Method, system, device and storage medium for simplifying transmission constraints of scuc lines

Abstract

This invention discloses a method, system, device, and storage medium for simplifying SCUC (Self-Controlled Grid Union) line transmission constraints. Belonging to the field of power system optimization scheduling and smart grid technology, the invention acquires historical SCUC operation data of the power system and constructs a model training dataset. A bus graph is constructed with buses as nodes and transmission lines as edges, mapping to obtain a line graph with lines as nodes and shared bus line topological associations as edges. Based on the electrical coupling similarity calculated using the power transfer distribution factor of transmission lines, the line graph is sparsified and reconstructed to obtain a sparse line graph. A graph attention network model is constructed based on the sparse line graph, using line graph features extracted from the dataset as input and historical active states of line constraints as labels for training, resulting in a prediction model. The model infers the constraint active probability, selects candidate constraints, constructs a simplified SCUC model for solution, and outputs the final scheduling scheme after full constraint verification and backfilling iteration.

Description

Technical Field

[0001] This invention belongs to the field of power system optimization scheduling and smart grid technology, specifically involving a method, system, equipment and storage medium for simplifying transmission constraints of SCUC lines. Background Technology

[0002] Security-Constrained Unit Commitment (SCUC) is a key optimization problem in day-ahead dispatch and power market clearing. It requires simultaneous consideration of factors such as unit start-up and shutdown logic, minimum start-up and shutdown times, ramping constraints, output limits, and network security constraints across multiple time periods. It is typically modeled as a Mixed-Integer Linear Programming (MILP) problem. In SCUC using the DC power flow approximation, line transmission constraints are expanded in a line-time period manner. This leads to a rapid increase in the number of constraints as the product of the number of lines and time periods increases, significantly increasing the computational burden of linear relaxation during branch-bounding. This is one of the main bottlenecks affecting solution speed and scalability.

[0003] To reduce the aforementioned computational overhead, existing technologies typically improve upon these technologies through two paths: 1) Optimization theory-driven constraint simplification methods: These methods, such as those based on dual information, Lagrange relaxation, cutting planes, and iterative constraint generation, dynamically identify and retain only potentially active cybersecurity constraints during the solution process, thereby reducing the search space. However, these methods typically require multiple iterations or additional sub-problem evaluations during the solution process, resulting in a complex overall process. Furthermore, they may still face high iteration overhead in large-scale systems and long-term scenarios.

[0004] 2) Machine Learning-Driven Constraint Reduction Methods: These methods utilize historical operating samples to learn the active patterns of line transmission constraints, pre-eliminating a large number of inactive constraints to reduce the initial size of the MILP and improve solution speed. While this approach significantly reduces the number of initial constraints, it also presents key problems: First, prediction errors may lead to the omission of truly active constraints, resulting in infeasible solutions or insufficient security. Second, the active patterns of constraints vary considerably under different power grid scenarios, load fluctuations, and changes in unit combinations; insufficient model generalization and interpretability can easily affect reliability.

[0005] In recent years, Graph Neural Networks (GNNs) have been widely used in tasks such as predicting generator start-up and shutdown and critical line transmission constraints due to their ability to explicitly utilize power grid topology information. However, existing GNN methods are typically based on bus graph modeling, making it difficult to directly represent the coupling relationships between lines for identifying the activity of line transmission constraints. Furthermore, the Power Transfer Distribution Factor (PTDF) directly reflects the impact of node-injected disturbances on line power flow and is an important prior information characterizing the coupling relationships between lines, but existing GNN methods do not fully utilize it in graph structure construction and neighborhood feature aggregation. Therefore, there is an urgent need for a technical solution that can integrate the PTDF physical prior with the graph attention mechanism to simplify line transmission constraints in SCUC while ensuring safety and feasibility. Summary of the Invention

[0006] The purpose of this invention is to overcome the problems of existing constraint simplification methods based on experience screening or pure data-driven prediction, such as structural redundancy, insufficient interpretability and generalization ability, and easy omission of real active constraints, which leads to infeasible solutions. The invention proposes a SCUC line transmission constraint simplification method, system, device and storage medium.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for simplifying SCUC line transmission constraints, comprising the following steps: Obtain historical operating data of power system safety-constrained unit combinations, and build a model training dataset based on the historical operating data; A bus graph is constructed using the power system's busbars as nodes and its transmission lines as edges. A line graph is then constructed based on the bus graph mapping, where the nodes of the line graph correspond to the transmission lines of the bus graph, and the edges of the line graph correspond to the topological associations of two transmission lines sharing the same busbar in the bus graph. Based on the electrical coupling similarity between adjacent nodes of the transmission lines in the line graph, the line graph is reconstructed by topological sparsification to obtain a sparse line graph; the electrical coupling similarity is calculated based on the power transfer distribution factor vector similarity of the transmission lines. A graph attention network model is constructed based on sparse line graphs. The node features of the line graph are extracted from the model training dataset and used as the input of the graph attention network model. The historical active state of line transmission constraints is used as the supervision label to train the graph attention network model and obtain a line transmission constraint activity prediction model. A line transmission constraint activity prediction model is used to perform forward reasoning in the safety constraint unit combination scenario to be solved, and obtain the activity probability of each transmission line transmission constraint. A candidate active constraint set is obtained by filtering according to the activity probability. A simplified safety constraint unit combination optimization model is constructed based on the candidate active constraint set and solved to obtain the initial scheduling scheme. The initial scheduling scheme is verified by using full line transmission constraints. The violations identified in the verification are then backfilled into the simplified safety constraint unit combination optimization model for iterative solution until all line transmission constraints meet the safety limits, and the final unit combination scheduling scheme is output.

[0008] Furthermore, the model training dataset is constructed, including: The acquired historical operational data is sequentially cleaned and continuous feature normalization is performed to obtain the processed data. The processed data is then divided into training set, validation set and test set according to a preset ratio. The normalization parameters for normalization are calculated solely from the statistics of the training set, and are used on both the validation and test sets.

[0009] Furthermore, the node features of the bus diagram are the load sequences of the corresponding bus, and the edge features of the bus diagram are the thermal stability limits and impedance parameters of the corresponding transmission lines; the node features of the line diagram are constructed based on the load sequences of the bus at both ends of the corresponding transmission line and their line thermal stability limits and impedance parameters, and the connection relationship between nodes in the line diagram is determined based on the topological association of the corresponding two transmission lines.

[0010] Furthermore, based on the electrical coupling similarity between transmission lines of adjacent nodes in the line graph, the line graph is reconstructed using topology sparsification, employing an adaptive threshold sparsification strategy: The electrical coupling similarity between transmission lines is calculated based on the vector similarity of the power transfer distribution factor. The electrical coupling similarity between transmission lines is mapped to a standardized similarity weight distribution. Adjacent edges are sorted in descending order of weight, and the minimum set of adjacent edges whose cumulative similarity reaches a preset threshold is retained. At the same time, upper and lower limits are set for the number of adjacent edges.

[0011] Furthermore, the graph attention network model is specifically as follows: On the sparse line graph, a graph attention network is constructed based on the characteristics of line nodes and their neighborhood information. The features of each neighboring node in the neighborhood of the target node are adaptively weighted and aggregated. The graph attention network model adopts a multi-head attention mechanism to perform neighborhood information interaction and feature fusion to obtain the node representation corresponding to the transmission line. The node representation is then input into a binary classification head to output the active probability of the corresponding line transmission constraint.

[0012] Furthermore, the model training dataset includes historical active state labels for line transmission constraints; these labels are determined based on whether the line power flow has reached the transmission capacity limit in the historical power flow calculation results; during model training, a binary cross-entropy loss function with class weights is used as the training objective.

[0013] In a second aspect, the present invention provides a SCUC line transmission constraint simplification system, comprising: The dataset construction module is used to acquire historical operating data of power system safety-constrained unit combinations and build a model training dataset based on the historical operating data; The line graph construction module is used to construct a bus graph by using the busbars of the power system as nodes and the transmission lines of the power system as edges; and to construct a line graph based on the bus graph mapping, where the nodes of the line graph correspond to the transmission lines of the bus graph, and the edges of the line graph correspond to the topological associations of two transmission lines sharing the same busbar in the bus graph. The sparse line graph reconstruction module is used to perform topological sparsification reconstruction of the line graph based on the electrical coupling similarity between adjacent nodes of the transmission lines in the line graph, so as to obtain a sparse line graph; the electrical coupling similarity is calculated based on the power transfer distribution factor vector similarity of the transmission lines; The model training module is used to construct a graph attention network model based on the sparse line graph. The features of the line graph nodes are extracted from the model training dataset and used as the input of the graph attention network model. The historical active state of line transmission constraints is used as the supervision label to train the graph attention network model, so as to realize the adaptive weighted aggregation of line neighborhood information and obtain the line transmission constraint activity prediction model. The module for solving the initial scheduling scheme is used to perform forward reasoning on the safety constraint unit combination scenario to be solved using the line transmission constraint activity prediction model to obtain the activity probability of each transmission line transmission constraint; based on the activity probability, a set of candidate active constraints is obtained; based on the set of candidate active constraints, a simplified safety constraint unit combination optimization model is constructed and solved to obtain the initial scheduling scheme. The final scheduling scheme output module is used to perform a safety check on the initial scheduling scheme using all line transmission constraints. The violation constraints identified in the check are backfilled into the simplified safety constraint unit combination optimization model for iterative solution until all line transmission constraints meet the safety limits, and the final unit combination scheduling scheme is output.

[0014] Thirdly, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the SCUC line transmission constraint simplification method.

[0015] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the SCUC line transmission constraint simplification method.

[0016] Compared with the prior art, the present invention has the following beneficial technical effects: The proposed SCUC line transmission constraint simplification method combines power transfer distribution factor and graph attention network, and utilizes PTDF physical similarity to guide line graph sparsification, constructing a sparse and effective line coupling line graph. This significantly improves the accuracy and generalization ability of constraint selection. By using PTDF to characterize the physical correlation of line power flow response and constructing a sparse and effective line coupling line graph accordingly, it achieves more reliable simplification of potentially active transmission constraints. Combined with necessary verification and iteration mechanisms to control the risk of misjudgment, it significantly reduces the number of transmission constraints that need to be explicitly added to the optimization model while ensuring that line safety constraints are met. This reduces the solution scale of SCUC and improves the overall solution efficiency and engineering applicability. Attached Figure Description

[0017] The accompanying drawings described herein are for illustrative purposes only and are not intended to limit the scope of the invention in any way. Furthermore, the shapes and proportions of the components in the drawings are merely schematic to aid in understanding the invention and do not specifically limit the shapes and proportions of the components. In the drawings: Figure 1 A flowchart for the simplification method of transmission constraints for SCUC lines.

[0018] Figure 2 A simplified structural diagram of the SCUC line transmission constraint simplified system.

[0019] Figure 3 A diagram of an electronic device for the SCUC line transmission constraint simplification method. Detailed Implementation

[0020] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0021] Example 1 See Figure 1 The SCUC line transmission constraint simplification method includes the following steps: Historical operating data of power system safety-constrained unit combinations are acquired, and a model training dataset is constructed based on the historical operating data. The construction of the model training dataset includes: sequentially cleaning the acquired historical operating data and normalizing continuous features to obtain the processed data, and dividing the processed data into training set, validation set and test set according to a preset ratio; the normalization parameters of the normalization process are calculated only from the statistics of the training set, and the normalization parameters are used in the validation set and test set.

[0022] A bus graph is constructed using the power system's busbars as nodes and its transmission lines as edges. A line graph is then constructed based on the bus graph mapping, where the nodes of the line graph correspond to the transmission lines in the bus graph, and the edges of the line graph correspond to the topological associations of two transmission lines sharing the same busbar in the bus graph. The node features of the bus graph are the load sequences of the corresponding busbars, and the edge features are the thermal stability limits and impedance parameters of the corresponding transmission lines. The node features of the line graph are constructed based on the load sequences of the buses at both ends of the corresponding transmission lines and their line thermal stability limits and impedance parameters. The connection relationships between nodes in the line graph are determined based on the topological associations of the two corresponding transmission lines, and no separate edge features are designed for the line graph.

[0023] Based on the electrical coupling similarity between adjacent nodes of the transmission lines in the line graph, the line graph is reconstructed by topological sparsification to obtain a sparse line graph; based on the electrical coupling similarity between adjacent nodes of the transmission lines in the line graph, the line graph is reconstructed by topological sparsification using an adaptive threshold sparsification strategy; the electrical coupling similarity is calculated based on the power transfer distribution factor vector similarity of the transmission lines. A graph attention network model is constructed based on a sparse line graph. Node features of the line graph are extracted from the model training dataset and used as input to the graph attention network model. The historical active states of line transmission constraints are used as supervision labels to train the graph attention network model, resulting in a prediction model for the activity of line transmission constraints. Specifically, the graph attention network model involves: learning representations of the input features of the corresponding nodes of each transmission line based on the sparse line graph; for each target node, a multi-head attention mechanism is used to weighted aggregate information from its neighboring nodes to characterize the topological association features and operational status features related to the target transmission line, forming an updated representation of the target node; the aggregated results from multiple attention heads are fused and input into a binary classification head to obtain the predicted probability that the transmission constraints of each transmission line are in an active state; the model training dataset includes historical active state labels for line transmission constraints, determined based on whether the line power flow has reached the transmission capacity limit in the historical power flow calculation results; during model training, a binary cross-entropy loss function with class weights is used as the training objective.

[0024] A line transmission constraint activity prediction model is used to perform forward reasoning in the safety constraint unit combination scenario to be solved, and obtain the activity probability of each transmission line transmission constraint. A candidate active constraint set is obtained by filtering according to the activity probability. A simplified safety constraint unit combination optimization model is constructed based on the candidate active constraint set and solved to obtain the initial scheduling scheme. The initial scheduling scheme is verified by using full line transmission constraints. The violations identified in the verification are then backfilled into the simplified safety constraint unit combination optimization model for iterative solution until all line transmission constraints meet the safety limits, and the final unit combination scheduling scheme is output.

[0025] By constructing effective coupling relationships at the line level by using the sensitivity response of line power flow to node injection as a physical prior, the scale of line graph connections is reduced while ensuring the ability to express key mutual influence information. Based on this, the constraints on line transmission are effectively simplified, thereby significantly reducing the number of constraints explicitly entering the SCUC model and improving the overall solution efficiency while meeting safety and feasibility requirements.

[0026] The specific steps include: Step 1: Data Acquisition and Preprocessing. A physical model of the unit combination is established, and a commercial solver is used to obtain historical unit start-up and shutdown status and unit output data. A systematic data preprocessing workflow is then employed to preprocess the historical data of the unit combination: First, data cleaning is performed, removing or correcting missing records, duplicate records, and obvious outliers in the original data, and ensuring that all types of data are aligned across time periods. Next, continuous features are normalized to scale features of different dimensions to a uniform numerical range for model training stability. The normalization parameters are obtained only from the training set statistics to avoid data leakage. Finally, the dataset is split into training, validation, and test sets, with proportions of 75%, 15%, and 15%, respectively.

[0027] Data feature normalization employs Min-Max normalization, and the parameters are uniformly calculated on the training set and then fixed for use on the validation and test sets to avoid data leakage. Specifically, Min-Max normalization scales features to the [0,1] interval according to the minimum and maximum values ​​of the training set: for any feature, let... This represents the original value of the feature in the current sample. and These represent the minimum and maximum values ​​of the feature in the training set samples, respectively. Let be the normalized eigenvalues, then we have: (1) in, To prevent stable terms with a denominator of zero.

[0028] The unit composition physical model includes the objective function and its corresponding constraints: The optimization objective of the generator set combination problem is usually to minimize the total system operating cost, including the operating cost and start-up and shutdown cost of the generator sets, which can be specifically represented as shown in equation (2): (2) In the formula: Indicates a time index. This represents the various time periods within a 24-hour scheduling cycle. This represents the total number of time periods in the scheduling cycle. Indicates the generator set index. Represents each generator set in the system; This represents the total number of generator sets in the system. Represents time From the first period to the second period Summing is performed over a time period, i.e., calculating the summation for a specific unit. Total cost throughout the entire scheduling cycle; Representative on the unit From the 1st to the 1st The platform performs summation, that is, it calculates the summation of all units during a certain time period. The total cost. Let be a continuous variable, representing the first... Each unit during the time period The active power output is measured in MW (megawatt). It is a binary variable, representing the first... Each unit at time The start / stop status is indicated by 1 for running and 0 for stopping. Indicates the unit During the period A binary variable indicating whether to start; 1 indicates during the time period. An activation action occurred; 0 indicates that the activation occurred during the specified time period. It did not start; Indicates the unit During the period A binary variable indicating whether the system is shut down; 1 indicates during the specified time period. A shutdown occurred; 0 indicates that a shutdown occurred during the specified time period. No shutdown; , and They are the generator sets The costs of no-load operation, startup, and downtime; It is a generator set The linear cost is usually expressed as a quadratic function of the unit's output, as shown below: (3) in, This is a cost factor, representing the unit's cost. Inherent constant parameters, when the unit is shut down At that time, this cost is 0.

[0029] The constraints include: 1) System power balance constraints (4) in, For the bus index; N Represents the system bus set; Indicates the first Each busbar during the time period The active power demand is expressed in MW (megawatts). Indicates in Time period, all in the system G The sum of the active power of each generator set; Indicates in All in the time period system N The sum of active power load requirements at each node.

[0030] 2) Unit output constraints (5) in, and For the unit The upper and lower limits of output.

[0031] 3) Climbing constraints (6) (7) in, and The units The limitations of power output decrease and upward climbing. and The units Startup and shutdown restrictions.

[0032] 4) Minimum start-stop time constraint (8) (9) (10) in, and The units Minimum start-stop time; Indicates time At the time within, total During this period, the unit The sum of the operating states; Indicates time At the time within, total During this period, the unit The sum of all shutdown states.

[0033] 5) Line transmission capacity constraints (11) in, Indicates the first The maximum transmission capacity of the line; Indicates busbar For the line The power transfer distribution factor. busbar exist The net injected power at time t. From equation (12), we obtain: (12) in, For connection to the busbar The collection of units on the platform.

[0034] Step 2: Construct a bus graph using buses as nodes and lines as edges. Use the load sequence on the bus side as node features and the line reactance and thermal limit as edge features to obtain a bus graph that expresses topological connectivity and operational characteristics. (Bus Graph) Using busbars as nodes and lines as edges, the node characteristics are the standardized intraday load sequence of the busbars. The edge features use static line parameters: .

[0035] To improve comparability across systems and scenarios, load sequences and line parameters are standardized using system statistics, and line reactance and thermal limit are shared as static features across 24 time periods to avoid meaningless time duplication coding.

[0036] Step 3: Based on the original bus graph, construct a line graph, mapping each line to a node in the line graph. If two lines share an endpoint bus in the bus graph, connect the corresponding nodes in the line graph to form an edge, thus obtaining a line-to-line level structure. This allows labels such as transmission line constraint activity to be directly aligned with line graph nodes, facilitating subsequent learning and prediction of the activity of line transmission constraints. (Line graph) The nodes correspond to the lines in the bus diagram. A 50-dimensional input vector is constructed based on the features of the bus diagram nodes. .

[0037] Step 4: Calculate the similarity between each line in the line diagram and its adjacent lines, and record the similarity of the lines in the PTDF matrix after fixing the reference bus. , The corresponding row vector is , For the two lines sharing the busbar and Their similarity Defined as equation (13): (13) To address the issues of edge explosion and excessive weak connections at higher-order busbars in standard line graphs, this paper proposes a sparsification method to reconstruct the graph's topological structure. The method stipulates that each edge has the same initial weight after sparsification and employs an adaptive threshold sparsification strategy. Specifically, for any line Let its candidate adjacent edge set be For any First, the absolute value of the PTDF cosine distance shown in equation (13) is calculated, and then mapped to the Softmax distribution of the standardized similarity weights: (14) in This is a hyperparameter used to control the weight distribution. Based on this, candidate adjacent edges are sorted according to... Sort by size from largest to smallest, and select those whose cumulative similarity reaches a threshold. Minimum set of adjacent edges : (15) in, This is the cumulative similarity threshold, i.e., the sparsity rate. When the line... When it only exhibits strong coupling with a few adjacent lines, the smaller That will satisfy However, when the coupling relationship is distributed across many adjacent lines, It will automatically increase in size to avoid over-pruning. To ensure scalability, this paper further sets upper and lower limits on the number of edges. and Soon Constraints This ensures that the edge complexity of the sparse line graph is always no greater than [value missing]. .

[0038] Step 5: In the sparse line graph The self-attention mechanism of Graph Attention Network (GAT) is introduced to learn the effective weights of edges, specifically as follows: For two adjacent nodes in a sparse line graph... and First, the unnormalized attention coefficients are calculated based on node features. : (16) in It is a linear transformation matrix. For attention vector parameters, This represents vector concatenation. This part is used to characterize the degree of association between adjacent line nodes based on sample features learned from them.

[0039] The result At the node Normalization within the neighborhood yields the node. Its neighboring nodes Attention weights: (18) Furthermore, a multi-head attention mechanism is employed to enhance stability and expressiveness: for each attention head... Calculate separately The aggregation results are then combined using an average fusion method to obtain the output representation. (19) in For the number of attention heads, This is the ReLU activation function.

[0040] Characterize the line nodes output by GAT Input the binary classification header to get the line During the period The probability of activity. The binary classification head is as follows: (20) in This is the Sigmoid function. The line-time label is constructed using historical trend data. exist The outcome of historical trends at any moment If the positive constraints are active, they are denoted as active line labels and used for supervised training. In this invention, predicting the activity of line transmission constraints is a binary classification task. The positive class represents active constraints during the corresponding time period, and the negative class represents inactive constraints. Since the number of active constraints is far less than the total number of line transmission constraints in most operating scenarios, the distribution of positive class samples in the sample set is sparse, resulting in significant class imbalance. To ensure that the model can effectively learn positive class features and improve its recall ability for positive classes, a binary cross-entropy loss function with class weights is chosen as the training objective. Specifically, the binary cross-entropy loss function with class weights is defined as: (twenty one) in These are the negative class weight coefficients, used as the loss weights for negative class samples; This is the positive class weight coefficient, used to increase the contribution of positive class samples in the loss function. For the first The true label of each sample For the model to the first The probability that a sample is predicted to be of the positive class.

[0041] Step 6: After training is complete, perform forward inference on a given new sample to obtain all the line constraints. The probability of being active at any given moment, and then The active constraints are generated as a set of candidate constraints: (twenty two) Based on candidate constraint set A simplified SCUC constraint model is constructed and solved to obtain the scheduling scheme. Subsequently, a post-hoc verification is performed on all line transmission constraint sets, and the missing active constraints found in the verification are backfilled according to the degree of violation and iteratively solved until all line transmission constraints are satisfied, thereby ensuring the safety and feasibility of the final solution.

[0042] To verify the effectiveness and scalability of the proposed SCUC line transmission constraint simplification method, experiments were conducted on both the IEEE 118-node and 500-node systems. All experiments were performed on computers configured with AMD Ryzen 5 7500F processors, NVIDIA GeForce RTX 4060 graphics cards, and 32GB (16GB×2) dual-channel memory. To comprehensively evaluate the effectiveness of different methods in identifying active constraints and their impact on SCUC solution performance, recall, area under the precision-recall curve (PR-AUC), base-normalized time saved (BNTS), constraint reduction rate, and number of infeasible samples were selected as evaluation metrics. Specifically, recall measures the completeness of active constraint identification, PR-AUC assesses the overall discriminative ability of the model under sparse positive class samples, constraint reduction rate indicates the degree of transmission constraint reduction, and the number of infeasible samples is used to verify the safety and feasibility of the reduced model.

[0043] The formula for calculating BNTS is: (twenty three) in, Indicates the solution time of the baseline SCUC model. This indicates the solution time of the model after constraint reduction. The larger the BNTS value, the more solution time is saved compared to the baseline model, that is, the more significant the model acceleration effect.

[0044] Table 1 Experimental results of the 118-node system

[0045] As shown in Table 1, compared to the bus diagram representation, the standard line diagram significantly improves the performance of active transmission constraint identification, indicating that modeling with lines as nodes is more suitable than modeling with buses as nodes in the line constraint screening task. Furthermore, the GAT model outperforms GCN, demonstrating that the attention mechanism can more effectively extract key neighborhood information.

[0046] However, the additional benefits of PTDF sparsification on the IEEE 118-node system are not significant. Under the GAT model, the PR-AUC of the standard line plot and the PTDF sparsification line plot are 0.9695 and 0.9693, respectively, almost at the same level; under the GCN model, the PTDF sparsification line plot is even slightly lower than the standard line plot.

[0047] The experimental results of the 118-node system described above demonstrate the effectiveness of the method described in this invention in small-scale systems. However, due to the relatively limited network size and topological complexity of the 118-node system, it cannot fully reflect the complex coupling relationships between lines and the large number of redundant connections in large-scale power systems. Therefore, the advantages of PTDF-guided sparsity have not been fully realized. To further verify the scalability of the method described in this invention and to examine the applicability of line graph representation, adaptive threshold sparsity, and graph attention mechanisms in larger-scale scenarios, further experimental analysis was conducted on a 500-node system.

[0048] Table 2 Experimental Results of the 500-Node System

[0049] As shown in Table 2, in a 500-node system, after further introducing PTDF sparsification on top of the standard line graph, the PR-AUC of the GCN model further improved to 0.8947. This indicates that in large-scale systems, structural redundancy has become a significant factor limiting model performance, and appropriately weakening weakly correlated adjacent edges and redundant connections helps improve the model's discriminative ability. The GAT model achieved the best results on the PTDF sparsified line graph, with a PR-AUC of 0.9552 and a Recall of 0.9988. This demonstrates that the combination of structural layer physical priors and graph attention mechanisms can significantly enhance the ability to identify active constraints and improve the representation of complex line coupling relationships.

[0050] This invention further compares the standard line graph, the randomized sparsified line graph, and the PTDF sparsified line graph on a 500-node system. To ensure a fair comparison, the randomized sparsified line graph randomly retains the same number of adjacent edges as the PTDF sparsified line graph at each higher-order bus, thereby ensuring that different sparsification methods can be compared and analyzed under the condition of similar edge size.

[0051] Table 3 Validation results of different sparsification strategies

[0052] Table 3 shows that, with the same number of edges as the PTDF-sparsed line graph, the PR-AUC of the randomized sparsed line graph is only 0.9261, basically the same as the 0.9275 of the standard line graph, showing no effective improvement; while the PR-AUC of the PTDF-sparsed line graph reaches 0.9552, significantly better than the former two. This result indicates that the performance gain of the proposed method is difficult to attribute to a simple edge reduction operation, but rather that PTDF similarity can more effectively preserve the line coupling relationships important for identifying active constraints, thereby improving the model's ability to identify key constraints and enhancing the representation of complex line association information. To further analyze the impact of the proposed method on model performance under different sparsity settings, further experimental analysis was conducted on a 500-node system.

[0053] Table 4 Validation results for different sparsity rates

[0054] As shown in Table 4, the model performance exhibits a clear non-monotonic pattern with increasing sparsity. As the sparsity increases from 0.70 to 0.90, the PR-AUC significantly improves from 0.8279 to 0.9552; however, when the sparsity further increases to 0.95, the performance drops back to 0.9242. This result indicates that PTDF sparsification strikes a balance between information preservation and redundancy suppression.

[0055] When the sparsity ratio is too low, the graph structure becomes excessively sparse, and some valuable line coupling relationships for identifying active constraints are overly reduced, leading to a decline in recognition performance. When the sparsity ratio is too high, weakly related neighboring edges are retained again, and the redundancy of the graph structure is enhanced again, thus weakening the advantages of sparsity. For this 500-node system, a sparsity ratio of 0.9 strikes a balance between information preservation and redundancy suppression.

[0056] Example 2 See Figure 2 The SCUC line transmission constraint simplification system includes: The dataset construction module is used to acquire historical operating data of power system safety-constrained unit combinations and to build a model training dataset based on the historical operating data; The line graph construction module is used to construct a bus graph by using the busbars of the power system as nodes and the transmission lines of the power system as edges; and to construct a line graph based on the bus graph mapping, where the nodes of the line graph correspond to the transmission lines of the bus graph, and the edges of the line graph correspond to the topological associations of two transmission lines sharing the same busbar in the bus graph. The sparse line graph reconstruction module is used to perform topological sparsification reconstruction of the line graph based on the electrical coupling similarity between adjacent nodes of the transmission lines in the line graph, so as to obtain a sparse line graph; the electrical coupling similarity is calculated based on the power transfer distribution factor vector similarity of the transmission lines; The model training module is used to build a graph attention network model based on sparse line graphs. The node features of the line graph are extracted from the model training dataset and used as the input of the graph attention network model. The historical active state of line transmission constraints is used as supervision labels to train the graph attention network model and obtain the line transmission constraint activity prediction model. The module for solving the initial scheduling scheme is used to perform forward reasoning on the safety constraint unit combination scenario to be solved using the line transmission constraint activity prediction model to obtain the activity probability of each transmission line transmission constraint; based on the activity probability, a set of candidate active constraints is obtained; based on the set of candidate active constraints, a simplified safety constraint unit combination optimization model is constructed and solved to obtain the initial scheduling scheme. The final scheduling scheme output module is used to perform a safety check on the initial scheduling scheme using all line transmission constraints. The violation constraints identified in the check are backfilled into the simplified safety constraint unit combination optimization model for iterative solution until all line transmission constraints meet the safety limits, and the final unit combination scheduling scheme is output.

[0057] Example 3 See Figure 3 An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the SCUC line transmission constraint simplification method when executing the computer program.

[0058] Example 4 A computer-readable storage medium storing a computer program that, when executed by a processor, implements the SCUC (Simplified Line Transmission Constraint) method.

[0059] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, read-only optical discs, optical storage, etc.) containing computer-usable program code.

[0060] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0061] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0062] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0063] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for reducing the transmission constraints of SCUC lines, characterized in that, Includes the following steps: Historical operating data of power system safety-constrained unit combinations are obtained, and a model training dataset is constructed based on the historical operating data; A bus graph is constructed using the power system's busbars as nodes and the power system's transmission lines as edges. A line graph is then constructed based on the bus graph mapping, where the nodes of the line graph correspond to the transmission lines of the bus graph, and the edges of the line graph correspond to the topological associations of two transmission lines sharing the same busbar in the bus graph. Based on the electrical coupling similarity between transmission lines of adjacent nodes in the line graph, the line graph is reconstructed by topological sparsification to obtain a sparse line graph. The electrical coupling similarity is calculated based on the power transfer distribution factor vector similarity of the transmission lines; A graph attention network model is constructed based on the sparse line graph. The node features of the line graph are extracted from the model training dataset and used as the input of the graph attention network model. The historical active state of the line transmission constraint is used as the supervision label to train the graph attention network model and obtain the line transmission constraint activity prediction model. Using the aforementioned line transmission constraint activity prediction model, forward reasoning is performed on the safety constraint unit combination scenario to be solved to obtain the activity probability of each transmission line transmission constraint; a candidate active constraint set is obtained by filtering according to the activity probability; a simplified safety constraint unit combination optimization model is constructed based on the candidate active constraint set and solved to obtain the initial scheduling scheme; The initial scheduling scheme is subjected to safety verification using full line transmission constraints. The violations identified during the verification are then backfilled into the simplified safety constraint unit combination optimization model for iterative solution until all line transmission constraints meet the safety limits, and the final unit combination scheduling scheme is output.

2. The SCUC transmission constraint compaction method of claim 1, wherein, The dataset used to build the model training dataset includes: The acquired historical operational data is sequentially cleaned and continuous feature normalization is performed to obtain processed data. The processed data is then divided into training set, validation set and test set according to a preset ratio. The normalization parameters for the normalization process are calculated solely from the statistics of the training set, and are used in the validation and test sets.

3. The SCUC transmission constraint compaction method of claim 1, wherein, The node features of the bus diagram are the load sequences of the corresponding bus, and the edge features of the bus diagram are the thermal stability limits and impedance parameters of the corresponding transmission lines. The node features of the line diagram are constructed based on the load sequences of the bus at both ends of the corresponding transmission line and their line thermal stability limits and impedance parameters. The connection relationship between nodes in the line diagram is determined based on the topological association of the two corresponding transmission lines.

4. The SCUC transmission constraint compaction method of claim 1, wherein, Based on the electrical coupling similarity between adjacent nodes in the line graph, the line graph is reconstructed using topology sparsification, employing an adaptive threshold sparsification strategy. The electrical coupling similarity between transmission lines is calculated based on the vector similarity of the power transfer distribution factor. The electrical coupling similarity between transmission lines is mapped to a standardized similarity weight distribution. Adjacent edges are sorted in descending order of weight, and the minimum set of adjacent edges whose cumulative similarity reaches a preset threshold is retained. At the same time, upper and lower limits are set for the number of adjacent edges.

5. The SCUC transmission constraint compaction method of claim 1, wherein, The graph attention network model is specifically as follows: On the sparse line graph, a graph attention network is constructed based on the characteristics of line nodes and their neighborhood information. The features of each neighboring node in the neighborhood of the target node are adaptively weighted and aggregated. The graph attention network model adopts a multi-head attention mechanism to perform neighborhood information interaction and feature fusion to obtain the node representation corresponding to the transmission line. The node representation is then input into a binary classification head to output the active probability of the corresponding line transmission constraint.

6. The SCUC transmission constraint compaction method of claim 1, wherein, The model training dataset includes historical active state labels of line transmission constraints, which are determined based on whether the line power flow in the historical power flow calculation results has reached the transmission capacity limit. During model training, a binary cross-entropy loss function with class weights is used as the training objective.

7. A system for reducing the complexity of transmission constraints in SCUC, characterized in that, include: The dataset construction module is used to acquire historical operating data of power system safety-constrained unit combinations and to build a model training dataset based on the historical operating data. The line graph construction module is used to construct a bus graph by using the busbars of the power system as nodes and the transmission lines of the power system as edges; and to construct a line graph based on the bus graph mapping, wherein the nodes of the line graph correspond to the transmission lines of the bus graph, and the edges of the line graph correspond to the topological associations of two transmission lines sharing the same busbar in the bus graph. The sparse line graph reconstruction module is used to perform topological sparsification reconstruction on the line graph based on the electrical coupling similarity between transmission lines of adjacent nodes in the line graph, so as to obtain a sparse line graph. The electrical coupling similarity is calculated based on the power transfer distribution factor vector similarity of the transmission lines; The model training module is used to construct a graph attention network model based on the sparse line graph. The node features of the line graph are extracted from the model training dataset and used as the input of the graph attention network model. The historical active state of the line transmission constraint is used as the supervision label to train the graph attention network model and obtain the line transmission constraint activity prediction model. The module for solving the initial scheduling scheme is used to perform forward reasoning on the safety constraint unit combination scenario to be solved using the line transmission constraint activity prediction model to obtain the activity probability of each transmission line transmission constraint; to obtain a candidate active constraint set based on the activity probability; to construct and solve a simplified safety constraint unit combination optimization model based on the candidate active constraint set to obtain the initial scheduling scheme. The final scheduling scheme output module is used to perform a safety check on the initial scheduling scheme using full line transmission constraints. The violation constraints identified in the check are backfilled into the simplified safety constraint unit combination optimization model for iterative solution until all line transmission constraints meet the safety limits, and the final unit combination scheduling scheme is output.

8. An electronic device, comprising: It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the SCUC line transmission constraint simplification method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the SCUC line transmission constraint simplification method as described in any one of claims 1-6.

Images