A power balance scheduling method and device fusing neural network learning and Monte Carlo simulation, a terminal device and a storage medium

Abstract

The application discloses a power balance scheduling method and device fusing neural network learning and Monte Carlo simulation, a terminal equipment and a storage medium, and belongs to the technical field of power scheduling. The method is as follows: acquiring topological structure data of each load node in a distribution network, historical load time series data, maximum transmission power limit values of each line, current power operation data, and predicted meteorological data and date information of a to-be-scheduled period; performing feeder block division on all load nodes, simultaneously determining first input features, and inputting the first input features into a preset load prediction model to obtain a benchmark load prediction value; generating second input features through Monte Carlo sampling, and then determining load fluctuation boundaries of each feeder block; and constructing a scheduling optimization model based on the load fluctuation boundaries, and solving to obtain a power balance scheduling scheme. Through implementation of the application, the problem that a load prediction result in the prior art has a large deviation, and then leads to an unreasonable power balance scheduling result can be solved.

Description

Technical Field

[0001] This invention relates to the field of power dispatching technology, and in particular to a power balance dispatching method, apparatus, terminal equipment, and storage medium that integrates neural network learning and Monte Carlo simulation. Background Technology

[0002] In recent years, to meet carbon emission reduction targets, my country's new energy industry has developed rapidly, with wind power and photovoltaic installed capacity ranking among the world's top. Various new energy power generation devices are continuously being connected to the distribution network. As the proportion of new energy connected continues to increase, its inherent intermittency and volatility bring greater uncertainty to power system planning. The analysis of power generation capacity balance across the entire network has become more complex, and the workload has increased significantly. To address the uncertainties in distribution network planning, research has gradually shifted its focus to the balance between power supply and demand, aiming to improve the allocation efficiency of power resources while ensuring the reliability of power supply from the distribution network.

[0003] Traditional power balancing methods based on neural network learning typically only combine fixed expected values ​​(such as fixed weather forecast data for the date to be predicted) for load forecasting. However, in reality, the weather on the date to be predicted may not be the same as the current weather forecast data. Therefore, when faced with sudden abnormal weather, the forecast results may have large deviations, which in turn leads to unreasonable power balancing and dispatching results based on such load forecast results. Summary of the Invention

[0004] This invention provides a power balance dispatching method, device, terminal equipment, and storage medium that integrates neural network learning and Monte Carlo simulation. It can solve the problem that the existing technology only combines fixed expected values ​​for load forecasting, which makes the forecast results deviate greatly due to abnormal weather and sudden events, resulting in unreasonable power balance dispatching results.

[0005] One embodiment of the present invention provides a power balance dispatching method that integrates neural network learning and Monte Carlo simulation, comprising: Acquire topology data, historical load time series data, maximum transmission power limit values ​​of each line, current power operation data, and forecast meteorological data and date information for the period to be dispatched in the distribution network for each load node; Based on the topology data and the maximum transmission power limit, all load nodes are divided into feeder blocks to obtain several feeder blocks; Based on the historical load time series data, predicted meteorological data and date information, the first input feature is calculated and input into the preset load prediction model based on neural network to obtain the baseline load prediction value of each load node in the scheduling period. Using the baseline load forecast as the mean reference benchmark, a probability distribution is constructed based on the forecast meteorological data, the baseline load forecast, and the date information. The forecast meteorological data in the probability distribution is randomly sampled several times using the Monte Carlo simulation method. Several second input features are generated based on the sampling results, historical load time series data, and the date information. Each second input feature is input into the preset load prediction model to obtain several final load prediction values, and the load fluctuation boundary of each feeder block is determined based on all the final load prediction values ​​and each feeder block. Based on the load fluctuation boundary, power operation data, and several feeder blocks, a scheduling optimization model with the goal of minimizing total operating cost and invalid power generation, along with corresponding constraints, is constructed. The scheduling optimization model is solved under the constraints to obtain a power balance scheduling scheme.

[0006] Furthermore, based on the aforementioned topology data and maximum transmission power limit, all load nodes are divided into feeder blocks, resulting in several feeder blocks, including: Obtain the rated power of each load node, as well as the net load forecast under different scenarios; Based on the topology data, the shortest geographical distance between any two load nodes is calculated, and a load node spacing distance matrix is ​​constructed based on the shortest geographical distance. Based on the net load forecast values ​​of all load nodes, the source-load characteristic matching degree between any two load nodes is calculated, and a source-load characteristic matching degree matrix is ​​constructed based on the source-load characteristic matching degree. The maximum power limit value of the feeder block is calculated based on the maximum transmission power limit value. Based on the rated power, the maximum power limit of the feeder block, the distance matrix between load points, and the source-load characteristic matching degree matrix, the node merging operation is repeatedly performed to obtain several feeder blocks; The node merging operation includes: Get the current cluster set; where the initial cluster set is all load nodes, and each load node is a cluster; Based on the current cluster set, rated power, load point spacing matrix, and source-load characteristic matching degree matrix, the comprehensive score of the current cluster pair to be merged and each current cluster pair to be merged is calculated. If the current cluster pair to be merged is empty, then each cluster in the current cluster set is taken as a feeder block, and then several feeder blocks are obtained. Otherwise, the current cluster pair with the lowest overall score is merged into a new cluster; the target cluster that does not contain the load nodes in the new cluster is extracted from the current cluster set, and the cluster set for the next time step is generated based on the new cluster and the target cluster.

[0007] Furthermore, based on the current cluster set, rated power, load point spacing matrix, and source-load characteristic matching matrix, the comprehensive score of the current cluster pair to be merged and each current cluster pair to be merged is calculated, including: Based on the current cluster set, generate several pairs of current clusters; The current total power of each current cluster pair is calculated based on the rated power, and cluster pairs whose current total power exceeds the maximum power limit of the feeder block are removed to obtain the currently selected cluster pairs; Based on the load point spacing matrix and the source load characteristic matching degree matrix, the comprehensive score of each currently selected cluster pair is calculated. Remove currently selected cluster pairs whose overall score exceeds a preset score threshold to obtain the current cluster pairs to be merged.

[0008] Furthermore, based on the historical load time-series data, predicted meteorological data, and date information, a first input feature is calculated and input into a preset load prediction model built on a neural network to obtain the baseline load prediction value for each load node in the scheduling period, including: The historical load time series data is processed by imputing missing values, removing outliers, and normalizing to obtain load input characteristics; Based on the predicted meteorological data, climate seasonality features are generated to characterize the season and weather of the scheduled period. Based on the date information, generate time-series features to characterize whether the scheduling period is a weekday and the morning / evening time period to which it belongs; The load input characteristics, climate seasonality characteristics, and time series characteristics are concatenated into the first input characteristic; The first input feature is input into a preset load prediction model to obtain the baseline load prediction value of each load node in the scheduling period.

[0009] Furthermore, the first input feature is input into a preset load prediction model to obtain the baseline load prediction value for each load node in the scheduling period, including: The first input feature is input into a preset load prediction model, so that the convolutional neural network in the preset load prediction model performs convolution, max pooling and fully connected operations on the first input feature to obtain a first output feature vector. Based on the bidirectional gated loop unit in the preset load prediction model, long-term time-series regularity features are extracted from the first output feature vector to obtain the baseline load prediction value of each load node in the scheduling period.

[0010] Furthermore, the training of the preset load prediction model includes: Acquire historical load sample data under historical scheduling cycles, as well as historical meteorological data, historical date information, and actual historical load data for the corresponding forecast period of the historical load sample data; Based on the historical load sample data, historical meteorological data, historical date information, and real historical load data, several input sample data with real labels are calculated and generated; wherein, the real labels are used to characterize the real historical load data corresponding to the input sample data; The input sample data is input into the load prediction model to be trained for iterative training until the loss function converges, and a trained preset load prediction model is obtained. In each iteration of training, the current load sample prediction value is obtained based on the current input sample data; the current physical constraint penalty value is calculated based on the current load sample prediction value, the preset maximum capacity of the energy storage device, and the preset maximum ramp rate of the generator; and the current loss function is calculated based on the current physical constraint penalty value, the current load sample prediction value, and the corresponding true label. Determine whether the current loss function has converged; if the current loss function has converged, then use the current load forecasting model as the preset load forecasting model; otherwise, adjust the parameters in the current load forecasting model and continue training.

[0011] Furthermore, the objective function of the scheduling optimization model is: In the formula, T represents the total number of time periods in the period to be scheduled. This indicates the total number of conventional generators within the feeder block. This represents the operating cost of unit g. The function representing the operating cost of unit g. This represents the output of the conventional generator unit g within the feeder block at time t. This indicates a high penalty coefficient for invalid electricity usage. This represents the invalid electricity consumption during time period t.

[0012] Based on the above method embodiments, the present invention provides corresponding apparatus embodiments; This invention provides a power balance dispatching device that integrates neural network learning and Monte Carlo simulation, comprising: The module includes a data acquisition module, a feeder block module, a baseline load prediction module, a random sampling module, a boundary generation module, a model building module, and a scheduling scheme generation module. The data acquisition module is used to acquire topology data of each load node in the distribution network, historical load time series data, maximum transmission power limit value of each line, current power operation data, and predicted meteorological data and date information for the scheduling period. The feeder block module is used to divide all load nodes into feeder blocks according to the topology data and the maximum transmission power limit value, thereby obtaining several feeder blocks. The baseline load prediction module is used to calculate the first input feature based on the historical load time series data, predicted meteorological data and date information, and input it into the preset load prediction model based on the neural network to obtain the baseline load prediction value of each load node in the scheduling period. The random sampling module is used to construct a probability distribution based on the mean reference benchmark of the baseline load forecast value, the predicted meteorological data, the baseline load forecast value and the date information, and to perform several random samplings of the predicted meteorological data in the probability distribution using the Monte Carlo simulation method, and to generate several second input features based on the sampling results, historical load time series data and the date information. The boundary generation module is used to input each second input feature into the preset load prediction model to obtain several final load prediction values, and to determine the load fluctuation boundary of each feeder block based on all the final load prediction values ​​and each feeder block. The model building module is used to construct a scheduling optimization model and corresponding constraints based on the load fluctuation boundary, power operation data and several feeder blocks, with the goal of minimizing total operating cost and invalid power. The scheduling scheme generation module is used to solve the scheduling optimization model under the constraints to obtain a power balance scheduling scheme.

[0013] Based on the above method embodiments, the present invention provides a corresponding terminal device embodiment; The present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the power balance scheduling method integrating neural network learning and Monte Carlo simulation described in any embodiment of the present invention.

[0014] Based on the above method embodiments, the present invention provides a corresponding storage medium embodiment; The present invention provides a storage medium including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the power balance scheduling method integrating neural network learning and Monte Carlo simulation described in any embodiment of the present invention.

[0015] The embodiments of the present invention have the following beneficial effects: This invention provides a power balance dispatching method, apparatus, terminal equipment, and storage medium that integrates neural network learning and Monte Carlo simulation. The method includes: acquiring topology data, historical load time-series data, maximum transmission power limits for each line, current power operation data, and predicted meteorological data and date information for the dispatching period for each load node in the distribution network; dividing all load nodes into feeder blocks based on the topology data and maximum transmission power limits to obtain several feeder blocks; calculating a first input feature based on the historical load time-series data, predicted meteorological data, and date information, and inputting it into a preset load prediction model based on a neural network to obtain a baseline load prediction value for each load node in the dispatching period; and using the baseline load prediction value as the mean parameter. Based on the benchmark, a probability distribution is constructed using the predicted meteorological data, the benchmark load forecast value, and the date information. Several random samples of the predicted meteorological data from this probability distribution are then performed using Monte Carlo simulation. Based on the sampling results, historical load time-series data, and the date information, several second input features are generated. Each second input feature is then input into the preset load forecast model to obtain several final load forecast values. Based on all final load forecast values ​​and each feeder block, the load fluctuation boundary of each feeder block is determined. Based on the load fluctuation boundary, power operation data, and several feeder blocks, a scheduling optimization model is constructed with the objective of minimizing total operating cost and ineffective power generation, along with corresponding constraints. The scheduling optimization model is then solved under these constraints to obtain a power balance scheduling scheme.

[0016] Therefore, this invention uses Monte Carlo simulation to sample possible weather conditions for future scheduling periods, and performs load forecasting based on these sampling results to obtain load forecasting results under different weather conditions, generating a load fluctuation boundary. This load fluctuation boundary is subsequently used to construct the scheduling optimization model, enabling the final power balance scheduling scheme to fully consider the node load conditions under different weather conditions, thereby obtaining a more reasonable power balance scheduling result. Attached Figure Description

[0017] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a power balance scheduling method that integrates neural network learning and Monte Carlo simulation, according to an embodiment of the present invention.

[0019] Figure 2 This is a schematic diagram of the structure of a power balance dispatching device that integrates neural network learning and Monte Carlo simulation according to an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in this application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are some embodiments of this application, but not all embodiments.

[0021] Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this application.

[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application; The terms “comprising” and “having”, and any variations thereof, used in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion.

[0023] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly indicating the number, specific order, or primary and secondary relationship of the indicated technical features.

[0024] In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0025] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application.

[0026] The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that is mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0027] In the description of the embodiments of this application, the term "and / or" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0028] Additionally, the character " / " in this article generally indicates that the objects before and after it are in an "or" relationship.

[0029] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0030] In the description of the embodiments of this application, unless otherwise expressly specified and limited, the technical terms such as "installation", "connection", "linking", and "fixing" should be interpreted broadly.

[0031] For example, it can be a fixed connection, a detachable connection, or an integral part; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be a connection within two components or an interaction between two components.

[0032] Those skilled in the art can understand the specific meaning of the above terms in the embodiments of this application according to the specific circumstances.

[0033] See Figure 1 To address the problem that existing technologies rely solely on fixed expected values ​​for load forecasting, leading to significant deviations in forecast results due to abnormal weather or unforeseen events, and consequently, unreasonable power balance dispatching outcomes, this invention provides a power balance dispatching method that integrates neural network learning and Monte Carlo simulation, comprising: Step S101: Obtain the topology data of each load node in the distribution network, historical load time series data, maximum transmission power limit value of each line, current power operation data, and predicted meteorological data and date information for the scheduling period; Specifically, the aforementioned topology data includes the electrical connections and geographical locations between each load node, and the aforementioned power operation data includes: energy storage charging power and energy storage discharging power.

[0034] Specifically, meteorological data includes season, temperature, humidity, and wind speed. The date information mentioned above refers to the type of date in the corresponding scheduling period (e.g., weekday or holiday) and the time of day (morning, noon, or evening).

[0035] Step S102: Based on the topology data and the maximum transmission power limit, divide all load nodes into feeder blocks to obtain several feeder blocks; Specifically, in this invention, the feeder blocks are divided using the concept of clustering to obtain several feeder blocks.

[0036] In a preferred embodiment, based on the topology data and the maximum transmission power limit, all load nodes are divided into feeder blocks to obtain several feeder blocks, including: Obtain the rated power of each load node, as well as the net load forecast under different scenarios; Based on the topology data, the shortest geographical distance between any two load nodes is calculated, and a load node spacing distance matrix is ​​constructed based on the shortest geographical distance. Specifically, the shortest geographical distance is calculated based on the following formula: In the formula, This represents the shortest geographical distance between node i and node j. This represents the actual distance of the optimal path from the starting load node i to the target load node k. This represents the estimated distance of the optimal path from the starting load node i to the target load node k.

[0037] Specifically, the optimal path actual distance is obtained by summing the electrical connection relationships between load nodes and the shortest path algorithm, through traversing the switches and tower connection relationships between nodes, and is used to represent the line length consumed from the starting load node to the target load node along the actual cable.

[0038] Specifically, the optimal path estimation distance is calculated directly based on the geographic coordinates of the load nodes and the Euclidean distance formula, yielding the straight-line geometric length between two load nodes.

[0039] Based on the net load forecast values ​​of all load nodes, the source-load characteristic matching degree between any two load nodes is calculated, and a source-load characteristic matching degree matrix is ​​constructed based on the source-load characteristic matching degree. Specifically, since the time steps of the net load characteristics among nodes are consistent, the differences in predicted values ​​at adjacent times can be used to reflect trend changes. Based on this, and combined with the affine model of the node net load, the degree of matching between source and load characteristics can be calculated. In the formula, This indicates the degree of source-load characteristic matching between load node i and load node j in scenario C at time t. This represents the predicted net load value of load node i under scenario C at time t+1. Let represent the predicted net load of load node i under scenario C at time t. This represents the predicted net load value of load node j under scenario C at time t+1. This represents the predicted net load of load node j under scenario C at time t.

[0040] Specifically, scenario C refers to a random scenario generated through Monte Carlo simulation. Each scenario corresponds to a set of possible weather, date information, and net load fluctuation parameters, which are used to evaluate the degree of source-load matching under different uncertainty conditions.

[0041] The maximum power limit value of the feeder block is calculated based on the maximum transmission power limit value. Specifically, the maximum power limit value of the feeder block is calculated using the following formula: In the formula, This indicates the maximum power limit value for the feeder block. This represents a technical parameter coefficient. In distribution network planning, it is usually stipulated that the feeder load rate of standard wiring should not exceed 50%. In this invention, it is taken as 0.5. This indicates the maximum transmission power limit for line l.

[0042] Based on the rated power, the maximum power limit of the feeder block, the distance matrix between load points, and the source-load characteristic matching degree matrix, the node merging operation is repeatedly performed to obtain several feeder blocks; The node merging operation includes: Get the current cluster set; where the initial cluster set is all load nodes, and each load node is a cluster; Specifically, initially, each load node is a separate cluster, and all load nodes together form the initial cluster set. Outside of the initial stage, a cluster can contain multiple load nodes that can be merged.

[0043] Based on the current cluster set, rated power, load point spacing matrix, and source-load characteristic matching degree matrix, the comprehensive score of the current cluster pair to be merged and each current cluster pair to be merged is calculated. If the current cluster pair to be merged is empty, then each cluster in the current cluster set is taken as a feeder block, and then several feeder blocks are obtained. If there are no cluster pairs to be merged, it means that the feeder block operation for all load nodes has been completed under the condition of meeting all constraints, and several feeder blocks can be generated directly based on the current cluster set.

[0044] Specifically, all load nodes within each cluster are combined into a feeder block, so several feeder blocks can be obtained based on the current cluster set.

[0045] Otherwise, the current cluster pair with the lowest overall score is merged into a new cluster; the target cluster that does not contain the load nodes in the new cluster is extracted from the current cluster set, and the cluster set for the next time step is generated based on the new cluster and the target cluster.

[0046] Specifically, if a load node has been merged, the corresponding cluster needs to be deleted from the original cluster set to avoid the problem of the same load node belonging to different feeder blocks at the same time.

[0047] In this preferred embodiment, based on the topology data and the maximum transmission power limit, all load nodes are divided into feeder blocks, resulting in several feeder blocks.

[0048] In another preferred embodiment, based on the current cluster set, rated power, load point spacing matrix, and source-load characteristic matching matrix, a comprehensive score for the current cluster pairs to be merged and each current cluster pair to be merged is calculated, including: Based on the current cluster set, generate several pairs of current clusters; Specifically, by combining all clusters in the current cluster set in pairs, several pairs of current clusters are obtained.

[0049] The current total power of each current cluster pair is calculated based on the rated power, and cluster pairs whose current total power exceeds the maximum power limit of the feeder block are removed to obtain the currently selected cluster pairs; Specifically, the sum of the rated power of all load nodes within the current cluster is calculated to obtain the corresponding current total power.

[0050] Based on the load point spacing matrix and the source load characteristic matching degree matrix, the comprehensive score of each currently selected cluster pair is calculated. Specifically, before performing node merging operations, a pre-defined comprehensive scoring matrix can be calculated by weighting the load point spacing matrix and the source-load characteristic matching degree matrix of each pair of load node pairs. Subsequently, during the node merging operation, if the current cluster pairs to be merged are all composed of a single load node, the comprehensive score of the corresponding load node pair can be directly read from the pre-defined comprehensive scoring matrix as the comprehensive score of the current cluster pair to be merged.

[0051] If a cluster consisting of multiple load nodes exists in the current cluster pair to be merged, then all load nodes in the current cluster pair are combined across clusters. Based on these combinations, the corresponding comprehensive scores are extracted from the preset comprehensive score matrix. Then, the sum of these comprehensive scores and the total number of pairs are calculated to obtain the average comprehensive score between clusters. Subsequently, this average comprehensive score is used as the comprehensive score of the current cluster pair to be merged.

[0052] For illustration, if the current cluster pairs to be merged are {A, B} and {C, D}, that is, the first cluster contains load nodes A and B, and the second cluster contains load nodes C and D, then after cross-cluster combination, we get the combinations "A, C", "A, D", "B, C" and "B, D". We extract the comprehensive scores corresponding to these combinations from the preset comprehensive score matrix, and then calculate the average comprehensive score to obtain the comprehensive score of the current cluster pairs to be merged, {A, B} and {C, D}.

[0053] Remove currently selected cluster pairs whose overall score exceeds a preset score threshold to obtain the current cluster pairs to be merged.

[0054] In this preferred embodiment, the current cluster pair to be merged and the comprehensive score of each current cluster pair to be merged are calculated based on the current cluster set, rated power, load point spacing matrix and source-load characteristic matching degree matrix.

[0055] Step S103: Based on the historical load time series data, predicted meteorological data and date information, calculate the first input feature and input it into the preset load prediction model based on the neural network to obtain the baseline load prediction value of each load node in the scheduling period. Specifically, the preset load forecasting model mainly consists of two layers: a convolutional neural network (CNN) and a bidirectional gated recurrent unit (BiGRU). Through neural network learning, it takes features constructed based on historical load time series data, forecast meteorological data, and date information as input and outputs load forecast values.

[0056] In a preferred embodiment, based on the historical load time-series data, predicted meteorological data, and date information, a first input feature is calculated and input into a preset load prediction model built on a neural network to obtain a baseline load prediction value for each load node in the scheduling period, including: The historical load time series data is processed by imputing missing values, removing outliers, and normalizing to obtain load input characteristics; Specifically, for missing values, the principle of date similarity is used to find the corresponding similar days, and the missing values ​​are filled by averaging the data of the relative time periods of the similar days. The principle of similarity here can be to select the same date corresponding to the previous year, or the same date corresponding to a preset past year.

[0057] Specifically, when removing outliers, the Laida criterion is first applied to the historical load time-series data after missing value imputation to determine the data's outlier status. Outliers that satisfy the following formula are removed: In the formula, This represents the i'th data point in the historical load time series data after missing value imputation. This represents the average of historical load time series data after missing values ​​were filled. This represents the residual error of the i'th data point in the historical load time series data after missing value imputation. It represents the standard deviation.

[0058] Finally, the data after outlier removal is normalized to address the issue of different data units. The normalization formula is as follows: In the formula, This represents the normalized i'-th historical load, i.e., the load input characteristic. This represents the minimum value in the historical load time series data after missing value imputation. This represents the maximum value in the historical load time series data after missing values ​​have been filled.

[0059] Based on the predicted meteorological data, climate seasonality features are generated to characterize the season and weather of the scheduled period. Based on the date information, generate time-series features to characterize whether the scheduling period is a weekday and the morning / evening time period to which it belongs; The load input characteristics, climate seasonality characteristics, and time series characteristics are concatenated into the first input characteristic; The first input feature is input into a preset load prediction model to obtain the baseline load prediction value of each load node in the scheduling period.

[0060] In this preferred embodiment, a first input feature is calculated based on historical load time series data, predicted meteorological data, and date information, and then input into a preset load prediction model to obtain the baseline load prediction value for each load node in the scheduling period.

[0061] In another preferred embodiment, the first input feature is input to a preset load prediction model to obtain a baseline load prediction value for each load node in the scheduling period, including: The first input feature is input into a preset load prediction model, so that the convolutional neural network in the preset load prediction model performs convolution, max pooling and fully connected operations on the first input feature to obtain a first output feature vector. Specifically, a convolutional neural network (CNN) consists of one-dimensional convolutional layers, pooling layers, and fully connected layers, capable of extracting feature information from workload data. After the first input feature arrives at the CNN, the one-dimensional convolutional layer first performs a convolution operation on the first input feature using convolution kernels, capturing the features of the workload sequence. In the formula, This represents the output of the k-th convolutional kernel at time t. This represents the activation function. This represents the i''th weight of the k-th convolutional kernel. Indicates the number of convolution kernel weights. Indicates the first input feature of the first input feature The value at time, This represents the bias term of the k-th convolution kernel.

[0062] Next, max pooling is used in the pooling layer to filter information and reduce the possibility of overfitting: In the formula, This represents the output of the pooling layer at time t. This represents the output of each convolution kernel at time t.

[0063] The output of the pooling layer is flattened and fed into the fully connected layer, and finally the first output feature vector is obtained through the sigmoid activation function. In the formula, This represents the first output feature vector at time t. Indicates the number of weights in a CNN layer. This represents the j'-th weight of the CNN layer. This represents the j'-th output of the pooling layer. This represents the bias term of the CNN layer.

[0064] Based on the bidirectional gated loop unit in the preset load prediction model, long-term time-series regularity features are extracted from the first output feature vector to obtain the baseline load prediction value of each load node in the scheduling period.

[0065] Specifically, after feature extraction through the CNN layer, a Bidirectional Gated Recurrent Unit (BiGRU) is used to further extract features from the first output feature vector of the CNN layer. BiGRU can uncover patterns in time series data and capture long-term dependencies through bidirectional flow.

[0066] Specifically, the output of BiGRU can be represented by the following formula: In the formula, This represents the output of the BiGRU at time t, i.e., the baseline load forecast value of the load node. Indicates BiGRU, This represents the result output by the forward GRU in the BiGRU at time t. This represents the result output by the backward GRU in the BiGRU at time t. This represents the value at time t in the first input feature.

[0067] Specifically, the forward GRU processes the sequence in the forward direction of time, encoding the temporal dependencies from the beginning to the current time; the backward GRU processes the sequence in the reverse direction of time, encoding the temporal dependencies from the end to the current time.

[0068] Preferably, the CNN model optimized by BiGRU is used to fully extract the temporal and local features of the load data, obtain more accurate prediction results, reduce the number of redundant power generation and energy storage adjustments caused by prediction bias, and improve the regional supply and demand matching degree and the accuracy of dynamic scheduling.

[0069] In this preferred embodiment, the first input feature is input into a preset load prediction model, and the feature is extracted through the CNN layer and BiGRU in the preset load prediction model to obtain the baseline load prediction value of each load node in the scheduling period.

[0070] In another preferred embodiment, the training of the preset load prediction model includes: Acquire historical load sample data under historical scheduling cycles, as well as historical meteorological data, historical date information, and actual historical load data for the corresponding forecast period of the historical load sample data; Based on the historical load sample data, historical meteorological data, historical date information, and real historical load data, several input sample data with real labels are calculated and generated; wherein, the real labels are used to characterize the real historical load data corresponding to the input sample data; Specifically, the generation of input sample data is similar to that of the first input features, except that the input sample data will have real labels for supervised learning.

[0071] The input sample data is input into the load prediction model to be trained for iterative training until the loss function converges, and a trained preset load prediction model is obtained. In each iteration of training, the current load sample prediction value is obtained based on the current input sample data; the current physical constraint penalty value is calculated based on the current load sample prediction value, the preset maximum capacity of the energy storage device, and the preset maximum ramp rate of the generator; and the current loss function is calculated based on the current physical constraint penalty value, the current load sample prediction value, and the corresponding true label. Specifically, to limit variations in load forecasts and ensure that forecasts meet constraints related to generator ramp rate and energy storage capacity, the basic forecast loss item is included. Mean squared error is used to measure the overall deviation between the model's predicted values ​​and the actual load labels. The formula for its calculation is as follows: In the formula, The total number of time points participating in the current training iteration. The model's predicted load sample value at time t. The true label at time t.

[0072] Based on the aforementioned fundamental loss, and combined with physical constraints, the resulting loss function is as follows: In the formula, This represents the value of the loss function. and All represent penalty coefficients. This represents the model's predicted load sample value at time t-1. This indicates the preset maximum climbing speed of the generator. This indicates the maximum capacity of the preset energy storage device.

[0073] Determine whether the current loss function has converged; if the current loss function has converged, then use the current load forecasting model as the preset load forecasting model; otherwise, adjust the parameters in the current load forecasting model and continue training.

[0074] In this preferred embodiment, a pre-defined load prediction model is trained using supervised learning based on historical load sample data under historical scheduling cycles, historical meteorological data, historical date information, and real historical load data for the corresponding prediction period.

[0075] Step S104: Using the baseline load forecast value as the mean reference benchmark, construct a probability distribution based on the forecast meteorological data, the baseline load forecast value, and the date information, and perform several random samplings on the forecast meteorological data in the probability distribution using the Monte Carlo simulation method. Generate several second input features based on the sampling results, historical load time series data, and the date information. Specifically, the preset load forecasting model is represented as: Where Y represents the baseline load forecast, This represents the preset load prediction model obtained after training. , represents the input variable corresponding to the first input feature, W represents the variable corresponding to the seasonal feature, and H represents the variable corresponding to the time series feature.

[0076] Considering the uncertainty of the input variables, they are modeled as probability distributions: The input distribution was analyzed using the Monte Carlo simulation method. Multiple random samples were performed to obtain multiple meteorological data points, which were then concatenated with the corresponding historical load time series data and date information to generate input variables (i.e., the second input feature): .

[0077] Preferably, random sampling of model prediction results based on Monte Carlo simulation is used to obtain the probability distribution characteristics of load prediction, which fully reflects uncertainty, optimizes unit and energy storage scheduling, reduces ineffective power output, effectively copes with load fluctuations and extreme situations, and improves the stability and economy of system operation.

[0078] Step S105: Input each of the second input features into the preset load prediction model to obtain several final load prediction values, and determine the load fluctuation boundary of each feeder block based on all the final load prediction values ​​and each feeder block; Specifically, these second input features are input into a preset load forecasting model to obtain the corresponding load forecasting results and form a set of final load forecast values.

[0079] Subsequently, for each feeder block, based on preset quantiles of the predicted distribution (such as the 2.5% and 97.5% quantiles), the load fluctuation boundary of the feeder block is constructed from the final load prediction values ​​of the load nodes within the feeder block. The predicted quantiles are calculated using the following formula: In the formula, Indicates the preset quantile. This indicates the quantile calculation operation. This represents the set of several final load forecast values ​​obtained after performing load forecasting on different randomly selected input variables.

[0080] Step S106: Based on the load fluctuation boundary, power operation data, and several feeder blocks, construct a scheduling optimization model with the goal of minimizing total operating cost and invalid power, along with corresponding constraints. Specifically, a scheduling optimization model with the goal of minimizing total operating costs and corresponding constraints is constructed using load fluctuation boundaries as the core input and feeder blocks as the unit.

[0081] The scheduling optimization model coordinates the output of conventional generators, renewable energy units, and energy storage devices within this feeder block to meet dynamic load demands while minimizing total system operating costs and ineffective power generation, thereby achieving economical and reliable regional power balance.

[0082] Specifically, the above constraints include robust power balance constraints, unit operation constraints, energy storage operation constraints, and invalid electricity definition constraints.

[0083] Among them, the robust power balance constraint requires that the dispatch output is not lower than the upper limit of the load fluctuation boundary, so as to ensure that in most scenarios, the output can not only meet the baseline load demand, but also cope with the positive fluctuation of the load.

[0084] Specifically, the robust power balance constraint is: In the formula, This represents the predicted output of renewable energy sources (such as wind power and solar power) within the feeder block at time t. This represents the net discharge power of the energy storage device s at time t (positive value indicates discharge state, negative value indicates charging state). This represents the upper bound of the load fluctuation boundary.

[0085] Specifically, the unit operating constraints are: , In the formula, This represents the minimum output of the conventional generator set g within the feeder block. This represents the maximum output of the conventional generator unit g within the feeder block. This represents the output of the conventional generator set g within the feeder block at time t-1.

[0086] Specifically, the constraints for energy storage operation are: In the formula, This represents the energy storage capacity at time t+1. This represents the energy storage capacity at time t. Indicates the energy storage discharge efficiency. This represents the discharge power of the energy storage device s at time t. This represents the charging power of the energy storage device s at time t. Indicates time interval, Indicates energy storage charging efficiency. Indicates the upper limit of energy storage capacity. Indicates the lower limit of energy storage capacity. Indicates the maximum charging power of the energy storage. This indicates the maximum discharge efficiency of the energy storage.

[0087] Specifically, the constraint for defining invalid electricity is as follows: Specifically, when the total available power generation and energy storage regulation capacity exceeds the upper limit of load demand, invalid power is defined.

[0088] Specifically, the decision variables of the scheduling optimization model include: the output of conventional generators in the feeder block, the predicted output of renewable energy in the feeder block, the net discharge power of the energy storage device, the energy storage capacity, and the ineffective power.

[0089] In a preferred embodiment, the objective function of the scheduling optimization model is: In the formula, T represents the total number of time periods in the period to be scheduled. This indicates the total number of conventional generators within the feeder block. This represents the operating cost of unit g. The function representing the operating cost of unit g. This represents the output of the conventional generator unit g within the feeder block at time t. This indicates a high penalty coefficient for invalid electricity usage. This represents the invalid electricity consumption during time period t.

[0090] In this preferred embodiment, the objective function of the scheduling optimization model is constructed with the goal of minimizing total operating cost and invalid power.

[0091] Step S107: Solve the scheduling optimization model under the constraints to obtain the power balance scheduling scheme.

[0092] Specifically, a linear programming solver is used to solve the convex optimization model of the scheduling optimization model to obtain the optimal values ​​of each decision variable in a future scheduling cycle, forming the scheduling plan for the corresponding feeder block in that future scheduling cycle. This scheduling plan takes into account the uncertainty of load forecasting (in this invention, it is...). Within the boundaries of the characterization, the power balance of the corresponding zone can be achieved at the lowest economic cost, and the output of ineffective power can be minimized through the flexible adjustment of the energy storage device, thereby improving the overall economic efficiency and reliability of operation.

[0093] Finally, the results of each feeder block and the corresponding scheduling plan are used as the final power balance scheduling scheme to achieve power balance in the region.

[0094] Preferably, in actual execution, the actual load of the scheduled period can be compared with the predicted value. If the deviation continues to exceed the preset confidence interval, the model is re-optimized, and the scheduling plan is updated using the latest data, forming a closed-loop control of "prediction-optimization-execution-feedback", which further enhances the robustness and adaptability of the scheme.

[0095] Based on the above method embodiments, the present invention provides corresponding apparatus embodiments.

[0096] like Figure 2 As shown, one embodiment of the present invention provides a power balance dispatching device that integrates neural network learning and Monte Carlo simulation, comprising: The module includes a data acquisition module, a feeder block module, a baseline load prediction module, a random sampling module, a boundary generation module, a model building module, and a scheduling scheme generation module. The data acquisition module is used to acquire topology data of each load node in the distribution network, historical load time series data, maximum transmission power limit value of each line, current power operation data, and predicted meteorological data and date information for the scheduling period. The feeder block module is used to divide all load nodes into feeder blocks according to the topology data and the maximum transmission power limit value, thereby obtaining several feeder blocks. The baseline load prediction module is used to calculate the first input feature based on the historical load time series data, predicted meteorological data and date information, and input it into the preset load prediction model based on the neural network to obtain the baseline load prediction value of each load node in the scheduling period. The random sampling module is used to construct a probability distribution based on the mean reference benchmark of the baseline load forecast value, the predicted meteorological data, the baseline load forecast value and the date information, and to perform several random samplings of the predicted meteorological data in the probability distribution using the Monte Carlo simulation method, and to generate several second input features based on the sampling results, historical load time series data and the date information. The boundary generation module is used to input each second input feature into the preset load prediction model to obtain several final load prediction values, and to determine the load fluctuation boundary of each feeder block based on all the final load prediction values ​​and each feeder block. The model building module is used to construct a scheduling optimization model and corresponding constraints based on the load fluctuation boundary, power operation data and several feeder blocks, with the goal of minimizing total operating cost and invalid power. The scheduling scheme generation module is used to solve the scheduling optimization model under the constraints to obtain a power balance scheduling scheme.

[0097] It should be noted that the device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0098] Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any inventive effort.

[0099] The above diagram is merely an example of a power balance dispatching device that integrates neural network learning and Monte Carlo simulation, and does not constitute a limitation on a power balance dispatching device that integrates neural network learning and Monte Carlo simulation. It may include more or fewer components than shown in the diagram, or combine certain components, or use different components.

[0100] Based on the above method embodiments, the present invention provides corresponding terminal device embodiments.

[0101] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor.

[0102] When the processor executes the computer program, it implements the power balance scheduling method that integrates neural network learning and Monte Carlo simulation as described in any embodiment of the present invention.

[0103] For example, in this embodiment, the computer program described above may be divided into one or more modules, one or more of which are stored in the memory and executed by the processor to complete the present invention.

[0104] The aforementioned one or more module elements may be a series of computer program instruction segments capable of performing a specific function, which are used to describe the execution process of the aforementioned computer program in the aforementioned device; The aforementioned terminal devices may be computing devices such as desktop computers, laptops, handheld computers, and cloud servers. These devices may include, but are not limited to, processors and memory. The processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0105] The general-purpose processor can be a microprocessor or any conventional processor, etc. The processor is the control center of the device and connects various parts of the device through various interfaces and lines. The aforementioned memory can be used to store the aforementioned computer programs and / or modules. The aforementioned processor implements various functions of the aforementioned device by running or executing the computer programs and / or modules stored in the aforementioned memory, and by calling the data stored in the memory.

[0106] The aforementioned memory may mainly include a program storage area and a data storage area, wherein the program storage area may store the operating system, at least one application program required for a function, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0107] Based on the above method embodiments, the present invention provides corresponding storage medium embodiments.

[0108] Another embodiment of the present invention provides a storage medium including a stored computer program, wherein, when the computer program is running, it controls the device where the storage medium is located to execute the power balance scheduling method integrating neural network learning and Monte Carlo simulation described in any embodiment of the present invention.

[0109] In this embodiment, the storage medium is a computer-readable storage medium, and the computer program includes computer program code, which may be in the form of source code, object code, executable file, or some intermediate form.

[0110] The aforementioned computer-readable media may include: any entity or device capable of carrying the aforementioned computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0111] The above are preferred embodiments of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A power balance dispatching method integrating neural network learning and Monte Carlo simulation, characterized in that, include: Acquire topology data, historical load time series data, maximum transmission power limit values ​​of each line, current power operation data, and forecast meteorological data and date information for the period to be dispatched in the distribution network for each load node; Based on the topology data and the maximum transmission power limit, all load nodes are divided into feeder blocks to obtain several feeder blocks; Based on the historical load time series data, predicted meteorological data and date information, the first input feature is calculated and input into the preset load prediction model based on neural network to obtain the baseline load prediction value of each load node in the scheduling period. Using the baseline load forecast as the mean reference benchmark, a probability distribution is constructed based on the forecast meteorological data, the baseline load forecast, and the date information. The forecast meteorological data in the probability distribution is randomly sampled several times using the Monte Carlo simulation method. Several second input features are generated based on the sampling results, historical load time series data, and the date information. Each second input feature is input into the preset load prediction model to obtain several final load prediction values, and the load fluctuation boundary of each feeder block is determined based on all the final load prediction values ​​and each feeder block. Based on the load fluctuation boundary, power operation data, and several feeder blocks, a scheduling optimization model with the goal of minimizing total operating cost and invalid power generation, along with corresponding constraints, is constructed. The scheduling optimization model is solved under the constraints to obtain a power balance scheduling scheme.

2. The power balance dispatching method integrating neural network learning and Monte Carlo simulation according to claim 1, characterized in that, Based on the topology data and maximum transmission power limit, all load nodes are divided into feeder blocks, resulting in several feeder blocks, including: Obtain the rated power of each load node, as well as the net load forecast under different scenarios; Based on the topology data, the shortest geographical distance between any two load nodes is calculated, and a load node spacing distance matrix is ​​constructed based on the shortest geographical distance. Based on the net load forecast values ​​of all load nodes, the source-load characteristic matching degree between any two load nodes is calculated, and a source-load characteristic matching degree matrix is ​​constructed based on the source-load characteristic matching degree. The maximum power limit value of the feeder block is calculated based on the maximum transmission power limit value. Based on the rated power, the maximum power limit of the feeder block, the distance matrix between load points, and the source-load characteristic matching degree matrix, the node merging operation is repeatedly performed to obtain several feeder blocks; The node merging operation includes: Get the current cluster set; where the initial cluster set is all load nodes, and each load node is a cluster; Based on the current cluster set, rated power, load point spacing matrix, and source-load characteristic matching degree matrix, the comprehensive score of the current cluster pair to be merged and each current cluster pair to be merged is calculated. If the current cluster pair to be merged is empty, then each cluster in the current cluster set is taken as a feeder block, and then several feeder blocks are obtained. Otherwise, the current cluster pair with the lowest overall score is merged into a new cluster; the target cluster that does not contain the load nodes in the new cluster is extracted from the current cluster set, and the cluster set for the next time step is generated based on the new cluster and the target cluster.

3. The power balance dispatching method integrating neural network learning and Monte Carlo simulation according to claim 2, characterized in that, Based on the current cluster set, rated power, load point spacing matrix, and source-load characteristic matching matrix, the comprehensive score of the current cluster pairs to be merged and each current cluster pair to be merged is calculated, including: Based on the current cluster set, generate several pairs of current clusters; The current total power of each current cluster pair is calculated based on the rated power, and cluster pairs whose current total power exceeds the maximum power limit of the feeder block are removed to obtain the currently selected cluster pairs; Based on the load point spacing matrix and the source load characteristic matching degree matrix, the comprehensive score of each currently selected cluster pair is calculated. Remove currently selected cluster pairs whose overall score exceeds a preset score threshold to obtain the current cluster pairs to be merged.

4. The power balance dispatching method integrating neural network learning and Monte Carlo simulation according to claim 3, characterized in that, Based on the historical load time-series data, predicted meteorological data, and date information, a first input feature is calculated and input into a preset load prediction model built on a neural network to obtain the baseline load prediction value for each load node in the scheduling period, including: The historical load time series data is processed by imputing missing values, removing outliers, and normalizing to obtain load input characteristics; Based on the predicted meteorological data, climate seasonality features are generated to characterize the season and weather of the scheduled period. Based on the date information, generate time-series features to characterize whether the scheduling period is a weekday and the morning / evening time period to which it belongs; The load input characteristics, climate seasonality characteristics, and time series characteristics are concatenated into the first input characteristic; The first input feature is input into a preset load prediction model to obtain the baseline load prediction value of each load node in the scheduling period.

5. The power balance dispatching method integrating neural network learning and Monte Carlo simulation according to claim 4, characterized in that, The first input feature is input into a preset load prediction model to obtain the baseline load prediction value for each load node in the scheduling period, including: The first input feature is input into a preset load prediction model, so that the convolutional neural network in the preset load prediction model performs convolution, max pooling and fully connected operations on the first input feature to obtain a first output feature vector. Based on the bidirectional gated loop unit in the preset load prediction model, long-term time-series regularity features are extracted from the first output feature vector to obtain the baseline load prediction value of each load node in the scheduling period.

6. The power balance dispatching method integrating neural network learning and Monte Carlo simulation according to claim 5, characterized in that, The training of the preset load prediction model includes: Acquire historical load sample data under historical scheduling cycles, as well as historical meteorological data, historical date information, and actual historical load data for the corresponding forecast period of the historical load sample data; Based on the historical load sample data, historical meteorological data, historical date information, and real historical load data, several input sample data with real labels are calculated and generated; wherein, the real labels are used to characterize the real historical load data corresponding to the input sample data; The input sample data is input into the load prediction model to be trained for iterative training until the loss function converges, and a trained preset load prediction model is obtained. In each iteration of training, the current load sample prediction value is obtained based on the current input sample data; the current physical constraint penalty value is calculated based on the current load sample prediction value, the preset maximum capacity of the energy storage device, and the preset maximum ramp rate of the generator; and the current loss function is calculated based on the current physical constraint penalty value, the current load sample prediction value, and the corresponding true label. Determine whether the current loss function has converged; if the current loss function has converged, then use the current load forecasting model as the preset load forecasting model; otherwise, adjust the parameters in the current load forecasting model and continue training.

7. The power balance dispatching method integrating neural network learning and Monte Carlo simulation according to claim 6, characterized in that, The objective function of the scheduling optimization model is: In the formula, T represents the total number of time periods in the period to be scheduled. This indicates the total number of conventional generators within the feeder block. This represents the operating cost of unit g. The function representing the operating cost of unit g. This represents the output of the conventional generator unit g within the feeder block at time t. This indicates a high penalty coefficient for invalid electricity usage. This represents the invalid electricity consumption during time period t.

8. A power balance dispatching device integrating neural network learning and Monte Carlo simulation, characterized in that, include: The module includes a data acquisition module, a feeder block module, a baseline load prediction module, a random sampling module, a boundary generation module, a model building module, and a scheduling scheme generation module. The data acquisition module is used to acquire topology data of each load node in the distribution network, historical load time series data, maximum transmission power limit value of each line, current power operation data, and predicted meteorological data and date information for the scheduling period. The feeder block module is used to divide all load nodes into feeder blocks according to the topology data and the maximum transmission power limit value, thereby obtaining several feeder blocks. The baseline load prediction module is used to calculate the first input feature based on the historical load time series data, predicted meteorological data and date information, and input it into the preset load prediction model based on the neural network to obtain the baseline load prediction value of each load node in the scheduling period. The random sampling module is used to construct a probability distribution based on the mean reference benchmark of the baseline load forecast value, the predicted meteorological data, the baseline load forecast value and the date information, and to perform several random samplings of the predicted meteorological data in the probability distribution using the Monte Carlo simulation method, and to generate several second input features based on the sampling results, historical load time series data and the date information. The boundary generation module is used to input each second input feature into the preset load prediction model to obtain several final load prediction values, and to determine the load fluctuation boundary of each feeder block based on all the final load prediction values ​​and each feeder block. The model building module is used to construct a scheduling optimization model and corresponding constraints based on the load fluctuation boundary, power operation data and several feeder blocks, with the goal of minimizing total operating cost and invalid power. The scheduling scheme generation module is used to solve the scheduling optimization model under the constraints to obtain a power balance scheduling scheme.

9. A terminal device, characterized in that, The device includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements a power balance scheduling method that integrates neural network learning and Monte Carlo simulation as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device where the storage medium is located to perform a power balance scheduling method that integrates neural network learning and Monte Carlo simulation as described in any one of claims 1 to 7.

Images