Power grid unit dispatching method, device, equipment, medium and product

Abstract

The application relates to a power grid unit scheduling method, device, equipment, medium and product. The method comprises the following steps: constructing an adjacency matrix corresponding to a power grid equipment connection relationship and a space-time feature matrix corresponding to a multi-period operation state of the power grid equipment according to power grid operation data of the power grid equipment; performing influence probability prediction on the adjacency matrix and the space-time feature matrix to obtain a predicted probability corresponding to each physical line in the power grid equipment; wherein the predicted probability is used to indicate the probability of the standard physical line affecting the unit scheduling of the power grid equipment; determining a correction capacity corresponding to at least one key line in each physical line according to the predicted probability; and determining a power grid unit scheduling scheme according to each key line and the correction capacity of each key line. The method can improve the efficiency of power grid unit scheduling.

Description

Technical Field

[0001] This application relates to the field of power grid dispatching technology, and in particular to a power grid unit dispatching method, apparatus, equipment, medium and product. Background Technology

[0002] With the high proportion of new energy sources connected to the grid in new power systems and the increasingly complex grid structure, smart grid dispatching and optimization technologies have developed rapidly, and safety-constrained unit combination technology has become the core supporting technology for day-ahead dispatching and economic dispatching of the power grid.

[0003] In traditional power grid unit dispatching methods, when constructing a safety-constrained unit combination model, the power flow constraints of all physical lines in the entire network are usually included in the optimization model and solved directly to obtain the initial unit start-up and shutdown and output dispatching scheme.

[0004] However, traditional power grid unit scheduling methods incorporate all line constraints across the entire network into the model without differentiation, resulting in a large number of constraints and a large model dimension, leading to low power grid unit scheduling efficiency. Summary of the Invention

[0005] Therefore, it is necessary to provide a power grid unit dispatching method, device, equipment, medium, and product that can improve the dispatching efficiency of power grid units, addressing the aforementioned technical problems.

[0006] Firstly, this application provides a power grid unit dispatching method, including:

[0007] Based on the power grid operation data of power grid equipment, construct the adjacency matrix corresponding to the connection relationship of power grid equipment and the spatiotemporal feature matrix corresponding to the multi-period operation status of power grid equipment;

[0008] Impact probability prediction is performed on the adjacency matrix and spatiotemporal feature matrix to obtain the predicted probability of each physical line in the power grid equipment; whereby the predicted probability is used to characterize the probability that the physical line will affect the unit scheduling in the power grid equipment.

[0009] Based on the predicted probability, determine the correction capacity corresponding to at least one critical path in each physical path;

[0010] Based on each critical path and its corrected capacity, determine the power grid unit dispatching scheme.

[0011] In one embodiment, step A includes: performing influence probability prediction on the adjacency matrix and the spatiotemporal feature matrix to obtain the predicted probability corresponding to each physical line in the power grid equipment, including:

[0012] Feature extraction is performed on the adjacency matrix and the spatiotemporal feature matrix to obtain the node features of each physical line in the power grid;

[0013] Based on the node characteristics of each physical line, the probability of influence is predicted to obtain the predicted probability of each physical line in the power grid equipment.

[0014] In one embodiment, the influence probability is predicted based on the node characteristics of each physical line to obtain the predicted probability corresponding to each physical line in the power grid equipment, including:

[0015] Based on the bus number corresponding to each physical line, the node features corresponding to each physical line are concatenated to obtain the line feature vector.

[0016] The predicted probability of each physical line in the power grid is obtained by predicting the impact probability based on the line feature vector.

[0017] In one embodiment, determining the correction capacity corresponding to at least one critical path among the physical paths based on the predicted probability includes:

[0018] The shrinkage factor is determined based on the preset capacity shrinkage coefficient and the predicted probability corresponding to each critical path.

[0019] The rated capacity of each critical path is corrected based on the shrinkage factor to obtain the corrected capacity for each critical path.

[0020] In one embodiment, a grid unit scheduling scheme is determined based on each critical line and its corrected capacity, including:

[0021] Based on each critical path and its modified capacity, construct the initial unit combination result;

[0022] Power flow verification is performed on each critical path in the initial unit combination results;

[0023] If, based on the power flow verification results, it is determined that there is at least one power flow exceeding the limit among each critical path, then the constraint compensation priority for each power flow exceeding the limit is constructed based on the degree of exceeding the limit and the predicted probability of each exceeding the limit.

[0024] Based on the priority of each constraint, a new initial unit combination result is constructed;

[0025] Based on the new initial unit combination results, return to the step of performing power flow verification on each critical path in the initial unit combination results. If, based on the power flow verification results, it is determined that there are no power flow exceeding the limit among the critical paths, then the initial unit combination results are used as the grid unit dispatching scheme.

[0026] In one embodiment, based on the power flow verification results, it is determined that at least one power flow violation path exists among the critical paths, including:

[0027] Based on the power flow verification results, determine the net injected power of each bus in each time period within the target scheduling cycle;

[0028] Based on the preset line reactance parameters, balancing node configuration, and net injected power, determine the actual power flow value of each physical line;

[0029] Among multiple physical lines, at least one physical line whose actual power flow value is greater than the rated capacity is selected, and based on the selected physical line, at least one power flow over-limit line is determined.

[0030] Secondly, this application also provides a power grid unit dispatching device, comprising:

[0031] The construction module is used to construct the adjacency matrix corresponding to the connection relationship of the power grid equipment and the spatiotemporal feature matrix corresponding to the multi-period operation status of the power grid equipment based on the power grid operation data of the power grid equipment.

[0032] The prediction module is used to predict the impact probability of the adjacency matrix and the spatiotemporal feature matrix to obtain the predicted probability of each physical line in the power grid equipment; wherein, the predicted probability is used to characterize the probability that the physical line will affect the unit scheduling in the power grid equipment.

[0033] The processing module is used to determine the correction capacity corresponding to at least one critical line in each physical line based on the predicted probability.

[0034] The scheduling module is used to determine the grid unit scheduling scheme based on each critical line and its corrected capacity.

[0035] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0036] Based on the power grid operation data of power grid equipment, construct the adjacency matrix corresponding to the connection relationship of power grid equipment and the spatiotemporal feature matrix corresponding to the multi-period operation status of power grid equipment;

[0037] Impact probability prediction is performed on the adjacency matrix and spatiotemporal feature matrix to obtain the predicted probability of each physical line in the power grid equipment; whereby the predicted probability is used to characterize the probability that the physical line will affect the unit scheduling in the power grid equipment.

[0038] Based on the predicted probability, determine the correction capacity corresponding to at least one critical path in each physical path;

[0039] Based on each critical path and its corrected capacity, determine the power grid unit dispatching scheme.

[0040] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0041] Based on the power grid operation data of power grid equipment, construct the adjacency matrix corresponding to the connection relationship of power grid equipment and the spatiotemporal feature matrix corresponding to the multi-period operation status of power grid equipment;

[0042] Impact probability prediction is performed on the adjacency matrix and spatiotemporal feature matrix to obtain the predicted probability of each physical line in the power grid equipment; whereby the predicted probability is used to characterize the probability that the physical line will affect the unit scheduling in the power grid equipment.

[0043] Based on the predicted probability, determine the correction capacity corresponding to at least one critical path in each physical path;

[0044] Based on each critical path and its corrected capacity, determine the power grid unit dispatching scheme.

[0045] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0046] Based on the power grid operation data of power grid equipment, construct the adjacency matrix corresponding to the connection relationship of power grid equipment and the spatiotemporal feature matrix corresponding to the multi-period operation status of power grid equipment;

[0047] Impact probability prediction is performed on the adjacency matrix and spatiotemporal feature matrix to obtain the predicted probability of each physical line in the power grid equipment; whereby the predicted probability is used to characterize the probability that the physical line will affect the unit scheduling in the power grid equipment.

[0048] Based on the predicted probability, determine the correction capacity corresponding to at least one critical path in each physical path;

[0049] Based on each critical path and its corrected capacity, determine the power grid unit dispatching scheme.

[0050] The aforementioned power grid unit scheduling methods, devices, equipment, media, and products construct an adjacency matrix representing the grid connection relationships and a spatiotemporal feature matrix reflecting the multi-period operating status of the units based on power grid operation data. Then, based on the adjacency matrix and spatiotemporal feature matrix, the probability of each physical line's impact on power grid unit scheduling is predicted. Critical lines are selected based on the predicted probabilities, and corresponding corrected capacities are determined. Finally, by combining the critical lines and their corrected capacities, a power grid unit scheduling scheme is obtained. This method simplifies the constraint scale of the scheduling model, reduces the computational load and time spent on optimization, and thus improves the computational efficiency of power grid unit scheduling. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is an application environment diagram of the power grid unit scheduling method in one embodiment;

[0053] Figure 2 This is a flowchart illustrating a power grid unit scheduling method in one embodiment;

[0054] Figure 3 This is a flowchart illustrating the power grid unit scheduling method in yet another embodiment;

[0055] Figure 4 This is a flowchart illustrating the power grid unit scheduling method in another embodiment;

[0056] Figure 5 This is a structural block diagram of a power grid unit dispatching device in one embodiment;

[0057] Figure 6 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0059] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0060] The power grid unit dispatching method provided in this application embodiment can be applied to, for example, Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or located in the cloud or on other network servers. Terminal 102 sends a power grid unit scheduling request to server 104. Server 104 receives the power grid unit scheduling request, executes the power grid unit scheduling method, and feeds back the power grid unit scheduling scheme to terminal 102. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, drones, low-altitude aircraft, IoT devices, and portable wearable devices. IoT devices can be smart speakers, smart TVs, smart air conditioners, smart vehicle devices, projection devices, etc. Portable wearable devices can be smartwatches, smart bracelets, head-mounted devices, etc. Head-mounted devices can be virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, etc. Server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0061] In one exemplary embodiment, such as Figure 2 As shown, a power grid unit scheduling method is provided, which can be applied to... Figure 1 Taking the server in the example of this, the explanation includes:

[0062] Step 201: Based on the power grid operation data of the power grid equipment, construct the adjacency matrix corresponding to the connection relationship of the power grid equipment and the spatiotemporal feature matrix corresponding to the multi-period operation status of the power grid equipment.

[0063] Among them, power grid equipment can be all key entities that constitute the basis of the physical connection and operation status of the power grid, including buses, physical lines and new energy power plants (such as wind farms and photovoltaic power plants); power grid operation data includes but is not limited to the topology parameters, line parameters, bus load forecast data and new energy output forecast data of the power grid; the adjacency matrix can be a dimensionless normalized matrix constructed based on the power grid topology and line parameters, with the number of rows and columns equal to the total number of power grid buses, and the matrix elements are the normalized electrical coupling weights between buses, used to characterize the electrical connection relationship of power grid equipment (such as buses); the spatiotemporal feature matrix can be a two-dimensional matrix constructed based on load and new energy forecast data, with the dimension being the number of buses × 24 time periods, and the matrix elements being the net load value of each bus for the corresponding time period, used to characterize the operation status characteristics of power grid equipment (such as buses) in multiple time periods.

[0064] Optionally, the topology parameters, line parameters, bus load forecast data, and renewable energy output forecast data of the power grid are obtained; an adjacency matrix of the power grid is constructed based on the topology parameters and line parameters, and a multi-period net load sequence of the bus is constructed based on the bus load forecast data and renewable energy output forecast data, and the multi-period net load sequence of the bus is used as a spatiotemporal feature matrix.

[0065] Step 202: Perform influence probability prediction on the adjacency matrix and spatiotemporal feature matrix to obtain the predicted probability of each physical line in the power grid equipment.

[0066] Among them, the predicted probability is used to characterize the probability that physical lines will affect the scheduling of generating units in the power grid.

[0067] Among them, the physical line can be a power transmission line connecting two busbars.

[0068] Optionally, a trained graph neural network constraint classifier can be used to process the adjacency matrix and spatiotemporal feature matrix to output the predicted probability of each physical line becoming an active constraint within the target scheduling period, thus obtaining the predicted probability of each physical line in the power grid equipment.

[0069] Optionally, the standard bus list in the network model library of the power grid energy management system is read; the bus load forecast data and renewable energy output forecast data are mapped to the numbering system of the standard bus list; for the i-th bus and the t-th time period, the load forecast value of the bus is read, and then the sum of the forecast output of all renewable energy connected to the bus in the same time period is subtracted to obtain the net load of the bus in that time period; the calculated net load is arranged according to the bus order and time period to generate a spatiotemporal feature matrix with a size of the number of buses multiplied by 24.

[0070] Optionally, the power transmission distribution factor of each physical line under unit node injection disturbance can be calculated using a DC power flow calculation program; the absolute value of the power transmission distribution factor is used as the original edge weight, and normalized according to the sum of the original edge weights of the adjacent lines of the first bus to obtain the line-level normalized edge weight; according to the bus number of the first and last bus, the line-level normalized edge weight is written into the corresponding position. If there are multiple parallel lines between the two buses, the sum of the edge weights of each physical line is written in; if there is no direct connection, the corresponding position is set to zero, thus forming a dimensionless normalized adjacency matrix.

[0071] Step 203: Based on the predicted probability, determine the correction capacity corresponding to at least one critical path in each physical path.

[0072] Among them, the critical line can be the line that plays a key constraint role in power grid security and unit scheduling; the corrected capacity can be the value obtained after adaptively and safely shrinking the original rated capacity of the critical line, which serves as the upper limit of the actual available transmission power of the line in the dimensionality reduction model.

[0073] Optionally, based on the predicted probability of physical lines, a risk ranking quantity is constructed by combining the rated capacity of physical lines with electrical coupling information to screen critical lines. Then, the rated capacity of critical lines is adaptively and safely reduced by combining the predicted probability to obtain a corrected capacity that adapts to scheduling risks.

[0074] Step 204: Determine the grid unit dispatching scheme based on each critical path and its corrected capacity.

[0075] Among them, the power grid unit dispatching scheme can be the final unit combination result that meets the safety constraints of all lines in the entire network and takes into account the operating characteristics of the units, including the start-up and shutdown status of each generator unit and the output plan for multiple time periods.

[0076] Optionally, a dimension-reduced safety-constrained unit combination model is constructed based on the set of retained lines (i.e. each critical line) and the corresponding modified capacity; the dimension-reduced safety-constrained unit combination model is solved to obtain the unit combination result, i.e., the power grid unit dispatching scheme.

[0077] In the aforementioned power grid unit scheduling method, an adjacency matrix representing the grid connection relationship and a spatiotemporal feature matrix reflecting the multi-period operating status of the units are constructed based on power grid operation data. Then, the probability of each physical line's impact on power grid unit scheduling is predicted based on the adjacency matrix and spatiotemporal feature matrix. Critical lines are selected based on the predicted probabilities, and corresponding corrected capacities are determined. Finally, by combining the critical lines and their corrected capacities, a power grid unit scheduling scheme is obtained. This method simplifies the constraint scale of the scheduling model, reduces the computational load and time consumption of optimization solutions, and thus improves the computational efficiency of power grid unit scheduling.

[0078] In one exemplary embodiment, such as Figure 3 As shown, the influence probability prediction is performed on the adjacency matrix and spatiotemporal feature matrix to obtain the predicted probability corresponding to each physical line in the power grid equipment, including:

[0079] Step 301: Extract features from the adjacency matrix and the spatiotemporal feature matrix to obtain the node features of each physical line in the power grid.

[0080] Among them, node features can be the feature vectors of the integrated topology association and multi-time period operation status carried by a bus, that is, the features that characterize the node attributes at both ends of each physical line.

[0081] Optionally, the spatiotemporal feature matrix is ​​normalized according to the baseline power and used as the initial node feature matrix of the graph neural network; the adjacency matrix is ​​used as the weight matrix of the graph structure. Node features are propagated and aggregated through two layers of graph convolution. Each layer of graph convolution sequentially performs weighted summation of the adjacency matrix, linear mapping operations, and uses a linear rectified activation function to complete the nonlinear transformation, refining the deep features of the nodes layer by layer. After the two layers of graph convolution operations, the high-order node features of all buses in the power grid are obtained, that is, the node features corresponding to the first and last buses of each physical line.

[0082] Step 302: Based on the node characteristics of each physical line, predict the impact probability to obtain the predicted probability of each physical line in the power grid equipment.

[0083] Optionally, the line feature vector is input into a fully connected classification structure, where linear transformation and activation function processing are performed sequentially to achieve a high-order classification mapping of the line features. Then, the Sigmoid activation function is used to map the output of the fully connected layer to the 0-1 interval, ultimately obtaining the predicted impact probability for each physical line.

[0084] In this embodiment, feature extraction is performed on the adjacency matrix and spatiotemporal feature matrix to obtain the node features of each physical line in the power grid. Based on the node features of each physical line, influence probability prediction is performed to obtain the predicted probability of each physical line in the power grid equipment. This enables automatic identification and quantification of the constraint impact of each physical line on unit scheduling from the power grid topology and operating status, eliminating redundant line constraints that have no impact on scheduling from the source, thereby reducing the number of constraints and modeling dimensionality of subsequent safety constraint unit combination models.

[0085] In one exemplary embodiment, such as Figure 4 As shown, based on the node characteristics of each physical line, the influence probability is predicted to obtain the predicted probability corresponding to each physical line in the power grid equipment, including:

[0086] Step 401: Based on the bus number corresponding to each physical line, concatenate the node features corresponding to each physical line to obtain the line feature vector.

[0087] Optionally, retrieve the starting bus number and ending bus number corresponding to each physical line from the power grid network model library to determine the node affiliation at both ends of the line; based on the starting and ending bus numbers, accurately extract the node feature vector of the corresponding bus from the high-order node feature matrix output by graph convolution; concatenate and splice the node feature vectors of the starting and ending bus nodes of the same physical line in the feature dimension to generate the line feature vector corresponding to the physical line.

[0088] Step 402: Based on the line feature vector, predict the impact probability to obtain the predicted probability of each physical line in the power grid equipment.

[0089] Optionally, the spatiotemporal feature matrix is ​​normalized to serve as the initial node feature matrix, and the adjacency matrix is ​​used as the weight matrix of the graph structure. The node features are propagated and aggregated through two layers of graph convolution. Each layer of graph convolution operation includes weighted summation of the adjacency matrix and linear mapping, and a nonlinear transformation is performed using a linear rectified activation function. Based on the bus number of the first and last ends of the line, the corresponding vector is extracted from the node features output by the second layer of graph convolution and concatenated to form the line feature vector. The line feature vector is input into a fully connected classification structure, and after linear transformation and Sigmoid function mapping, the probability value of the line becoming an active constraint is output, that is, the influence prediction probability corresponding to each physical line is obtained.

[0090] In this embodiment, a unique line feature vector is generated by splicing the node features of the busbars at both ends of the physical line, and the line impact probability prediction is completed based on the feature vector. This can accurately characterize the line constraint attributes and predict the degree of constraint impact of the line on unit scheduling, providing a reliable quantitative basis for subsequent key line selection and model dimensionality reduction.

[0091] In an exemplary embodiment, determining the corrected capacity corresponding to at least one critical line among the physical lines based on the predicted probability includes: determining a shrinkage factor based on a preset capacity shrinkage coefficient and the predicted probability corresponding to each critical line; and correcting the rated capacity of each critical line based on the shrinkage factor to obtain the corrected capacity corresponding to each critical line.

[0092] Among them, the preset capacity shrinkage coefficient is used to control the shrinkage range of the line capacity; the shrinkage factor can be a dimensionless coefficient calculated from the shrinkage coefficient and the predicted probability, which is used to adaptively scale the rated capacity of the line.

[0093] Optionally, a capacity shrinkage factor (i.e., a preset capacity shrinkage factor) can be set; the rated capacity of the reserved line (i.e. the critical line) is multiplied by the shrinkage factor to generate the corrected capacity, and the corrected capacity is used as the actual available transmission limit of the reserved line in the dimensionality reduction model; wherein, the shrinkage factor can be determined by the product of 1 minus the capacity shrinkage factor and 1 minus the difference of the predicted probability.

[0094] In this embodiment, by predicting probabilities to adaptively generate shrinkage factors and correct line capacity, the line capacity can be dynamically adjusted according to line constraint risks. This achieves dimensionality reduction of the safety constraint unit combination model while ensuring the operational safety of the scheduling scheme and significantly improving the accuracy of the dimensionality reduction model.

[0095] In an exemplary embodiment, determining a power grid unit scheduling scheme based on each critical line and its corrected capacity includes: constructing an initial unit combination result based on each critical line and its corrected capacity; performing power flow verification on each critical line in the initial unit combination result; if, based on the power flow verification result, it is determined that at least one power flow exceeding the limit exists among each critical line, then constructing a constraint compensation priority for each power flow exceeding the limit based on the line exceeding the limit degree and predicted probability corresponding to each power flow exceeding the limit; constructing a new initial unit combination result based on each constraint compensation priority; and, based on the new initial unit combination result, returning to the step of performing power flow verification on each critical line in the initial unit combination result; if, based on the power flow verification result, it is determined that no power flow exceeding the limit exists among each critical line, then the initial unit combination result is used as the power grid unit scheduling scheme.

[0096] The initial unit combination result can be obtained by solving the dimensionality-reduced safety constraint unit combination model, resulting in the start-up and shutdown status of each generator unit and the output plan for multiple time periods; the power flow verification result can be the result of verifying whether the actual power flow of each line exceeds the rated capacity; the power flow exceeding the limit line can be the physical line whose actual power flow value is greater than the original rated capacity after power flow verification; the line exceeding the limit degree can be the maximum exceeding ratio of the exceeding line in 24 time periods, i.e., the maximum value of (actual power flow value - original rated capacity) / original rated capacity; the constraint compensation priority can be a quantitative index constructed by combining the line exceeding the limit degree and the prediction probability, used to determine the order of constraint compensation for exceeding the limit line.

[0097] Optionally, a dimensionality-reduced safety-constrained unit combination model is constructed based on the retained line set and the corresponding corrected capacity. The dimensionality-reduced safety-constrained unit combination model is solved to obtain the unit combination result (i.e., the initial unit combination result). The maximum over-limit ratio of each over-limit line in multiple time periods (e.g., 24 time periods) is calculated (the maximum over-limit ratio is the difference between the actual power flow value and the original rated capacity divided by the original rated capacity; the maximum over-limit ratio is multiplied by 1 and the sum of the predicted probabilities), i.e., the degree of line over-limit, and a compensation priority index (i.e., constraint compensation priority) is generated. According to the compensation priority index from largest to smallest, several over-limit lines (i.e., power flow over-limit lines) are selected and added to the compensation set. For each line in the compensation set, its power transmission distribution factor is read from the network model library, and a power flow constraint expression is constructed. The constructed power flow constraint expression is written into the dimensionality-reduced safety-constrained unit combination model in an incremental manner, and the constraint upper limit is set to the original rated capacity of the line. After each write-back of the over-limit line constraints and re-solve, the DC power flow calculation for the entire network is executed again. If all lines satisfy the condition that the actual power flow value is less than or equal to the original rated capacity in all time periods, the safety constraint conditions for the entire network are met, the backfilling process is stopped, and the current unit combination result is output as the final dispatching scheme (i.e., the grid unit dispatching scheme). If there are still over-limit lines, the constraint backfilling process based on the backfilling priority index is continued until convergence.

[0098] Optionally, the method to determine that there is at least one power flow exceeding the limit among each critical path can be as follows: read the normalized edge weight of each physical path recorded in the path edge table; multiply the predicted probability by a gain term, which includes 1 plus the ratio of the line's rated capacity to the network's reference capacity and then multiplied by the normalized edge weight, to calculate the path risk ranking quantity; rank all lines in the network according to the path risk ranking quantity, and determine the set of retained paths, i.e., at least one power flow exceeding the limit, in combination with a preset configuration threshold.

[0099] In this embodiment, by using dimensionality reduction modeling for rapid solution, priority constraint replenishment, and iterative power flow verification, the computational load of solving the safety constraint unit combination model is reduced while ensuring that the final scheduling scheme meets the safety requirements of the entire network.

[0100] In an exemplary embodiment, determining that at least one power flow exceeding the limit exists among each critical line based on the power flow verification results includes: determining the net injected power of each bus in each time period within the target scheduling cycle based on the power flow verification results; determining the actual power flow value of each physical line based on preset line reactance parameters, balancing node configuration, and net injected power; selecting at least one physical line whose actual power flow value is greater than the rated capacity among multiple physical lines, and determining at least one power flow exceeding the limit based on the selected physical line.

[0101] Among them, net injected power can be the total power generation of each bus in the corresponding time period minus the total load power; line reactance parameters can be the inherent reactance values ​​of physical lines in the power grid network model library; balancing node configuration can be the power balance reference node parameters set in the DC power flow calculation; actual power flow value can be the actual active power transmitted by the physical lines in each time period.

[0102] Optionally, the output of each unit in the unit combination result is summarized by bus to obtain the net injected power of each bus in each time period; the DC power flow calculation module is called, and the actual power flow value of each physical line in the whole network in each time period is calculated in combination with the line reactance parameters and slack node settings in the network model library; the actual power flow value is compared with the original rated capacity in the network model library time by time period to identify the lines and time periods that meet the condition that the actual power flow value is greater than the original rated capacity.

[0103] In this embodiment, by accurately calculating the net injected power of the bus, efficiently carrying out DC power flow calculation and comparison screening, the power flow verification and over-limit line location of the unit combination results are accurately completed, providing a precise basis for subsequent constraint replenishment, and effectively ensuring the accuracy and timeliness of power grid dispatch safety verification.

[0104] In an exemplary embodiment, the power grid unit scheduling method may further include: acquiring the power grid's topology parameters, line parameters, bus load forecast data, and renewable energy output forecast data; constructing the power grid's adjacency matrix based on the topology parameters and line parameters, constructing a multi-period net load sequence for the bus based on the bus load forecast data and renewable energy output forecast data, using the multi-period net load sequence for the bus as a spatiotemporal feature matrix, and using the adjacency matrix and the spatiotemporal feature matrix as input data; processing the input data using a trained graph neural network model to output the predicted probability that each line in the entire network will become an active constraint within a target scheduling period (e.g., 24 hours); constructing a line risk ranking quantity based on the predicted probability, the line's rated capacity, and the electrical coupling information in the adjacency matrix; ranking all lines in the entire network according to the line risk ranking quantity, and determining the set of retained lines (i.e., multiple key lines) in conjunction with a preset configuration threshold. The process involves: (1) Adaptively shrinking the rated capacity of each line in the reserved line set according to the predicted probability to generate a corrected capacity; constructing a dimensionality-reduced safety-constrained unit combination model based on the reserved line set and the corresponding corrected capacity; solving the dimensionality-reduced safety-constrained unit combination model to obtain the unit combination result (i.e., the initial unit combination result); calculating the DC power flow of the entire network based on the unit combination result, identifying lines that exceed power flow limits (i.e., power flow exceeding limit lines), and constructing a replenishment priority index (i.e., constraint replenishment priority) based on the degree of exceeding limits and the corresponding predicted probability; selecting high-priority exceeding limit lines according to the replenishment priority index, writing the original network constraints corresponding to the selected exceeding limit lines back into the dimensionality-reduced safety-constrained unit combination model, and resolving and verifying until the safety constraints of the entire network are met, and outputting the final scheduling scheme (i.e., the power grid unit scheduling scheme).

[0105] Optionally, the topology parameters, line parameters, bus load forecast data, and renewable energy output forecast data of the power grid are obtained; an adjacency matrix of the power grid is constructed based on the topology parameters and line parameters; a multi-period net load sequence of the bus is constructed based on the bus load forecast data and renewable energy output forecast data; the multi-period net load sequence of the bus is used as a spatiotemporal feature matrix; and the adjacency matrix and the spatiotemporal feature matrix are used as input data.

[0106] This embodiment is used to organize the scattered power grid operation data stored in day-ahead scheduling into a graph structure input that can be directly received by the subsequent constraint classifier. The processing result still maintains the core form of the original scheme, that is, it consists of a spatiotemporal feature matrix. and adjacency matrix This data collectively characterizes the multi-period operating status and electrical connection relationships of the power grid. The input data comes directly from the original data source of the dispatching business system and mainly includes four parts: First, power grid topology and line parameter data, sourced from the network model library of the energy management system or the dispatch automation master station database, typically obtained through the model export interface. This data includes bus number, line head bus number, line end bus number, line reactance, line rated transmission capacity, and generator connection bus information. Second, 24-hour day-ahead load forecast data, sourced from the load forecasting module or dispatching planning system interface, commonly in the format of bus number, time period number, and forecasted load value, in MW. Third, 24-hour day-ahead renewable energy forecast output data, sourced from wind power forecasting systems, photovoltaic power forecasting systems, or renewable energy centralized control platforms, commonly in the format of grid connection point number, time period number, and forecasted output value, in MW. Fourth, power flow parameters used to calculate the power transmission distribution factor, including balancing node settings and line impedance parameters, which are directly obtained by the DC power flow module from the network model library. After the above data enters the processing flow, the topology and parameter data are used to build the graph structure, the load and new energy prediction data are used to form node characteristics, and the power flow parameters are used to form the adjacency matrix. All kinds of inputs have a clear purpose in subsequent calculations and can be directly connected with the existing data links at the dispatch site.

[0107] Optionally, the standard bus list in the network model library can be used as a unified index to map all data from the load forecasting system, renewable energy forecasting system, and network model library to the same bus numbering system. In actual engineering, different subsystems often use different equipment coding methods. For example, the load forecasting side may use the plant name-bus segment number, the renewable energy side may use the plant grid connection point number, and the network model library uses the internal bus number of the EMS (Energy Management System). Therefore, during processing, the grid connection relationship table in the equipment ledger is read first, each renewable energy plant is mapped to its access bus, and then the time-series forecast values ​​of all renewable energy plants under the same bus are merged. After the numbering is unified, a 24-hour net load sequence is generated according to two dimensions: bus and time period. The first busbar and the first For each time period, the predicted load value of the bus is read, and then the sum of the predicted output of all new energy sources connected to the bus in the same time period is subtracted to obtain the net load of the bus in that time period. The calculation formula is written as:

[0108]

[0109] in, Indicates the first The busbar at the first The net load for each time period, in MW, is calculated from the forecast data in this step and written into the spatiotemporal feature matrix. The Line 1 List; Indicates the first The busbar at the first The load forecast value for each time period, in MW, is derived from the load forecast module interface. Indicates the first The busbar at the first The total predicted output of new energy sources for each time period, in MW, is obtained from the new energy prediction system by summing the predicted values ​​of all wind power and photovoltaic power stations under the same bus line for each time period. This represents the bus index, determined by the standard bus order exported from the network model library. This represents the time period index, with values ​​ranging from 1 to 24, corresponding to the 24 discrete time periods of the day-ahead scheduling. Dimension check: The dimensions are , The dimensions are ,therefore The dimensions are The dimensions are consistent. After completing the calculation according to this rule for each busbar and each time period, the dimension is obtained as the number of busbars. 24 spatiotemporal characteristic matrix For example, in the day-ahead plan of a provincial power grid, if the predicted load of 500kV bus B17 in the 9th period is 620MW, and this bus is connected to one wind farm and two photovoltaic stations with predicted outputs of 110MW, 35MW, and 25MW respectively in the 9th period, then the total predicted output of new energy sources for this bus in that period is 170MW, and the net load is 450MW. The system will write this result into the matrix. The calculations for the remaining 23 time periods are performed in the same manner, corresponding to row B17 and column 9, thus forming the complete daily net load trajectory for the bus.

[0110] After obtaining the spatiotemporal feature matrix Then, continue constructing the adjacency matrix. This embodiment uses the power transmission distribution factor as the weight for the node connection relationships in the diagram to reflect the electrical and physical connections between buses. Specifically, it calls a DC power flow calculation program to read the bus, line, reactance, and slack node settings in the network model, and calculates the power transmission distribution factor for each physical line under a disturbance injected at a unit node. For the [missing information]... For each physical line, first take the absolute value of its corresponding power transmission distribution factor as the original edge weight, and then normalize it according to the sum of the original edge weights of the adjacent lines at the head end of the line to obtain the line-level normalized edge weights. Then, based on the busbar numbers at the beginning and end of the line, Write to adjacency matrix The corresponding busbar position; if there are multiple parallel lines between the same pair of buses, then the position of each physical line is retained in the line edge table. and in the adjacency matrix Write these at the corresponding positions The summation value; if there is no direct line connection between the two busbars, the corresponding position is zero. This forms... It is a dimensionless normalized adjacency matrix, and the line edge table synchronously retains the corresponding physical line. This preserves both the graph structure at the bus level and the independent edge weight information at the line level. During project implementation, this process is typically completed by a script that runs automatically before the day-ahead schedule begins. The script first reads a snapshot of the network model from the scheduling master station, exports the bus and line lists, then calls the power flow calculation module to generate power transmission distribution factors in batches, completes line-level edge weight normalization, and writes back the adjacency matrix based on the bus numbers at the beginning and end of the lines. and the line side table.

[0111] The final output consists of two variables: one is the spatiotemporal feature matrix. The first is derived from the hourly calculation of the 24-hour net load of the busbars and the arrangement of data in busbar order. The data format is a two-dimensional floating-point matrix, with the number of rows equal to the number of buses and a fixed number of 24 columns, in MW; the second is the adjacency matrix. The data is constructed from the normalized weighted average of the power grid topology and power transmission distribution factors. The data format is a two-dimensional floating-point matrix, with the number of rows and columns equal to the number of buses. It is a dimensionless weight matrix used to aggregate adjacent node information in the subsequent constraint classifier. Simultaneously, the normalized edge weights corresponding to each physical line are recorded in the line edge table. This data is then used for subsequent online screening at the line level. After the above processing, the original load forecast data, renewable energy forecast data, network model data, and power flow parameters in the dispatch system are uniformly converted into a graph input format consistent with the physical structure of the power grid. This retains both the temporal characteristics of the operating status for the next 24 hours and the spatial characteristics of the power propagation paths in the network, providing a directly calculable and directly callable input basis for subsequent identification of line constraint activity.

[0112] Optionally, this embodiment uses a trained graph neural network constraint classifier to process the input data and output the predicted probability that each line in the entire network will become an active constraint within the target scheduling period. Based on the predicted probability, the rated capacity of the line, and the electrical coupling information in the adjacency matrix, a line risk ranking quantity is constructed. The lines in the entire network are ranked according to the line risk ranking quantity, and a set of retained lines is determined in combination with a preset configuration threshold. The rated capacity of each line in the set of retained lines is adaptively and safely reduced according to the predicted probability to generate a corrected capacity. A dimensionality-reduced safety constraint unit combination model is constructed based on the set of retained lines and the corresponding corrected capacity.

[0113] After obtaining the spatiotemporal feature matrix Adjacency Matrix Building upon this foundation, a graph neural network model for predicting the activity of line constraints is constructed. Historical operational data is used to train the model parameters, enabling the model to map "multi-period net injection changes at the bus + grid topology coupling relationships" into a probabilistic result of "whether a line becomes an active constraint." The modeling logic here originates from DC power flow theory and graph signal processing methods: in the DC power flow model, node injection changes propagate through the network topology and form power flow distributions on the lines. Graph convolution is essentially a neighborhood-weighted summation operation on a graph structure, and both share the commonality of "adjacency relationships driving information propagation." Therefore, the classic graph convolution expression is combined with the power propagation mechanism of the power grid to propagate and aggregate node features layer by layer, and then line-level features are obtained through line endpoint mapping, ultimately completing the classification judgment.

[0114] First, the spatiotemporal feature matrix According to the system's unified benchmark power Normalization is performed to obtain the initial node feature matrix. ,in The baseline power used in the power flow calculation of the dispatching system is typically 100MW. After normalization, The Row represents busbar The dimensionless net load characteristic sequence over 24 time periods; the resulting spatiotemporal characteristic matrix The weight matrix of the graph structure consists of dimensionless normalized edge weights. The propagation of node features on the graph employs a discrete form derived from spectral graph convolution theory. This form can be derived from the Laplacian operator expansion and, in engineering implementation, is simplified to a combination of weighted summation of adjacency matrices and linear mapping. In this embodiment, a two-layer graph convolution structure is used to extract node features, where single-layer propagation can be written as:

[0115]

[0116] in, Indicates the first The node feature matrix of the layer, initially For the reason according to The dimensionless matrix obtained by normalization; It is an adjacency matrix, derived from the normalized edge weights, and is a dimensionless matrix; For the first The weight matrix of the layer is obtained through training and is a dimensionless parameter. This represents the linear rectified activation function, which is implemented by performing the operation "set to 0 if less than 0, otherwise leave unchanged" on each element in the matrix; This outputs the features for the next layer of nodes. In dimensional analysis, , and All are dimensionless, therefore Maintaining a dimensionless dimension facilitates subsequent line feature splicing and probability classification.

[0117] To facilitate understanding, this embodiment provides a local calculation example. That is, let's assume... MW, a certain node After normalization, the first The value of a certain feature dimension of the layer is 3.0, and the two adjacent nodes are... and The corresponding eigenvalues ​​are 2.0 and 1.0, respectively, and the corresponding weights in the adjacency matrix are... , Then the median value of that dimension after weighted summation is Then multiply by the weight matrix The coefficient 2.0 corresponds to 1.0, and after ReLU, it remains 1.0, completing one feature update. Through this node-by-node, dimension-by-dimensional propagation calculation, node features gradually integrate neighborhood information, thereby reflecting the power diffusion characteristics in the power grid topology.

[0118] After completing two layers of graph convolution, a node-level high-order feature matrix is ​​obtained. Since the goal is to determine whether a line is an active constraint, node features need to be mapped to line features. Specifically, this involves reading the start and end bus numbers of each line in the topology data, and correspondingly... Extract two row vectors from the data and concatenate them along the feature dimension to form the feature vector of the route. This process is directly based on the line connection relationships in the network model library, ensuring that the line characteristics correspond one-to-one with the actual power grid structure.

[0119] The line features are input into a fully connected classification structure, whose computational form is derived from the classic Multilayer Perceptron (MLP) structure. A linear transformation followed by nonlinear activation is used to map features to probabilities. The final output probability expression is:

[0120]

[0121] in, Indicates the first The probability that a line becomes an active constraint is a dimensionless value; This is the line feature vector, obtained by concatenating node features, and is dimensionless. , These are the weight parameters of the two fully connected network layers, obtained through training; , These are bias parameters; It is a linear rectified function; The Sigmoid function has the following form: This is used to map the output to a range of 0 to 1. In dimensional analysis, all input features are dimensionless, as are the weights and biases; therefore, the output... It is a dimensionless probability value with consistent dimensions.

[0122] The model training process employs supervised learning, specifically: training samples consist of historical operational data. For each historical day, using the unit combination results and power flow calculation results stored in the scheduling system, the actual power flow value of each line within 24 hours is extracted and compared with its rated capacity. If the power flow exceeds 95% of the capacity during a certain period, the line is marked as 1; otherwise, it is marked as 0. This label generation process can be implemented by calling the historical power flow calculation module, for example, by reading SCADA / EMS historical power flow records or offline power flow simulation results. Multiple historical day samples are combined into a training set, with each sample corresponding to a set. , And a set of line labels. During training, a mini-batch iterative approach is used, inputting samples into the model and updating parameters through backpropagation. , , etc., to make the predicted probability The model gradually approximates the true label. The training process can be completed on a conventional computing platform, such as using a GPU server to run a deep learning framework for parameter optimization, and the model parameters are then fixed after training. Through this process, the resulting model can perform training on new inputs. and At that time, output the active probability of all lines. This result is directly derived from the joint modeling of the power grid topology and operating status, and can reflect the impact of node injection changes on line constraints after propagation through the network, thus providing basic data for subsequent constraint selection.

[0123] Optionally, the reduced-dimensional safety-constrained unit combination model can be solved to obtain the unit combination results.

[0124] The spatiotemporal feature matrix regenerated from the previous day's scheduling time is used. and adjacency matrix The trained constraint classification model is fed into the data to obtain the predicted probability that each line in the entire network will become an active constraint within the target scheduling period. .in, It is still formed by aggregating the load forecasting interface and the new energy forecasting interface in the dispatching master station according to the bus. The data is still calculated using the network model library and the DC power flow program. Therefore, this stage is a continuous closed loop with the previous two stages in terms of data structure: the first stage is responsible for organizing the raw business data from the dispatching site into a graph input; the second stage is responsible for mapping the graph input into line activity probabilities; and the current stage transforms the probability results into a set of network constraints that can be directly written into the safety constraint unit combination solver. During online execution, the dispatch automation system first reads the 24-hour bus load forecast for the next day, the forecast output of each new energy power station, the rated capacity of the line, and the bus numbers at the beginning and end of the line, and forms a network constraint set according to the bus numbering system. and Then, the graph neural network model with fixed parameters is invoked to output the corresponding parameters for each line. Here The Sigmoid output, derived from the end of the fully connected classification layer in the previous stage, ranges from 0 to 1. A larger value indicates a higher probability that the line will become an active constraint within the next 24 hours due to power flow approaching its capacity limit. Since the current problem is a safety-constrained unit combination problem, rather than a simple line congestion warning, the online screening stage cannot simply retain lines based on probability. Instead, it needs to unify the predicted risk of the line, the propagation sensitivity of the line's electrical location, and the safety margin after entering the optimization model. This will form a constraint screening and dimensionality reduction modeling process that can significantly compress the model size while conforming to the scheduling operation mechanism.

[0125] The online screening process first constructs a risk ranking for each route. This ranking is based on... Based on this, an adjacency matrix is ​​then superimposed. The electrical coupling information corresponding to the first and last buses of the line allows the screening logic to reflect not only "the probability that the line itself is judged as active," but also "the sensitivity of the corridor containing this line to power propagation." In power grid scenarios, many constraints do not become tense due to local fluctuations in a single line, but rather because the power coupling between adjacent buses on a connecting corridor is very strong, and certain net injection changes can be quickly propagated to critical sections. Therefore, [the following text is incomplete and requires further context to translate accurately]. Incorporating the endpoint coupling strength into the ranking parameters allows the structural information learned by the graphical model in the first two stages to be passed on to the optimization stage. This is combined with the line's rated capacity. And the normalized edge weights recorded in the edge table in step one. Then, the risk ranking of the line is calculated using the following formula. :

[0126]

[0127] in, Indicates the first The online screening and ranking quantity of each line is a dimensionless value, calculated based on the predicted probability, line capacity, and line-level edge weights. Indicates the first The predicted probability of a line becoming an active constraint within the target scheduling period is derived from the online forward computation results of the trained constraint classification model; Indicates the first The rated transmission capacity of each line, in MW, is derived from the line setting data in the energy management system network model library; This represents the reference capacity of the entire network, in MW, and is obtained by the dispatching system by taking the average value of the rated capacity of the entire network in the network model library. Indicates the first The normalized edge weights of each physical path are derived from the construction of the adjacency matrix. The line edge table generated synchronously contains dimensionless values, and each physical line is recorded separately. Dimensionality check: Dimensionless Dimensionless Since it is dimensionless, It is dimensionless, yet has consistent dimensions. The derivation of this formula originates from the combination of the binary classification probability ranking idea and the power system sensitivity analysis idea: the original probability ranking only uses... In this embodiment, the edge weights are normalized at the line level. Relative quantity to line capacity By writing the gain term together, physical lines located in strongly coupled channels and with larger capacities receive higher priority under the same probability level. Taking the day-ahead dispatch of a provincial power grid as an example, if the reference capacity of the entire network... MW, a certain line The predicted probability is Rated capacity is MW, Line-level Normalized Edge Weights ,but Another route The predicted probability is Rated capacity is MW, Line-level Normalized Edge Weights ,but According to this sorting result, although The original predicted probability is higher, but The channel in question has a higher line-level edge weight and capacity, making it more sensitive to system security constraints; therefore, it is given priority for retention. The network constraints are consistent with the dispatching site's handling habits for critical sections.

[0128] In formation Then, the scheduling system sorts all network lines in descending order and combines this with configured thresholds. Generate a set of reserved lines. In actual implementation, the optimization program first reads the entire network line table and sets the reserved lines for each line. The data is written to the filtering cache in the scheduling database, and then the constraint generation module executes a two-level retention strategy: the first level directly retains all data that meets the requirements. The first level involves the lines; the second level targets key sections, inter-regional connecting lines, and stability control-related channels pre-maintained by the dispatcher, directly writing them into the retention set by line ID. This retention set is not simply a probability-based selection result, but a combined result of "graph neural network risk judgment + electrical coupling amplification + solidification of key dispatch channels." Subsequently, when constructing the safety-constrained unit combination model, unit operation constraints such as power balance constraints, unit start-up and shutdown constraints, minimum start-up and shutdown time constraints, ramp-up constraints, and spinning reserve constraints are generated according to the standard templates of existing dispatch solvers; the network constraint part only generates power flow constraint expressions for the lines in the retention set, and introduces a probability-related safety contraction amount to the line capacity when writing it into the solver. This contraction amount originates from the classic safety margin concept, that is, reserving a stricter safety boundary for the upper limit of capacity when the operating risk is high; in the current scenario, the risk information is directly written into the right-hand side of the reduced-dimensional constraints, forming the following capacity modification form:

[0129]

[0130] in, Indicates the first The available transmission limit of each reserved line in the reduced-dimensional safety-constrained unit combination model, in MW, is calculated in this stage and written into the solver. Indicates the first The rated transmission capacity of each line, in MW, is derived from the network model library; Indicates the first The probability of activity on each line; This represents the capacity contraction coefficient, a dimensionless configuration parameter derived from the scheduling strategy configuration table. It can be set between 0.03 and 0.10 based on historical back-substitution experiments. Dimension check: and All are dimensionless, therefore Dimensionless and Both The dimensions are consistent. This formula originates from the traditional SCUC's approach to reserving a safety margin for line capacity. The derivation extends the fixed margin to an adaptive margin related to the line's activity probability: when... When the value is close to 1, it indicates that the model judges the line to be highly active and that the line has been preserved. Its constraints mainly serve as explicit boundary conditions, and the capacity shrinkage is small. When a line is due to... It was higher and thus retained, but At a medium level, it indicates that it is retained more due to channel coupling sensitivity or critical channel properties, at which point... Retaining a larger safety margin within the capacity limit can improve the robustness of the dimensionality reduction model without significantly increasing the number of constraints. For example, if the rated capacity of a certain reserved line is 500MW, the predicted probability is... ,but MW; if the other reserved line also has a rated capacity of 500MW, but the predicted probability is... ,but MW. With this processing, the available transmission limit for the second line in the SCUC is automatically more conservative, which is more in line with its "structure sensitive but moderate probability" operating characteristics.

[0131] The above-mentioned set of reserved lines and their corresponding After being written into the mixed-integer optimization solver, a dimensionality-reduced safety-constrained unit combination model is formed. During online solving, the constraint generation module of the scheduling system first reads the start-stop state boundaries, minimum start-stop times, ramp rates, and upper and lower output limits of thermal power, hydropower, and gas turbine units from the unit library, and then extracts the start and end buses, power transmission distribution factors, and corrected capacities of each line from the reserved line set. The power flow constraints are discretized into standard linear constraint terms and submitted to the solver, for example, in each time period. For each reserved line, construct a PTDF (Power Transfer Distribution Factor) weighted node injection with a sum not exceeding [a certain value]. "and its reverse constraints. Because network constraints are only generated for the retained lines, the constraint size of the model will decrease significantly with the size of the retained set; and because Structural coupling information and predicted probabilities were both incorporated into the screening and ranking process. The risk level is further incorporated into the constraint boundaries. Therefore, dimensionality reduction is not simply about reducing the number of lines, but rather about translating the graph structure knowledge learned in the first two stages into constraint configurations that the solver can directly use. The resulting output includes two outcomes: first, a set of retained lines for over-limit positioning and constraint remediation during the next stage of verification; and second, a dimensionality-reduced safety-constrained unit composition model already incorporated into the corrected capacity boundaries, allowing the solver to directly calculate unit start-up, shutdown, and power output plans. The entire process links "graph input construction - active probability identification - online risk ranking - constraint writing with safety contraction" into a complete chain, ensuring that the model learning results from the first two stages truly translate into SCUC (Security Constrained Unit Commitment) optimization modeling, thereby achieving a unified goal of reducing the number of network constraints, increasing solution speed, and ensuring controllable operational safety boundaries.

[0132] Optionally, the DC power flow of the entire network is calculated based on the unit combination results, the lines that have exceeded the power flow limit are identified, and a compensation priority index is constructed based on the degree of exceeding the limit and the corresponding prediction probability. The high-priority lines that have exceeded the limit are selected according to the compensation priority index, the original network constraints corresponding to the selected lines that have exceeded the limit are written back into the reduced-dimensional safety constraint unit combination model, and the solution and verification are performed again until the safety constraint conditions of the entire network are met, and the final scheduling scheme is output.

[0133] The input comes directly from three types of results generated by the preceding calculations: First, the solution results of the safety-constrained unit combination model after dimensionality reduction, which are in the form of the output plan and start-up / shutdown status of each unit in 24 time periods. This result is output by the mixed integer programming solver and stored in the scheduling system's memory or database; Second, the generated set of reserved lines and the corrected capacity corresponding to the lines already included in the model. Third, during online screening, predictive probabilities are generated and cached for each physical line in the entire network. And the original rated capacity of all network lines in the network model library. This phase revolves around "restoring the physical security consistency of the entire network under dimensionality reduction constraints." Through iterative processes of DC power flow verification, limit violation identification, risk ranking, and constraint replenishment, the optimization results are gradually corrected to a state that satisfies the security constraints of all lines. The entire calculation process is based on the classic DC power flow model of power systems. This model originates from the basic formula of "linear relationship between node injection and line power flow" in power system analysis theory. Its core idea is to solve for the line power flow distribution through the linear relationship between node injected power and the network admittance matrix.

[0134] Optionally, the output of each unit in the unit combination result is first summarized by bus to obtain the net injected power of each bus in each time period. Then, the DC power flow calculation module in the dispatching system is called, and the actual power flow value of each physical line in the entire network in each time period is calculated by combining the line reactance parameters and slack node settings in the network model library. The unit is MW. DC power flow calculation is essentially derived from Kirchhoff's current law and the linearized power-phase relationship. Its engineering implementation typically involves multiplying the power transfer distribution factor matrix by the node injection vector, a process already well-established in scheduling systems. This yields... Then, it is compared with the original rated capacity in the network model library. Perform time-by-time comparisons to identify all those that meet the criteria. The system identifies the lines and time slots and calculates the degree of limit exceedance. To differentiate the importance of different lines during subsequent replenishment, a normalized expression based on relative limit exceedance is introduced. This expression is derived from the commonly used engineering definition of "limit exceedance ratio," which is the ratio of the limit exceedance value to the capacity limit. Based on this, the predicted probability of the cache is combined... Construct a priority index for line replenishment Its form is:

[0135]

[0136] The above formula is based on the over-limit ratio. Based on this, a probability amplification term is introduced. The improved form formed, in which For the first The priority index for replenishment of each route is dimensionless. For the first The routes during the time period The power flow value, in MW, is calculated by the DC power flow module; This is the original rated capacity of the line, in MW, sourced from the network model library; The active probability of the cached data during the online screening phase of this route is calculated by the model in the online forward pass. This indicates selecting the maximum percentage exceeding the limit among 24 time periods. During dimensional checks, and All The difference and ratio are dimensionless, and multiplied by It remains dimensionless, therefore It is a dimensionless quantity. The derivation logic of this improved formula is as follows: First, the over-limit ratio relative to the original rated capacity is used to measure the actual operating risk. Then, the risk is weighted using probability information, so that lines that are "predicted to be high risk and actually exceed the limit" have higher priority in the ranking, thus prioritizing the correction of key constraints under a limited number of replenishment times.

[0137] After obtaining all the over-limit lines Then, the scheduling system sorts the lines according to this indicator and selects several top-ranked lines to add to the replenishment set. This selection process can be achieved by setting a threshold or a fixed number, for example, selecting... For lines with a value greater than 0.05, or the top 10 lines selected in each round. Then, the constraint generation phase begins: for each selected line... The power transmission distribution factor is read from the network model library, the corresponding power flow constraints are rewritten into the safety constraint unit combination model, and the original capacity of the line is used. This serves as an upper limit constraint. This process is implemented in the optimization program by dynamically adding linear constraint terms, specifically in each time period. Construct "node injection linear combination based on power transfer distribution factor not exceeding The constraint expression is written into the solver model.

[0138] After constraint backfilling is completed, the updated model is submitted to the solver again for solution, and the power flow calculation, limit violation identification, and backfilling sorting process are repeated, forming a closed-loop iteration. This iterative process usually has fast convergence under actual power grid scale because the initial screening has retained most of the critical lines, and only a few missing constraints need to be added to meet the overall network safety conditions. For example, a line has a maximum power flow of 520MW in 24 time periods, and its original rated capacity... If the capacity is 500MW, then the over-limit ratio is... If the predicted probability of this route is... Then its replenishment index is The other line's over-limit ratio relative to its original rated capacity is 0.05, but... ,but Therefore, the former is given priority in entering the replacement set. Through this calculation process, the application of the formula in engineering and its impact on the sorting results can be clearly demonstrated.

[0139] As the iteration progresses, when all lines satisfy the requirements in all time periods... When the time comes, the backfilling process stops, and the final scheduling scheme is output. While maintaining the dimensionality reduction advantage of the previous stage, this scheme gradually restores the complete network security constraints through closed-loop verification, realizing the overall technical path of "first reducing dimensions and accelerating, then verifying and correcting". This achieves both efficient solution capability and strict operational safety requirements in large-scale power grid scenarios.

[0140] To more comprehensively demonstrate this solution, this embodiment presents a power grid unit dispatching method, specifically including:

[0141] 1. Based on the power grid operation data of the power grid equipment, construct the adjacency matrix corresponding to the connection relationship of the power grid equipment and the spatiotemporal feature matrix corresponding to the multi-period operation status of the power grid equipment;

[0142] 2. Extract features from the adjacency matrix and spatiotemporal feature matrix to obtain the node features of each physical line in the power grid;

[0143] 3. Based on the bus number corresponding to each physical line, the node features corresponding to each physical line are concatenated to obtain the line feature vector.

[0144] 4. Based on the line feature vector, predict the impact probability to obtain the predicted probability of each physical line in the power grid equipment; where the predicted probability is used to determine the probability of the standard physical line affecting the unit scheduling in the power grid equipment.

[0145] 5. Determine the shrinkage factor based on the preset capacity shrinkage coefficient and the predicted probability corresponding to each critical path;

[0146] 6. Adjust the rated capacity of each critical path according to the shrinkage factor to obtain the adjusted capacity of each critical path.

[0147] 7. Based on each critical path and its corrected capacity, construct the initial unit combination result;

[0148] 8. Perform power flow verification on each critical path in the initial unit combination results;

[0149] 9. If, based on the power flow verification results, it is determined that there is at least one power flow exceeding the limit among each critical path, then based on the degree of exceeding the limit and the predicted probability of each power flow exceeding the limit, the constraint compensation priority of each power flow exceeding the limit is constructed.

[0150] 10. Based on the priority of each constraint compensation, construct a new initial unit combination result;

[0151] 11. Based on the new initial unit combination results, return to the step of performing power flow verification on each critical path in the initial unit combination results. If, based on the power flow verification results, it is determined that there are no power flow exceeding the limit paths among the critical paths, then the initial unit combination results are used as the power grid unit dispatching scheme.

[0152] The specific process of the above steps can be found in the description of the above method embodiments. The implementation principle and technical effect are similar, and will not be repeated here.

[0153] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0154] Based on the same inventive concept, this application also provides a power grid unit scheduling device for implementing the power grid unit scheduling method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more power grid unit scheduling device embodiments provided below can be found in the limitations of the power grid unit scheduling method described above, and will not be repeated here.

[0155] In one exemplary embodiment, such as Figure 5 As shown, a power grid unit dispatching device is provided, comprising: a construction module 51, a prediction module 52, a processing module 53, and a dispatching module 54, wherein:

[0156] Module 51 is used to construct an adjacency matrix corresponding to the connection relationship of power grid equipment and a spatiotemporal feature matrix corresponding to the multi-period operation status of power grid equipment based on the power grid operation data of power grid equipment.

[0157] The prediction module 52 is used to predict the impact probability of the adjacency matrix and the spatiotemporal feature matrix to obtain the predicted probability of each physical line in the power grid equipment; wherein, the predicted probability is the probability of the standard physical line affecting the unit scheduling in the power grid equipment.

[0158] Processing module 53 is used to determine the correction capacity corresponding to at least one critical line in each physical line based on the predicted probability.

[0159] The scheduling module 54 is used to determine the grid unit scheduling scheme based on each critical line and the corrected capacity of each critical line.

[0160] In one embodiment, the prediction module 52 is further configured to:

[0161] Feature extraction is performed on the adjacency matrix and the spatiotemporal feature matrix to obtain the node features of each physical line in the power grid;

[0162] Based on the node characteristics of each physical line, the probability of influence is predicted to obtain the predicted probability of each physical line in the power grid equipment.

[0163] In one embodiment, the prediction module 52 is further configured to:

[0164] Based on the bus number corresponding to each physical line, the node features corresponding to each physical line are concatenated to obtain the line feature vector.

[0165] The predicted probability of each physical line in the power grid is obtained by predicting the impact probability based on the line feature vector.

[0166] In one embodiment, the processing module 53 is further configured to:

[0167] The shrinkage factor is determined based on the preset capacity shrinkage coefficient and the predicted probability corresponding to each critical path.

[0168] The rated capacity of each critical path is corrected based on the shrinkage factor to obtain the corrected capacity for each critical path.

[0169] In one embodiment, the scheduling module 54 is further configured to:

[0170] Based on each critical path and its modified capacity, construct the initial unit combination result;

[0171] Power flow verification is performed on each critical path in the initial unit combination results;

[0172] If, based on the power flow verification results, it is determined that there is at least one power flow exceeding the limit among each critical path, then the constraint compensation priority for each power flow exceeding the limit is constructed based on the degree of exceeding the limit and the predicted probability of each exceeding the limit.

[0173] Based on the priority of each constraint, a new initial unit combination result is constructed;

[0174] Based on the new initial unit combination results, return to the step of performing power flow verification on each critical path in the initial unit combination results. If, based on the power flow verification results, it is determined that there are no power flow exceeding the limit among the critical paths, then the initial unit combination results are used as the grid unit dispatching scheme.

[0175] In one embodiment, the scheduling module 54 is further configured to:

[0176] Based on the power flow verification results, determine the net injected power of each bus in each time period within the target scheduling cycle;

[0177] Based on the preset line reactance parameters, balancing node configuration, and net injected power, determine the actual power flow value of each physical line;

[0178] Among multiple physical lines, at least one physical line whose actual power flow value is greater than the rated capacity is selected, and based on the selected physical line, at least one power flow over-limit line is determined.

[0179] Each module in the aforementioned power grid unit dispatching device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0180] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 6 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores power grid unit dispatching schemes. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a power grid unit dispatching method.

[0181] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0182] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0183] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0184] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0185] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0186] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0187] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0188] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A power grid unit dispatching method, characterized in that, The method includes: Based on the power grid operation data of the power grid equipment, construct the adjacency matrix corresponding to the connection relationship of the power grid equipment and the spatiotemporal feature matrix corresponding to the multi-period operation status of the power grid equipment; Influence probability prediction is performed on the adjacency matrix and the spatiotemporal feature matrix to obtain the predicted probability corresponding to each physical line in the power grid equipment; wherein, the predicted probability is used to characterize the probability that the physical line will affect the unit scheduling in the power grid equipment; Based on the predicted probability, determine the correction capacity corresponding to at least one critical line in each of the physical lines; The grid unit scheduling scheme is determined based on each critical path and its corrected capacity.

2. The method according to claim 1, characterized in that, The step of performing influence probability prediction on the adjacency matrix and the spatiotemporal feature matrix to obtain the predicted probability corresponding to each physical line in the power grid equipment includes: Feature extraction is performed on the adjacency matrix and the spatiotemporal feature matrix to obtain the node features of each physical line in the power grid; Based on the node characteristics of each physical line, the predicted probability of each physical line in the power grid equipment is obtained by predicting the impact probability.

3. The method according to claim 2, characterized in that, The step of predicting the impact probability based on the node characteristics of each physical line to obtain the predicted probability corresponding to each physical line in the power grid equipment includes: Based on the bus number corresponding to each physical line, the node features corresponding to each physical line are concatenated to obtain the line feature vector. Based on the line feature vector, the influence probability is predicted to obtain the predicted probability of each physical line in the power grid equipment.

4. The method according to claim 1, characterized in that, The step of determining the correction capacity corresponding to at least one critical path among the physical paths based on the predicted probability includes: The shrinkage factor is determined based on the preset capacity shrinkage coefficient and the predicted probability corresponding to each critical path. The rated capacity of each critical path is corrected according to the shrinkage factor to obtain the corrected capacity corresponding to each critical path.

5. The method according to any one of claims 1 to 4, characterized in that, The step of determining the power grid unit dispatching scheme based on each of the critical lines and the corrected capacity of each of the critical lines includes: Based on each critical path and the modified capacity of each critical path, construct the initial unit combination result; Power flow verification is performed on each of the critical paths in the initial unit combination results. If, based on the power flow verification results, it is determined that there is at least one power flow exceeding the limit among the critical paths, then, based on the degree of exceeding the limit and the predicted probability of each power flow exceeding the limit, the constraint compensation priority of each power flow exceeding the limit is constructed. Based on the priority of each constraint, a new initial unit combination result is constructed; Based on the new initial unit combination result, return to the step of performing power flow verification on each of the critical lines in the initial unit combination result. If, based on the power flow verification result, it is determined that there are no power flow exceeding the limit lines among the critical lines, then the initial unit combination result is used as the power grid unit scheduling scheme.

6. The method according to claim 5, characterized in that, The step of determining, based on the power flow verification results, that at least one power flow violation path exists among the critical paths includes: Based on the power flow verification results, determine the net injected power of each bus in each time period within the target scheduling cycle; Based on the preset line reactance parameters, balancing node configuration, and net injected power, the actual power flow value of each physical line is determined; Among multiple physical lines, at least one physical line whose actual power flow value is greater than the rated capacity is selected, and at least one power flow over-limit line is determined based on the selected physical line.

7. A power grid unit dispatching device, characterized in that, The device includes: The construction module is used to construct an adjacency matrix corresponding to the connection relationship of the power grid equipment and a spatiotemporal feature matrix corresponding to the multi-period operation status of the power grid equipment based on the power grid operation data of the power grid equipment; The prediction module is used to predict the impact probability of the adjacency matrix and the spatiotemporal feature matrix to obtain the predicted probability of each physical line in the power grid equipment; wherein, the predicted probability is used to characterize the probability that the physical line will affect the unit scheduling in the power grid equipment. The processing module is used to determine the correction capacity corresponding to at least one critical line among the physical lines based on the predicted probability. The scheduling module is used to determine the grid unit scheduling scheme based on each of the critical lines and the corrected capacity of each of the critical lines.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images