time series correlation feature vectors of the power load at the second time scale;

[0021] Performing one-dimensional convolution coding based on the third convolution layer on the preprocessed historical power load data of each region to obtain a sequence of time series correlation feature vectors of the power load at the third time scale;

[0022] The scales of the first convolutional layer, the second convolutional layer and the third convolutional layer are 7×1, 30×1 and 90×1 respectively.

[0023] For example, according to the method for controlling the dynamic distribution of electric loads in an embodiment of the present application, the sequence of the first time scale electric load time series associated feature vectors, the sequence of the second time scale electric load time series associated feature vectors, and the sequence of the third time scale electric load time series associated feature vectors are respectively subjected to contribution feature enhancement aggregation based on cluster analysis to obtain the first time scale electric load time series significant aggregation feature vector, the second time scale electric load time series significant aggregation feature vector, and the third time scale electric load time series significant aggregation feature vector, including:

[0024] Performing cluster analysis on the sequence of the first time scale power load time series correlation feature vectors to obtain a first time scale power load time series correlation feature self-supervised cluster representation vector;

[0025] Based on the self-supervised clustering representation vector of the first time scale power load timing association feature, the sequence of the first time scale power load timing association feature vectors is subjected to feature clustering field modulation aggregation to obtain the first time scale power load timing significant aggregation feature vector.

[0026] For example, according to the electric power load dynamic distribution control method of an embodiment of the present application, a cluster analysis is performed on the sequence of the first time scale electric load timing correlation feature vectors to obtain a self-supervised clustering representation vector of the first time scale electric load timing correlation feature, including: calculating the mean vector of the sequence of the first time scale electric load timing correlation feature vectors to obtain a self-supervised clustering representation vector of the first time scale electric load timing correlation feature.

[0027] For example, according to the power load dynamic distribution control method of an embodiment of the present application, based on the self-supervised clustering representation vector of the first time scale power load time series correlation feature, the sequence of the first time scale power load time series correlation feature vector is subjected to feature clustering field modulation aggregation to obtain the first time scale power load time series significant aggregation feature vector, including:

[0028] Calculating the implicit clustering contribution factor of each first time scale power load time series association feature vector in the sequence of the first time scale power load time series association feature vector relative to the first time scale power load time series association feature self-supervised clustering representation vector to obtain a sequence of implicit clustering contribution factors of the first time scale power load time series association feature;

[0029] Vectorizing the sequence of implicit clustering contribution factors of the first time scale electric load timing correlation feature to obtain a first time scale electric load timing correlation feature clustering contribution field distribution vector;

[0030] Inputting the distribution vector of the cluster contribution field of the first time scale power load timing correlation feature into the cluster contribution field explicit modeling module based on the self-attention mechanism to obtain the first time scale power load timing correlation feature cluster contribution field modulation weight vector;

[0031] Taking each eigenvalue in the first time scale power load timing association feature clustering contribution field modulation weight vector as weight, calculate the position weighted sum of the sequence of the first time scale power load timing association feature vector to obtain the first time scale power load timing significant aggregation feature vector.

[0032] For example, according to the power load dynamic distribution control method of an embodiment of the present application, the implicit clustering contribution factor of each first time scale power load time series association feature vector in the sequence of the first time scale power load time series association feature vector relative to the first time scale power load time series association feature self-supervised clustering representation vector is calculated to obtain a sequence of implicit clustering contribution factors of the first time scale power load time series association feature, including:

[0033] Calculating the absolute value of the positional division between the first time scale power load time series correlation feature vector and the first time scale power load time series correlation feature self-supervised clustering representation vector to obtain the first time scale power load time series self-supervised clustering interactive correlation feature vector;

[0034] Calculate the logarithmic function value of each position characteristic value in the first time scale electric load time series self-supervised clustering interactive correlation characteristic vector with base 2 to obtain the first time scale electric load time series self-supervised clustering interactive logarithmic representation vector;

[0035] Taking each position characteristic value in the self-supervised clustering interaction logarithm representation vector of the first time scale electric load time series as a weighting coefficient, calculating the weighted sum of each position characteristic value in the first time scale electric load time series association characteristic vector to obtain the first time scale electric load time series association characteristic clustering contribution degree representation factor;

[0036] The exponential function value of the clustering contribution degree representation factor of the first time scale power load timing correlation feature is calculated with the natural constant e as the base to obtain the implicit clustering contribution factor of the first time scale power load timing correlation feature.

[0037] For example, according to the power load dynamic distribution control method of an embodiment of the present application, the first time scale power load timing correlation feature cluster contribution field distribution vector is input into the cluster contribution field explicit modeling module based on the self-attention mechanism to obtain the first time scale power load timing correlation feature cluster contribution field modulation weight vector, including:

[0038] Calculate the matrix multiplication between the first time scale power load time series correlation feature cluster contribution field distribution vector and the query weight matrix, key weight matrix and value weight matrix respectively to obtain the first time scale power load cluster contribution field distribution query vector, the first time scale power load cluster contribution field distribution key vector and the first time scale power load cluster contribution field distribution value vector;

[0039] After calculating the multiplication between the query vector of the first time scale power load cluster contribution field distribution and the transposed vector of the first time scale power load cluster contribution field distribution key vector, the obtained matrix is ​​divided by the square root of the length of the first time scale power load cluster contribution field distribution key vector by position to obtain the query-key first time scale power load cluster contribution field semantic interaction representation matrix;

[0040] The query-key first time scale power load cluster contribution field semantic interaction representation matrix is ​​normalized by using a Softmax function to obtain a normalized query-key first time scale power load cluster contribution field semantic interaction representation matrix;

[0041] The multiplication between the distribution value vector of the first time scale power load cluster contribution field and the query-key first time scale power load cluster contribution field semantic interaction representation matrix is ​​calculated to obtain the first time scale power load timing correlation feature cluster contribution field domain modulation weight vector.

[0042] For example, according to the power load dynamic distribution control method of an embodiment of the present application, the physical space topology matrix and the time series characteristics of the power loads in each region are input into the trained load forecasting model to obtain the load forecasting result, including:

[0043] Inputting the physical space topology matrix and the multi-time scale power load time series significant aggregation feature vectors of each region into the graph convolution layer of the load forecasting model to obtain the spatial topology enhanced power load global time series joint feature matrix;

[0044] The spatial topology enhanced electric load global time series joint characteristic matrix is ​​input into the decoding layer of the load forecasting model to obtain the load forecasting result.

[0045] The embodiment of the present application also provides a power load dynamic distribution control system, which includes:

[0046] Data collection module, used to collect historical power load data of various regions;

[0047] A data preprocessing module, used for preprocessing the historical power load data of each area to obtain the preprocessed historical power load data of each area;

[0048] A time series feature extraction module, used to extract the time series features of the power load of each region from the pre-processed historical power load data of each region;

[0049] A physical space topology matrix construction module is used to construct a physical space topology matrix between the regions, wherein the value of each non-diagonal position in the physical space topology matrix is ​​used to represent the Euclidean distance between the corresponding two regions;

[0050] A load forecasting module, used for inputting the physical space topology matrix and the time series characteristics of the power loads in each region into the trained load forecasting model to obtain a load forecasting result;

[0051] The scheduling module is used to schedule the power load according to the load forecast results and the actual situation of the current power system using an optimization algorithm, including generator set scheduling, energy storage device scheduling and transmission network scheduling.

[0052] According to the power load dynamic distribution control system and method of the embodiment of the present application, it pre-processes the historical power load data of each area, extracts the time series characteristics of the power load of each area, and then combines the spatial distance information between each area contained in the physical space topology matrix to more accurately predict the future power load, so as to schedule the power load. In this way, the optimized power load dynamic distribution control method is used for scheduling, including generator group scheduling, energy storage equipment scheduling and transmission network scheduling, which can reduce operating costs and improve system flexibility while ensuring power supply safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application are briefly introduced below. Obviously, the drawings described below only relate to some embodiments of the present application, and are not intended to limit the present application.

[0054] Figure 1 A schematic diagram of the application architecture of the power load dynamic distribution control method in an embodiment of the present application is shown;

[0055] Figure 2 A flow chart of a method for controlling dynamic distribution of electric loads in an embodiment of the present application is shown;

[0056] Figure 3 A flowchart of sub-step S530 of the power load dynamic distribution control method in an embodiment of the present application is shown;

[0057] Figure 4 A flowchart of sub-step S550 of the power load dynamic distribution control method in an embodiment of the present application is shown;

[0058] Figure 5 A schematic diagram of the structure of the power load dynamic distribution control system in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present application.

[0060] The terms used in this specification are those common terms currently widely used in the art in consideration of the functions of the present application, but these terms may vary according to the intentions of those skilled in the art, precedents, or new technologies in the art. In addition, specific terms may be selected, and in this case, their detailed meanings will be described in the detailed description of the present application. Therefore, the terms used in the specification should not be understood as simple names, but rather as a general description based on the meaning of the terms and the present application.

[0061] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0062] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. At the same time, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0063] Figure 1 A schematic diagram of the application architecture of the power load dynamic distribution control method in an embodiment of the present application is shown, including a server and a terminal device.

[0064] The terminal device and the server can be connected through the Internet to achieve mutual communication. Optionally, the above-mentioned Internet uses standard communication technology and / or protocol. The Internet is usually the Internet, but it can also be any network, including but not limited to any combination of local area network (Local Area Network, LAN), metropolitan area network (Metropolitan Area Network, MAN), wide area network (Wide Area Network, WAN), mobile, wired or wireless network, dedicated network or virtual private network. In some embodiments, the data exchanged through the network is represented by the technology and / or format including Hyper Text Markup Language (Hyper Text Markup Language, HTML), Extensible Markup Language (Extensible Markup Language, XML), etc. In addition, conventional encryption technologies such as Secure Socket Layer (Secure Socket Layer, SSL), Transport Layer Security (Transport Layer Security, TLS), Virtual Private Network (Virtual Private Network, VPN), Internet Protocol Security (Internet Protocol Security, IPsec) can also be used to encrypt all or some links. In other embodiments, customized and / or dedicated data communication technology can also be used to replace or supplement the above-mentioned data communication technology.

[0065] The server can provide various network services for the terminal device, wherein the server can be a single server, a server cluster consisting of several servers, or a cloud computing center. Specifically, the server may include a processor (Center Processing Unit, CPU), a memory, an input device, and an output device, etc. The input device may include a keyboard, a mouse, a touch screen, etc., and the output device may include a display device, such as a liquid crystal display (Liquid Crystal Display, LCD), a cathode ray tube (Cathode Ray Tube, CRT), etc.

[0066] The memory may include a read-only memory (ROM) and a random access memory (RAM), and provides the processor with program instructions and data stored in the memory. In the embodiment of the present application, the memory may be used to store the program of the power load dynamic distribution control method in the embodiment of the present application.

[0067] The processor calls the program instructions stored in the memory, and the processor is used to execute the steps of any one of the power load dynamic distribution control methods in the embodiments of the present application according to the obtained program instructions.

[0068] In addition, the application architecture diagram in the embodiment of the present application is intended to more clearly illustrate the technical solution in the embodiment of the present application, and does not constitute a limitation on the technical solution provided in the embodiment of the present application. Of course, for other application architectures and business applications, the technical solution provided in the embodiment of the present application is also applicable to similar problems.

[0069] The following is a non-restrictive description of the power load dynamic distribution control method provided according to at least one embodiment of the present application through several examples or embodiments. As described below, different features in these specific examples or embodiments can be combined with each other without conflicting with each other to obtain new examples or embodiments, and these new examples or embodiments also fall within the scope of protection of the present application.

[0070] In the power system, dynamic load distribution control is crucial. By real-time monitoring and forecasting of power load, the power dispatching center can optimize the dispatching of generators, energy storage equipment and transmission networks, and improve the overall efficiency and reliability of the system. In fact, when forecasting power load, the changing trends and patterns of power loads in physically adjacent areas may show a high correlation due to the influence of similar climatic conditions, socio-economic activities and other factors. Therefore, if physical space information can be introduced, the load forecasting results can be optimized. Specifically, since two places with close distances are often affected by similar environmental factors (such as weather), the changing trends and patterns of their power loads are more similar. For example, if two adjacent cities face the same high temperature weather, their air conditioning power consumption is likely to grow synchronously.

[0071] Based on this, in the technical solution of this application, a method for dynamic distribution control of power load is proposed. Figure 2 The flowchart of the power load dynamic distribution control method in the embodiment of the present application is shown. For example, the power load dynamic distribution control method can be executed by a server, which can be Figure 1 The server shown in . Figure 2As shown, according to the embodiment of the present application, the method for dynamic distribution control of electric load includes the following steps: S510, collecting historical electric load data of each area; S520, preprocessing the historical electric load data of each area to obtain the preprocessed historical electric load data of each area; S530, extracting the time series characteristics of the electric load of each area from the preprocessed historical electric load data of each area; S540, constructing a physical space topology matrix between the areas, and the values ​​of each position on the non-diagonal position in the physical space topology matrix are used to represent the Euclidean distance between the corresponding two areas; S550, inputting the physical space topology matrix and the time series characteristics of the electric load of each area into the trained load forecasting model to obtain the load forecasting result; S560, according to the load forecasting result and the actual situation of the current power system, using an optimization algorithm to schedule the electric load, including generator set scheduling, energy storage equipment scheduling and transmission network scheduling.

[0072] In one example, step S510 can be performed by installing smart meters and sensor devices in different areas to regularly collect historical power load data for each area. The collected data may include power consumption, power prices, weather conditions, special events, etc. The time granularity of data collection can be minute, hour or day, depending on the needs of the system and the availability of data. In step S520, the data is cleaned to remove outliers and noise. To process missing values, interpolation methods such as linear interpolation, seasonal interpolation, etc. can be used to convert non-numeric data into numeric data, such as date and time features. Data is standardized or normalized so that it can be used in model training. In step S530, time series features such as trends, seasonality, periodicity, etc. are extracted, and weather-related features such as temperature, humidity, wind speed, etc. are extracted. In this step, the impact of special events or holidays on power load is identified, and the correlation features between power load and other variables (such as price and temperature) are extracted. In step S540, geographic information system (GIS) data can be used to determine the physical location relationship between regions, and the distance between regions can be calculated. Euclidean distance, Manhattan distance or other appropriate distance measurement methods can be used to construct a topological matrix. Each element of the matrix represents the distance between two regions. The matrix is ​​symmetrical, and the diagonal elements are zero (indicating the distance between the region and itself). In step S550, the load forecasting model can be a machine learning model, such as a random forest, a support vector machine, a neural network, etc., and the historical power load data and the topological matrix are used as input features during model training. The model predicts future power loads and outputs load forecast results. In step S560, an optimization algorithm, such as linear programming, genetic algorithm, particle swarm optimization, etc., is used to determine the optimal power dispatching strategy. Consider the actual constraints of the power system, such as the maximum and minimum output limits of the generator, the charging and discharging limits of the energy storage device, and the transmission limits of the power grid. The dispatching strategy aims to optimize the operating efficiency of the power system, while meeting the power demand and ensuring the reliability of power supply, by monitoring the operating status of the power system in real time and adjusting the dispatching strategy as needed. These steps are combined to form a comprehensive power load dynamic distribution control method, which aims to improve the prediction accuracy and dispatching efficiency of the power system.

[0073] In particular, in the above-mentioned power load dynamic distribution control method, not only the historical power load data of each region is considered, but also the physical space relationship between regions is introduced to optimize the load forecasting results. By preprocessing the historical power load data of each region and extracting the time series characteristics of the power load of each region, the future power load can be more accurately predicted by combining the spatial distance information between each region contained in the physical space topology matrix, so as to schedule the power load. In this way, the optimized power load dynamic distribution control method is used for scheduling, including generator group scheduling, energy storage equipment scheduling and transmission network scheduling, which can reduce operating costs and improve system flexibility while ensuring power supply safety.

[0074] When processing the pre-processed historical power load data of each region to extract the time series characteristics of the power load of each region to provide a basis for the subsequent power load scheduling, the technical concept of this application is to use data processing and analysis algorithms based on artificial intelligence and machine learning to extract the time series characteristics of the power load of each region based on multi-scale, so as to capture the multi-scale time series characteristics of the power load of each region, and perform feature aggregation based on the contribution of cluster analysis on these multi-scale time series characteristics of the power load of each region, so as to characterize the time series characteristics of the power load of each region. This power load time series feature can be used to combine with the spatial topology information of each region to predict and schedule the power load, thereby realizing the dynamic distribution control of the power load and more efficient power system management, providing strong support for the sustainable development of the power industry.

[0075] Specifically, in the process of extracting the time series characteristics of the power load of each region from the pre-processed historical power load data of each region, it is considered that in the power load forecast, feature extraction at different time scales is very necessary, because short-term load changes may be affected by weather, emergencies, etc., while long-term load changes are more affected by macro factors such as seasonality and economic development. Therefore, in order to extract the time series correlation characteristics of the power load from different time scales, so as to more accurately and comprehensively perceive the dynamic changes and trends of the power load time series in each region, in the technical solution of the present application, the pre-processed historical power load data of each region is subjected to one-dimensional convolution encoding based on the first convolution layer to obtain a sequence of time series correlation feature vectors of the power load at the first time scale; the pre-processed historical power load data of each region is subjected to one-dimensional convolution encoding based on the second convolution layer to obtain a sequence of time series correlation feature vectors of the power load at the second time scale; the pre-processed historical power load data of each region is subjected to one-dimensional convolution encoding based on the third convolution layer to obtain a sequence of time series correlation feature vectors of the power load at the third time scale. In particular, the scales of the first convolutional layer, the second convolutional layer and the third convolutional layer are 7×1, 30×1 and 90×1 respectively. Specifically, the convolution kernels of different scales (7×1, 30×1 and 90×1) correspond to short, medium and long time windows respectively, so as to capture different periodic patterns of power load changes over time. Among them, the scale of the first convolutional layer (7×1) can be used to extract short-term power load characteristics, such as the load change pattern of each day in a week. Because the power consumption patterns of each day in a week usually have high similarity, such as the difference between weekdays and weekends, this scale can well reflect short-term periodic changes. The scale of the second convolutional layer (30×1) can be used to capture the power load change pattern within a month. Due to seasonal changes, festivals or temperature changes in a specific month, the power load in a month will show a unique pattern. The scale of the third convolutional layer (90×1) can be used to analyze the power load characteristics within a quarter or longer period of time. Long-term time spans help to discover power load patterns in a larger range, such as seasonal changes, annual load fluctuations, etc. In this way, characteristic information at different time scales can be integrated to enable the model to more comprehensively understand the temporal variation of power load, thereby improving the accuracy and robustness of load forecasting.

[0076] It should be understood that since the sequence of the first time scale power load time series associated feature vectors, the sequence of the second time scale power load time series associated feature vectors and the sequence of the third time scale power load time series associated feature vectors respectively contain different scale time series features of the power load in each region in the time dimension, for example, for the first time scale (7×1), the daily load change law may be captured; while for the second time scale (30×1), the monthly load change may be reflected; and the third time scale (90×1) may cover the quarterly or even annual load change. These different scales of power load time series features not only have characteristic information related to the subsequent power load time series representation and power load forecasting and scheduling tasks in each region when expressing the features, but also have some interference features and noise information. Therefore, in order to obtain the key features of power load changes from multiple time scales and extract significant information therefrom to enhance the performance of the prediction model, in the technical solution of the present application, the sequence of the first time scale power load time series associated feature vectors, the sequence of the second time scale power load time series associated feature vectors and the sequence of the third time scale power load time series associated feature vectors are further subjected to contribution feature enhancement aggregation based on cluster analysis to obtain the first time scale power load time series significant aggregation feature vector, the second time scale power load time series significant aggregation feature vector and the third time scale power load time series significant aggregation feature vector. By strengthening the key information in these feature vectors, it is possible to better understand and predict the changes in power load, especially the periodic and non-periodic patterns that may exist at different time scales. The feature vectors processed in this way can be used to train the load forecasting model to improve the prediction accuracy and reliability of the model.

[0077] Specifically, for the process of performing cluster analysis-based contribution feature reinforcement aggregation on the sequence of the first time scale power load time series associated feature vectors to obtain the first time scale power load time series significantly aggregated feature vectors, first, an unsupervised learning method is used to perform cluster analysis on the sequence of the first time scale power load time series associated feature vectors extracted using the first time scale. This self-supervised clustering can automatically discover patterns and structures in the data without labels, and generate a self-supervised clustering representation vector of the first time scale power load time series associated features. In this way, the inherent structure of the power load time series data under the first time scale can be revealed, and the model's understanding of the distribution of power load time series data can be enhanced. Furthermore, by calculating the implicit clustering contribution factor of each first time scale power load time series associated feature vector relative to the first time scale power load time series associated feature self-supervised clustering representation vector, it is used to reflect the contribution of each first time scale power load time series associated feature vector to the cluster to which it belongs. This step helps to identify and quantify which feature vectors and power load time series features play an important role in the clustering process, and provides a basis for subsequent feature selection and weight allocation. Next, the sequence of implicit clustering contribution factors of the first time scale power load time series correlation feature calculated is converted into the first time scale power load time series correlation feature clustering contribution field distribution vector, that is, these implicit clustering contribution factors are integrated into the vector field by data integration, so as to construct the clustering contribution field. This step utilizes the concept of field theory to visualize the importance of the feature vector in space, which is convenient for subsequent processing. Then, the first time scale power load time series correlation feature clustering contribution field distribution vector is explicitly modeled using the self-attention mechanism to generate the first time scale power load time series correlation feature clustering contribution field modulation weight vector. The self-attention mechanism allows the model to focus on those power load time series correlation features that contribute more to the clustering, while suppressing the expression of unimportant features, thereby improving the quality of feature representation and the prediction accuracy of the model. Finally, the first time scale power load time series correlation feature vector is weighted fused using the first time scale power load time series correlation feature clustering contribution field modulation weight vector to generate the first time scale power load time series significant aggregation feature vector. This optimized significant aggregation feature vector not only retains the information of the original power load time series characteristics, but also integrates the knowledge learned in the clustering process, so that the model can more accurately capture the key features of the data when dealing with complex power load forecasting and scheduling tasks, thereby improving the performance and generalization ability of the model.

[0078] Accordingly, in step S530, if Figure 3As shown, extracting the time series characteristics of the power load of each region from the pre-processed historical power load data of each region includes: S531, extracting the power load time series correlation characteristics based on the multi-dimensional time scale from the pre-processed historical power load data of each region to obtain a sequence of first time scale power load time series correlation feature vectors, a sequence of second time scale power load time series correlation feature vectors and a sequence of third time scale power load time series correlation feature vectors; S532, respectively extracting the sequence of the first time scale power load time series correlation feature vectors, the sequence of the second time scale power load time series correlation feature vectors and the sequence of the third time scale power load time series correlation feature vectors. The sequence of third time scale power load timing associated feature vectors is subjected to contribution feature intensive aggregation based on cluster analysis to obtain the first time scale power load timing significantly aggregated feature vector, the second time scale power load timing significantly aggregated feature vector and the third time scale power load timing significantly aggregated feature vector; S533, the first time scale power load timing significantly aggregated feature vector, the second time scale power load timing significantly aggregated feature vector and the third time scale power load timing significantly aggregated feature vector are integrated to obtain the multi-time scale power load timing significantly aggregated feature vector as the power load timing characteristics of each region.

[0079] Wherein, in step S531, the preprocessed historical power load data of each region is subjected to power load time series correlation feature extraction based on a multidimensional time scale to obtain a sequence of first time scale power load time series correlation feature vectors, a sequence of second time scale power load time series correlation feature vectors and a sequence of third time scale power load time series correlation feature vectors, including: performing one-dimensional convolution encoding based on a first convolution layer on the preprocessed historical power load data of each region to obtain a sequence of first time scale power load time series correlation feature vectors; performing one-dimensional convolution encoding based on a second convolution layer on the preprocessed historical power load data of each region to obtain a sequence of second time scale power load time series correlation feature vectors; performing one-dimensional convolution encoding based on a third convolution layer on the preprocessed historical power load data of each region to obtain a sequence of third time scale power load time series correlation feature vectors; wherein the scales of the first convolution layer, the second convolution layer and the third convolution layer are 7×1, 30×1 and 90×1, respectively.

[0080] Wherein, in step S532, the sequence of the first time scale power load time series associated feature vectors, the sequence of the second time scale power load time series associated feature vectors and the sequence of the third time scale power load time series associated feature vectors are respectively subjected to contribution feature enhancement aggregation based on cluster analysis to obtain the first time scale power load time series significantly aggregated feature vector, the second time scale power load time series significantly aggregated feature vector and the third time scale power load time series significantly aggregated feature vector, including: performing cluster analysis on the sequence of the first time scale power load time series associated feature vectors to obtain the first time scale power load time series associated feature self-supervised clustering representation vector; based on the first time scale power load time series associated feature self-supervised clustering representation vector, performing feature clustering field modulation aggregation on the sequence of the first time scale power load time series associated feature vectors to obtain the first time scale power load time series significantly aggregated feature vector.

[0081] Among them, cluster analysis is performed on the sequence of the first time scale power load timing correlation feature vectors to obtain the first time scale power load timing correlation feature self-supervised cluster representation vector, including: calculating the mean vector of the sequence of the first time scale power load timing correlation feature vectors to obtain the first time scale power load timing correlation feature self-supervised cluster representation vector.

[0082] Among them, based on the self-supervised clustering representation vector of the first time scale electric load timing association characteristic, the sequence of the first time scale electric load timing association characteristic vectors is subjected to feature clustering field modulation aggregation to obtain the first time scale electric load timing association significant aggregation characteristic vector, including: calculating the implicit clustering contribution factor of each first time scale electric load timing association characteristic vector in the sequence of the first time scale electric load timing association characteristic vector relative to the self-supervised clustering representation vector of the first time scale electric load timing association characteristic to obtain a sequence of the implicit clustering contribution factors of the first time scale electric load timing association characteristic; The sequence of characteristic implicit clustering contribution factors is vectorized to obtain the distribution vector of the characteristic clustering contribution field of the first time scale power load timing association feature; the distribution vector of the characteristic clustering contribution field of the first time scale power load timing association feature is input into the clustering contribution field explicit modeling module based on the self-attention mechanism to obtain the modulation weight vector of the characteristic clustering contribution field of the first time scale power load timing association feature; each characteristic value in the modulation weight vector of the characteristic clustering contribution field of the first time scale power load timing association feature is used as a weight, and the position-weighted sum of the sequence of the characteristic vector of the first time scale power load timing association feature is calculated to obtain the significant aggregation characteristic vector of the first time scale power load timing.

[0083] Specifically, the implicit clustering contribution factor of each first time scale power load time series association feature vector in the sequence of the first time scale power load time series association feature vector is calculated relative to the first time scale power load time series association feature self-supervised clustering representation vector to obtain a sequence of implicit clustering contribution factors of the first time scale power load time series association feature, including: calculating the absolute value of the positional division between the first time scale power load time series association feature vector and the first time scale power load time series association feature self-supervised clustering representation vector to obtain the first time scale power load time series self-supervised clustering interactive association feature vector; calculating the first time scale power load time series with a base of 2 The logarithmic function value of each position feature value in the self-supervised clustering interaction correlation feature vector is used to obtain the self-supervised clustering interaction logarithmic representation vector of the first time scale power load time series; using the each position feature value in the self-supervised clustering interaction logarithmic representation vector of the first time scale power load time series as the weighting coefficient, the weighted sum of the each position feature value in the first time scale power load time series correlation feature vector is calculated to obtain the first time scale power load time series correlation feature clustering contribution representation factor; calculate the exponential function value of the first time scale power load time series correlation feature clustering contribution representation factor with the natural constant e as the base to obtain the first time scale power load time series correlation feature implicit clustering contribution factor.

[0084] Specifically, the first time scale power load timing correlation feature cluster contribution field distribution vector is input into the cluster contribution field explicit modeling module based on the self-attention mechanism to obtain the first time scale power load timing correlation feature cluster contribution field domain modulation weight vector, including: calculating the first time scale power load timing correlation feature cluster contribution field distribution vector and the query weight matrix, key weight matrix and value weight matrix respectively to obtain the first time scale power load cluster contribution field distribution query vector, the first time scale power load cluster contribution field distribution key vector and the first time scale power load cluster contribution field distribution value vector; calculating the first time scale power load cluster contribution field distribution query vector and the first time scale power load cluster contribution field distribution key vector. After multiplying the transpose vectors of the vectors, the obtained matrix is ​​divided by the square root of the length of the first time scale power load clustering contribution field distribution key vector by position to obtain the query-key first time scale power load clustering contribution field semantic interaction representation matrix; the query-key first time scale power load clustering contribution field semantic interaction representation matrix is ​​normalized by using the Softmax function to obtain the normalized query-key first time scale power load clustering contribution field semantic interaction representation matrix; the multiplication between the first time scale power load clustering contribution field distribution value vector and the query-key first time scale power load clustering contribution field semantic interaction representation matrix is ​​calculated to obtain the first time scale power load timing correlation feature clustering contribution field domain modulation weight vector.

[0085] In a specific example, the sequence of the first time scale power load time series associated feature vectors, the sequence of the second time scale power load time series associated feature vectors and the sequence of the third time scale power load time series associated feature vectors are respectively subjected to the contribution feature enhancement aggregation based on cluster analysis to obtain the first time scale power load time series significant aggregation feature vector, the second time scale power load time series significant aggregation feature vector and the third time scale power load time series significant aggregation feature vector, including: the sequence of the first time scale power load time series associated feature vectors is subjected to the contribution feature enhancement aggregation based on cluster analysis to obtain the first time scale power load time series significant aggregation feature vector according to the following feature aggregation formula; wherein the feature aggregation formula is:

[0086] X={x1,x2,...,x k ,...,x n}

[0087]

[0088] v d ={d1; d2; ...; d j ;...;d n}

[0089]

[0090] v q =W q v d

[0091] v k =W k v d

[0092] v v =W b v d

[0093]

[0094] Where X is the sequence of the time series correlation feature vectors of the power load at the first time scale, x1, x2, x k ,x n are the first, second, kth and nth first time scale electric load time series correlation feature vectors in the sequence of the first time scale electric load time series correlation feature vectors, respectively, i is the i-th first time scale electric load time series associated characteristic vector in the sequence of the first time scale electric load time series associated characteristic vector, N is the number of characteristic vectors in the sequence of the first time scale electric load time series associated characteristic vector, x c is the self-supervised clustering representation vector of the first time scale power load time series correlation feature, is the characteristic value of the kth position in the jth first time scale electric load time series associated characteristic vector in the sequence of the first time scale electric load time series associated characteristic vector, is the eigenvalue of the kth position in the self-supervised clustering representation vector of the time series correlation characteristics of the power load at the first time scale, exp represents the natural exponential function, d j is the implicit clustering contribution factor of the first time scale power load time series correlation feature corresponding to the jth first time scale power load time series correlation feature vector, d1, d2, d n are the implicit clustering contribution factors of the first time scale power load time series correlation characteristics corresponding to the first, second and nth first time scale power load time series correlation characteristic vectors, v d is the first time scale power load time series correlation feature clustering contribution field distribution vector obtained by quantizing multiple first time scale power load time series correlation feature implicit clustering contribution factor vectors, W q , W k and W b are the query weight matrix, key weight matrix and value weight matrix respectively, vq 、v k and v v are respectively the query vector of the first time scale power load clustering contribution field distribution, the key vector of the first time scale power load clustering contribution field distribution and the value vector of the first time scale power load clustering contribution field distribution, L is the length of the key vector of the first time scale power load clustering contribution field distribution, v k T represents the transpose of the key vector of the distribution of the power load cluster contribution field at the first time scale, Softmax(·) is the normalized exponential function, is vector multiplication, v w The field modulation weight vector contributing to the clustering of the first time scale power load timing correlation characteristics is represents the eigenvalue of the first time scale power load time series correlation feature clustering contribution field modulation weight vector at the i-th position, x f It is the significant aggregated feature vector of the power load time series at the first time scale.

[0095] Furthermore, the first time scale power load time series significant aggregation feature vector, the second time scale power load time series significant aggregation feature vector and the third time scale power load time series significant aggregation feature vector are fused to obtain the multi-time scale power load time series significant aggregation feature vector as the power load time series feature of each region. It should be understood that the fusion of the power load time series significant aggregation features on these three time scales can ensure that the model can not only identify short-term power load fluctuations, but also capture medium-term and long-term trend changes. In this way, the model can better cope with various different situations, whether it is short-term emergencies or long-term seasonal changes, and can make more accurate power load forecasting and scheduling. In this way, the fused power load time series features of each region are used for subsequent combination with the spatial topology information of each region to predict and dispatch the power load, thereby realizing the dynamic distribution control of the power load and more efficient power system management, providing strong support for the sustainable development of the power industry.

[0096] In step S540, a physical space topology matrix is ​​constructed between the various regions, and the values ​​of each position on the non-diagonal position in the physical space topology matrix are used to represent the Euclidean distance between the corresponding two regions, including: obtaining the precise geographical location coordinates of each power load monitoring point through GIS (Geographic Information System), and then treating each power load monitoring point as a node in the graph, representing the power demand situation in a specific area. In order to construct a topological structure that reflects the relationship between regions, the connection mode and strength of the nodes must be defined. Usually, there is an edge between two nodes, and its weight is determined by multiple factors. In this application, the Euclidean distance is used, that is, the straight-line distance between the geographical locations of the two nodes.

[0097] Specifically, create a symmetric matrix of size N×N, where N is equal to the number of all monitoring points. For each element in the non-diagonal position of the matrix, calculate the Euclidean distance between the corresponding two nodes and fill this value into the corresponding position of the matrix. The formula is:

[0098]

[0099] Among them, d i,j represents the distance between the i-th node and the j-th node, and (a i ,b i ) and (a j ,b j ) are their respective latitude and longitude coordinates. The elements on the diagonal represent the distance between the same nodes and should be set to 0.

[0100] Furthermore, since the physical space topology matrix is ​​an important tool for describing the geographical relationship between various regions. By calculating the Euclidean distance between regions, additional spatial information can be provided for the load forecasting model, so that it can better understand the mutual influence between regions. This topological structure can not only enhance the learning ability of the model, but also improve the adaptability of the model to future load changes and improve the accuracy of the prediction results. Therefore, in order to introduce the spatial topological structure information between different regions when characterizing the time series of power loads in various regions, in the technical solution of the present application, the physical space topology matrix and the multi-time scale power load time series significant aggregation feature vectors of the various regions are further input into the graph convolution layer of the load forecasting model to obtain the spatial topology enhanced power load global time series joint feature matrix. Then, the spatial topology enhanced power load global time series joint feature matrix is ​​input into the decoding layer of the load forecasting model to obtain the load forecasting result. By combining the spatial distance information between the various regions contained in the physical space topology matrix, the future power load can be predicted more accurately, so as to schedule the power load. In this way, the dynamic allocation control of the power load and more efficient power system management are realized, providing strong support for the sustainable development of the power industry.

[0101] Accordingly, in step S550, if Figure 4As shown, the physical space topology matrix and the time series characteristics of the power load in each region are input into the trained load forecasting model to obtain the load forecasting result, including: S551, inputting the physical space topology matrix and the multi-time scale power load time series significant aggregation feature vector of each region into the graph convolution layer of the load forecasting model to obtain the spatial topology enhanced power load global time series joint feature matrix; S552, inputting the spatial topology enhanced power load global time series joint feature matrix into the decoding layer of the load forecasting model to obtain the load forecasting result.

[0102] Specifically, the load forecasting model includes a graph convolution layer and a decoding layer, wherein the graph convolution layer is good at processing data with complex topological structures and can effectively utilize the information in the physical space topological matrix. The graph convolution layer can capture the spatial correlation between different regions by treating each power load monitoring point as a node in the graph and propagating information based on the connections between these nodes and their weights. The graph convolution layer allows the model to integrate information from neighboring nodes. As the number of layers increases, each layer can capture spatial correlations in a larger range. The final spatial topology-enhanced power load global time series joint feature matrix not only contains features in the time dimension, but also integrates information in the spatial dimension. The decoding layer is responsible for converting high-dimensional abstract features after graph convolution processing into the final load forecasting results. The task of the decoding layer is to map these features back to the original problem domain, that is, to predict future power load values. In this application, the decoding layer is a fully connected neural network layer.

[0103] Specifically, the load forecasting model is trained, including: first, it is necessary to prepare the constructed physical space topology matrix and the extracted multi-time scale power load time series significant aggregation feature vectors of each region. Then, these input features are sent to the load forecasting model constructed above. At the same time, the annotated real load data is used as labels to train the model parameters through supervised learning. This means that historical power load data is used as input and the actual load values ​​are used as labels for training. The purpose is to minimize the gap between the predicted value and the true value. Loss functions such as mean square error (MSE) and mean absolute error (MAE) are commonly used to quantify this gap. Optimization algorithms such as gradient descent and its variants (such as Adam, RMSprop) reduce the value of the loss function by iteratively adjusting the model parameters until convergence is reached or the preset stop condition is met.

[0104] Here, considering that the physical space topology matrix and the multi-time scale power load time series significant aggregation feature vectors of each region respectively represent the spatial topology information between each region and the multi-scale correlation characteristics of the power load time series of each region, when they are input into the graph convolution layer of the load forecasting model, the feature heterogeneity and feature depth differences between the spatial topology information and the multi-scale correlation characteristics of the power load time series of each region will cause the spatial topology enhanced power load global time series joint feature matrix obtained by graph convolution encoding to have differences based on feature modality and dimensional representation, which in turn leads to an imbalance in the regression comprehension of the feature representation relative to the source semantics, affecting the accuracy of the decoding results.

[0105] In a preferred example, inputting the spatial topology enhanced electric load global time series joint characteristic matrix into the decoding layer of the load forecasting model to obtain the load forecasting result includes:

[0106] Determine the number of zero eigenvalues ​​in the spatial topology-enhanced electric load global time series joint characteristic matrix, and respectively calculate the reciprocal of the logarithm of the number of zero eigenvalues ​​with base two and the exponential value of the reciprocal of the number of zero eigenvalues ​​with base natural constants to obtain a first spatial topology-enhanced electric load global time series joint representation value and a second spatial topology-enhanced electric load global time series joint representation value;

[0107] Expanding the spatial topology enhanced electric load global time series joint characteristic matrix into a spatial topology enhanced electric load global time series joint characteristic vector;

[0108] Calculate each eigenvalue of the spatial topology enhanced electric load global time series joint eigenvector with the reciprocal of the number of zero eigenvalues ​​as the exponent to obtain the spatial topology enhanced electric load global time series joint structural state vector;

[0109] The spatial topology enhanced electric load global time series joint structure state vector is point-multiplied by the first spatial topology enhanced electric load global time series joint representation value to obtain a spatial topology enhanced electric load global time series joint information representation vector;

[0110] The autocorrelation matrix of the spatial topology-enhanced electric load global time series joint structural state vector is point-multiplied with the second spatial topology-enhanced electric load global time series joint representation value and weight hyperparameter to obtain the spatial topology-enhanced electric load global time series joint regression understanding matrix;

[0111] Performing matrix multiplication of the spatial topology enhanced electric load global time series joint information representation vector and the spatial topology enhanced electric load global time series joint regression understanding matrix to obtain an optimized spatial topology enhanced electric load global time series joint feature vector;

[0112] The optimized spatial topology reinforced electric load global time series joint feature vector is passed through the decoding layer of the load forecasting model to obtain the load forecasting result.

[0113] That is, the first spatial topology strengthened electric load global time series joint representation value and the second spatial topology strengthened electric load global time series joint representation value are expressed as:

[0114] α=[log2(n)] -1

[0115] β=exp[(n) -1 ]

[0116] Among them, n is the number of zero eigenvalues ​​in the joint characteristic matrix of the global time series of the spatial topology enhanced power load, α is the joint representation value of the global time series of the first spatial topology enhanced power load, β is the joint representation value of the global time series of the second spatial topology enhanced power load, and exp represents the natural exponential function.

[0117] And the optimized spatial topology strengthens the joint characteristic vector of the global time series of power load as follows:

[0118]

[0119] Among them, V is the spatial topology enhanced power load global time series joint feature vector, w is the weight hyperparameter, M is the spatial topology enhanced power load global time series joint regression understanding matrix, V1 is the spatial topology enhanced power load global time series joint information representation vector, n is the number of zero eigenvalues ​​in the spatial topology enhanced power load global time series joint feature matrix, α is the first spatial topology enhanced power load global time series joint representation value, β is the second spatial topology enhanced power load global time series joint representation value, represents a power function that calculates each eigenvalue of the spatial topology-enhanced global time series joint eigenvector with the inverse of the number of zero eigenvalues ​​as the exponent, T represents the transpose of the vector, ⊙ represents the point multiplication by position, is a matrix multiplication, and V' is the global time series joint eigenvector of the optimized spatial topology-enhanced power load.

[0120] Based on this, in the above-mentioned preferred example, by modeling the vector field zero dimension of the spatial topology enhanced power load global time series joint feature vector after the spatial topology enhanced power load global time series joint feature matrix is ​​expanded as the global field redundant dependency, the effective long-range modeling of the spatial topology enhanced power load global time series joint feature matrix under the field linear complexity is realized. In this way, in the feature regression process of the spatial topology enhanced power load global time series joint feature matrix, the structured correlation self-distillation between information representation and decoding regression understanding is performed in the selective feature enhancement state space based on feature distribution, so as to maintain the overall information complexity of the feature distribution of the spatial topology enhanced power load global time series joint feature matrix while promoting the long-range dynamic regression understanding balance of its feature convergence, so as to improve the regression understanding consistency between the feature regression and feature extraction processes, and improve the accuracy of the load prediction results obtained by the spatial topology enhanced power load global time series joint feature matrix through the decoding layer of the load prediction model.

[0121] Based on the above embodiments, see Figure 5 As shown, it is a structural schematic diagram of a power load dynamic distribution control system in an embodiment of the present application. The power load dynamic distribution control system includes: a data collection module for collecting historical power load data of each region; a data preprocessing module for preprocessing the historical power load data of each region to obtain the preprocessed historical power load data of each region; a time series feature extraction module for extracting the time series features of the power load of each region from the preprocessed historical power load data of each region; a physical space topology matrix construction module for constructing a physical space topology matrix between the regions, and the values ​​of each position on the non-diagonal position in the physical space topology matrix are used to represent the Euclidean distance between the corresponding two regions; a load prediction module for inputting the physical space topology matrix and the time series features of the power load of each region into the trained load prediction model to obtain the load prediction result; a scheduling module for scheduling the power load using an optimization algorithm according to the load prediction result and the actual situation of the current power system, including generator group scheduling, energy storage equipment scheduling and transmission network scheduling.

[0122] Here, those skilled in the art will appreciate that the specific functions and operations of each module in the above-mentioned power load dynamic distribution control system have been described in detail above. Figures 2 to 4 The description of the power load dynamic distribution control method has been introduced in detail, and therefore, its repeated description will be omitted.

[0123] Based on the above embodiments, another exemplary embodiment of an electronic device is also provided in the embodiments of the present application. In some possible implementations, the electronic device in the embodiments of the present application may include a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor can implement the steps of the power load dynamic distribution control method in the above embodiments when executing the program.

[0124] For example, in the case of electronic equipment Figure 1 Taking the server in as an example, the processor in the electronic device is the processor in the server, and the memory in the electronic device is the memory in the server.

[0125] The embodiment of the present application also provides a computer-readable storage medium, on which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, the power load dynamic distribution control method according to the embodiment of the present application described with reference to the above figures can be executed. The computer-readable storage medium includes, but is not limited to, for example, volatile memory and / or non-volatile memory. The volatile memory may, for example, include a random access memory (RAM) and / or a cache memory (cache), etc. The non-volatile memory may, for example, include a read-only memory (ROM), a hard disk, a flash memory, etc.

[0126] The embodiment of the present application also provides a computer program product or a computer program, which includes computer executable instructions, and the computer executable instructions are stored in a computer readable storage medium. The processor of the computer device reads the computer executable instructions from the computer readable storage medium, and the processor executes the computer executable instructions, so that the computer device executes the power load dynamic distribution control method according to the embodiment of the present application.

[0127] Those skilled in the art will appreciate that the contents disclosed in this application may be subject to various variations and improvements. For example, the various devices or components described above may be implemented by hardware, or by software, firmware, or a combination of some or all of the three.

[0128] In addition, although the present application makes various references to certain units in the system according to embodiments of the present application, any number of different units can be used and run on the client and / or server. The units are only illustrative, and different aspects of the system and method can use different units.

[0129] Those skilled in the art will appreciate that all or part of the steps in the above method can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a disk or an optical disk. Optionally, all or part of the steps in the above embodiment can also be implemented using one or more integrated circuits. Accordingly, each module / unit in the above embodiment can be implemented in the form of hardware or in the form of software function modules. The present application is not limited to any particular form of combination of hardware and software.

[0130] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those of ordinary skill in the art to which this application belongs. It should also be understood that terms such as those defined in common dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an idealized or extremely formal sense, unless explicitly defined as such herein.

[0131] The above is an explanation of the present application and should not be considered as limiting thereof. Although several exemplary embodiments of the present application have been described, those skilled in the art will readily appreciate that many modifications may be made to the exemplary embodiments without departing from the novel teachings and advantages of the present application.

Claims

1. A method for controlling dynamic distribution of electric loads, characterized in that: include: Collect historical power load data for each region; Preprocessing the historical power load data of each region to obtain preprocessed historical power load data of each region; Extracting the time series characteristics of power load in each region from the pre-processed historical power load data of each region; Constructing a physical space topology matrix between the regions, wherein the value of each non-diagonal position in the physical space topology matrix is ​​used to represent the Euclidean distance between the corresponding two regions; Inputting the physical space topology matrix and the time series characteristics of the power loads in each region into the trained load forecasting model to obtain a load forecasting result; According to the load forecast results and the actual situation of the current power system, an optimization algorithm is used to dispatch the power load, including generator group scheduling, energy storage equipment scheduling and transmission network scheduling.

2. The method for controlling the dynamic distribution of electric load according to claim 1, characterized in that: Extracting the time series characteristics of the power load in each region from the pre-processed historical power load data of each region includes: Extracting the time series correlation features of the power loads based on the multi-dimensional time scale from the pre-processed historical power load data of each region to obtain a sequence of time series correlation feature vectors of the power loads at the first time scale, a sequence of time series correlation feature vectors of the power loads at the second time scale, and a sequence of time series correlation feature vectors of the power loads at the third time scale; Performing contribution feature enhancement aggregation based on cluster analysis on the sequence of the first time scale electric load time series associated feature vectors, the sequence of the second time scale electric load time series associated feature vectors, and the sequence of the third time scale electric load time series associated feature vectors, respectively, to obtain the first time scale electric load time series significantly aggregated feature vectors, the second time scale electric load time series significantly aggregated feature vectors, and the third time scale electric load time series significantly aggregated feature vectors; The first time scale power load time series significant aggregation feature vector, the second time scale power load time series significant aggregation feature vector and the third time scale power load time series significant aggregation feature vector are fused to obtain a multi-time scale power load time series significant aggregation feature vector as the power load time series feature of each region.

3. The method for controlling the dynamic distribution of electric load according to claim 2, characterized in that: The pre-processed historical power load data of each region is subjected to power load time series correlation feature extraction based on a multi-dimensional time scale to obtain a sequence of first time scale power load time series correlation feature vectors, a sequence of second time scale power load time series correlation feature vectors, and a sequence of third time scale power load time series correlation feature vectors, including: Performing one-dimensional convolution coding based on the first convolution layer on the preprocessed historical power load data of each region to obtain a sequence of time series correlation feature vectors of the power load at the first time scale; Performing one-dimensional convolutional coding based on a second convolutional layer on the preprocessed historical power load data of each region to obtain a sequence of time series correlation feature vectors of the power load at the second time scale; Performing one-dimensional convolution coding based on the third convolution layer on the preprocessed historical power load data of each region to obtain a sequence of time series correlation feature vectors of the power load at the third time scale; The scales of the first convolutional layer, the second convolutional layer and the third convolutional layer are 7×1, 30×1 and 90×1 respectively.

4. The method for controlling the dynamic distribution of electric load according to claim 3, characterized in that: The sequence of the first time scale electric load time series associated feature vectors, the sequence of the second time scale electric load time series associated feature vectors and the sequence of the third time scale electric load time series associated feature vectors are respectively subjected to contribution feature enhancement aggregation based on cluster analysis to obtain the first time scale electric load time series significant aggregation feature vector, the second time scale electric load time series significant aggregation feature vector and the third time scale electric load time series significant aggregation feature vector, including: Performing cluster analysis on the sequence of the first time scale power load time series correlation feature vectors to obtain a first time scale power load time series correlation feature self-supervised cluster representation vector; Based on the self-supervised clustering representation vector of the first time scale power load timing association feature, the sequence of the first time scale power load timing association feature vectors is subjected to feature clustering field modulation aggregation to obtain the first time scale power load timing significant aggregation feature vector.

5. The method for controlling the dynamic distribution of electric load according to claim 4, characterized in that: A cluster analysis is performed on the sequence of the first time scale electric load timing correlation feature vectors to obtain a first time scale electric load timing correlation feature self-supervised cluster representation vector, including: calculating the mean vector of the sequence of the first time scale electric load timing correlation feature vectors to obtain the first time scale electric load timing correlation feature self-supervised cluster representation vector.

6. The method for controlling the dynamic distribution of electric load according to claim 5, characterized in that: Based on the self-supervised clustering representation vector of the first time scale electric load time series correlation feature, a sequence of the first time scale electric load time series correlation feature vectors is subjected to feature clustering field modulation aggregation to obtain a significant aggregation feature vector of the first time scale electric load time series, including: Calculating the implicit clustering contribution factor of each first time scale power load time series association feature vector in the sequence of the first time scale power load time series association feature vector relative to the first time scale power load time series association feature self-supervised clustering representation vector to obtain a sequence of implicit clustering contribution factors of the first time scale power load time series association feature; Vectorizing the sequence of implicit clustering contribution factors of the first time scale electric load timing correlation feature to obtain a first time scale electric load timing correlation feature clustering contribution field distribution vector; Inputting the distribution vector of the cluster contribution field of the first time scale power load timing correlation feature into the cluster contribution field explicit modeling module based on the self-attention mechanism to obtain the first time scale power load timing correlation feature cluster contribution field modulation weight vector; Taking each eigenvalue in the first time scale power load timing association feature clustering contribution field modulation weight vector as weight, calculate the position weighted sum of the sequence of the first time scale power load timing association feature vector to obtain the first time scale power load timing significant aggregation feature vector.

7. The method for controlling the dynamic distribution of electric load according to claim 6, characterized in that: Calculating the implicit clustering contribution factor of each first time scale power load time series association feature vector in the sequence of the first time scale power load time series association feature vector relative to the first time scale power load time series association feature self-supervised clustering representation vector to obtain a sequence of implicit clustering contribution factors of the first time scale power load time series association feature, including: Calculating the absolute value of the positional division between the first time scale power load time series correlation feature vector and the first time scale power load time series correlation feature self-supervised clustering representation vector to obtain the first time scale power load time series self-supervised clustering interactive correlation feature vector; Calculate the logarithmic function value of each position characteristic value in the first time scale electric load time series self-supervised clustering interactive correlation characteristic vector with base 2 to obtain the first time scale electric load time series self-supervised clustering interactive logarithmic representation vector; Taking each position characteristic value in the self-supervised clustering interaction logarithm representation vector of the first time scale electric load time series as a weighting coefficient, calculating the weighted sum of each position characteristic value in the first time scale electric load time series association characteristic vector to obtain the first time scale electric load time series association characteristic clustering contribution degree representation factor; The exponential function value of the clustering contribution degree representation factor of the first time scale power load timing correlation feature is calculated with the natural constant e as the base to obtain the implicit clustering contribution factor of the first time scale power load timing correlation feature.

8. The method for controlling the dynamic distribution of electric load according to claim 7, characterized in that: Inputting the distribution vector of the cluster contribution field of the first time scale power load timing correlation feature into the cluster contribution field explicit modeling module based on the self-attention mechanism to obtain the first time scale power load timing correlation feature cluster contribution field modulation weight vector, including: Calculate the matrix multiplication between the first time scale power load time series correlation feature cluster contribution field distribution vector and the query weight matrix, key weight matrix and value weight matrix respectively to obtain the first time scale power load cluster contribution field distribution query vector, the first time scale power load cluster contribution field distribution key vector and the first time scale power load cluster contribution field distribution value vector; After calculating the multiplication between the query vector of the first time scale power load cluster contribution field distribution and the transposed vector of the first time scale power load cluster contribution field distribution key vector, the obtained matrix is ​​divided by the square root of the length of the first time scale power load cluster contribution field distribution key vector by position to obtain the query-key first time scale power load cluster contribution field semantic interaction representation matrix; The query-key first time scale power load cluster contribution field semantic interaction representation matrix is ​​normalized by using a Softmax function to obtain a normalized query-key first time scale power load cluster contribution field semantic interaction representation matrix; The multiplication between the distribution value vector of the first time scale power load cluster contribution field and the query-key first time scale power load cluster contribution field semantic interaction representation matrix is ​​calculated to obtain the first time scale power load timing correlation feature cluster contribution field domain modulation weight vector.

9. The method for controlling the dynamic distribution of electric load according to claim 8, characterized in that: Inputting the physical space topology matrix and the time series characteristics of the power loads in each region into the trained load forecasting model to obtain a load forecasting result, including: Inputting the physical space topology matrix and the multi-time scale power load time series significant aggregation feature vectors of each region into the graph convolution layer of the load forecasting model to obtain the spatial topology enhanced power load global time series joint feature matrix; The spatial topology enhanced electric load global time series joint characteristic matrix is ​​input into the decoding layer of the load forecasting model to obtain the load forecasting result.

10. A power load dynamic distribution control system according to the power load dynamic distribution control method according to any one of claims 1 to 9, characterized in that: include: Data collection module, used to collect historical power load data of various regions; A data preprocessing module, used for preprocessing the historical power load data of each region to obtain the preprocessed historical power load data of each region; A time series feature extraction module, used to extract the time series features of the power load of each region from the pre-processed historical power load data of each region; A physical space topology matrix construction module is used to construct a physical space topology matrix between the regions, wherein the value of each non-diagonal position in the physical space topology matrix is ​​used to represent the Euclidean distance between the corresponding two regions; A load forecasting module, used for inputting the physical space topology matrix and the time series characteristics of the power load in each area into the trained load forecasting model to obtain a load forecasting result; The scheduling module is used to schedule the power load according to the load forecast results and the actual situation of the current power system using an optimization algorithm, including generator set scheduling, energy storage device scheduling and transmission network scheduling.

Images