Flexible load optimization scheduling method and device combined with load type analysis

Abstract

The present invention discloses a flexible load optimization scheduling method and device combined with load type analysis, which relates to the field related to smart grid scheduling. The method includes: obtaining the physical structure of the power grid, constructing a topological network, and generating a distribution topological network; classifying multiple distribution topological nodes into load types; obtaining multiple historical load data, constructing and generating M edge load prediction models; synchronizing the M edge load prediction models to a local server; pre-constructing M flexible load scheduling nodes; obtaining M load scheduling update cycles based on multiple historical load data analysis; performing load prediction, obtaining M groups of flexible load demands; and performing near-time load distribution updates. The method solves the technical problem that it is difficult to flexibly respond to load fluctuations and changes in the existing power scheduling, resulting in poor grid stability, and achieves the technical effect of improving the flexibility and stability of grid scheduling.

Description

Technical Field

[0001] The present application relates to the field related to smart grid dispatching, and in particular to a flexible load optimization dispatching method and device combined with load type analysis. Background Art

[0002] In the context of the development of smart grids, with the widespread access to renewable energy and the increasing diversification of electricity demand, the dispatching and management of power grids are facing unprecedented challenges. Traditional power dispatching methods are often based on rigid load models, which assume that load demand is fixed and cannot be adjusted. This makes it difficult to flexibly respond to load fluctuations and changes during grid operation, especially in the face of extreme weather, emergencies or large-scale changes in user behavior, which can easily lead to problems such as supply and demand imbalance and voltage instability.

[0003] Among the current related technologies, power dispatching has the technical problem of being unable to flexibly respond to load fluctuations and changes, resulting in poor grid stability. Summary of the invention

[0004] The present application provides a flexible load optimization scheduling method and device combined with load type analysis, adopts the method of constructing a distribution topology network of the target city, divides the distribution topology nodes into load types based on spatial analysis, constructs multiple edge load prediction models and maps them to the local server, pre-builds flexible load scheduling nodes and communicates with the distribution management system, uses the load scheduling update cycle as a constraint, performs load prediction based on the edge load prediction model, and updates the near-term load distribution of the target city according to the flexible load demand, etc., to achieve the technical effect of improving the flexibility and stability of power grid scheduling.

[0005] The present application provides a flexible load optimization scheduling method combined with load type analysis, including:

[0006] The physical structure of the power grid of the target city is interactively obtained, and a topological network is constructed according to the physical structure of the power grid to generate a distribution topological network; a spatial analysis is performed on the target city, and based on the spatial analysis result, multiple distribution topological nodes in the distribution topological network are classified into load types to obtain M groups of distribution topological nodes, wherein the M groups of distribution topological nodes have M types of load type identifiers; multiple historical load data of the multiple distribution topological nodes are interactively obtained, and the multiple historical load data are used as training data to construct and generate M types of edge load prediction models; the M types of edge load prediction models are mapped and synchronized to the local server of the M groups of distribution topological nodes; M flexible The M groups of distribution topology nodes are communicatively connected to the M flexible load dispatching nodes, wherein the M flexible load dispatching nodes are communicatively connected to the distribution management system of the target city; M load dispatching update periods are obtained based on the analysis of the multiple historical load data; the M groups of distribution topology nodes use the M load dispatching update periods as constraints, perform load forecasting based on the edge load forecasting model in the local server, and obtain M groups of flexible load demands; the M flexible load dispatching nodes transfer the M groups of flexible load demands to the distribution management system, and perform recent load distribution update of the target city based on the load feedback of the distribution management system.

[0007] In a possible implementation, a spatial analysis is performed on the target city, and based on the spatial analysis result, multiple distribution topology nodes in the distribution topology network are divided into load types to obtain M groups of distribution topology nodes, and the following processing is performed:

[0008] Interactively obtain the geospatial data of the target city, and restore the urban spatial model based on the geospatial data; use GIS technology to analyze the urban spatial model, and divide the urban spatial model into N load areas based on the analysis results, wherein the N load areas have N load type identifiers; load type identifiers are performed on the multiple distribution topology nodes in the distribution topology network according to the N load type identifiers of the N load areas to obtain multiple load type identifiers of the multiple distribution topology nodes; aggregate the multiple distribution topology nodes according to the M load types and multiple load type identifiers to obtain the M groups of distribution topology nodes.

[0009] In a possible implementation, load type identification is performed on the multiple power distribution topology nodes in the power distribution topology network according to the N load type identifications of the N load areas, multiple load type identifications of the multiple power distribution topology nodes are obtained, and the following processing is performed:

[0010] Aggregate the N load type identifiers to obtain the M load types; fit the distribution topology network to the urban space model to obtain a comprehensive power space model; in the comprehensive power space model, perform load type identification on the multiple distribution topology nodes according to the N load type identifiers of the N load areas to obtain the multiple load type identifiers.

[0011] In a possible implementation, multiple historical load data of the multiple power distribution topology nodes are interactively obtained, and the multiple historical load data are used as training data to construct and generate M types of edge load prediction models, and the following processing is performed:

[0012] Based on the multiple historical load data, load correlation characteristics are analyzed to obtain M groups of load correlation indicators for the M load types; data directional cleaning of the multiple historical load data is performed according to the M groups of load correlation indicators and the M groups of distribution topology nodes to obtain multiple sample load data; based on the multiple sample load data, multiple local load prediction models of the multiple distribution topology nodes are constructed and generated; based on the M groups of distribution topology nodes, federated learning of the multiple local load prediction models is performed to obtain the M edge load prediction models.

[0013] In a possible implementation, load correlation characteristics analysis is performed based on the plurality of historical load data to obtain M groups of load correlation indicators of the M load types, and the following processing is performed:

[0014] Preset a load fluctuation threshold and an associated fluctuation window; use the load fluctuation threshold to traverse the multiple historical load data to obtain multiple groups of load fluctuation nodes; use the associated fluctuation window as a data collection constraint, monitor the changes in the multiple historical load data at the multiple groups of load fluctuation nodes to obtain multiple groups of load fluctuation index sets; normalize the multiple groups of load fluctuation index sets based on the M load types to obtain M load fluctuation index sets; preset a fluctuation frequency threshold, and after aggregating the indicators of the M load fluctuation index sets, use the fluctuation frequency threshold to screen associated indicators to obtain the M groups of load associated indicators.

[0015] In a possible implementation, the federated learning of the multiple local load prediction models is performed according to the M groups of power distribution topology nodes to obtain the M edge load prediction models, and the following processing is performed:

[0016] According to the multiple sample data amounts of the multiple sample load data, model trustworthy weights are assigned to the multiple local load prediction models to obtain multiple model trustworthy weights; multiple model parameter sets of the multiple local load prediction models are extracted, and the multiple model parameter sets are divided into M groups of model parameter sets based on the M groups of distribution topology nodes; mapping weighting of the M groups of model parameter sets is performed according to the multiple model trustworthy weights to obtain M adjustment parameter sets; the M adjustment parameter sets are used to perform M model parameter adjustment and storage of any local load prediction model among the multiple local load prediction models to obtain the M edge load prediction models.

[0017] In a possible implementation, based on the analysis of the plurality of historical load data, M load scheduling update periods are obtained, and the following processing is performed:

[0018] Calculate the adjacent node intervals of multiple groups of load fluctuation nodes to obtain multiple groups of fluctuation interval durations; extract minimum values ​​of the multiple groups of fluctuation interval durations to obtain multiple fluctuation interval extreme values; aggregate the multiple fluctuation interval extreme values ​​according to the M groups of distribution topology nodes to obtain M groups of fluctuation interval extreme values; extract minimum values ​​of the M groups of fluctuation interval extreme values ​​to obtain the M load scheduling update cycles.

[0019] The present application also provides a flexible load optimization scheduling device combined with load type analysis, including:

[0020] A topology network construction module, which is used to interactively obtain the physical structure of the power grid of the target city, and construct a topology network based on the physical structure of the power grid to generate a distribution topology network; a load type classification module, which is used to perform spatial analysis on the target city, and based on the spatial analysis results, perform load type classification on multiple distribution topology nodes in the distribution topology network to obtain M groups of distribution topology nodes, wherein the M groups of distribution topology nodes have M types of load type identifiers; an edge load prediction model construction module, which is used to interactively obtain multiple historical load data of the multiple distribution topology nodes, and use the multiple historical load data as training data to construct and generate M types of edge load prediction models; a model mapping module, which is used to synchronize the mapping of the M types of edge load prediction models to the local server of the M groups of distribution topology nodes; a flexible load scheduling node construction module The flexible load dispatching node construction module is used to pre-construct M flexible load dispatching nodes, and communicatively connect the M groups of distribution topology nodes to the M flexible load dispatching nodes, wherein the M flexible load dispatching nodes are communicatively connected to the distribution management system of the target city; a load dispatching update cycle acquisition module, the load dispatching update cycle acquisition module is used to obtain M load dispatching update cycles based on the analysis of the multiple historical load data; a load forecasting module, the load forecasting module is used for the M groups of distribution topology nodes to perform load forecasting based on the edge load forecasting model in the local server with the M load dispatching update cycles as constraints, and obtain M groups of flexible load demands; a recent load distribution update module, the recent load distribution update module is used for the M flexible load dispatching nodes to transfer the M groups of flexible load demands to the distribution management system, and perform recent load distribution update of the target city based on the load feedback of the distribution management system.

[0021] The flexible load optimization scheduling method and device combined with load type analysis proposed in this application first interactively obtain the physical structure of the power grid in the target city, and construct a topological network based on the physical structure of the power grid to generate a distribution topology network. Then, a spatial analysis is performed on the target city, and based on the spatial analysis results, multiple distribution topology nodes in the distribution topology network are divided into load types to obtain M groups of distribution topology nodes, wherein the M groups of distribution topology nodes have M types of load type identifiers, and then multiple historical load data of multiple distribution topology nodes are interactively obtained, and the multiple historical load data are used as training data to construct and generate M types of edge load prediction models, and then the M types of edge load prediction models are mapped and synchronized to the local service of the M groups of distribution topology nodes. The server is used to pre-build M flexible load dispatching nodes, and M groups of distribution topology nodes are communicatively connected to the M flexible load dispatching nodes, wherein the M flexible load dispatching nodes are communicatively connected to the distribution management system of the target city, and then M load dispatching update cycles are obtained based on multiple historical load data analysis, and then the M groups of distribution topology nodes use the M load dispatching update cycles as constraints to perform load forecasting based on the edge load forecasting model in the local server to obtain M groups of flexible load demands, and finally the M flexible load dispatching nodes transfer the M groups of flexible load demands to the distribution management system, and perform the near-term load distribution update of the target city based on the load feedback of the distribution management system, thereby achieving the technical effect of improving the flexibility and stability of power grid dispatching. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the accompanying drawings of the embodiment of the present invention will be briefly introduced below. A flow chart is used in the present application to illustrate the operations performed by the device according to the embodiment of the present application. It should be understood that the previous or following operations are not necessarily performed accurately in order. On the contrary, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or more operations can be removed from these processes.

[0023] Figure 1 A flow chart of a flexible load optimization scheduling method combined with load type analysis provided in an embodiment of the present application.

[0024] Figure 2 A schematic diagram of the structure of a flexible load optimization scheduling device combined with load type analysis provided in an embodiment of the present application.

[0025] Explanation of the accompanying drawings: topology network construction module 10, load type classification module 20, edge load prediction model construction module 30, model mapping module 40, flexible load scheduling node construction module 50, load scheduling update cycle acquisition module 60, load prediction module 70, recent load distribution update module 80. DETAILED DESCRIPTION

[0026] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0027] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0028] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict. The terms "including" and "having" and any variations are intended to cover non-exclusive inclusions. For example, a process, method, device, product, or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field of this application. The terms used herein are for the purpose of describing the embodiments of the present application only.

[0029] The present application embodiment provides a flexible load optimization scheduling method combined with load type analysis, such as Figure 1 As shown, the method includes:

[0030] Step S100, interactively obtain the physical structure of the power grid of the target city, and construct a topological network based on the physical structure of the power grid to generate a distribution topological network. Specifically, the physical structure data of the power grid of the target city is obtained through GIS (geographic information system) data, CAD drawings or on-site surveys provided by the power company, including the geographical location and connection relationship of substation locations, transmission lines, distribution lines, transformers and other equipment. The collected data is cleaned to remove errors or redundant information and organized into a format that can be used to construct a topological network. Using graph theory and network analysis methods, a distribution topological network is constructed based on the connection relationship of power grid equipment. The nodes in the distribution topological network represent power equipment (such as transformers, switches, etc.), and the edges represent the electrical connections between the equipment.

[0031] Step S200, spatial analysis is performed on the target city, and based on the spatial analysis results, multiple distribution topology nodes in the distribution topology network are divided into load types to obtain M groups of distribution topology nodes, wherein the M groups of distribution topology nodes have M load type identifiers. Specifically, the spatial analysis tool of GIS is used to conduct a comprehensive analysis of the geographical, demographic, economic and other factors of the target city to identify the power consumption characteristics and load types of different regions. According to the spatial analysis results and combined with historical load data, the nodes in the distribution topology network are divided into M groups, each group corresponding to a load type (such as residential electricity, commercial electricity, industrial electricity, etc.), and different load types have different power consumption characteristics and requirements.

[0032] In a possible implementation, a spatial analysis is performed on the target city, and based on the spatial analysis results, multiple distribution topology nodes in the distribution topology network are divided into load types to obtain M groups of distribution topology nodes. Step S200 further includes step S210, interactively obtaining the geospatial data of the target city, and restoring the city spatial model based on the geospatial data. Specifically, the geospatial data of the target city, including terrain, landforms, buildings, roads, water systems, etc., are collected through various channels such as government public data, satellite images, and GIS databases. Using the GIS platform, a three-dimensional spatial model of the target city is constructed based on the processed geospatial data. This model is used to truly reflect the geographical characteristics and spatial layout of the city. Step S220, the city spatial model is analyzed using GIS technology, and the city spatial model is divided into N load areas based on the analysis results, wherein the N load areas have N load type identifiers. Specifically, the spatial analysis tool of GIS is used to conduct a comprehensive analysis of multiple factors such as population distribution, industrial structure, building type, energy consumption pattern, etc. on the urban spatial model. According to the spatial analysis results, the urban spatial model is divided into N areas with different load characteristics, and a load type identifier is assigned to each divided area to distinguish its load characteristics and electricity demand. Step S230, load type identification is performed on the multiple distribution topology nodes in the distribution topology network according to the N load type identifiers of the N load areas, and multiple load type identifiers of the multiple distribution topology nodes are obtained. Specifically, the specific position of each distribution topology node in the distribution topology network in the urban spatial model is determined, and according to the position of the distribution topology node, it is matched with the nearest load area, and the corresponding load type identifier of the area is assigned, and the load type identifier of each distribution topology node is updated in the database of the distribution topology network. Step S240, the multiple distribution topology nodes are aggregated according to the M load types and multiple load type identifiers to obtain the M groups of distribution topology nodes. Specifically, according to the pre-defined M load types, the distribution topology nodes with similar load characteristics are merged, that is, the N load type identifiers are mapped to the M load types, and according to the merged load types, the multiple distribution topology nodes in the distribution topology network are divided into M groups, each group of nodes has the same load type identifier, representing a specific load type and power demand. This implementation method analyzes and divides the urban space by combining geospatial data and GIS technology, accurately reflects the load characteristics and power demand of different regions, and achieves accurate division of the distribution topology network by matching and merging the distribution topology nodes with the load areas.

[0033] In a possible implementation, the multiple distribution topology nodes in the distribution topology network are identified by load type according to the N load type identifiers of the N load areas to obtain multiple load type identifiers of the multiple distribution topology nodes. Step S230 further includes step S231, aggregating the N load type identifiers to obtain the M load types. Specifically, the load type identifiers of the N load areas are counted to determine the frequency and distribution of each load type identifier. Similarity analysis is performed based on the attributes and characteristics of the load type identifiers. For example, commercial areas and office areas have higher load demands during peak hours, while residential areas have higher load demands at night. Based on the results of the similarity analysis, load type identifiers with similar load characteristics are aggregated to form M more general and universal load types. Step S232, the distribution topology network is fitted to the urban space model to obtain a comprehensive power space model. Specifically, the nodes, lines and other elements in the distribution topology network are mapped to the urban space model to form a corresponding relationship between power facilities and the urban geographical environment. The distribution topology network and the urban space model are integrated to generate a comprehensive power space model including power facilities and urban geographic information. Step S233, in the comprehensive power space model, the multiple distribution topology nodes are identified by load type according to the N load type identifiers of the N load areas to obtain the multiple load type identifiers. Specifically, in the comprehensive power space model, based on the geographical location of the node and the division result of the load area, the load area to which each distribution topology node belongs is determined, and according to the load type identifier of the load area to which the distribution topology node belongs, the corresponding load type identifier is assigned to the node, and the load type identifier information of each distribution topology node is updated in the database of the distribution topology network. This implementation method fits the distribution topology network to the urban space model to generate a comprehensive power space model, so that the relationship between power facilities and the urban geographical environment is more intuitive and clear. In the comprehensive power space model, the distribution topology nodes are identified by load type according to the load type identifier of the load area, ensuring that each node is correctly classified into the load type to which it belongs. At the same time, by first aggregating the load type identifier, the number of load types is reduced and the efficiency of load type division is improved.

[0034] Step S300, interactively obtain multiple historical load data of the multiple distribution topology nodes, and use the multiple historical load data as training data to construct and generate M edge load prediction models. Specifically, the historical load data of each distribution topology node is obtained from the power company database and the smart meter system, and the data is preprocessed such as cleaning, normalization, and feature selection for model training. Using machine learning methods, M edge load prediction models are constructed respectively with historical load data as training data, and each model corresponds to a load type.

[0035] In a possible implementation, multiple historical load data of the multiple distribution topology nodes are interactively obtained, and the multiple historical load data are used as training data to construct and generate M types of edge load prediction models. Step S300 further includes step S310, performing load association feature analysis based on the multiple historical load data to obtain M groups of load association indicators of the M types of load types. Specifically, the historical load data includes power consumption, voltage, current, etc. in different time periods. The collected historical load data of multiple distribution topology nodes are classified according to load type to ensure that the data of each load type is processed separately. For each load type, features related to its load change are extracted, including time features (such as periodic changes such as day, week, month, season, etc.), meteorological features (such as temperature, humidity, wind speed, etc.), socio-economic features (such as holidays, special events, etc.), etc. Statistical methods are used to analyze the correlation between each load type and its associated features to form a set of specific load association indicators. Step S320, data directional cleaning of the multiple historical load data is performed according to the M groups of load association indicators and the M groups of distribution topology nodes to obtain multiple sample load data. Specifically, according to the load-related index of each load type, the data directly related to the change of the load type is screened out, and those data that are irrelevant to the load change or have a very low correlation (such as some sensor data not related to power consumption) are eliminated. The screened data is further cleaned, including processing missing values, outliers, etc., to ensure the accuracy and reliability of the data. By directional cleaning of data based on load-related indicators, noise and redundant information are reduced, and the generalization ability of the model is improved. Step S330, based on the multiple sample load data, multiple local load prediction models of the multiple distribution topology nodes are constructed and generated. Specifically, for each distribution topology node, an independent BP neural network model is constructed. These models are predicted based on their respective load-related indicators. The cleaned sample load data is used as training data to train the BP neural network. The training data is features related to load prediction, including historical load data, time features, meteorological features, socio-economic features, etc. The output of the BP neural network is the load prediction value at a certain point in the future. The trained local load prediction model is deployed on the corresponding distribution topology node for real-time prediction of the load change of the node. Step S340, performing federated learning of the multiple local load forecasting models according to the M groups of distribution topology nodes to obtain the M edge load forecasting models. Specifically, without directly sharing the original data, each node sends the parameters or updates of the local model to the central server, and the central server generates a global load forecasting model through an aggregation algorithm based on the received parameters or updates. The global load forecasting model integrates the knowledge of multiple data sources and has stronger generalization ability.The central server distributes the global load forecasting model back to each node for updating the local model. This process is iterated until the performance of the global model meets the requirements or reaches the preset number of iterations. Repeat the above process until the performance of the global load forecasting model meets the requirements or reaches the preset number of iterations. This implementation method effectively utilizes the historical load data of multiple distribution topology nodes through load correlation feature analysis, data directional cleaning, local model construction and federated learning, and constructs an edge load forecasting model that is both privacy-preserving and accurate. This method not only improves the prediction accuracy, but also promotes data sharing and cooperation between different nodes.

[0036] In a possible implementation, load correlation characteristics are analyzed based on the multiple historical load data to obtain M groups of load correlation indicators of the M load types, and step S310 further includes step S311, presetting a load fluctuation threshold and an associated fluctuation window. Specifically, according to the fluctuation of the historical load data, a threshold is pre-set to determine whether the load data has fluctuated significantly at a certain moment. For example, the load change exceeding 5% of the average load can be set as a sign of significant fluctuation. A time range, i.e., a window, is set to further observe the changes in the data around the node after the load fluctuation node is detected. Step S312, traverse the multiple historical load data using the load fluctuation threshold to obtain multiple groups of load fluctuation nodes. Specifically, traverse the entire historical load data set, compare each data point with its previous data point, and if the difference between the two data points (such as the load change) exceeds the preset load fluctuation threshold, it is considered that a significant load fluctuation has occurred at this point, and the point is marked as a load fluctuation node, and the above process is repeated until all historical data points are traversed to obtain multiple groups of load fluctuation nodes. Step S313, taking the associated fluctuation window as a data collection constraint, monitoring the changes of the multiple historical load data at the multiple groups of load fluctuation nodes to determine the changes in the data, and obtaining multiple groups of load fluctuation index sets. Specifically, for each load fluctuation node, with it as the center, collect data in the associated fluctuation window, analyze the changes in the data in the window, such as the change rate, change trend, fluctuation period, etc., to form a set of load fluctuation indexes, repeat the above process for all load fluctuation nodes, and obtain multiple groups of load fluctuation index sets. Step S314, normalize the multiple groups of load fluctuation index sets based on the M load types to obtain M load fluctuation index sets. Specifically, since the data of different load types have different dimensions and ranges, it is necessary to normalize the load fluctuation index set of each load type to eliminate the dimension and range differences between different load types, so that the load fluctuation index sets of each load type are comparable. The load fluctuation index sets of the M load types are normalized respectively to obtain M normalized load fluctuation index sets. Step S315, preset a fluctuation frequency threshold, and after aggregating the indicators of the M load fluctuation indicator sets, use the fluctuation frequency threshold to screen the associated indicators to obtain the M groups of load associated indicators. Specifically, a fluctuation frequency threshold is preset to screen the associated indicators that frequently appear in multiple load fluctuation nodes and have an important impact on load changes. Aggregate the M normalized load fluctuation indicator sets and count the frequency of each indicator appearing in all load fluctuation nodes. According to the fluctuation frequency threshold, screen out the associated indicators with a frequency higher than the threshold to form M groups of load associated indicators.This implementation method accurately identifies the fluctuation nodes and associated features in the historical load data through preset load fluctuation thresholds and associated fluctuation windows. Through normalization processing, it eliminates the dimensional and range differences between data of different load types, making subsequent analysis more accurate and reliable. By screening the fluctuation frequency threshold, it retains those associated indicators that have a significant impact on load changes, thereby improving the pertinence and effectiveness of the analysis.

[0037] In a possible implementation, the federated learning of the multiple local load prediction models is performed according to the M groups of distribution topology nodes to obtain the M edge load prediction models. Step S340 further includes step S341, according to the multiple sample data amounts of the multiple sample load data, the model trust weights are assigned to the multiple local load prediction models to obtain multiple model trust weights. Specifically, the data amount of the sample load data corresponding to each local load prediction model is analyzed. The model with more data contains more information and has stronger prediction ability. Based on the analysis results, a trust weight is assigned to each local load prediction model, and the weight size reflects the credibility of the model in the current learning task. Step S342, extract multiple model parameter sets of the multiple local load prediction models, and divide the multiple model parameter sets into M groups of model parameter sets based on the M groups of distribution topology nodes. Specifically, all training parameters of each local load prediction model are extracted, including but not limited to weights, biases, etc. According to the corresponding relationship of the M groups of distribution topology nodes, the extracted model parameter sets are divided into M groups, and each group of parameter sets corresponds to a group of distribution topology nodes. Step S343, mapping and weighting the M groups of model parameter sets according to the multiple model trust weights to obtain M adjustment parameter sets. Specifically, the model trust weight obtained in step S341 is mapped to each grouped model parameter set, that is, the parameters in each parameter set are adjusted according to the weight of the model to which it belongs. The parameter sets in the same group are weighted summed (or other forms of weighted average) to obtain a new adjustment parameter set, that is, the adjustment parameter set is a parameter set that integrates multiple model parameters. Step S344, using the M adjustment parameter sets to perform M model parameter adjustment and preservation of any local load prediction model in the multiple local load prediction models, to obtain the M edge load prediction models. Specifically, using the M adjustment parameter sets obtained in step S343, M parameter adjustment operations are performed on any local load prediction model, and each adjustment parameter set will generate a fine-tuned model version. Each fine-tuned model version is saved as an independent edge load prediction model. This implementation method fully utilizes the advantages of each model by weighted fusion of the parameters of multiple models, reduces the deviations or deficiencies of a single model, and improves the accuracy of the overall prediction.

[0038] Step S400: Map and synchronize the M edge load prediction models to the local servers of the M groups of power distribution topology nodes. Specifically, the trained M edge load prediction model files are uploaded to the local servers corresponding to the M groups of power distribution topology nodes using a model deployment tool.

[0039] Step S500, pre-build M flexible load dispatching nodes, and connect the M groups of distribution topology nodes to the M flexible load dispatching nodes, wherein the M flexible load dispatching nodes are connected to the distribution management system of the target city. Specifically, according to the preset architecture and parameters, configure the hardware and software environment of the M flexible load dispatching nodes, wherein the flexible load dispatching nodes are intermediate nodes responsible for receiving the load forecast data of the distribution topology nodes and performing load dispatch according to the instructions of the distribution management system. Determine the communication protocol (such as TCP / IP, etc.) between the distribution topology node and the flexible load dispatching node to ensure that data can be transmitted reliably, set network routing and firewall rules, ensure that the flexible load dispatching node can safely receive data from the distribution topology node, and communicate with the distribution management system.

[0040] Step S600, based on the analysis of the multiple historical load data, M load scheduling update cycles are obtained. Specifically, a time series analysis is performed on the historical load data of multiple distribution topology nodes to identify the periodicity and trend of load changes, and based on the analysis results and in combination with the actual needs of power grid operation (such as peak and valley time periods, load fluctuation amplitude, etc.), M load scheduling update cycles are determined, and the load scheduling update cycle is the time interval for load forecasting and allocation.

[0041] In a possible implementation, based on the analysis of the multiple historical load data, M load scheduling update cycles are obtained, and step S600 further includes step S610, performing adjacent node interval calculations on multiple groups of load fluctuation nodes to obtain multiple groups of fluctuation interval durations. Specifically, for each load fluctuation node, the time interval between it and the adjacent load fluctuation node, that is, the fluctuation interval duration, is calculated, and these interval durations reflect the frequency and amplitude of load changes. Step S620, performing minimum value extraction on the multiple groups of fluctuation interval durations to obtain multiple fluctuation interval extreme values. Specifically, all the fluctuation interval durations calculated in step S610 are sorted to form a data set, and the minimum values ​​of all fluctuation interval durations, that is, the shortest fluctuation interval duration, are extracted from this data set. These minimum values ​​represent the time period when the load changes most frequently. Step S630, aggregating the multiple fluctuation interval extreme values ​​according to the M groups of distribution topology nodes to obtain M groups of fluctuation interval extreme values. Specifically, for each group of distribution topology nodes, all fluctuation interval extreme values ​​belonging to the group are aggregated to form a set of fluctuation interval extreme values ​​for the group. Step S640, extract the minimum value of the M groups of fluctuation interval extreme values ​​to obtain the M load scheduling update cycles. Specifically, extract the minimum value again from the fluctuation interval extreme value set of each group, that is, the shortest fluctuation interval duration in the group. These minimum values ​​are used as the load scheduling update cycle of the group of distribution topology nodes. This implementation method ensures that the load scheduling update cycle can cover all load fluctuation nodes by extracting the minimum value of the fluctuation interval extreme value, thereby improving the response speed and scheduling efficiency of the system.

[0042] Step S700: The M groups of distribution topology nodes use the M load scheduling update cycles as constraints, perform load prediction based on the edge load prediction model in the local server, and obtain M groups of flexible load demands. Specifically, at the beginning of each load scheduling update cycle, the edge load prediction model on the distribution topology node is triggered. The edge load prediction model predicts the load demand in the future based on historical load data, current operating status, and external environmental factors (such as weather forecasts, holiday information, etc.), and summarizes the prediction results into M groups of flexible load demands, which are sent to the flexible load scheduling node.

[0043] Step S800, the M flexible load dispatching nodes transfer the M groups of flexible load demands to the distribution management system, and update the near-term load distribution of the target city based on the load feedback of the distribution management system. Specifically, the flexible load dispatching node receives the flexible load demand data from the distribution topology node and forwards it to the distribution management system. The distribution management system distributes the received flexible load demand according to the global load situation, power grid operation rules and optimization goals, and formulates a near-term load distribution plan. The distribution management system feeds back the load distribution results to the flexible load dispatching node, and the flexible load dispatching node sends these instructions to the corresponding distribution topology node to guide it to adjust the load. The embodiment of the present application adopts the construction of the distribution topology network of the target city, the load type classification of the distribution topology node based on spatial analysis, the construction of multiple edge load prediction models and mapping to the local server, the pre-construction of the flexible load dispatching node and the communication connection with the distribution management system, the load dispatching update cycle is constrained, the load forecast is performed based on the edge load prediction model, and the near-term load distribution of the target city is updated according to the flexible load demand. Technical means such as updating, achieve the technical effect of improving the flexibility and stability of power grid dispatching.

[0044] In the above, refer to Figure 1 The flexible load optimization scheduling method combined with load type analysis according to an embodiment of the present invention is described in detail. Figure 2 A flexible load optimization scheduling device combined with load type analysis according to an embodiment of the present invention is described.

[0045] The flexible load optimization scheduling device combined with load type analysis according to the embodiment of the present invention is used to solve the technical problem that it is difficult to flexibly respond to load fluctuations and changes in the existing power dispatching, resulting in poor grid stability, and achieve the technical effect of improving the flexibility and stability of grid dispatching. The flexible load optimization scheduling device combined with load type analysis includes: a topology network construction module 10, a load type classification module 20, an edge load prediction model construction module 30, a model mapping module 40, a flexible load scheduling node construction module 50, a load scheduling update cycle acquisition module 60, a load prediction module 70, and a recent load distribution update module 80.

[0046] The topology network construction module 10 is used to interactively obtain the physical structure of the power grid of the target city, and construct the topology network according to the physical structure of the power grid to generate a distribution topology network; the load type classification module 20 is used to perform spatial analysis on the target city, and based on the spatial analysis results, perform load type classification on multiple distribution topology nodes in the distribution topology network to obtain M groups of distribution topology nodes, wherein the M groups of distribution topology nodes have M types of load type identifiers; the edge load prediction model construction module 30 is used to interactively obtain multiple historical load data of the multiple distribution topology nodes, and use the multiple historical load data as training data to construct and generate M types of edge load prediction models; the model mapping module 40 is used to map and synchronize the M types of edge load prediction models to the local server of the M groups of distribution topology nodes; the flexible load scheduling node construction module 30 is used to generate the M types of edge load prediction models; the load type classification module 20 is used to perform spatial analysis on the target city, and perform load type classification on multiple distribution topology nodes in the distribution topology network based on the spatial analysis results to obtain M groups of distribution topology nodes, wherein the M groups of distribution topology nodes have M types of load type identifiers; the edge load prediction model construction module 30 is used to interactively obtain multiple historical load data of the multiple distribution topology nodes, and use the multiple historical load data as training data to construct and generate M types of edge load prediction models; the model mapping module 40 is used to synchronize the M types of edge load prediction models to the local server of the M groups of distribution topology nodes; the flexible load scheduling node construction module 30 is used to generate the M types of edge load prediction models; the load type classification module 2 ... The construction module 50 is used to pre-construct M flexible load dispatching nodes, and communicatively connect the M groups of distribution topology nodes to the M flexible load dispatching nodes, wherein the M flexible load dispatching nodes are communicatively connected to the distribution management system of the target city; the load dispatching update cycle acquisition module 60 is used to obtain M load dispatching update cycles based on the analysis of the multiple historical load data; the load prediction module 70 is used for the M groups of distribution topology nodes to perform load prediction based on the edge load prediction model in the local server with the M load dispatching update cycles as constraints, and obtain M groups of flexible load demands; the recent load distribution update module 80 is used for the M flexible load dispatching nodes to transfer the M groups of flexible load demands to the distribution management system, and perform recent load distribution update of the target city based on the load feedback of the distribution management system.

[0047] The specific configuration of the load type classification module 20 will be described in detail below. As described above, a spatial analysis is performed on the target city, and based on the spatial analysis results, load types are classified for multiple distribution topology nodes in the distribution topology network to obtain M groups of distribution topology nodes. The load type classification module 20 may further include: an urban space model acquisition unit for interactively obtaining the geographic spatial data of the target city, and restoring the urban space model based on the geographic spatial data; a load area classification unit for analyzing the urban space model using GIS technology, and dividing the urban space model into N load areas based on the analysis results, wherein the N load areas have N load type identifiers; a load type identification unit for load type identification of the multiple distribution topology nodes in the distribution topology network according to the N load type identifiers of the N load areas, and obtaining multiple load type identifiers of the multiple distribution topology nodes; a distribution topology node aggregation unit for aggregating the multiple distribution topology nodes according to the M load types and multiple load type identifiers to obtain the M groups of distribution topology nodes.

[0048] Among them, the multiple distribution topology nodes in the distribution topology network are load type identified according to the N load type identifications of the N load areas to obtain multiple load type identifications of the multiple distribution topology nodes, and the load type identification unit may further include: a load type identification aggregation subunit is used to aggregate the N load type identifications to obtain the M load types; a comprehensive power space model acquisition subunit is used to fit the distribution topology network to the urban space model to obtain a comprehensive power space model; the load type identification subunit is used to, in the comprehensive power space model, identify the multiple distribution topology nodes according to the N load type identifications of the N load areas to obtain the multiple load type identifications.

[0049] The specific configuration of the edge load prediction model construction module 30 will be described in detail below. As described above, multiple historical load data of the multiple distribution topology nodes are interactively obtained, and the multiple historical load data are used as training data to construct and generate M types of edge load prediction models. The edge load prediction model construction module 30 may further include: a load correlation feature analysis unit is used to perform load correlation feature analysis based on the multiple historical load data to obtain M groups of load correlation indicators of the M types of load types; a data directional cleaning unit is used to perform data directional cleaning of the multiple historical load data according to the M groups of load correlation indicators and the M groups of distribution topology nodes to obtain multiple sample load data; a local load prediction model generation unit is used to construct and generate multiple local load prediction models of the multiple distribution topology nodes based on the multiple sample load data; and an edge load prediction model acquisition unit is used to perform federated learning of the multiple local load prediction models according to the M groups of distribution topology nodes to obtain the M types of edge load prediction models.

[0050] Among them, load correlation characteristics analysis is performed based on the multiple historical load data to obtain M groups of load correlation indicators of the M load types, and the load correlation characteristics analysis unit may further include: a fluctuation preset subunit is used to preset a load fluctuation threshold and an associated fluctuation window; a traversal subunit is used to traverse the multiple historical load data using the load fluctuation threshold to obtain multiple groups of load fluctuation nodes; a monitoring data change judgment subunit is used to use the associated fluctuation window as a data collection constraint, and perform monitoring data change judgment on the multiple historical load data at the multiple groups of load fluctuation nodes to obtain multiple groups of load fluctuation indicator sets; a normalization subunit is used to normalize the multiple groups of load fluctuation indicator sets based on the M load types to obtain M load fluctuation indicator sets; an associated indicator screening subunit is used to preset a fluctuation frequency threshold, and based on the aggregation of indicators for the M load fluctuation indicator sets, the fluctuation frequency threshold is used to perform associated indicator screening to obtain the M groups of load correlation indicators.

[0051] Among them, the federated learning of the multiple local load prediction models is performed according to the M groups of distribution topology nodes to obtain the M edge load prediction models. The edge load prediction model acquisition unit may further include: a trusted weight allocation subunit is used to allocate model trusted weights to the multiple local load prediction models according to the multiple sample data amounts of the multiple sample load data, and obtain multiple model trusted weights; a model parameter set extraction subunit is used to extract and obtain multiple model parameter sets of the multiple local load prediction models, and divide the multiple model parameter sets into M groups of model parameter sets based on the M groups of distribution topology nodes; a mapping weighting subunit is used to perform mapping weighting of the M groups of model parameter sets according to the multiple model trusted weights to obtain M adjustment parameter sets; a model parameter storage subunit is used to use the M adjustment parameter sets to perform M times of model parameter storage for any local load prediction model among the multiple local load prediction models to obtain the M edge load prediction models.

[0052] The specific configuration of the load scheduling update cycle acquisition module 60 will be described in detail below. As described above, based on the analysis of the multiple historical load data, M load scheduling update cycles are obtained, and the load scheduling update cycle acquisition module 60 may further include: an adjacent node interval calculation unit is used to perform adjacent node interval calculations on multiple groups of load fluctuation nodes to obtain multiple groups of fluctuation interval durations; a fluctuation interval duration minimum value extraction unit is used to perform minimum value extraction on the multiple groups of fluctuation interval durations to obtain multiple fluctuation interval extreme values; a fluctuation interval extreme value aggregation unit is used to aggregate the multiple fluctuation interval extreme values ​​according to the M groups of distribution topology nodes to obtain M groups of fluctuation interval extreme values; a fluctuation interval extreme value minimum value extraction unit is used to perform minimum value extraction on the M groups of fluctuation interval extreme values ​​to obtain the M load scheduling update cycles.

[0053] The flexible load optimization scheduling device combined with load type analysis provided in the embodiment of the present invention can execute the flexible load optimization scheduling method combined with load type analysis provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0054] Although the present application makes various references to certain modules in the device according to the embodiments of the present application, any number of different modules may be used and run on the user terminal and / or server, and the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0055] The above specific implementation manner does not constitute a limitation to the protection scope of the present application. It should be understood by those skilled in the art that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application. In some cases, the actions or steps recorded in the present application can be performed in an order different from that in the embodiment and can still achieve the desired results. In addition, the process depicted in the accompanying drawings does not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A flexible load optimization scheduling method combined with load type analysis is characterized by: The method comprises: Interactively obtain the physical structure of the power grid of the target city, and construct a topological network according to the physical structure of the power grid to generate a distribution topological network; Performing spatial analysis on the target city, and classifying multiple power distribution topology nodes in the power distribution topology network by load type based on the spatial analysis result to obtain M groups of power distribution topology nodes, wherein the M groups of power distribution topology nodes have M load type identifiers; Interactively obtain multiple historical load data of the multiple distribution topology nodes, and use the multiple historical load data as training data to construct and generate M types of edge load prediction models; Synchronize the M edge load prediction model mappings to the local servers of the M groups of power distribution topology nodes; Pre-constructing M flexible load dispatching nodes, and communicatively connecting the M groups of distribution topology nodes to the M flexible load dispatching nodes, wherein the M flexible load dispatching nodes are communicatively connected to the distribution management system of the target city; Obtaining M load scheduling update cycles based on the analysis of the multiple historical load data; The M groups of power distribution topology nodes use the M load scheduling update periods as constraints and perform load forecasting based on the edge load forecasting model in the local server to obtain the M groups of flexible load demands; The M flexible load dispatching nodes transfer the M groups of flexible load demands to the power distribution management system, and perform a recent load distribution update for the target city based on the load feedback from the power distribution management system; Based on the analysis of the plurality of historical load data, M load scheduling update periods are obtained, and the method further comprises: Calculate the intervals between adjacent nodes for multiple groups of load fluctuation nodes to obtain the durations of multiple groups of fluctuation intervals; Extracting minimum values ​​of the multiple groups of fluctuation interval durations to obtain multiple fluctuation interval extreme values; Aggregating the multiple fluctuation interval extreme values ​​according to the M groups of power distribution topology nodes to obtain M groups of fluctuation interval extreme values; Minimum values ​​are extracted from the M groups of fluctuation interval extreme values ​​to obtain the M load scheduling update cycles.

2. The flexible load optimization scheduling method combined with load type analysis according to claim 1 is characterized in that: Performing spatial analysis on the target city, and classifying multiple power distribution topology nodes in the power distribution topology network by load type based on the spatial analysis result to obtain M groups of power distribution topology nodes, the method further comprising: Interactively obtain the geospatial data of the target city, and restore the city spatial model based on the geospatial data; Analyzing the urban space model by using GIS technology, and dividing the urban space model into N load areas based on the analysis result, wherein the N load areas have N load type identifiers; Performing load type identification on the multiple power distribution topology nodes in the power distribution topology network according to the N load type identifications of the N load areas to obtain multiple load type identifications of the multiple power distribution topology nodes; The multiple power distribution topology nodes are aggregated according to the M load types and multiple load type identifiers to obtain the M groups of power distribution topology nodes.

3. The flexible load optimization scheduling method combined with load type analysis according to claim 2 is characterized in that: Performing load type identification on the multiple power distribution topology nodes in the power distribution topology network according to the N load type identifications of the N load areas to obtain multiple load type identifications of the multiple power distribution topology nodes, the method further comprising: Aggregating the N load type identifiers to obtain the M load types; Fitting the power distribution topology network to the urban space model to obtain a comprehensive power space model; In the comprehensive power space model, the multiple distribution topology nodes are load type identified according to the N load type identifications of the N load areas to obtain the multiple load type identifications.

4. The flexible load optimization scheduling method combined with load type analysis according to claim 1 is characterized in that: Performing spatial analysis on the target city, and classifying the M power distribution topology nodes in the power distribution topology network by load type based on the spatial analysis result to obtain multiple groups of power distribution topology nodes, the method further comprising: Perform load correlation characteristic analysis based on the plurality of historical load data to obtain M groups of load correlation indicators of the M load types; Performing data-oriented cleaning of the plurality of historical load data according to the M groups of load-related indicators and the M groups of power distribution topology nodes to obtain a plurality of sample load data; Constructing and generating a plurality of local load prediction models for the plurality of power distribution topology nodes based on the plurality of sample load data; Federated learning of the multiple local load prediction models is performed according to the M groups of power distribution topology nodes to obtain the M edge load prediction models.

5. The flexible load optimization scheduling method combined with load type analysis according to claim 4 is characterized in that: Performing load correlation feature analysis based on the plurality of historical load data to obtain M groups of load correlation indicators of the M load types, the method further includes: Preset load fluctuation thresholds and associated fluctuation windows; Using the load fluctuation threshold to traverse the plurality of historical load data, to obtain a plurality of groups of load fluctuation nodes; Taking the associated fluctuation window as a data collection constraint, performing monitoring data change judgment on the multiple historical load data at the multiple groups of load fluctuation nodes to obtain multiple groups of load fluctuation indicator sets; Normalizing the multiple groups of load fluctuation index sets based on the M load types to obtain M load fluctuation index sets; A fluctuation frequency threshold is preset, and after the M load fluctuation indicator sets are aggregated, the fluctuation frequency threshold is used to screen associated indicators to obtain the M groups of load associated indicators.

6. The flexible load optimization scheduling method combined with load type analysis according to claim 4 is characterized in that: Performing federated learning of the multiple local load prediction models according to the M groups of power distribution topology nodes to obtain the M edge load prediction models, the method further comprising: According to the multiple sample data amounts of the multiple sample load data, model credibility weights are assigned to the multiple local load forecasting models to obtain multiple model credibility weights; Extracting and obtaining a plurality of model parameter sets of the plurality of local load prediction models, and dividing the plurality of model parameter sets into M groups of model parameter sets based on the M groups of distribution topology nodes; Perform mapping weighting of the M groups of model parameter sets according to the multiple model trust weights to obtain M adjustment parameter sets; The M adjustment parameter sets are used to perform M model parameter adjustments and saves on any local load forecasting model among the multiple local load forecasting models to obtain the M edge load forecasting models.

7. A flexible load optimization dispatching device combined with load type analysis is characterized in that: The device is used to implement the flexible load optimization scheduling method combined with load type analysis according to any one of claims 1 to 6, and the device comprises: A topology network construction module, wherein the topology network construction module is used to interactively obtain the physical structure of the power grid of the target city, and construct a topology network according to the physical structure of the power grid to generate a distribution topology network; A load type classification module, the load type classification module is used to perform spatial analysis on the target city, and based on the spatial analysis result, perform load type classification on multiple distribution topology nodes in the distribution topology network to obtain M groups of distribution topology nodes, wherein the M groups of distribution topology nodes have M load type identifiers; An edge load prediction model construction module, wherein the edge load prediction model construction module is used to interactively obtain multiple historical load data of the multiple distribution topology nodes, and use the multiple historical load data as training data to construct and generate M types of edge load prediction models; A model mapping module, the model mapping module is used to synchronize the M edge load prediction model mappings to the local servers of the M groups of power distribution topology nodes; A flexible load dispatching node construction module, wherein the flexible load dispatching node construction module is used to pre-construct M flexible load dispatching nodes, and communicatively connect the M groups of distribution topology nodes to the M flexible load dispatching nodes, wherein the M flexible load dispatching nodes are communicatively connected to the distribution management system of the target city; A load scheduling update period acquisition module, the load scheduling update period acquisition module is used to obtain M load scheduling update periods based on the analysis of the multiple historical load data; A load prediction module, wherein the load prediction module is used for performing load prediction on the M groups of distribution topology nodes based on the M load scheduling update cycles and an edge load prediction model in a local server to obtain M groups of flexible load demands; A recent load distribution update module, which is used for the M flexible load dispatching nodes to transfer the M groups of flexible load demands to the distribution management system, and to update the recent load distribution of the target city based on the load feedback of the distribution management system; The load scheduling update cycle acquisition module further includes: an adjacent node interval calculation unit for performing adjacent node interval calculations on multiple groups of load fluctuation nodes to obtain multiple groups of fluctuation interval durations; a fluctuation interval duration minimum value extraction unit for performing minimum value extraction on the multiple groups of fluctuation interval durations to obtain multiple fluctuation interval extreme values; a fluctuation interval extreme value aggregation unit for aggregating the multiple fluctuation interval extreme values ​​according to the M groups of distribution topology nodes to obtain M groups of fluctuation interval extreme values; a fluctuation interval extreme value minimum value extraction unit for performing minimum value extraction on the M groups of fluctuation interval extreme values ​​to obtain the M load scheduling update cycles.

Images