Power distribution system source load scene generation method

Abstract

The invention discloses a power distribution system source load scene generation method. The method comprises the steps of calculating correlation between new energy output and load according to preprocessed new energy output historical data and load historical data of a power distribution system, and generating an extended adjacency matrix; based on a pre-constructed message passing neural network model and a time sequence generative adversarial network model, determining a power distribution system source load scene generation model; training a power distribution system source load scene generation model by using the new energy output historical data, the load historical data and the extended adjacency matrix; and generating a source load scene according to the trained power distribution system source load scene generation model. The method can adapt to the change of the topological structure caused by network reconstruction of the power distribution system, effectively captures the source load characteristics of the reconstructed network, and enhances the flexibility and adaptability of the model. And the space-time correlation between the new energy output and the load can be simulated more accurately, and the precision of power distribution system source load scene generation is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution networks, and in particular, to a method for generating source-load scenarios of a distribution system. Background Art

[0002] With the continuous progress of renewable energy technologies and the rapid development of new types of loads such as electric vehicles, the distribution network is undergoing profound changes. The installed capacity of new energy is continuously climbing, and diverse flexible loads such as electric vehicles are also developing on a large scale. This has gradually transformed the traditional distribution network into an active distribution network, showing the characteristics of random fluctuations and multi-source interactions between sources and loads. This change has brought new challenges, especially the uncertainty of new energy generation and the dynamic change problem of load demand, significantly enhancing the uncertainty of the power system. The output of new energy generation is unstable due to weather conditions, and the access of flexible loads such as electric vehicles makes the load demand more variable and difficult to predict. These factors are intertwined, making the operation and scheduling of the power system extremely complex and also posing higher requirements for the stability and reliability of the power system.

[0003] Traditional source-load scenario generation methods ignore the internal relationship between new energy output and load, and it is difficult to accurately capture the spatio-temporal correlation between sources and loads, resulting in a deviation between the generated scenarios and the actual operation situation. Moreover, due to data limitations, traditional source-load scenario generation methods cannot accurately predict the source-load characteristics of the network after the reconstruction of the distribution system. Summary of the Invention

[0004] The present invention provides a method for generating source-load scenarios of a distribution system to solve the problems that the prior art is difficult to accurately capture the spatio-temporal correlation between sources and loads and cannot accurately predict the source-load characteristics of the network after the reconstruction of the distribution system.

[0005] According to one aspect of the present invention, there is provided a method for generating source-load scenarios of a distribution system, including:

[0006] Calculating the correlation between the new energy output and the load according to the historical data of new energy output and the historical data of load of the preprocessed distribution system, and generating an extended adjacency matrix;

[0007] Determining a source-load scenario generation model for the distribution system based on a pre-constructed message passing neural network model and a time series generative adversarial network model;

[0008] Training the source-load scenario generation model for the distribution system by using the historical data of new energy output, the historical data of load, and the extended adjacency matrix;

[0009] Generating source-load scenarios according to the trained source-load scenario generation model for the distribution system.

[0010] Optionally, calculating the correlation between the new energy output and the load based on the historical data of new energy output and the load history data of the preprocessed distribution system, and generating an extended adjacency matrix includes:

[0011] Sort the magnitudes of the observed values of the preprocessed new energy output and the load history data;

[0012] Calculate the Spearman rank correlation coefficient between the new energy output and the load according to the sorting sequence numbers of the observed values;

[0013] Generate an extended adjacency matrix according to the Spearman rank correlation coefficient and a preset threshold.

[0014] Optionally, calculating the Spearman rank correlation coefficient between the new energy output and the load according to the sorting sequence numbers of the observed values includes:

[0015] Calculate the Spearman rank correlation coefficient between the new energy output and the load using the following formula:

[0016]

[0017] where d i is the difference between the sorting sequence numbers corresponding to two observed values; n is the number of historical data.

[0018] Optionally, generating an extended adjacency matrix according to the Spearman rank correlation coefficient and a preset threshold includes:

[0019] Determine the source nodes with correlation as the nodes whose absolute value of the Spearman rank correlation coefficient is greater than or equal to the preset threshold;

[0020] Add the horizontal and vertical dimension information of the source nodes in a row and a column to the pre-established initial adjacency matrix to generate an extended adjacency matrix.

[0021] Optionally, determining the source-load scenario generation model of the distribution system based on the pre-constructed message passing neural network model and time series generative adversarial network model includes:

[0022] Use the output of the message passing neural network model as the input of the time series generative adversarial network model to determine the source-load scenario generation model of the distribution system.

[0023] Optionally, before determining the source-load scenario generation model of the distribution system based on the pre-constructed message passing neural network model and time series generative adversarial network model, it further includes:

[0024] Construct a message passing neural network model to extract the spatial dimension correlation features between new energy and load;

[0025] Build a time series generative adversarial network model to extract time dimension features.

[0026] Optionally, the building of the message passing neural network model to extract the spatial dimension correlation features between new energy and load includes:

[0027] Learn the topological structure and feature representation of the graph by exchanging and updating information between nodes;

[0028] Map the features of the entire graph into a feature vector describing the features of the whole graph; where, the feature vector should satisfy the following formula:

[0029]

[0030] In the formula, represents the feature vector of the whole graph; R represents the readout function; T represents the number of rounds of message passing; G represents the whole graph; represents the hidden state of node v after T rounds of message passing.

[0031] Optionally, learning the topological structure and feature representation of the graph by exchanging and updating information between nodes includes:

[0032] Define the hidden state of each node before message passing starts, where the hidden state should satisfy the following formula:

[0033] In the formula, represents the hidden state of node v before message passing starts; represents the sequence composed of the historical power data of the v-th node at n moments, which is the initial feature vector of node v;

[0034] Map the element values in the extended adjacency matrix into edge features in the graph structure;

[0035] Initialize the time step;

[0036] Generate the message to be sent according to the hidden states of each node and each neighbor node of each node, and the edge features between each node and the neighbor node, where the message should satisfy the following formula:

[0037] In the formula, m v→w is the message sent by node v to neighbor node w; and are the hidden states of node v and node w at time step t respectively; M is the message function; e vw is the edge feature between node v and node w;

[0038] Aggregate the messages received by each node from all its neighbor nodes, with the formula:

[0039] In the formula, is the message aggregated by node v, and N(v) is the set of neighbor nodes of node v;

[0040] Update the hidden state of the node according to the received information, with the formula:

[0041] In the formula, is the new hidden state of the node at time step t + 1; U is the state update function;

[0042] Calculate the change in the hidden state, and the change in the hidden state should satisfy the following formula:

[0043] In the formula, Δh v is the change in the hidden state of node v between time steps t and t + 1;

[0044] Calculate the average change in the change in the hidden state, and the average change should satisfy the following formula:

[0045] In the formula, is the average value of the changes in the hidden states of all nodes.

[0046] Optionally, constructing the time series generative adversarial network model to extract time dimension features includes:

[0047] Build an embedding network to map the high-dimensional feature vector to a low-dimensional space and obtain the information with the optimal value, where the embedding network function should satisfy the following formula:

[0048] where s is the static feature vector, x t is the dynamic feature vector, h s represents the low-dimensional static feature vector after being mapped by the embedding network, and h t represents the low-dimensional dynamic feature vector after being mapped by the embedding network;

[0049] Build a recovery network to reconstruct the low-dimensional feature vector into a high-dimensional feature vector, where the recovery network function should satisfy the following formula:

[0050] In the formula, is the static feature vector output by the recovery network, is the dynamic feature vector output by the recovery network;

[0051] Build a generator network, and the generator network function should satisfy the following formula:

[0052] In the formula, z s , z t is the noise of the generator, and are the static feature vector and the dynamic feature vector generated by the generator respectively;

[0053] Concatenate the output of the generator network and the output of the embedded network to obtain the hidden features as the input of the discriminator network;

[0054] Build a discriminator network, and the discriminator network function should satisfy the following formula:

[0055] In the formula, is the discrimination value, is the input of the discriminator network, d s , d x represents the classification function of the output layer of the discriminator network.

[0056] Optionally, the training of the source-load scenario generation model of the distribution system by using the new energy output historical data, the load historical data, and the extended adjacency matrix includes:

[0057] Import the new energy output historical data, the load historical data, and the extended adjacency matrix into the optimizer, set the activation function and initialize the network parameters;

[0058] Construct a loss function training network according to the structure input of the source-load scenario generation model of the distribution system; among them, the loss function of the message passing neural network should satisfy the following formula:

[0059] In the formula, L MPNN is the loss function of the message passing neural network, y i is the true value of node i, p i is the predicted value of node i, and n is the number of samples;

[0060] The joint loss function of the embedded network and the recovery network should satisfy the following formula:

[0061] In the formula, L R is the joint loss function of the embedded network and the recovery network;

[0062] The joint loss function of the generator and the discriminator should satisfy the following formula:

[0063] In the formula, L gd is the joint loss function of the generator and the discriminator;

[0064] The supervision loss function should satisfy the following formula:

[0065] In the formula, L eg is the supervision loss function;

[0066] Determine whether the source-load scenario generation model of the distribution system is completed according to each loss function and preset conditions, and save the trained model parameters when the source-load scenario generation model of the distribution system is completed.

[0067] An embodiment of the present invention provides a method for generating a source-load scenario of a distribution system, including: calculating the correlation between new energy output and load according to the historical data of new energy output and the historical data of load of the preprocessed distribution system, and generating an extended adjacency matrix; determining a source-load scenario generation model of the distribution system based on a pre-constructed message passing neural network model and a time series generative adversarial network model; training the source-load scenario generation model of the distribution system by using the historical data of new energy output, the historical data of load and the extended adjacency matrix; generating a source-load scenario according to the trained source-load scenario generation model of the distribution system. The technical solution provided by the embodiment of the present invention incorporates factors related to new energy output and factors related to load as "elements" into the matrix construction on the basis of the traditional adjacency matrix concept, constructs an extended adjacency matrix, so as to more comprehensively display the complex relationship and respective characteristics between new energy output and load, can adapt to the topological structure changes brought about by the network reconstruction of the power system, generate diversified operation scenarios, expand the sample data set, effectively capture the source-load characteristics of the reconstructed network, and enhance the flexibility and adaptability of the model. Since the message passing neural network model can extract the spatial dimension correlation features between new energy and load, and the time series generative adversarial network model can extract the time dimension features, and the combination of the two constitutes the source-load scenario generation model of the distribution system, which includes the spatio-temporal correlation between new energy output and load, so it can more accurately simulate the spatio-temporal correlation between new energy output and load and improve the accuracy of generating the source-load scenario of the distribution system.

[0068] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present invention, nor is it used to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. Description of the Drawings

[0069] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained according to these drawings.

[0070] Figure 1 Flow chart of a method for generating source-load scenarios of a power distribution system provided by an embodiment of the present invention;

[0071] Figure 2 Flow chart of another method for generating source-load scenarios of a power distribution system provided by an embodiment of the present invention;

[0072] Figure 3 Structural schematic diagram of a device for generating source-load scenarios of a power distribution system provided by an embodiment of the present invention;

[0073] Figure 4 Structural schematic diagram of an electronic device for a method for generating source-load scenarios of a power distribution system provided by an embodiment of the present invention. Detailed implementation manners

[0074] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0075] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0076] Figure 1 Flow chart of a method for generating source-load scenarios of a power distribution system provided by an embodiment of the present invention. This method can be executed by a device for generating source-load scenarios of a power distribution system. The device for generating source-load scenarios of a power distribution system can be implemented in the form of hardware and / or software, and the device for generating source-load scenarios of a power distribution system can be configured in any electronic device with communication functions. Refer to Figure 1 and the method for generating source-load scenarios of a power distribution system includes:

[0077] S110. Calculate the correlation between new energy output and load according to the historical data of new energy output and the historical data of load of the preprocessed power distribution system, and generate an extended adjacency matrix.

[0078] Among them, the historical data of new energy output includes power generation power data, power generation volume data, equipment operation data, meteorological and environmental data, grid connection and consumption data; the power generation power data includes wind power, photovoltaic power, and hydropower; the power generation volume data includes daily power generation, monthly power generation, and annual power generation; the equipment operation data includes the operation status of wind turbines, such as parameters like the start and stop time, operating speed, pitch angle, nacelle temperature, and gearbox oil temperature of the wind turbine; the operation status of photovoltaic modules, such as parameters like the open-circuit voltage, short-circuit current, fill factor, photoelectric conversion efficiency, and surface temperature of the photovoltaic module; the operation status of hydropower equipment, such as parameters like the speed, flow rate, water head, and guide vane opening of the water turbine, as well as electrical parameters like the voltage, current, and power factor of the generator; the meteorological and environmental data includes parameters like wind speed and direction, light intensity and sunshine duration, temperature and humidity. The historical load data includes power load data, such as active power load, reactive power load, and apparent power load; time dimension data, such as daily load data, weekly load data, monthly load data, and annual load data, etc.; industry and user classification data, such as industrial load data: historical records of electricity consumption loads for different industrial fields, such as industries like steel, chemical, and electronics; commercial load data: covering the electricity consumption load situations of commercial places such as shopping malls, hotels, and office buildings; residential load data: reflecting the changes in electricity consumption loads of residential households, which is closely related to the living habits of residents. For example, the electricity consumption load will increase during periods such as cooking and watching TV at night; agricultural load data: including the load situations of electricity consumption related to agriculture such as farmland irrigation, agricultural product processing, and breeding. The extended adjacency matrix is a matrix representation form further expanded based on the concept of the adjacency matrix in graph theory. For a graph (which can be an undirected graph or a directed graph), the traditional adjacency matrix is a matrix used to describe the connection relationships between vertices in the graph. If the graph has n vertices, then the adjacency matrix is an n*n matrix A, where the element a ij represents the situation of the edge between vertex i and vertex j, and the extended adjacency matrix, on the basis of the adjacency matrix, adds more characteristic information about the graph. Commonly, some attributes of the vertices themselves and other attributes of the edges are also incorporated into the matrix representation, enabling the matrix to more comprehensively reflect the structure and related properties of the graph.

[0079] Specifically, in an actual power distribution system, there are often some problems with the historical data of new energy output and load history data obtained. For example, the data may have missing values, outliers, or the data formats, dimensions, etc. are not unified. Preprocessing is to perform a series of operations such as data cleaning and data normalization on these original data, so that the data reaches a cleaner and more standardized state for subsequent accurate correlation analysis and other operations. There is a correlation between new energy output and load. For example, when there is sufficient sunlight during the day, the photovoltaic output increases, and at the same time, the electricity loads of residents and industrial and commercial users may also be at a relatively high level due to various production and living activities, and the two may show a positive correlation. Or when the wind power output is relatively large at night, the load may be at a low valley compared to the day, showing a certain negative correlation. Therefore, the Pearson correlation coefficient or the Spearman rank correlation coefficient can be calculated to characterize the correlation between new energy output and load. For example, the Pearson correlation coefficient measures the degree of linear correlation between two variables (here, new energy output and load) by calculating the ratio of the covariance of the two variables to the product of their standard deviations. Its value range is between -1 and 1. A value close to 1 indicates a strong positive correlation, a value close to -1 indicates a strong negative correlation, and a value close to 0 indicates a weak correlation. By calculating the correlation coefficients of different time periods, different new energy types and various loads, the degree of connection between them can be clearly understood. The traditional adjacency matrix is often used to describe the connection relationship between vertices in a graph structure. For example, in a graph represented by a simple power grid topology structure, the vertices represent each node (such as a substation, a distribution room, etc.), the edges represent the line connections, and the matrix elements reflect whether there is a connection between nodes and the connection situation (such as directed / undirected edges, edge weights, etc.). In the context of correlation analysis based on new energy output and load, the extended adjacency matrix integrates new energy output and load into an adjacency matrix on the basis of the traditional adjacency matrix concept. For example, the rows and columns of the matrix can correspond to different new energy power generation units (such as each wind farm, photovoltaic power station, etc.) and load nodes, and the elements in the matrix can reflect the correlation between new energy output and load, thus constructing a matrix that can reflect the connection situation between new energy and load in the power distribution system.

[0080] S120. Determine the source-load scenario generation model of the power distribution system based on the pre-constructed message passing neural network model and the time series generative adversarial network model.

[0081] Among them, the Message Passing Neural Network (MPNN) is a deep learning model specifically designed for processing graph-structured data. In a distribution system, its structure can be regarded as a graph composed of numerous nodes (such as substations, distribution rooms, distributed power generation access points, etc.) and edges (representing power line connections, etc.). The core idea of MPNN is to update the feature representation of nodes through information passing between nodes. In each iteration, it allows each node to collect "messages" from its adjacent nodes, and then updates its own state based on these messages. After multiple rounds of such processes, finally each node can comprehensively consider the influence of surrounding nodes and obtain a feature representation that integrates local and global information. In the analysis of the distribution system, it can be used to capture the complex interrelationships between different electrical devices (corresponding to nodes), such as how the access of distributed power generation affects the surrounding load conditions, and the power interaction between different substations, etc., so as to effectively model and analyze the operating state of the distribution system. The Time Generative Adversarial Network (TimeGAN) is a variant of GAN specifically designed for time series data. It takes into account the characteristics of time series data such as sequentiality and periodicity, and can learn the internal distribution law of time series data, so as to generate new time series data that conforms to this law. The data such as new energy output and load in the distribution system are often time series data with obvious time-varying characteristics. For example, the load shows fluctuations such as morning and evening peaks within a day, and the wind power output is affected by wind speed changes over time. TimeGAN can learn the distribution law of these data, and then simulate and generate various time series scenarios that may occur in the future, providing a reference for the planning, scheduling, etc. of the distribution system.

[0082] Specifically, based on the previously pre-constructed Message Passing Neural Network model (used to process the graph-structured information of the distribution system and depict the structural relationship between sources and loads) and the Time Generative Adversarial Network model (used to learn the time series change law of source-load data and generate time series data that conforms to the law), combine the advantages of the two. Through reasonable integration methods, such as using the structural features extracted by MPNN as part of the input of TimeGAN, or further fusing the outputs of the two, etc., construct a model specifically for generating source-load scenarios of the distribution system. This model can simulate source-load scenarios under different conditions based on the existing operating data of the distribution system, such as simulating complex scenarios where new energy output sharply decreases and load peaks occur under extreme weather, or scenarios when new energy is booming and the load is at a low valley, etc., providing a powerful analysis tool for many aspects such as the stable operation, optimal scheduling, and reliability assessment of the distribution system.

[0083] S130. Use the historical data of new energy output, the historical data of load, and the extended adjacency matrix to train the source-load scenario generation model of the distribution system.

[0084] Specifically, integrate the historical data of new energy output, the historical data of load, and the extended adjacency matrix into the overall data input system in a format that meets the model input standard, so that the source-load scenario generation model of the distribution system can obtain the information contained in these data. For example, the extended adjacency matrix can be directly input in matrix form or the key feature vectors can be extracted for input; the training objective is to make the generated source-load scenarios as close to the real situation as possible, that is, to hope that the combination of the new energy output situation and the load situation generated by the model conforms to the scenarios that actually occurred in history in terms of data distribution, mutual relationship, etc. During the training process, the model performs operations based on the given input data (the historical data of new energy output, the historical data of load, and the extended adjacency matrix), evaluates the quality of the generated results through a loss function (such as a function that measures the difference between the generated scenario and the real scenario, like the mean square error, etc.), and then uses an optimization algorithm (such as gradient descent, etc.) to continuously adjust the internal parameters of the model to reduce the loss value, so that the model is continuously optimized and the generated source-load scenarios are more and more in line with the actual situation. After multiple rounds of such iterative training, the model can finally better master the internal relationship and variation law between the source and load of the distribution system, and thus have the ability to accurately generate different source-load scenarios.

[0085] S140. Generate source-load scenarios according to the trained source-load scenario generation model of the distribution system.

[0086] Specifically, by using the historical data of new energy output, load historical data, and the extended adjacency matrix for training, the trained distribution system source-load scenario generation has learned the internal relationships and laws between the sources and loads in the distribution system. For example, it may have learned the power generation variation laws of different types of new energy (such as wind power, photovoltaic power, etc.) under different seasons, weather conditions, and times, as well as the load demand laws of different time periods and different types of users (such as industrial, commercial, residential, etc.), and includes the correlation between sources and loads, the connections between different attributes, etc. During the training process, the internal parameters have been optimized and adjusted. These parameters determine the behavior and output of the trained distribution system source-load scenario generation model. They are obtained by continuously adjusting according to the training data through optimization algorithms (such as the gradient descent algorithm, etc.) in a way that minimizes the loss function (used to measure the difference between the generation results of the trained distribution system source-load scenario generation model and the real historical data). The trained distribution system source-load scenario generation model can generate outputs closer to the actual situation when corresponding information is input. Different information can be input into the trained distribution system source-load scenario generation model. This information may include some basic conditions, such as the current time, date, season, weather conditions (such as predicted wind speed, light intensity, etc.), or some initial system state information (such as the current grid voltage, frequency, some existing source-load data, etc.). The specific input information depends on the model design and application requirements. After the required information is input, the trained distribution system source-load scenario generation model, based on its internal structure and optimized parameters, uses the source-load relationships and laws it has learned, and through a series of calculation and data processing operations, generates one or more source-load scenarios. These scenarios are predictions and simulations of the possible future source-load states of the distribution system, providing valuable decision-making support information for aspects such as the planning, operation, management, and guarantee of the distribution system.

[0087] The technical solution provided by the embodiments of the present invention incorporates factors related to new energy output and load-related factors as "elements" into the matrix construction on the basis of the traditional adjacency matrix concept to generate an extended adjacency matrix, thereby more comprehensively demonstrating the complex relationships and respective characteristics between new energy output and load, being able to adapt to the topological structure changes brought about by the network reconstruction of the power system, generating diverse operation scenarios, expanding the sample data set, effectively capturing the source-load characteristics of the reconstructed network, and enhancing the flexibility and adaptability of the model. Since the message passing neural network model can extract the spatial dimension correlation features between new energy and load, and the time series generative adversarial network model can extract the time dimension features, and the combination of the two constitutes the distribution system source-load scenario generation model, which includes the spatio-temporal correlation between new energy output and load, so it can more accurately simulate the spatio-temporal correlation between new energy output and load and improve the accuracy of distribution system source-load scenario generation.

[0088] Optionally, according to the historical data of new energy output and load history data of the preprocessed power distribution system, calculate the correlation between new energy output and load, and generate an extended adjacency matrix including:

[0089] Sort the magnitudes of the observed values of the preprocessed new energy output and load history data.

[0090] Among them, the observed value refers to the specific data points in the historical data of new energy output and load history data after preprocessing. For the historical data of new energy output, the observed value can be the power generation of a wind farm at a certain moment, the real-time power generation of a photovoltaic power station; for the load history data, the observed value can be the total load at a certain moment, the active power of a certain type of user, etc.

[0091] Specifically, sort the observed values of the historical data of new energy output. For example, sort the observed values of the power generation of wind power from small to large or from large to small over a period of time to intuitively see the strength of the power generation capacity at different times and find the maximum output period and the minimum output period. For the data of photovoltaic power stations, the power generation situation at different times can be understood after sorting, and the time when the maximum or minimum power generation is reached can be found. Sorting the observed values of the load history data can help us find the peak and valley values of the power consumption load. Sorting from large to small can quickly find the moment of the largest historical power consumption load, while sorting from small to large can find the moment of the smallest power consumption load. This is very useful for analyzing the operating state of the power system, understanding the distribution range and fluctuation of the load. Common sorting algorithms such as bubble sort, quick sort, and merge sort can be used to store the observed values in an array or list and arrange them in ascending or descending order according to the logic of different algorithms. For example, bubble sort gradually "floats" the largest or smallest element to one end of the array by repeatedly comparing and swapping adjacent elements; quick sort selects a pivot element, places the elements smaller than it on the left and the elements larger than it on the right, and then recursively sorts the left and right sub-arrays; merge sort continuously splits the array into smaller arrays and then merges the ordered sub-arrays into a larger ordered array. By sorting the magnitudes of the observed values of the preprocessed historical data of new energy output and load history data, data can be analyzed from different perspectives, which helps to discover information such as the extreme values, distribution characteristics, and fluctuation ranges of new energy output and load, providing a basis for further analyzing the performance of the power system, formulating reasonable dispatching strategies, and evaluating the stability of the system. For example, by observing the sorting results of new energy output, the high-output periods of new energy output can be found, and the charging and discharging operations of energy storage systems can be reasonably arranged; by the load sorting results, preparations can be made in advance to cope with peak loads, adjust the power generation plan or take corresponding grid regulation measures to ensure the stable operation of the power system.

[0092] Calculate the Spearman rank correlation coefficient between the new energy output and the load according to the sorting serial numbers of the observed values.

[0093] Specifically, the following formula is used to calculate the Spearman rank correlation coefficient between the new energy output and the load:

[0094]

[0095] In the formula, d i is the difference between the sorting serial numbers corresponding to two observed values; n is the number of historical data.

[0096] Generate an extended adjacency matrix according to the Spearman rank correlation coefficient and a preset threshold.

[0097] Optionally, generating an extended adjacency matrix according to the Spearman rank correlation coefficient and a preset threshold includes: determining the nodes with the absolute value of the Spearman rank correlation coefficient greater than or equal to the preset threshold as the source nodes with correlation; adding the horizontal and vertical dimension information of the source nodes in a row and a column to the initially established initial adjacency matrix to generate an extended adjacency matrix.

[0098] Specifically, first, according to the network topology structure of the power system, establish the initial adjacency matrix among the load nodes to represent the physical connection relationship; subsequently, the wind-solar nodes with the absolute value of the Spearman rank correlation coefficient |ρ| exceeding the preset threshold ρ s are identified as the source nodes with significant correlation, add a row and a column to it in the adjacency matrix to generate an extended adjacency matrix, and mark the corresponding elements as 1 in the extended adjacency matrix.

[0099] Optionally, determining the source-load scenario generation model of the distribution system based on the pre-constructed message passing neural network model and the time series generative adversarial network model includes:

[0100] Taking the output of the message passing neural network model as the input of the time series generative adversarial network model to determine the source-load scenario generation model of the distribution system.

[0101] Among them, the node numbers of different nodes are used as the static features input of the embedding network; the feature vectors output by the MPNN are used as the dynamic features input of the embedding network.

[0102] Specifically, by using the output of the message passing neural network model as the input of the time series generative adversarial network model, the structural information and time series information of the distribution system can be combined to form a more powerful and practical distribution system source-load scenario generation model, providing more accurate and comprehensive support for the analysis, optimization, and decision-making of the distribution system. For example, when simulating the operating states of the power system under different network structures and environments, this model can help predict the impacts of different source-load scenarios on the power grid, formulate corresponding countermeasures in advance, and ensure the stable and efficient operation of the power grid.

[0103] Figure 2 FIG. is a flowchart of another distribution system source-load scenario generation method provided by an embodiment of the present invention. The embodiment of the present invention further refines the foregoing embodiments on the basis of the above embodiments. Refer to Figure 2 and the distribution system source-load scenario generation method includes:

[0104] S210. Preprocess the historical data of new energy output and load historical data of the distribution system.

[0105] Specifically, in an actual distribution system, the obtained historical data of new energy output and load historical data often have some problems. For example, the data may have missing values, outliers, or the data formats, dimensions, etc. are not unified. Preprocessing is to perform a series of operations such as data cleaning and data normalization on these original data to make the data reach a cleaner and more standardized state, so as to accurately perform correlation analysis and other operations subsequently.

[0106] S220. Calculate the correlation between the new energy output and the load according to the preprocessed historical data of new energy output and load historical data of the distribution system, and generate an extended adjacency matrix.

[0107] S230. Construct a message passing neural network model to extract the spatial dimension correlation features between new energy and load.

[0108] Specifically, learn the topological structure and feature representation of the graph in the exchange and update of information between nodes; map the features of the entire graph into a feature vector describing the features of the whole graph; where the feature vector should satisfy the following formula:

[0109] In the formula, represents the feature vector of the whole graph; R represents the readout function; T represents the number of rounds of message passing; G represents the entire graph; represents the hidden state of node v after T rounds of message passing.

[0110] Optionally, learning the topological structure and feature representation of the graph in the exchange and update of information between nodes includes:

[0111] Define the hidden state of each node before the start of message passing, where the hidden state should satisfy the following formula:

[0112] In the formula, represents the hidden state of node v before the start of message passing; represents the sequence of historical power data of the v-th node at n moments, which is the initial feature vector of node v.

[0113] Map the element values in the extended adjacency matrix to the edge features in the graph structure.

[0114] Initialize the time step t = 0.

[0115] Generate the message to be sent according to the hidden states of each node and its neighbor nodes, and the edge features between each node and its neighbor nodes, where the message should satisfy the following formula:

[0116] In the formula, m v→w is the message sent by node v to neighbor node w; and are the hidden states of node v and node w at time step t respectively; M is the message function; e vw is the edge feature between node v and node w.

[0117] Aggregate the messages received by each node from all its neighbor nodes. The formula is:

[0118] In the formula, is the aggregated message of node v, and N(v) is the set of neighbor nodes of node v.

[0119] Update the hidden state of the node according to the received information. The formula is:

[0120] In the formula, is the new hidden state of the node at time step t + 1; U is the state update function.

[0121] Calculate the change in the hidden state. The change in the hidden state should satisfy the following formula:

[0122] In the formula, Δh v is the change in the hidden state of node v between time steps t and t + 1.

[0123] Calculate the average change in the change in the hidden state. The average change should satisfy the following formula:

[0124] In the formula, It is the average value of the change in the hidden state of all nodes. If or t = T, that is, the termination condition is met, then go to the step of mapping the features of the entire graph into a feature vector describing the features of the whole graph; otherwise, t = t + 1, go to the step of generating the message to be sent according to the hidden states of each node and its neighbor nodes, and the edge features between each node and its neighbor nodes; where ε represents the change threshold, and when the change in the hidden state of the node is lower than this threshold, it is considered that the model has tended to be stable; T is the preset maximum number of iterations.

[0125] S240. Construct a time series generative adversarial network model to extract time dimension features.

[0126] Specifically, build an embedding network to map the high-dimensional feature vector into a low-dimensional space to obtain the information with the optimal value, where the embedding network function should satisfy the following formula:

[0127] where s is the static feature vector, x t is the dynamic feature vector, h s represents the low-dimensional static feature vector after being mapped by the embedding network, and h t represents the low-dimensional dynamic feature vector after being mapped by the embedding network.

[0128] Build a recovery network to reconstruct the low-dimensional feature vector into a high-dimensional feature vector, where the recovery network function should satisfy the following formula:

[0129] In the formula, is the static feature vector output by the recovery network, is the dynamic feature vector output by the recovery network.

[0130] Build a generator network, and the generator network function should satisfy the following formula:

[0131] In the formula, z s , z t are the noises of the generator, and are respectively the static feature vector and the dynamic feature vector generated by the generator.

[0132] Concatenate the output of the generator network with the output of the embedding network to obtain the hidden features as the input of the discriminator network.

[0133] Build a discriminator network, and the discriminator network function should satisfy the following formula:

[0134] In the formula, is the discrimination value, is the input of the discriminator network, ds , d x represents the classification function of the output layer of the discriminator network.

[0135] S250. Determine the source-load scenario generation model of the distribution system based on the pre-constructed message passing neural network model and the time series generative adversarial network model.

[0136] S260. Use the historical data of new energy output, the historical data of load, and the extended adjacency matrix to train the source-load scenario generation model of the distribution system.

[0137] Specifically, import the historical data of new energy output, the historical data of load, and the extended adjacency matrix into the optimizer, set the activation function and initialize the network parameters;

[0138] Construct a loss function to train the network according to the structural input of the source-load scenario generation model of the distribution system; among them, the loss function of the message passing neural network should satisfy the following formula:

[0139] In the formula, L MPNN is the loss function of the message passing neural network, y i is the true value of node i, p i is the predicted value of node i, and n is the number of samples;

[0140] The combined loss function of the embedded network and the recovery network should satisfy the following formula:

[0141] In the formula, L R is the combined loss function of the embedded network and the recovery network;

[0142] The combined loss function of the generator and the discriminator should satisfy the following formula:

[0143] In the formula, L gd is the combined loss function of the generator and the discriminator;

[0144] The supervised loss function should satisfy the following formula:

[0145] In the formula, L eg is the supervised loss function;

[0146] Determine whether the source-load scenario generation model of the distribution system is completed according to each loss function and the preset conditions, and save the trained model parameters when the source-load scenario generation model of the distribution system is completed.

[0147] Specifically, if the loss function L of the message passing neural network MPNN , the combined loss function L of the embedded network and the recovery network RThe combined loss function L of the generator and discriminator gd and the supervised loss function L eg are both not less than the preset threshold of the loss value, or the number of iterations is not greater than the maximum preset number of iterations, then backpropagation is performed to update the network parameters, and the number of iterations is incremented by one; if the loss function L of the message passing neural network MPNN The combined loss function L of the embedded network and the recovery network R The combined loss function L of the generator and discriminator gd and the supervised loss function L eg are both less than the preset threshold of the loss value, or the number of iterations exceeds the maximum preset number of iterations, then the training is completed and the model parameters are saved.

[0148] S270. Generate source-load scenarios according to the trained source-load scenario generation model of the power distribution system.

[0149] The technical solution provided by the embodiment of the present invention can more accurately simulate the spatio-temporal correlation and source-load correlation between new energy output and load, thereby improving the accuracy of source-load scenario generation; it can also adapt to the changes brought about by the network reconstruction of the power system, effectively capture the source-load characteristics of the reconstructed network, and enhance the flexibility and adaptability of the model.

[0150] Figure 3 It is a schematic structural diagram of a source-load scenario generation device for a power distribution system provided by an embodiment of the present invention. Refer to Figure 3 , the source-load scenario generation device for the power distribution system includes: a calculation module 310, a determination module 320, a training module 330, and a generation module 340.

[0151] The calculation module 310 is used to calculate the correlation between new energy output and load according to the historical data of new energy output and the historical data of load of the preprocessed power distribution system, and generate an extended adjacency matrix.

[0152] The determination module 320 is used to determine the source-load scenario generation model of the power distribution system based on the pre-constructed message passing neural network model and the time series generative adversarial network model.

[0153] The training module 330 is used to train the source-load scenario generation model of the power distribution system by using the historical data of new energy output, the historical data of load, and the extended adjacency matrix.

[0154] The generation module 340 is used to generate source-load scenarios according to the trained source-load scenario generation model of the power distribution system.

[0155] The source-load scenario generation device provided by the embodiment of the present invention can execute the source-load scenario generation method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the method.

[0156] Figure 4 Schematic diagram of the structure of an electronic device for a method of generating source-load scenarios in a power distribution system provided by an embodiment of the present invention. The electronic device is intended to represent various forms of digital computers, such as, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as, personal digital processors, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present invention described herein and / or claimed.

[0157] As Figure 4 shown, the electronic device 10 includes at least one processor 11, and a memory communicatively connected to the at least one processor 11, such as read-only memory (ROM) 12, random access memory (RAM) 13, etc. Among them, the memory stores a computer program executable by the at least one processor. The processor 11 can execute various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, ROM 12, and RAM 13 are connected to each other through a bus 14. The input / output (I / O) interface 15 is also connected to the bus 14.

[0158] Multiple components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0159] The processor 11 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as a method of generating source-load scenarios in a power distribution system.

[0160] In some embodiments, the method for generating a power distribution system source-load scenario can be implemented as a computer program tangibly embodied in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the method for generating a power distribution system source-load scenario described above can be performed. Alternatively, in other embodiments, the processor 11 can be configured to execute the method for generating a power distribution system source-load scenario by any other suitable means (e.g., by means of firmware).

[0161] The various embodiments of the systems and techniques described above in this document can be implemented in digital electronic circuitry, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on a chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special-purpose or general-purpose programmable processor that receives data and instructions from a storage system, at least one input device, and at least one output device, and transmits the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0162] The computer programs for implementing the methods of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the computer programs are executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer programs can be executed entirely on the machine, partially on the machine, as a stand-alone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0163] In the context of the present invention, a computer-readable storage medium can be a tangible medium that can contain or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. The computer-readable storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, the computer-readable storage medium can be a machine-readable signal medium. More specific examples of the machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0164] To provide for interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the electronic device. Other kinds of devices can also be used to provide for interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0165] The systems and techniques described herein can be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0166] A computing system may include a client and a server. The client and the server are generally far from each other and usually interact via a communication network. The client-server relationship is created by computer programs running on respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system, solving the defects of difficult management and weak business scalability existing in traditional physical hosts and VPS services.

[0167] It should be understood that various forms of the processes shown above can be used, steps can be reordered, added, or deleted. For example, the steps recited in the present invention can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved, and no limitation is made herein.

[0168] The above specific embodiments do not constitute a limitation on the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for generating source-load scenarios in a power distribution system, characterized in that: include: According to the preprocessed new energy output historical data and load historical data of the power distribution system, the correlation between the new energy output and the load is calculated to generate an extended adjacency matrix; Determine the distribution system source-load scenario generation model based on the pre-built message passing neural network model and time series generative adversarial network model; Using the new energy output historical data, the load historical data and the extended adjacency matrix, training the distribution system source-load scenario generation model; Generate source-load scenarios based on the trained distribution system source-load scenario generation model.

2. The method for generating source-load scenarios of a power distribution system according to claim 1, characterized in that: The step of calculating the correlation between the new energy output and the load based on the pre-processed new energy output historical data and load historical data of the power distribution system to generate an extended adjacency matrix includes: Sorting the order of magnitude of the observed values ​​of the new energy output and the load historical data after preprocessing; Calculate the Spearman rank correlation coefficient between the new energy output and the load according to the sorting sequence number of the observation value; An extended adjacency matrix is ​​generated according to the Spearman rank correlation coefficient and a preset threshold.

3. The method for generating source-load scenarios of a power distribution system according to claim 2, characterized in that: The calculating of the Spearman rank correlation coefficient between the new energy output and the load according to the sorting sequence number of the observation value comprises: The Spearman rank correlation coefficient between the new energy output and the load is calculated using the following formula: Where, d i is the difference between the sorting numbers corresponding to the two observations; n is the number of historical data.

4. The method for generating source-load scenarios of a power distribution system according to claim 2, characterized in that: The step of generating an extended adjacency matrix according to the Spearman rank correlation coefficient and a preset threshold comprises: Determine the nodes whose absolute values ​​of the Spearman rank correlation coefficients are greater than or equal to the preset threshold as source nodes with correlation; A row and a column of horizontal and vertical dimension information of the source node are added to the pre-established initial adjacency matrix to generate an extended adjacency matrix.

5. The method for generating source-load scenarios of a power distribution system according to claim 1, characterized in that: The method of determining the distribution system source-load scenario generation model based on the pre-built message passing neural network model and the time series generative adversarial network model includes: The output of the message passing neural network model is used as the input of the time series generative adversarial network model to determine the source-load scenario generation model of the power distribution system.

6. The method for generating source-load scenarios of a power distribution system according to claim 1, characterized in that: Before determining the distribution system source-load scenario generation model based on the pre-built message passing neural network model and the time series generative adversarial network model, the method further includes: A message passing neural network model is constructed to extract the spatial dimension correlation features between new energy and loads; Construct a time series generative adversarial network model to extract time dimension features.

7. The method for generating source-load scenarios of a power distribution system according to claim 6, characterized in that: The construction of a message passing neural network model to extract spatial dimension correlation features between new energy and loads includes: Learn the topological structure and feature representation of the graph by exchanging and updating information between nodes; The features of the entire image are mapped into a feature vector that describes the features of the entire image; wherein the feature vector should satisfy the following formula: In the formula, represents the feature vector of the entire graph; R represents the readout function; T represents the number of rounds of message passing; G represents the entire graph; Represents the hidden state of node v after T rounds of message passing.

8. The method for generating source-load scenarios of a power distribution system according to claim 7, characterized in that: Learning the topological structure and feature representation of the graph in exchanging and updating information between nodes includes: Define the hidden state of each node before message passing, where the hidden state should satisfy the following formula: In the formula, represents the hidden state of node v before message passing; Represents the sequence of historical power data of the vth node at n moments, which is the initial feature vector of node v; Map the element values ​​in the extended adjacency matrix to the edge features in the graph structure; Initialize the time step; A message to be sent is generated according to the hidden state of each node and each neighbor node of the node, and the edge features between each node and the neighbor node, wherein the message should satisfy the following formula: In the formula, m v→w is the message sent by node v to its neighbor node w; and are the hidden states of node v and node w at time step t; M is the message function; e vw is the edge feature between node v and node w; Aggregate the messages received by each node from all of its neighboring nodes, the formula is: In the formula, is the aggregated message of node v, N(v) is the set of neighbor nodes of node v; Update the hidden state of the node according to the received information. The formula is: In the formula, is the new hidden state of the node at time step t+1; U is the state update function; Calculate the hidden state change, which should satisfy the following formula: In the formula, Δh v is the hidden state change of node v between time steps t and t+1; The average change of the hidden state change is calculated, and the average change should satisfy the following formula: In the formula, It is the average value of hidden state change of all nodes.

9. The method for generating source-load scenarios of a power distribution system according to claim 6, characterized in that: The constructing of a time series generative adversarial network model to extract time dimension features includes: Build an embedded network to map high-dimensional feature vectors to low-dimensional space to obtain information of optimal value, where the embedded network function should satisfy the following formula: Among them, s is the static eigenvector, x t is the dynamic feature vector, h s represents the low-dimensional static feature vector after embedding network mapping, h t Represents the low-dimensional dynamic feature vector after embedding network mapping; Build a recovery network to reconstruct the low-dimensional feature vector into a high-dimensional feature vector, where the recovery network function should satisfy the following formula: In the formula, is the static feature vector output by the recovery network, is the dynamic feature vector output by the recovery network; Build a generator network, the generator network function should satisfy the following formula: The static feature vector and dynamic feature vector formed; Concatenate the output of the generator network with the output of the embedded network to obtain hidden features as the input of the discriminator network; Build a discriminator network, the discriminator network function should satisfy the following formula: In the formula, is the discriminant value, is the input of the discriminator network, d s d x Represents the classification function of the discriminator network output layer.

10. The method for generating source-load scenarios of a power distribution system according to claim 1, characterized in that: The training of the power distribution system source-load scenario generation model using the new energy output historical data, the load historical data and the extended adjacency matrix includes: Importing the new energy output historical data, the load historical data and the extended adjacency matrix into the optimizer, setting the activation function and initializing the network parameters; According to the structural input of the distribution system source-load scenario generation model, a loss function training network is constructed; wherein the message passing neural network loss function should satisfy the following formula: Where, L MPNN is the message passing neural network loss function, y i is the true value of node i, p i is the predicted value of node i, n is the number of samples; The joint loss function of the embedded network and the restored network should satisfy the following formula: Where, L R is the joint loss function of the embedded network and the restored network; The joint loss function of the generator and the discriminator should satisfy the following formula: Where, L gd is the joint loss function of the generator and the discriminator; The supervision loss function should satisfy the following formula: Where, L eg is the supervision loss function; Determine whether the power distribution system source-load scenario generation model has completed training according to each loss function and preset conditions, and save the trained model parameters when the power distribution system source-load scenario generation model has completed training.

Images