Optimized scheduling method and system for power restoration of power grid

Abstract

The invention relates to the technical field of power grid system dispatching, and discloses a power grid power restoration optimization dispatching method and system.The power grid power restoration optimization dispatching method comprises the steps that power grid operation data and environment data are obtained and preprocessed to generate feature vectors, fault detection point information is dynamically detected through a target detection model, and a power restoration path is optimized to generate an optimal power restoration strategy; and a target dynamic scheduling decision is generated through dynamic adjustment, a power restoration optimization scheduling decision is determined by adopting fault detection point information and the target dynamic scheduling decision, and power restoration operation is executed. According to the invention, rapid fault identification and power restoration decision are realized through an intelligent algorithm; by using multi-source data and a machine learning algorithm, the accuracy of fault positioning is improved; and through dynamic scheduling and decision optimization, the power failure time and the power recovery cost are reduced. The technical problems that when a power grid topological structure is adjusted, a new energy access proportion changes or a load fluctuates in real time, the prediction precision of an existing model is remarkably reduced, and dynamic optimization of a power recovery scheme is difficult to achieve are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid system dispatching, and in particular to a power grid restoration optimization dispatching method and system. Background Art

[0002] As the construction of new power systems advances, grid structures become increasingly complex. High-proportion integration of new energy sources and the interaction of diverse loads place higher demands on the intelligence of power restoration dispatching. Breaking through the data barriers and static optimization limitations of traditional methods to achieve deep fusion of multi-source data, accurate fault prediction, and dynamic optimization of power restoration plans has become a key technical requirement for improving power restoration efficiency and reliability.

[0003] However, traditional power system dispatching relies heavily on the manual experience of dispatchers, requiring manual intervention from fault identification and restoration plan development to the issuance of execution instructions. This results in a lengthy power restoration decision-making process. Especially in complex fault scenarios, decision-making can take hours or even longer, making it difficult to meet the rapid response requirements of modern power systems. For example, in the event of a large-scale power outage, manually troubleshooting each section of the fault and manually arranging restoration routes can miss the optimal restoration time, further expanding the scope of the outage.

[0004] Therefore, existing technologies attempt to incorporate data analysis and machine learning methods to optimize power restoration scheduling. For example, classification models are trained using historical fault data to initially locate faults, or heuristic algorithms are used to generate power restoration path plans. However, these models are often trained based on static historical data and lack the ability to perceive dynamic changes in the grid's operating status. When the grid topology adjusts, the proportion of new energy access changes, or the load fluctuates in real time, the model's prediction accuracy decreases significantly, making it difficult to dynamically optimize the power restoration plan. Summary of the Invention

[0005] This invention provides a method and system for optimizing power restoration scheduling. This solves the technical problem that existing models, which are mostly trained based on static historical data, lack the ability to perceive dynamic changes in the grid's operating status. This problem is compounded by the significant decrease in model prediction accuracy when the grid topology adjusts, the proportion of new energy access changes, or real-time load fluctuations occur, making it difficult to dynamically optimize power restoration plans.

[0006] A first aspect of the present invention provides a method for optimizing power restoration scheduling in a power grid, comprising:

[0007] Obtaining grid operation data and environmental data of the power grid system and preprocessing them to generate feature vectors;

[0008] Inputting the feature vector into a preset target detection model, and detecting fault detection point information corresponding to the feature vector through the target detection model;

[0009] Determine a power restoration path based on the fault detection point information, and optimize the power restoration path to generate an optimal power restoration strategy;

[0010] Dynamically adjusting the optimal power restoration strategy to generate a target dynamic scheduling decision;

[0011] Based on the fault detection point information and the target dynamic scheduling decision, a power restoration optimization scheduling decision is determined, and a power restoration operation is performed according to the power restoration optimization scheduling decision.

[0012] Optionally, acquiring grid operation data and environmental data of the grid system and preprocessing the data to generate a feature vector includes:

[0013] Set up multiple detection points in the power grid system;

[0014] Collecting grid operation data at each detection point through high-precision sensors;

[0015] Collecting environmental data of each detection point through environmental sensors;

[0016] Performing a weighted operation on the power grid operation data and the environmental data using a weighted average method to generate updated power grid operation data and updated environmental data;

[0017] Using a filtering algorithm to filter the updated power grid operation data and the updated environmental data to generate target power grid operation data and target environmental data;

[0018] The target power grid operation data and the target environment data are fused to generate fused data, and features of the fused data are extracted to generate a feature vector.

[0019] Optionally, inputting the feature vector into a preset target detection model, and detecting fault detection point information corresponding to the feature vector by the target detection model includes:

[0020] Inputting the feature vector into a fault type model corresponding to a preset target detection model, and detecting the fault type of the feature vector through the fault category model;

[0021] Inputting the feature vector into a decision tree model corresponding to the target detection model, and detecting the fault location of the feature vector through the decision tree model;

[0022] The fault type and the fault location are integrated to generate fault detection point information.

[0023] Optionally, determining a power restoration path based on the fault detection point information, optimizing the power restoration path, and generating an optimal power restoration strategy includes:

[0024] Calculating a power restoration path for fault location of the fault detection point information using the Dijkstra algorithm;

[0025] A genetic algorithm is used to optimize the power restoration path to generate an optimal power restoration strategy.

[0026] Optionally, dynamically adjusting the optimal power restoration strategy to generate a target dynamic scheduling decision includes:

[0027] Reinforcement learning is used to dynamically adjust the optimal power restoration strategy to generate an initial dynamic scheduling decision;

[0028] A load balancing algorithm is used to balance the load of the initial dynamic scheduling decision to generate a target dynamic scheduling decision.

[0029] Optionally, determining a power restoration optimization scheduling decision based on the fault detection point information and the target dynamic scheduling decision, and performing a power restoration operation according to the power restoration optimization scheduling decision, includes:

[0030] Establishing a mapping relationship between the fault detection point information and the target dynamic scheduling decision, and assigning priority weights according to different types of faults and fault areas;

[0031] Normalizing the fault detection point information and target dynamic scheduling decision of the mapping relationship according to priority weights;

[0032] Constructing a power restoration optimization objective function based on the fault handling timeliness, power restoration cost, and system stability of the power grid system;

[0033] An automated decision rule base is established using normalized fault detection point information and target dynamic scheduling decisions;

[0034] Formulate multiple power restoration plans based on the power restoration optimization objective function and automated decision rule library;

[0035] Calculating the difference between the predicted value and the actual value of each power restoration plan to generate a plurality of error values;

[0036] Determine whether each error value is greater than a preset error threshold, and select a power restoration plan with the smallest error value according to the judgment result to determine the power restoration optimization scheduling decision;

[0037] Execute the power restoration operation according to the power restoration optimization scheduling decision.

[0038] Optionally, the determining whether each error value is greater than a preset error threshold, and selecting a power restoration scheme with a minimum error value according to the determination result to determine a power restoration optimization scheduling decision, includes:

[0039] Determining whether each of the error values ​​is greater than a preset error threshold;

[0040] If so, update the priority weight and jump to the step of normalizing the fault detection point information and target dynamic scheduling decision of the mapping relationship according to the priority weight;

[0041] If not, the power restoration plan with the smallest error value is selected to determine the power restoration optimization scheduling decision.

[0042] A second aspect of the present invention provides a power grid restoration optimization dispatching system, comprising:

[0043] An acquisition module is used to acquire grid operation data and environmental data of the power grid system and perform preprocessing to generate feature vectors;

[0044] An input module, configured to input the feature vector into a preset target detection model, and detect fault detection point information corresponding to the feature vector through the target detection model;

[0045] an optimization module, configured to determine a power restoration path based on the fault detection point information, and optimize the power restoration path to generate an optimal power restoration strategy;

[0046] An adjustment module, configured to dynamically adjust the optimal power restoration strategy and generate a target dynamic scheduling decision;

[0047] An execution module is used to determine a power restoration optimization scheduling decision based on the fault detection point information and the target dynamic scheduling decision, and perform a power restoration operation according to the power restoration optimization scheduling decision.

[0048] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements the power grid restoration optimization scheduling method as described in any one of the above items.

[0049] A fourth aspect of the present invention provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions, wherein when the program instructions are executed by a computer, the computer executes the power grid restoration optimization scheduling method as described in any one of the above items.

[0050] It can be seen from the above technical solutions that the present invention has the following advantages:

[0051] This method acquires grid operation and environmental data, preprocesses it to generate feature vectors, and uses a target detection model to dynamically detect fault detection point information. Based on this information, it determines and optimizes the power restoration path to generate the optimal power restoration strategy. Dynamic adjustments are then made to generate a target dynamic scheduling decision. Finally, the optimized power restoration scheduling decision is determined by combining the fault detection point information with the target dynamic scheduling decision, and the power restoration operation is executed. This method uses intelligent algorithms to rapidly identify faults and make power restoration decisions, significantly improving scheduling efficiency. It also leverages multi-source data and machine learning algorithms to improve fault location accuracy. It optimizes the power restoration path and sequence to ensure load balancing and system stability. Through dynamic scheduling and decision optimization, it reduces outage time and power restoration costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0053] Figure 1 A flowchart of the steps of a power grid restoration optimization scheduling method provided in Example 1 of the present invention;

[0054] Figure 2 A flowchart of the steps of a power grid restoration optimization scheduling method provided in the second embodiment of the present invention;

[0055] Figure 3 This is a structural block diagram of a power grid restoration optimization and dispatching system provided in Example 3 of the present invention. DETAILED DESCRIPTION

[0056] Embodiments of the present invention provide a method and system for optimizing power restoration scheduling. This approach addresses the technical problem that existing models, which are mostly trained based on static historical data, lack the ability to perceive dynamic changes in the grid's operating status. This problem is compounded by the significant decrease in model prediction accuracy when the grid topology adjusts, the proportion of renewable energy access changes, or real-time load fluctuations occur, making it difficult to dynamically optimize power restoration plans.

[0057] In order to make the purpose, features, and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0058] See also Figure 1 , Figure 1 This is a flowchart of the steps of a power grid restoration optimization scheduling method provided in Example 1 of the present invention.

[0059] The present invention provides a method for optimizing power restoration and dispatching of a power grid, comprising the following steps:

[0060] Step 101: Obtain grid operation data and environmental data of the grid system and perform preprocessing to generate a feature vector.

[0061] It should be noted that multiple detection points are set in the power grid system.

[0062] In the embodiments of the present invention, grid operation data refers to real-time data that directly reflects the grid's operational status. This includes electrical parameters such as operating current, operating voltage, operating frequency, and workload at each detection point. This data, collected by high-precision sensors, enables real-time monitoring of grid power quality, equipment load, and power flow distribution, serving as a key basis for determining grid operation.

[0063] Environmental data refers to external environmental factors that affect grid operation, including ambient temperature, humidity, and wind speed at the monitoring point. For example, high temperatures can cause equipment to overheat and trip, while strong winds can cause line shorts. This data, collected through environmental sensors, is used to analyze the potential impact of external factors on grid failures.

[0064] A feature vector is a numerical vector generated by fusing preprocessed grid operation data with environmental data and reducing its dimensionality. After removing outliers through weighted averaging and filtering algorithms, the data is converted into a feature vector recognizable by the model. Its dimensionality depends on the number of key features after data fusion (such as normalized indicators such as current, voltage, and temperature).

[0065] The pre-processed grid operation data is fused with environmental data, and the numerical vector generated after dimensionality reduction is used as the input of the target detection model. This can comprehensively reflect the grid operation status and environmental impact, and provide data support for dynamic fault detection and power restoration strategy optimization.

[0066] Step 102: Input the feature vector into a preset target detection model, and use the target detection model to detect fault detection point information corresponding to the feature vector.

[0067] In the embodiment of the present invention, the target detection model refers to a two-level fault analysis model built based on machine learning, which is used to parse fault information from feature vectors, specifically including:

[0068] SVM fault type model: A classification model built using the support vector machine algorithm. Its core goal is to accurately identify fault types by maximizing sample intervals.

[0069] Decision tree model: A tree-like decision structure is constructed based on the attribute values ​​of the feature vector for fault area location.

[0070] Fault detection point information refers to the core parameters of the power grid fault output after analysis by the target detection model. It is used to guide the formulation of power restoration strategies and includes:

[0071] Fault type: This characterizes the physical nature and severity of the fault. For example, a short circuit fault manifests as a sudden surge in current and a sudden drop in voltage, while an overload fault manifests as sustained high load operation. Accurately identifying the fault type helps the dispatch system select the appropriate power restoration strategy (e.g., isolating the fault point for a short circuit and adjusting the load distribution for an overload).

[0072] Fault location: This method determines the specific location of a fault, such as the kth detection point on a line or a specific device in a substation. Combined with the grid topology, the fault location results can be used to calculate the shortest path to restore power (e.g., searching for the shortest path from the fault point to a normally powered area).

[0073] Machine learning algorithms are used to extract fault signatures from feature vectors, enabling fault type determination and location localization. By inputting feature vectors into a two-tiered model architecture, this system achieves progressive detection, from fault presence determination to precise fault type and location analysis. Compared to traditional fault identification methods that rely on manual experience, this significantly improves detection efficiency and accuracy. In particular, it addresses the challenge of fault signature extraction during dynamic grid operation, often caused by high-dimensional data and strong noise interference. This provides a reliable basis for the dynamic optimization of subsequent power restoration strategies.

[0074] Step 103: Determine a power restoration path based on the fault detection point information, optimize the power restoration path, and generate an optimal power restoration strategy.

[0075] In this embodiment of the present invention, the power restoration path refers to the power transmission path from a functioning power supply node to the faulted area after a power grid failure. Essentially, it is a connected path in a weighted graph. Purpose: The selection of a power restoration path directly impacts power restoration time, energy costs, and system stability, and is the physical foundation of the power restoration strategy.

[0076] The optimal power restoration strategy refers to a global optimal power restoration plan generated through algorithm optimization by comprehensively considering multiple objectives such as power restoration efficiency, cost, and system stability. It includes two parts: path selection and execution sequence.

[0077] Based on the identified fault type and location (e.g., a short circuit on a transmission line), the grid topology is abstracted into a weighted graph (nodes are substations, test points, etc., edges are lines, and weights are line length, load capacity, etc.). Using the fault point as the demarcation point, an algorithm is used to calculate the shortest path from the normal power supply area to the faulted area. Based on this shortest path, an algorithm is introduced to optimize the power restoration sequence (e.g., the order in which multiple fault points are addressed and the sequence of sectional closing). A fitness function, Fitness(sequence), is defined to comprehensively evaluate indicators such as power restoration efficiency and load balance, generating a globally optimal power restoration strategy.

[0078] Step 104: Dynamically adjust the optimal power restoration strategy to generate a target dynamic scheduling decision.

[0079] In the embodiment of the present invention, the target dynamic scheduling decision refers to a real-time execution plan generated by modifying the optimal power restoration strategy in combination with the real-time power grid status and the dynamic adjustment algorithm.

[0080] In the process of implementing the optimal power restoration strategy, the power grid may face dynamic factors such as sudden changes in real-time load, fluctuations in renewable energy output, and changes in weather conditions (such as new line failures caused by thunderstorms). Strategy adjustments need to be made through state perception, action evaluation, and strategy updates to obtain the target dynamic scheduling decision.

[0081] Step 105: Determine a power restoration optimization scheduling decision based on the fault detection point information and the target dynamic scheduling decision, and perform a power restoration operation according to the power restoration optimization scheduling decision.

[0082] In this embodiment of the present invention, optimized power restoration scheduling refers to a multi-objective optimization algorithm that integrates fault detection point information (fault type and location) with the target dynamic scheduling decision (dynamically adjusted routing, sequence, and load control strategy) to generate a globally optimal execution plan. Essentially, it aims to achieve coordinated optimization of power restoration efficiency, cost, and stability within grid operating constraints (such as equipment capacity and safety regulations).

[0083] Power restoration refers to the physical process of isolating faults, switching paths, and restoring loads based on optimized power restoration scheduling decisions, using automated control equipment or manual intervention. Its core is to translate digital decisions into actual actions of grid equipment to achieve power restoration.

[0084] Through multi-dimensional data fusion and decision rule integration, fault diagnosis results and dynamic scheduling strategies are converted into executable power restoration plans, and power grid equipment is driven to complete the power restoration task.

[0085] The decision-making and execution phase of optimized power restoration dispatching is the bridge connecting the algorithm model and the physical power grid. Through the deep collaboration of data, algorithms, and equipment, it has achieved a revolutionary upgrade from "manual experience-led" to "intelligent and precise control" in power restoration, providing key technical support for improving the reliability and resilience of new power systems.

[0086] See also Figure 2 , Figure 2 A flowchart of the steps of a power grid restoration optimization scheduling method provided in the second embodiment of the present invention.

[0087] The present invention provides a method for optimizing power restoration and dispatching in a power grid, comprising:

[0088] Step 201: Obtain grid operation data and environmental data of the grid system and perform preprocessing to generate a feature vector.

[0089] In an embodiment of the present invention, the grid operation data of each detection point of the grid system is collected by high-precision sensors, and the external environment data at each detection point of the grid system is obtained by environmental sensors. The grid operation data and the external environment data are fused and preprocessed to generate a feature vector.

[0090] Furthermore, step 201 further includes the following sub-steps:

[0091] S11. Set up multiple detection points in the power grid system.

[0092] In this embodiment of the present invention, a monitoring point refers to a pre-defined physical node or monitoring location within a functioning power grid system that collects real-time grid operation data and external environmental data. This is the fundamental unit for grid status awareness, enabling comprehensive monitoring and data collection of the grid's operating status through the deployment of various sensors.

[0093] Several detection points are set in a normally operating power grid system, where the detection points are physical nodes or monitoring locations for power grid operation data and external environment data.

[0094] S12. Collect power grid operation data at each detection point through high-precision sensors.

[0095] In the embodiment of the present invention, the high-precision sensor refers to a high-precision electrical sensor (such as a current transformer, a voltage transformer, etc.).

[0096] The grid operation data of each detection point is collected through high-precision sensors. , where i represents the detection point number, Expressed as the operating current, Expressed as the operating voltage, Expressed as the operating frequency, Expressed as workload.

[0097] S13. Collect environmental data of each detection point through environmental sensors.

[0098] In the embodiment of the present invention, the environmental sensor refers to a sensor such as a thermometer, a hygrometer, an anemometer, etc.

[0099] Obtain external environmental data at the detection point through environmental sensors, , where i represents the detection point number, It is expressed as the ambient temperature of the detection point, Expressed as the ambient humidity at the detection point, It is expressed as the ambient wind speed at the detection point.

[0100] S14. Perform weighted calculation on the power grid operation data and the environmental data using a weighted average method to generate updated power grid operation data and updated environmental data.

[0101] In the embodiment of the present invention, the weighted average method refers to a statistical method that assigns different weight coefficients according to the importance of data items, and optimizes the processing of original data by weighted summation.

[0102] Updating grid operation data refers to generating an optimized electrical parameter set by fusing and calculating the original grid operation data (such as the current and voltage at each detection point) using the weighted average method.

[0103] Updated environmental data refers to the environmental impact assessment indicators generated by weighted calculation of original environmental data (such as temperature, humidity, wind speed, etc.).

[0104] The calculation formula of the weighted average method is:

[0105]

[0106] Where, Represents the i-th original data (such as the current value or temperature value of a certain detection point); Represents the weight factor of the i-th data item ( ,and ), the larger the weight, the more significant the impact of the data item on the result.

[0107] Weight setting principles:

[0108] ① Grid operation data weight:

[0109] Key electrical parameters (such as operating current, voltage) have higher weights (such as , , because it directly reflects the power quality of the power grid and the equipment load status;

[0110] Auxiliary parameters (such as operating frequency, load) have relatively low weight (such as , ).

[0111] ② Environmental data weight:

[0112] Factors that have a significant impact on grid security have a higher weight (such as ambient temperature , high temperature may cause the device to overheat);

[0113] Secondary environmental factors (e.g. humidity, wind speed) have lower weights (e.g. , ).

[0114] After the original grid operation data (such as current and voltage at each detection point) and the original environmental data (such as temperature, humidity, wind speed, etc.) are integrated and calculated using the weighted average method, updated grid operation data and updated environmental data are obtained.

[0115] S15. Use a filtering algorithm to filter the updated power grid operation data and the updated environmental data to generate target power grid operation data and target environmental data.

[0116] In the present invention, a filtering algorithm refers to a technique for processing data sequences through specific mathematical operations. Its core goal is to suppress or eliminate noise components (such as random interference and abnormal jump values) in the data while retaining the effective components that reflect the true signal characteristics. Common filtering algorithms used in power grid data processing include sliding average filtering, Kalman filtering, and wavelet filtering.

[0117] Target grid operation data refers to high-precision, low-noise grid operating status data generated by filtering the "updated grid operating data" (electrical parameters after weighted averaging). Essentially, it further extracts effective signals that truly reflect grid operating patterns from the weighted electrical parameters.

[0118] Target environmental data refers to the environmental impact signature data generated by filtering the "updated environmental data" (environmental parameters processed by weighted averaging). Its core purpose is to isolate environmental signals that have a substantial impact on grid operation from the weighted environmental data and eliminate accidental interference factors.

[0119] The calculation formula of the filtering algorithm is:

[0120]

[0121] The filtering algorithm is a key filter for the transition of power grid data from the "acquisition layer" to the "analysis layer". The target power grid operation data and target environment data obtained after filtering the updated power grid operation data and updated environmental data using the filtering algorithm provide a solid data foundation for the precise operation of subsequent intelligent algorithms. It is a necessary prerequisite for realizing the "intelligent and reliable" power grid restoration optimization and scheduling.

[0122] S16. Fusing the target power grid operation data and the target environment data to generate fused data and extracting features of the fused data to generate a feature vector.

[0123] In the embodiment of the present invention, data fusion refers to integrating heterogeneous data from different data sources (such as power grid operation detection points and environmental sensors) into a unified data set through operations such as spatiotemporal alignment and feature association.

[0124] Fusion methods and implementation:

[0125] Space-time alignment:

[0126] Time dimension: Align grid operation data and environmental data with the same sampling period (e.g., once per second) to ensure consistent timestamps on the fused data.

[0127] Spatial dimension: Associate the grid data of the same detection point (such as the current at detection point 5) with the environmental data at the corresponding location (such as the temperature at detection point 5) to form a three-dimensional data unit of "detection point-time-space-parameter".

[0128] Feature-level fusion:

[0129] Direct splicing: target power grid operation data (such as 、 、 、 ) and target environment data (such as 、 、 ) splicing by dimension to form a joint dataset containing electrical and environmental characteristics;

[0130] Derivative feature construction: Generate new features through mathematical operations, such as:

[0131] Load temperature coefficient: L i ×T i , reflecting the comprehensive impact of load on equipment under high temperature;

[0132] Humidity resistance factor: H i 2 / V i , quantifying the impact of humid environment on line insulation impedance.

[0133] Feature extraction involves filtering and refining the key indicators that best reflect the grid's operating status and fault characteristics from fused data, forming a low-dimensional, efficient feature representation and generating a feature vector. Essentially, this approach uses dimensionality reduction to address the "curse of dimensionality" problem, improving model training efficiency and generalization capabilities.

[0134] Step 202: Input the feature vector into a preset target detection model, and use the target detection model to detect fault detection point information corresponding to the feature vector.

[0135] In an embodiment of the present invention, the feature vector is input into a preset target detection model, the target detection model is used to detect the fault type and fault location of the feature vector respectively, and the fault type and fault location are fused to obtain fault detection point information.

[0136] Furthermore, step 202 further includes the following sub-steps:

[0137] S21 , inputting the feature vector into a fault type model corresponding to a preset target detection model, and detecting the fault type of the feature vector through the fault category model.

[0138] In the embodiment of the present invention, the fault type model refers to the SVM fault type model. The calculation formula of the SVM fault type model is:

[0139]

[0140] Where w represents the weight vector, which determines the direction of the hyperplane; b represents the bias term, which determines the position of the hyperplane; C represents the penalty parameter, which balances the classification error and model complexity; Represents a slack variable that allows a small number of samples to be misclassified.

[0141] Model input: Feature vectors generated through data fusion and feature extraction (e.g., numerical vectors containing dimensions such as current, voltage, temperature, and load);

[0142] Model output: no fault indicator (such as output "0" or "normal") or fault type label (such as "short circuit", "overload", "broken wire", "ground", etc.).

[0143] Key logic for fault type detection:

[0144] 1. Mapping between feature vectors and failure modes

[0145] The fault type model establishes a mapping relationship between feature vectors and fault types by learning historical fault data. For example:

[0146] ① Short circuit fault: The current eigenvalue in the eigenvector increases significantly (e.g., >0.9 after normalization), the voltage eigenvalue decreases significantly (e.g., <0.1 after normalization), and the frequency characteristic fluctuates slightly (<0.2). The model classifies the fault as "short circuit" using the SVM hyperplane.

[0147] ② Overload fault: The current and load characteristic values ​​in the characteristic vector are continuously high (for example, both > 0.7), and the temperature characteristic value increases simultaneously (> 0.6). The model identifies it as an "overload" category, which is different from a short-term impact load (without the characteristic of continuous temperature increase).

[0148] ③ Line break fault: The current eigenvalue in the eigenvector approaches 0, and the line impedance derived eigenvalue increases abnormally (for example, the voltage / current ratio is greater than the threshold value). The model classifies it as a "line break" fault.

[0149] 2. Decision Boundary and Confidence Assessment

[0150] ① Decision boundary: The SVM model determines the separation boundary between categories (such as the hyperplane equation) through training. The relative position of the feature vector and the boundary determines the fault type.

[0151] ② Confidence output: Some implementations output the distance between the sample and the hyperplane (e.g., geometric margin) as a confidence indicator for the classification result. For example, the farther the distance from the hyperplane, the more reliable the classification result. If the distance is less than a threshold, the model may mark it as "uncertain," triggering manual review.

[0152] S22. Input the feature vector into a decision tree model corresponding to the target detection model, and detect the fault location of the feature vector through the decision tree model.

[0153] In the embodiment of the present invention, the calculation formula of the decision tree model is:

[0154]

[0155] Where L represents the loss function (such as classification error rate); Indicates the real location label; Represents the predicted location label.

[0156] The feature vector is input into the decision tree model and the decision tree algorithm is used to locate the fault.

[0157] In a specific embodiment, 1. Technical logic of the decision tree model:

[0158] 1. Model positioning and function: The decision tree model is the second-layer module of the target detection model. Based on the known fault type, it focuses on extracting spatial topological features and parameter gradient features from the feature vector to solve the problem of "where the fault occurs."

[0159] Input: A feature vector containing the fault type label (such as a feature vector that has been determined as "short circuit" by the SVM model); additional power grid topology information (such as the spatial coordinates of the detection point and the line connection relationship).

[0160] Output: The specific location where the fault occurred (e.g., “the line section between detection point X and detection point Y,” “the A-phase equipment at a certain substation Z”).

[0161] 2. Core algorithm: decision tree algorithm

[0162] Principle: By recursively partitioning the feature space, a tree-like decision structure is constructed. Each internal node represents a judgment on a feature (such as "Is the current greater than 50A"), branches represent the judgment results, and leaf nodes represent the final fault location label.

[0163] Key steps:

[0164] ① Feature selection: Use information gain (such as the ID3 algorithm), information gain ratio (C4.5 algorithm), or Gini index (CART algorithm) to select the most discriminative features for fault location (such as current difference and voltage gradient at the detection point).

[0165] ②Tree generation: Starting from the root node, recursively split the node until the stopping condition is met (such as the number of leaf node samples is less than the threshold, the purity reaches the upper limit).

[0166] ③ Pruning optimization: Avoid overfitting and improve model generalization ability through pre-pruning (limiting the depth of the tree) or post-pruning (removing redundant branches).

[0167] 2. Key Logic of Fault Location

[0168] 1. Hierarchical decision-making based on topological features:

[0169] The decision tree model decomposes the power grid topology into hierarchical judgment conditions by simulating manual troubleshooting logic. For example:

[0170] Root node: “Did the fault occur on a transmission line?” (based on the line type label in the feature vector);

[0171] Branch Node:

[0172] If so, further determine whether it is a trunk line (based on the importance of the line, such as transmission capacity > 100MW).

[0173] If not, determine "Is it an internal fault in the distribution area?" (based on the area number characteristics of the detection point).

[0174] Leaf node: After multiple layers of judgment, it outputs the specific fault location (such as "section 12-15 of detection points on trunk line L1").

[0175] 2. Spatial Correlation Analysis of Eigenvectors

[0176] The model locates faults by analyzing the spatial neighbor relationship and parameter differences of detection points in the feature vector:

[0177] ① Current gradient characteristics:

[0178] If the current at detection point A is normal (e.g., 50A) and the current at the adjacent detection point B is 0A, the decision tree determines that the fault is located in the line between A and B based on the splitting condition of "current difference > 40A."

[0179] ②Voltage attenuation characteristics:

[0180] If the fault type is "broken line," the voltage at each detection point in the eigenvector shows a gradient of "normal on the power supply side and sudden drop on the fault side." The model uses the condition of "voltage attenuation rate > 80%" to locate the broken line point close to the detection point where the voltage suddenly drops.

[0181] ③ Environmental feature assistance:

[0182] If the feature vector contains environmental data with a wind speed greater than 10 m / s, the decision tree will prioritize determining that the fault may be located in an elevated line section susceptible to strong winds, thereby narrowing the search range.

[0183] 3. Technical Implementation and Application Scenarios

[0184] 1. Model Training Process

[0185] Topology data preparation:

[0186] Construct a power grid topology map, marking the spatial coordinates of each detection point, the line / substation to which it belongs, the upstream and downstream connection relationships, etc.

[0187] Fault location marking:

[0188] Collect historical fault records and mark the detection point interval or equipment corresponding to each fault (such as "between towers 3 and 5 of line L1").

[0189] Feature Engineering:

[0190] Generate topology-related features (such as detection point spacing and line length) and parameter gradient features (such as current difference and voltage difference between adjacent detection points) as input to the decision tree.

[0191] Model training and validation:

[0192] Use historical data to train a decision tree and use a confusion matrix to evaluate positioning accuracy (e.g., requiring 90% of fault positioning errors to be less than one detection point interval).

[0193] 2. Typical Application Cases

[0194] Scenario 1: Transmission Line Short Circuit Location

[0195] The feature vector shows that the fault type is "short circuit", the current at detection point 5 is 100A (normal range 50-80A), and the current at detection point 6 is 0A. The decision tree locates the fault through the following path:

[0196] "Is it a transmission line?" → "Yes" → "Is the current difference between adjacent detection points greater than 50A?" → "Yes" → "Locate the line between detection points 5 and 6."

[0197] Scenario 2: Ground fault in the distribution station area

[0198] The eigenvector indicates a ground fault, and the zero-sequence current at detection point 8 within the substation is abnormally high (>5A). All other detection points are normal. The decision tree proceeds through the following steps: "Is this a distribution substation?" → "Yes" → "Zero-sequence current >3A?" → "Yes" → "Locate the substation branch line where detection point 8 is located."

[0199] S23. Fault type and fault location are integrated to generate fault detection point information.

[0200] In this embodiment of the present invention, the fusion of fault type and fault location refers to the aggregation of information between the fault attribute identified by the SVM model (such as "short circuit") and the physical location output by the decision tree model (such as "line between detection points 5 and 6") to generate a unified description that includes "what the fault is" and "where the fault is located."

[0201] The semantic association between fault type and location is achieved through preset rules. Typical rules include:

[0202] 1. Type-position association rules:

[0203] If the fault type is "overload" and is located in "distribution station area X", it will be automatically marked as "area overload", triggering the load balancing policy;

[0204] If the fault type is "line break" and is located in the "mountainous section of transmission line L2", then combined with environmental data (such as wind speed > 10m / s), it is marked as "line break caused by strong winds".

[0205] 2. Severity assessment rules:

[0206] The severity of a trunk line fault (such as one located at the "hub substation outlet") is automatically set to "high" and is handled with priority;

[0207] Branch line faults (such as "User Branch Line C") are set to "Medium" or "Low" and handled in the normal order.

[0208] 3. Rules for deducing the scope of influence:

[0209] According to the power grid topology, the system automatically generates an "impact range" list (such as "the fault in the detection point 5-6 section affects 200 users") by tracing back from the fault location to the power source point upstream and traversing the affected users downstream.

[0210] Step 203: Use the Dijkstra algorithm to calculate the power restoration path for fault location based on the fault detection point information.

[0211] In the embodiment of the present invention, the Dijkstra algorithm refers to a greedy strategy that starts from the starting point (normal power supply node) and gradually expands the shortest path tree to ensure that the path selected at each step is the shortest known one. The specific process is as follows:

[0212] 1. Initialization:

[0213] Set the distance of the starting point s to 0 and the distances of the remaining nodes to infinity; use a priority queue (min heap) to sort the nodes by distance.

[0214] 2. Iterative expansion:

[0215] Take out the node u with the smallest distance in the queue and traverse its neighbor node v; if the distance to v through u is shorter, update the distance of v and add v to the queue.

[0216] 3. Termination conditions:

[0217] When the shortest distance to the target node (edge ​​node of the fault area) is determined, the algorithm terminates and the path is backtracked to obtain the shortest power restoration path.

[0218] It is worth mentioning that the calculation formula of Dijkstra algorithm is:

[0219]

[0220] Where, Indicates the power restoration path.

[0221] Step 204: Optimize the power restoration path using a genetic algorithm to generate an optimal power restoration strategy.

[0222] In an embodiment of the present invention, by simulating the biological evolution process, a multi-objective optimization is performed on the initially calculated power restoration path (such as the shortest path generated by the Dijkstra algorithm), balancing multiple constraints such as power restoration efficiency, cost, and system stability, and generating a more globally adaptable power restoration solution.

[0223] 1. Adaptability Analysis of Genetic Algorithm

[0224] 1. Multi-objective characteristics of power restoration path optimization

[0225] Core objectives:

[0226] Shortest path (minimize physical distance or transmission loss);

[0227] Lowest cost (the sum of labor, equipment, and energy costs is minimized);

[0228] Load balancing (avoiding local line overload after power restoration);

[0229] Risk avoidance (preferring routes with low environmental risks, such as avoiding areas with high lightning strike rates).

[0230] Limitations of traditional algorithms:

[0231] The Dijkstra algorithm can only optimize a single objective (such as the shortest distance) and cannot meet multiple objectives simultaneously. Genetic algorithms, through population evolution and multi-objective fitness functions, can achieve global optimization under complex constraints.

[0232] 2. Genetic encoding of the regeneration pathway

[0233] Map the complex electrical path to the chromosome of the genetic algorithm:

[0234] Encoding method:

[0235] Binary encoding: Each bit represents whether a certain line is selected (e.g. "1" indicates selection, "0" indicates non-selection), which is suitable for simple topologies;

[0236] Integer encoding: A path is represented by a node sequence (such as "[s, 2, 4, t]"), which is more consistent with the grid topology.

[0237] Example chromosome:

[0238] If there are 5 nodes in the power grid graph, the path “s→1→3→t” can be encoded as the integer sequence [0, 1, 3, 4] (assuming the node numbers are 0=s, 1, 2, 3, 4=t).

[0239] 2. Key Operation Process of Genetic Algorithm

[0240] 1. Population initialization

[0241] Generate an initial set of paths:

[0242] Generate several shortest paths as initial individuals based on Dijkstra algorithm;

[0243] Randomly generate some paths to increase population diversity (e.g., generate 10% random paths through a random walk algorithm).

[0244] 2. Fitness function design

[0245] The fitness value of an individual is calculated by integrating multiple objectives. The typical formula is:

[0246]

[0247] Where, Indicates the path length (normalized to 0-1); Represents the cost of power restoration (normalized to a value between 0 and 1); Indicates the load balance degree (value range is 0-1, 1 means complete balance); Indicates the risk index (value range is 0-1, 0 means no risk); The weight of each target (such as the peak period of the power grid Prioritize load balancing)

[0248] 3. Genetic Operations

[0249] Selection:

[0250] Roulette wheel selection method: select individuals based on fitness ratio, the path with higher fitness has a greater probability of being selected;

[0251] Elite retention strategy: retain the current best individual and directly enter the next generation to avoid the loss of high-quality solutions.

[0252] Crossover:

[0253] Single-point crossover: randomly select a crossover point and exchange some genes (such as path segments) of the two chromosomes;

[0254] Example: Chromosome A [s,1,3,t] and chromosome B [s,2,4,t] are crossed to generate [s,1,4,t] and [s,2,3,t].

[0255] Mutation:

[0256] Randomly changing genes in chromosomes (e.g. replacing a node in a path with an adjacent node) to maintain population diversity;

[0257] The mutation probability is usually set to 1%-5% to avoid excessive mutation that destroys high-quality solutions.

[0258] 4. Termination Conditions

[0259] Reach the preset number of iterations (e.g. 500 generations);

[0260] The fitness value of the population tends to be stable (for example, the optimal solution remains unchanged for 50 consecutive generations);

[0261] Output the current optimal individual as the optimal power restoration strategy.

[0262] Step 205: Dynamically adjust the optimal power restoration strategy to generate a target dynamic scheduling decision.

[0263] In an embodiment of the present invention, during the execution of the optimal power restoration strategy, the power grid may face dynamic factors such as real-time load mutations, fluctuations in renewable energy output, and changes in weather conditions (such as new line failures caused by thunderstorms). Strategy adjustments need to be implemented through state perception, action evaluation, and strategy updates to obtain target dynamic scheduling decisions.

[0264] Furthermore, step 205 further includes the following sub-steps:

[0265] S31. Use reinforcement learning to dynamically adjust the optimal power restoration strategy and generate an initial dynamic scheduling decision.

[0266] In this embodiment of the present invention, reinforcement learning (Q-learning algorithm) refers to a classic offline policy reinforcement learning algorithm that evaluates the long-term benefits of actions under each state by maintaining a Q-value table (state-action value function). The Q-learning algorithm is expressed as:

[0267]

[0268] Where, represents the state-action value function, represents the learning rate (trading off immediate rewards vs. future rewards), represents the discount factor (weighing immediate rewards against future rewards).

[0269] The initial dynamic scheduling decision refers to the scheduling plan generated after the first dynamic correction based on the optimal power restoration strategy through the reinforcement learning algorithm.

[0270] In the specific implementation, 1. Strategy initialization: input the preset optimal power restoration strategy (such as the path and sequence generated by the genetic algorithm) as the initial strategy π0 of reinforcement learning.

[0271] 2. State perception: collecting real-time status (such as sudden changes in current at the detection point or increased wind speed) to determine the dimensions that need to be adjusted (such as decreased path reliability).

[0272] 3. Action selection: based on Greedy strategy selects actions:

[0273] by Probabilistically randomly exploring new actions (e.g., trying unused alternative paths);

[0274] by Probabilistically select the action with the largest current Q value (such as maintaining the original path).

[0275] 4. Execution and feedback:

[0276] Execute an action , observe the environment feedback reward and new status (such as whether the load is balanced);

[0277] Update the Q-value table to gradually increase the priority of high-reward actions.

[0278] 5. Decision output:

[0279] After a finite number of iterations (e.g., 5 steps), the strategy with the highest current cumulative reward is output as the initial dynamic scheduling decision.

[0280] S32. Use a load balancing algorithm to balance the load of the initial dynamic scheduling decision and generate a target dynamic scheduling decision.

[0281] In the embodiments of the present invention, the load balancing algorithm refers to an optimization algorithm that adjusts the load distribution of each node or line in the power system to make the overall load as uniform as possible. Its goal is to minimize the deviation of the load of each node from the average load of the system. The load balancing calculation formula is:

[0282]

[0283] Where, It is expressed as the average load of detection point i.

[0284] Technical process of load balancing adjustment:

[0285] 1. Input data and initial conditions

[0286] ① Initial dynamic scheduling decision:

[0287] Includes preliminary plans such as the power restoration path and closing sequence (such as dynamic adjustment strategies generated by Q-learning).

[0288] ②Real-time load data:

[0289] Real-time parameters such as active power, reactive power, and current at each detection point are collected through the SCADA system or edge computing devices.

[0290] ③Grid topology constraints:

[0291] Physical constraints such as line capacity limit and transformer rated load.

[0292] 2. Algorithm execution steps

[0293] ①Load status assessment:

[0294] Calculate the current system load average and the load deviation of each node ;

[0295] Identify overload risk nodes (e.g. ) or underloaded nodes (such as , which may lead to waste of resources)

[0296] ② Adjustment strategy generation:

[0297] Path switching: Shifting part of the load from a heavily loaded line to a lightly loaded line (e.g., enabling a backup path to share traffic);

[0298] Sequence adjustment: Delay the restoration of power to high-load areas and prioritize the restoration of power to low-load areas to balance the overall distribution;

[0299] Resource scheduling: Invest in energy storage equipment or adjust the output of distributed power sources (such as photovoltaics and energy storage) to compensate for load deviations.

[0300] ③ Constraint verification and iterative optimization:

[0301] Verify whether the adjusted plan meets the constraints such as line capacity and voltage limit;

[0302] If not, the strategy is adjusted repeatedly until the load balance meets the standard or the maximum number of iterations is reached.

[0303] The initial dynamic scheduling decision is optimized and adjusted according to the load balancing adjustment technology to generate the target dynamic scheduling decision.

[0304] Step 206: Determine a power restoration optimization scheduling decision based on the fault detection point information and the target dynamic scheduling decision, and perform the power restoration operation according to the power restoration optimization scheduling decision.

[0305] In this embodiment of the present invention, a multi-objective optimization algorithm generates a global optimal execution plan by integrating fault detection point information (fault type and location) with target dynamic scheduling decisions (dynamically adjusted paths, sequences, and load control strategies). Based on this global optimal execution plan, automated control equipment or manual intervention are used to complete the actual actions of fault isolation, path switching, and load restoration, thereby achieving power restoration.

[0306] Furthermore, step 206 includes the following sub-steps:

[0307] S41. Establish a mapping relationship between fault detection point information and target dynamic scheduling decision, and assign priority weights according to different types of faults and fault areas.

[0308] In this embodiment of the present invention, a mapping relationship between fault detection point information and dynamic scheduling decisions is established, and priority weights are assigned to different types of faults (such as short circuits, overloads, and disconnections) and fault areas (such as substations, transmission lines, and distribution substations). For example:

[0309] The priority of trunk line failure is higher than that of branch line, and the weight coefficient is set to =0.6;

[0310] Short-circuit faults pose a greater threat to grid stability, and the weight coefficient is set to =0.5.

[0311] S42: normalize the fault detection point information and the target dynamic scheduling decision in the mapping relationship according to the priority weight.

[0312] In an embodiment of the present invention, the fault type and regional information corresponding to the fault detection point information of the mapping relationship are normalized with the power restoration path, sequence, and load balancing indicators in the target dynamic scheduling decision, and the dimensions are unified for fusion.

[0313] S43. Construct a power restoration optimization objective function based on the fault handling timeliness, power restoration cost and system stability of the power grid system.

[0314] In the embodiment of the present invention, the power restoration optimization objective function F is defined, taking the following factors into consideration:

[0315] Troubleshooting timeliness: (The shorter the fault duration, the higher the priority);

[0316] Power restoration cost: (including manpower, equipment, and energy costs);

[0317] System stability: (Load balance, voltage fluctuation range).

[0318] The objective function expression is:

[0319] in , the weights can be dynamically adjusted according to the grid operation status (e.g., prioritizing stability during peak hours, increased to 0.7).

[0320] S44. Establish an automated decision rule base using the normalized fault detection point information and target dynamic scheduling decision.

[0321] In an embodiment of the present invention, the automated decision rule base refers to a set of structured decision rules pre-defined in the power grid restoration optimization and dispatching system based on expert experience, historical cases, and power grid operation laws, which is used to achieve automated decision-making from fault detection, path optimization to dynamic scheduling.

[0322] Specifically, based on expert experience and historical cases, the normalized fault detection point information and target dynamic scheduling decision-making are used to establish an automated decision rule base to achieve automated decision-making. For example:

[0323] Rule 1: If the fault area involves important users (such as hospitals and transportation hubs), the "fast power restoration path" is preferred, allowing short-term load overload (not exceeding 120% of the rated value).

[0324] Rule 2: If multiple faults exist simultaneously, sort them by "fault priority x impact range" and handle the fault with the largest product value first.

[0325] Use fuzzy logic algorithms to handle uncertainties such as ambient humidity When the humidity is high, the power restoration path needs to avoid sections prone to waterlogging, and the edge weights of the Dijkstra algorithm need to be adjusted (by adding a humidity penalty term).

[0326] S45. Develop multiple power restoration plans based on the power restoration optimization objective function and the automated decision rule library.

[0327] In this embodiment of the present invention, multiple power restoration plans are formulated based on the priority of the fault location in the power restoration optimization objective function and the real-time execution results of dynamic scheduling decisions corresponding to the automated decision-making rule base (such as the load curves of each line and the power restoration progress bar). These power restoration plans include the fastest power restoration plan, the lowest cost plan, and the safest plan.

[0328] S46. Calculate the difference between the predicted value and the actual value of each power restoration plan to generate multiple error values.

[0329] In this embodiment of the present invention, after the power restoration task is executed, data such as actual power restoration time, cost, and load fluctuation are collected and compared with the predicted values ​​of the optimization model to calculate multiple error values. The error value calculation formula is:

[0330] .

[0331] S47. Determine whether each error value is greater than a preset error threshold, and select the power restoration plan with the smallest error value based on the judgment result to determine the power restoration optimization scheduling decision.

[0332] In the embodiment of the present invention, the preset error threshold is 15%.

[0333] Determine whether the error value is greater than the preset error threshold. If both are smaller than the preset error threshold, select the power restoration plan with the smallest error value to determine the power restoration optimization scheduling decision.

[0334] Furthermore, step S47 further includes the following sub-steps:

[0335] S471. Determine whether each error value is greater than a preset error threshold.

[0336] In the embodiment of the present invention, it is determined Is it greater than 15%?

[0337] S472: If yes, update the priority weight, and jump to the step of normalizing the fault detection point information of the mapping relationship and the target dynamic scheduling decision according to the priority weight.

[0338] In the embodiment of the present invention, if the error value exceeds a threshold value (such as >15%), the model update mechanism is triggered:

[0339] A reinforcement learning algorithm (such as Q-learning) is used to update the fault priority weights, and the execution jumps to the step of normalizing the mapping relationship between the fault detection point information and the target dynamic scheduling decision according to the priority weights.

[0340] Retrain the SVM fault type model and decision tree location model based on new data to improve the accuracy of subsequent decisions.

[0341] S473. If not, select the power restoration plan with the smallest error value to determine the power restoration optimization scheduling decision.

[0342] In an embodiment of the present invention, if the error value does not exceed the threshold, the power restoration plan with the smallest error value is selected to determine the global optimal execution plan corresponding to the power restoration optimization scheduling decision.

[0343] S48. Execute the power restoration operation according to the power restoration optimization scheduling decision.

[0344] In an embodiment of the present invention, according to the global optimal execution plan corresponding to the power restoration optimization scheduling decision, the actual actions of fault isolation, path switching, and load restoration are completed through automated control equipment or manual intervention to achieve the power restoration task.

[0345] See also Figure 3 , Figure 3 This is a structural block diagram of a power grid restoration optimization and dispatching system provided in Example 3 of the present invention.

[0346] The present invention provides a power grid restoration optimization dispatching system, comprising:

[0347] An acquisition module 301 is used to acquire grid operation data and environmental data of the grid system and perform preprocessing to generate a feature vector;

[0348] An input module 302 is used to input the feature vector into a preset target detection model, and detect the fault detection point information corresponding to the feature vector through the target detection model;

[0349] The optimization module 303 is used to determine the power restoration path based on the fault detection point information, optimize the power restoration path, and generate the optimal power restoration strategy;

[0350] An adjustment module 304 is configured to dynamically adjust the optimal power restoration strategy and generate a target dynamic scheduling decision;

[0351] The execution module 305 is used to determine the power restoration optimization scheduling decision based on the fault detection point information and the target dynamic scheduling decision, and perform the power restoration operation according to the power restoration optimization scheduling decision.

[0352] Furthermore, the acquisition module 301 includes:

[0353] A setting submodule, used to set multiple detection points in the power grid system;

[0354] The first acquisition submodule is used to collect power grid operation data of each detection point through high-precision sensors;

[0355] The second collection submodule is used to collect environmental data of each detection point through environmental sensors;

[0356] The weighted submodule is used to perform weighted operations on the power grid operation data and the environmental data using a weighted average method to generate updated power grid operation data and updated environmental data;

[0357] A filtering submodule is used to filter the updated power grid operation data and the updated environmental data using a filtering algorithm to generate target power grid operation data and target environmental data;

[0358] The first fusion submodule is used to fuse the target power grid operation data and the target environment data to generate fused data and extract features of the fused data to generate a feature vector.

[0359] Furthermore, the input module 302 includes:

[0360] An input submodule is used to input the feature vector into the fault type model corresponding to the preset target detection model, and detect the fault type of the feature vector through the fault category model;

[0361] The detection submodule is used to input the feature vector into the decision tree model corresponding to the target detection model, and detect the fault location of the feature vector through the decision tree model;

[0362] The second fusion submodule is used to fuse the fault type and fault location to generate fault detection point information.

[0363] Furthermore, the optimization module 303 includes:

[0364] A path submodule is used to calculate the power restoration path for fault location based on the fault detection point information using the Dijkstra algorithm;

[0365] The optimization submodule is used to optimize the power restoration path using a genetic algorithm and generate the optimal power restoration strategy.

[0366] Furthermore, the adjustment module 304 includes:

[0367] The adjustment submodule is used to dynamically adjust the optimal power restoration strategy using reinforcement learning to generate the initial dynamic scheduling decision;

[0368] The balance adjustment submodule is used to use a load balancing algorithm to balance the load of the initial dynamic scheduling decision and generate a target dynamic scheduling decision.

[0369] Furthermore, the execution module 305 includes:

[0370] Establish a submodule to establish the mapping relationship between fault detection point information and target dynamic scheduling decision, and assign priority weights according to different types of faults and fault areas;

[0371] Normalization submodule, used to normalize the fault detection point information and target dynamic scheduling decision of the mapping relationship according to the priority weight;

[0372] Construct submodules to construct the power restoration optimization objective function based on the fault handling timeliness, power restoration cost and system stability of the power grid system;

[0373] The rule base submodule is used to establish an automated decision rule base using normalized fault detection point information and target dynamic scheduling decisions;

[0374] A formulation submodule is used to formulate multiple power restoration plans based on the power restoration optimization objective function and the automated decision rule library;

[0375] A calculation submodule is used to calculate the difference between the predicted value and the actual value of each power restoration plan to generate multiple error values;

[0376] The judgment submodule is used to judge whether each error value is greater than a preset error threshold, and select the power restoration plan with the smallest error value according to the judgment result to determine the power restoration optimization scheduling decision;

[0377] The execution submodule is used to execute the power restoration operation according to the power restoration optimization scheduling decision.

[0378] Furthermore, the judgment submodule includes:

[0379] An error judgment submodule is used to judge whether each error value is greater than a preset error threshold;

[0380] An updating submodule, configured to update the priority weight if yes, and jump to the step of normalizing the fault detection point information of the mapping relationship and the target dynamic scheduling decision according to the priority weight;

[0381] The submodule is selected to select the power restoration plan with the smallest error value to determine the power restoration optimization scheduling decision if no.

[0382] A fourth embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed, the power grid restoration optimization scheduling method according to any embodiment of the present invention is implemented.

[0383] Embodiment 5 of the present invention provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer executes the power grid restoration optimization scheduling method as described in any embodiment of the present invention.

[0384] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0385] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0386] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0387] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0388] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0389] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optimizing power grid restoration scheduling, characterized in that: include: Obtaining grid operation data and environmental data of the power grid system and preprocessing them to generate feature vectors; Inputting the feature vector into a preset target detection model, and detecting fault detection point information corresponding to the feature vector through the target detection model; Determine a power restoration path based on the fault detection point information, and optimize the power restoration path to generate an optimal power restoration strategy; Dynamically adjusting the optimal power restoration strategy to generate a target dynamic scheduling decision; Based on the fault detection point information and the target dynamic scheduling decision, a power restoration optimization scheduling decision is determined, and a power restoration operation is performed according to the power restoration optimization scheduling decision.

2. The power grid restoration optimization scheduling method according to claim 1, characterized in that: The obtaining of grid operation data and environmental data of the grid system and preprocessing thereof to generate a feature vector includes: Set up multiple detection points in the power grid system; Collecting grid operation data at each detection point through high-precision sensors; Collecting environmental data of each detection point through environmental sensors; Performing a weighted operation on the power grid operation data and the environmental data using a weighted average method to generate updated power grid operation data and updated environmental data; Using a filtering algorithm to filter the updated power grid operation data and the updated environmental data to generate target power grid operation data and target environmental data; The target power grid operation data and the target environment data are fused to generate fused data, and features of the fused data are extracted to generate a feature vector.

3. The power grid restoration optimization scheduling method according to claim 1, characterized in that: Inputting the feature vector into a preset target detection model and detecting fault detection point information corresponding to the feature vector through the target detection model includes: Inputting the feature vector into a fault type model corresponding to a preset target detection model, and detecting the fault type of the feature vector through the fault category model; Inputting the feature vector into a decision tree model corresponding to the target detection model, and detecting the fault location of the feature vector through the decision tree model; The fault type and the fault location are integrated to generate fault detection point information.

4. The power grid restoration optimization scheduling method according to claim 1, characterized in that: The determining of a power restoration path based on the fault detection point information, optimizing the power restoration path, and generating an optimal power restoration strategy includes: Calculating a power restoration path for fault location of the fault detection point information using the Dijkstra algorithm; A genetic algorithm is used to optimize the power restoration path to generate an optimal power restoration strategy.

5. The power grid restoration optimization scheduling method according to claim 1, characterized in that: The dynamically adjusting the optimal power restoration strategy to generate a target dynamic scheduling decision includes: Reinforcement learning is used to dynamically adjust the optimal power restoration strategy to generate an initial dynamic scheduling decision; A load balancing algorithm is used to balance the load of the initial dynamic scheduling decision to generate a target dynamic scheduling decision.

6. The power grid restoration optimization scheduling method according to claim 1, characterized in that: The determining of a power restoration optimization scheduling decision based on the fault detection point information and the target dynamic scheduling decision, and performing a power restoration operation according to the power restoration optimization scheduling decision, includes: Establishing a mapping relationship between the fault detection point information and the target dynamic scheduling decision, and assigning priority weights according to different types of faults and fault areas; Normalizing the fault detection point information and target dynamic scheduling decision of the mapping relationship according to priority weights; Constructing a power restoration optimization objective function based on the fault handling timeliness, power restoration cost, and system stability of the power grid system; An automated decision rule base is established using normalized fault detection point information and target dynamic scheduling decisions; Formulate multiple power restoration plans based on the power restoration optimization objective function and automated decision rule library; Calculating the difference between the predicted value and the actual value of each power restoration plan to generate a plurality of error values; Determine whether each error value is greater than a preset error threshold, and select a power restoration plan with the smallest error value according to the judgment result to determine the power restoration optimization scheduling decision; Execute the power restoration operation according to the power restoration optimization scheduling decision.

7. The power grid restoration optimization dispatching method according to claim 6, characterized in that: The determining whether each error value is greater than a preset error threshold, and selecting a power restoration plan with the smallest error value according to the determination result to determine a power restoration optimization scheduling decision, includes: Determining whether each of the error values ​​is greater than a preset error threshold; If so, update the priority weight and jump to the step of normalizing the fault detection point information and target dynamic scheduling decision of the mapping relationship according to the priority weight; If not, the power restoration plan with the smallest error value is selected to determine the power restoration optimization scheduling decision.

8. A power grid restoration optimization dispatching system, characterized in that: include: An acquisition module is used to acquire grid operation data and environmental data of the power grid system and perform preprocessing to generate feature vectors; An input module, configured to input the feature vector into a preset target detection model, and detect fault detection point information corresponding to the feature vector through the target detection model; an optimization module, configured to determine a power restoration path based on the fault detection point information, and optimize the power restoration path to generate an optimal power restoration strategy; An adjustment module, configured to dynamically adjust the optimal power restoration strategy and generate a target dynamic scheduling decision; An execution module is used to determine a power restoration optimization scheduling decision based on the fault detection point information and the target dynamic scheduling decision, and perform a power restoration operation according to the power restoration optimization scheduling decision.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed, the power grid restoration optimization scheduling method according to any one of claims 1 to 7 is implemented.

10. A computer program product, characterized in that The computer program product includes a computer program stored on a non-transitory computer-readable storage medium, wherein the computer program includes program instructions, wherein when the program instructions are executed by a computer, the computer is caused to execute the power grid restoration optimization scheduling method according to any one of claims 1 to 7.

Images