Hydropower centralized control electronic speed controller monitoring system diagnosis method and system

Abstract

The invention relates to a hydroelectric centralized control electronic speed controller monitoring system diagnosis method and system, and the method comprises the steps: obtaining monitoring data, converting the monitoring data into a weighted directed graph structure, and constructing and extracting fault propagation path data based on an improved interior point method; based on the fault propagation path data, an approximate Submod Matandroid constraint partitioning method is adopted to generate an optimized equipment monitoring grouping result; based on the equipment monitoring grouping result, executing a multi-dimensional diagnosis method in parallel, performing fusion by adopting an improved interior point method based on a logarithm barrier function, and outputting a comprehensive diagnosis conclusion; calculating and diagnosing performance indexes, monitoring system state changes, and adaptively adjusting and improving interior point method parameters and an equipment partitioning scheme. Incremental shortest path analysis of a fault propagation path is realized by improving an interior point method, sparse matrix storage and a preprocessing conjugate gradient algorithm are combined, calculation complexity is reduced, real-time diagnosis of large-scale water and electricity monitoring is realized, and diagnosis accuracy and efficiency are improved.

Description

Technical Field

[0001] This invention relates to the field of power system monitoring technology, specifically to a diagnostic method and system for a hydropower centralized control and dispatch monitoring system. Background Technology

[0002] The hydropower centralized control and dispatch monitoring system is the core management platform for hydropower energy production and operation, undertaking the tasks of real-time monitoring, status assessment, and fault early warning of hydropower station power generation equipment, control and protection equipment, and auxiliary equipment. With the expansion of single-unit capacity, the broadening of centralized control scope, and the improvement of intelligence level of hydropower stations, the amount of analog, digital, and SOE event data that the monitoring system needs to process is growing exponentially, and the system topology and operation mode are becoming increasingly complex.

[0003] Existing fault diagnosis technologies for centralized control and dispatch monitoring systems in hydropower mainly rely on threshold alarm mechanisms, static expert rule bases, or simple causal relationship matrix analysis. Some advanced systems employ a layered architecture, integrating multi-dimensional diagnostic methods such as expert knowledge rules, equipment mechanism models, signal feature analysis, and event sequence correlation, which improves diagnostic reliability to some extent. However, significant shortcomings remain: First, traditional diagnostic algorithms struggle to meet the real-time diagnostic needs of large-scale centralized control systems when faced with massive monitoring data and complex system topologies, leading to delayed fault response. Second, static diagnostic models cannot dynamically adapt when system topology is adjusted, equipment parameters change, or operating modes switch, requiring manual intervention to remodel, resulting in decreased diagnostic accuracy and even misdiagnosis or missed diagnosis. Third, the complex relationships between physical connections, functional associations, and fault propagation between devices are not fully considered, and the lack of optimization in zonal diagnosis leads to low efficiency in fault isolation and location. Fourth, the heterogeneous results from different diagnostic methods lack a unified optimization and fusion mechanism, making it difficult to integrate the advantages of each method, resulting in insufficient confidence and reliability of diagnostic conclusions. Summary of the Invention

[0004] The technical problem of this invention is to propose a diagnostic method and system for a hydropower centralized control and dispatch monitoring system, which reduces the computational complexity of the diagnostic algorithm, dynamically adapts to changes in system topology, equipment parameters and operating modes, and achieves efficient partitioning based on fault propagation characteristics, thereby improving fault isolation and location efficiency and increasing the accuracy of diagnostic conclusions.

[0005] The purpose of this invention is to solve the above-mentioned problems and propose a diagnostic method for a hydropower centralized control and dispatch monitoring system, comprising the following steps: S1: Acquire monitoring data from the hydropower centralized control and dispatch monitoring system, convert the monitoring data into a weighted directed graph structure, and then extract fault propagation path data based on the improved interior point method; the monitoring data includes analog quantities, switch quantities, and SOE events; the fault propagation path data includes fault source nodes, propagation paths, and affected terminal nodes; S2: Construct a Submodular objective function and design Matroid constraints using fault propagation path data, and use an approximate Submod Matroid constraint partitioning method to generate optimized device monitoring grouping results; S3: Based on the equipment monitoring grouping results, expert rule diagnosis, equipment mechanism model analysis, signal feature analysis and event sequence analysis are performed in parallel. The above analysis results are constructed into a multi-objective optimization problem, and then the improved interior point method based on the logarithmic barrier function is used for fusion to output a comprehensive diagnostic conclusion and confidence score. S4: Based on the comprehensive diagnostic conclusions and confidence scores, calculate the diagnostic accuracy, recall, and timeliness, and monitor changes in system topology, equipment parameters, and operating modes to generate diagnostic effectiveness evaluation results and system status change detection results; S5: Based on the diagnostic effect evaluation results and system state change detection results, adaptively adjust and improve the internal point method parameters and equipment partitioning scheme.

[0006] Preferably, in step S1, the monitoring data is converted into a weighted directed graph structure, including the following sub-steps: 1) Extract the device nodes corresponding to physical devices, the measurement point nodes corresponding to sensors, and the logic unit nodes corresponding to control and protection function units based on monitoring data, and assign a unique identifier and type attribute to each node to construct a node set; 2) Based on the node set, identify the physical connection relationships, functional association relationships, and fault propagation relationships between nodes, and establish an edge set; 3) Based on the edge set, calculate the physical connection strength, logical association strength and historical correlation of each edge to generate an edge weight matrix and obtain a weighted directed graph structure.

[0007] Furthermore, in step S1, fault propagation path data is extracted based on the improved interior point method, including the following sub-steps: 1) Based on the weighted directed graph structure, construct the linear programming form of the standard interior point method, introduce slack variables and logarithmic barrier functions, and establish an improved mathematical model of the interior point method; 2) Detect local changes in the network topology or edge weights of a weighted directed graph structure, determine the affected region, and construct an incremental computation subproblem; 3) Based on the incremental computation subproblem, an adaptive step size selection and preprocessing conjugate gradient algorithm is used to solve the problem and output the fault propagation path data.

[0008] Preferably, in step S2, constructing a submodular objective function using fault propagation path data includes the following sub-steps: 1) Based on the fault propagation path data, extract the equipment type matching degree, functional parameter similarity, and control logic correlation degree, and construct a functional correlation function; 2) Based on the functional correlation function, calculate the degree of concentration of physical locations of equipment and construct a location proximity function; 3) Based on the location proximity function and fault propagation path data, evaluate the effect of grouping on fault isolation, construct the fault propagation correlation function, and form a submodular objective function through weighted combination.

[0009] Preferably, the design of the Matroid constraint conditions in step S2 includes: 1) Based on the Submodular objective function, set the range of the number of devices in each group and construct a balanced partitioning Matroid constraint; 2) Based on the balanced partitioning Matroid constraint and the redundancy backup requirements of critical functional devices, construct a redundancy protection Matroid constraint; 3) Based on the redundancy guarantee Matroid constraint, and based on the plant-unit-equipment hierarchical structure of the hydropower system, a diagnostic efficiency Matroid constraint is constructed to obtain the Matroid constraint conditions.

[0010] Preferably, in step S3, the diagnostic accuracy, recall, and timeliness are calculated, and changes in system topology, device parameters, and operating modes are monitored, including the following sub-steps: 1) Based on the device monitoring grouping results, identify potentially abnormal device groups and generate a list of abnormal device groups; 2) Based on the list of abnormal device groups, create diagnostic tasks and allocate computing resources to generate a list of parallel diagnostic tasks; 3) Based on the parallel diagnostic task list, a subset of expert rules is introduced and logical reasoning is performed through the forward inference engine to output the rule diagnostic results; 4) Based on the rule-based diagnostic results, establish a device mechanism model and calculate the theoretical values ​​under the current operating conditions, and output the deviation analysis results; 5) Based on the deviation analysis results, perform time-domain and frequency-domain analysis on the high-frequency acquired data, and output the signal characteristic analysis results; 6) Based on the signal feature analysis results, use SOE records to perform event time window clustering, output event sequence analysis results, and obtain the results of the multi-dimensional diagnostic method.

[0011] Preferably, in step S3, the above analysis results are constructed into a multi-objective optimization problem, and then fused using an improved interior-point method based on a logarithmic barrier function, including the following sub-steps: 1) Collect the results of multi-dimensional diagnostic methods, map the heterogeneous diagnostic results to a unified diagnostic hypothesis space, and construct a set of diagnostic hypotheses; 2) Based on the set of diagnostic hypotheses, comprehensively evaluate the credibility of each diagnostic hypothesis and establish a multi-objective optimization problem; 3) Based on the multi-objective optimization problem, a logarithmic barrier function is introduced to construct a fusion model based on the improved interior point method; 4) Based on the improved interior point method, the fusion model is used to set the constraint relationships between assumptions and obtain the constrained optimization model; 5) Based on the constrained optimization model, the confidence level of each diagnostic hypothesis is adjusted by an iterative solution method, and the main fault hypothesis, supporting evidence list and uncertainty factors are output to obtain the comprehensive diagnostic conclusion and confidence score.

[0012] Preferably, in step S4, the diagnostic accuracy, recall, and timeliness are calculated based on the comprehensive diagnostic conclusion and confidence score, including the following sub-steps: 1) Collect comparative data of comprehensive diagnostic conclusions and confidence scores with actual fault conditions, calculate diagnostic accuracy, recall, precision and F1 score, and generate a set of performance indicators; 2) Based on the set of performance indicators, decompose and analyze them according to fault type, equipment type and fault severity to construct a classification performance analysis result; 3) Based on the classification performance analysis results, identify misdiagnosed and missed cases and establish a knowledge base of failure cases; 4) Based on the failure case knowledge base, classify the causes of failure and analyze the time interval of fault handling to generate timeliness analysis results; 5) Based on the timeliness analysis results, key indicators are displayed in a visual format, and the diagnostic effect evaluation results are output.

[0013] Furthermore, step S4 involves monitoring changes in system topology, equipment parameters, and operating modes, including the following sub-steps: 1) Monitor the topology, equipment parameters, and operating modes of the hydropower system, compare the current state with the baseline state, and identify the types of changes; 2) Based on the type of change, the topology, equipment parameters, and operating modes of the changes are analyzed to identify three key types of changes; 3) Based on the three types of key changes, assess the criticality and connectivity of the changed elements to determine the importance of the topological changes; 4) Based on the importance of topology changes, set multiple levels of change thresholds to obtain the degree of impact of equipment parameter changes; 5) Based on the degree of impact of changes in equipment parameters, identify the current operating mode and detect mode transitions, generating a record of operating mode changes; 6) Based on the record of changes in operating mode, a multi-confirmation mechanism is used to perform cross-validation and evaluate the cumulative effect, and output the system state change detection results.

[0014] A diagnostic method for a hydropower centralized control and dispatch monitoring system includes a diagnostic system comprising: The data acquisition module is used to acquire monitoring data from the hydropower centralized control and dispatch monitoring system and convert the monitoring data into a weighted directed graph structure. The path analysis module is used to extract fault propagation path data based on the improved interior point method. The equipment grouping module is used to generate optimized equipment monitoring grouping results based on fault propagation path data and an approximate Submod Matroid constrained partitioning method; the equipment includes power generation equipment, control and protection equipment, and auxiliary equipment of hydropower stations; The diagnostic fusion module is used to execute multi-dimensional diagnostic methods in parallel based on the device monitoring grouping results, fuse the diagnostic results, and output a comprehensive diagnostic conclusion and confidence score. The effectiveness evaluation module is used to calculate diagnostic performance indicators and monitor system state changes, and generate diagnostic effectiveness evaluation results and system state change detection results. The adaptive optimization module is used to adaptively adjust and improve the interior point method parameters and equipment partitioning scheme based on the diagnostic effect evaluation results and system state change detection results. Compared with the prior art, the beneficial effects of the present invention include: 1) The present invention proposes a diagnostic method for a hydropower centralized control and dispatch monitoring system. By improving the interior point method, the incremental shortest path analysis of the fault propagation path is realized. Combined with sparse matrix storage and preprocessing conjugate gradient algorithm, the computational complexity is reduced and the diagnostic response time is shortened from the second level to the millisecond level, which meets the real-time diagnostic requirements of large-scale hydropower centralized control systems.

[0015] 2) This invention constructs a diagnostic effect evaluation and system status monitoring mechanism, which can dynamically sense changes in system topology, equipment parameters and operating modes, and adaptively adjust algorithm parameters and equipment partitioning schemes without manual intervention, ensuring the stability of the system's diagnostic performance in complex and ever-changing scenarios.

[0016] 3) This invention adopts an approximate Submod Matroid constrained partitioning method to achieve optimal grouping of devices, taking into account functional correlation, location concentration and fault isolation effect. Through the optimization and fusion of multi-dimensional diagnostic results, it integrates the advantages of expert rules, mechanism models, signal features and event sequences, thereby improving the accuracy of diagnosis.

[0017] 4) This invention is applicable to hydropower centralized control and dispatch monitoring systems of all sizes. It can cover all types of fault diagnosis of power generation equipment, control and protection equipment and auxiliary equipment. The diagnosis results include confidence scores and a list of supporting evidence, providing a clear basis for operation and maintenance decisions, significantly shortening fault handling time, and improving the safety and economy of hydropower system operation. Attached Figure Description

[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0019] Figure 1 This is a schematic diagram of the diagnostic method for a hydropower centralized control and dispatch monitoring system according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the incremental shortest path algorithm based on the improved interior point method according to an embodiment of the present invention. Figure 3 This is a schematic diagram of the approximate Submod Matroid constrained partitioning method according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the architecture of a hydropower centralized control and dispatch monitoring system according to an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0021] Example 1: Taking Unit 1 as an example, this unit includes equipment such as a turbine, generator, governor, and excitation system. The implementation process of the diagnostic method of this invention is explained in detail below: like Figure 1 As shown, a diagnostic method for a hydropower centralized control and dispatch monitoring system includes the following steps: S1: Acquire monitoring data from the hydropower centralized control and dispatch monitoring system, convert the monitoring data into a weighted directed graph structure, and then extract fault propagation path data based on the improved interior point method; the monitoring data includes analog quantities, switch quantities and SOE events; the fault propagation path data includes fault source nodes, propagation paths and affected terminal nodes.

[0022] In this embodiment, monitoring data from the hydropower centralized control and dispatch monitoring system is collected, including analog quantities, digital quantities, and high-precision event sequence records (SOEs). Analog quantities include voltage, current, temperature, and water level; digital quantities include circuit breaker status and valve position; and high-precision SOEs are used for data collection. This heterogeneous data is then uniformly converted into a graph model. In the graph model, nodes represent physical devices such as generators, turbines, transformers, and circuit breakers; measurement points include various sensors; and logic units include protection devices and control units. Edges represent the relationships between nodes, mainly including three types: physical connections, functional relationships, and fault propagation relationships. Each edge has a weight, representing the probability of a fault propagating from one node to another.

[0023] S101. Based on the monitoring data, extract the device nodes corresponding to the physical devices, the measurement point nodes corresponding to the sensors, and the logic unit nodes corresponding to the control and protection function units, and assign a unique identifier and type attribute to each node to construct a node set.

[0024] In this embodiment, physical device nodes include motors, water turbines, and transformers; measurement point nodes include various types of sensors; and logic unit nodes include protection devices and control units.

[0025] Specifically, node set construction: Extract physical device nodes: Generator No. 1 GEN01, Water Turbine No. 1 TRB01, Main Transformer No. 1 TRF01; Extracted measurement points: Stator temperature sensor GST01T and rotor current sensor GRC01 of generator No. 1; Extract the logic unit nodes: Unit 1 protection device PROT01, speed controller SPD01.

[0026] S102. Based on the node set, identify the physical connection relationship, functional association relationship and fault propagation relationship between nodes, and establish the edge set.

[0027] In this embodiment, the physical connection relationship refers to the actual connection between devices, the functional association relationship refers to devices that are functionally related, and the fault propagation relationship refers to the possibility that a fault in one device may affect another device.

[0028] Specifically, turbine TRB01 and generator GEN01 are connected by their main shafts and have a physical connection. Generator No. 1 GEN01 and generator No. 1 stator temperature sensor GST01T have a functional relationship because the sensors monitor the generator status; The governor of Unit 1 and the guide vane opening of Unit 1 may be abnormal due to governor failure, and there is a fault propagation relationship.

[0029] S103. Based on the edge set, calculate the physical connection strength, logical association strength, and historical correlation of each edge to generate an edge weight matrix and obtain a weighted directed graph structure.

[0030] In this embodiment, a weight is calculated for each edge. The weight calculation comprehensively considers three factors: physical connection strength, logical association strength, and historical relevance. The final weight value is obtained by weighted summation. A sparse matrix is ​​used to store the weighted directed graph to save storage space and improve computational efficiency.

[0031] Specifically, the formula for calculating the weight of each edge is: Weight = 0.4 × Physical Connection Strength + 0.3 × Logical Association Strength + 0.3 × Historical Correlation. For edge (GEN01, TRB01), the physical connection strength is 0.9, the logical association strength is 0.7, and the historical correlation is 0.8. Using the weighting formula, the final weight is 0.4×0.9 + 0.3×0.7 + 0.3×0.8 = 0.81.

[0032] S104. Based on the weighted directed graph structure, construct the linear programming form of the standard interior-point method, introduce slack variables and logarithmic barrier functions, and establish an improved mathematical model of the interior-point method.

[0033] In this embodiment, the shortest path problem is represented as a linear programming problem. Finding the shortest path from the starting point to the ending point in a given weighted directed graph is represented as a flow minimization problem, with constraints including flow conservation and non-negative flow requirements. The core of the improved interior-point method is the introduction of a barrier function, transforming inequality constraints into penalty terms in the objective function. A logarithmic barrier function is designed, a Lagrangian function is constructed, and the computational framework is obtained by solving for the optimal conditions.

[0034] The model is constructed by transforming the device nodes and connections in the network into a mathematical optimization problem. By defining flow variables and constraints, the possible propagation paths of faults in the system are accurately described: each device in the system corresponds to a node in the graph, the physical connections and functional associations between devices constitute directed edges, and the weight of the edges integrates multiple factors such as connection strength and historical fault correlation.

[0035] In each iteration, the algorithm gradually approaches the boundary of the feasible region by adjusting the value of the barrier parameter μ. When applied to fault propagation analysis, firstly, the barrier function ensures that the solution always remains within the feasible region, avoiding the generation of invalid solutions; secondly, the logarithmic barrier term generates a sufficiently large gradient as it approaches the boundary, guiding the algorithm to converge quickly to the optimal path. Practice shows that this approach is particularly suitable for physical networks with strictly non-negative properties, such as hydropower systems.

[0036] The linear system established by KKT conditions is solved using the Newton-Raphson iteration method. To address the sparse characteristics typically found in hydropower monitoring networks, a compressed row storage (CSR) format is employed to efficiently process large sparse matrices. To accelerate the convergence of the conjugate gradient method, a preprocessor based on incomplete Cholesky decomposition is designed. To meet real-time diagnostic requirements, an adaptive step size selection mechanism is implemented.

[0037] The shortest path problem can be expressed in the following linear programming form: ; In the formula, E represents the set of all edges in the graph. This represents the weight of edge (i,j). Let (i,j) represent the flow variable on edge (i,j).

[0038] The expression for the constraint is: ; In the formula, V represents the set of all nodes, s represents the fault source node, and t represents the affected terminal node.

[0039] The formula for handling non-negativity constraints using a logarithmic barrier function is as follows: ; In the formula, Indicates the obstacle parameter.

[0040] The expression for the Lagrange function is: ; In the formula, A represents the v*E-dimensional flow conservation constraint matrix, b represents the corresponding right-hand vector, λ represents the Lagrange multiplier vector, and x represents the vector composed of all flow variables.

[0041] S105. Detect local changes in the network topology or edge weights of a weighted directed graph structure, determine the affected region, and construct an incremental computation subproblem. Specifically, changes are detected and classified, including changes in edge weights, additions or deletions of edges, and additions or deletions of nodes. Then, the affected area of ​​the change is determined, that is, by constructing an incremental computation subproblem through the set of nodes that may cause changes in the shortest path.

[0042] A fault was detected in the voltage feedback loop of the excitation system of Unit 1, causing a change in the weight of the relevant edges. The weight of edge (EXC01, GEN01) increased from 0.65 to 0.85, indicating an increased probability that the excitation system fault would propagate to the generator. The affected area was determined to include the excitation system node EXC01 and its adjacent nodes GEN01 and AVR01. A subgraph centered on these nodes was constructed as an incremental computational subproblem.

[0043] S106. Based on the incremental computation subproblem, an adaptive step size selection and preprocessing conjugate gradient algorithm is used to solve the problem and output the fault propagation path data.

[0044] In this embodiment, an adaptive step size selection strategy is adopted to dynamically adjust the step size based on the characteristics of the current solution, thereby accelerating convergence; a preconditional conjugate gradient method is used to solve the linear equation system, reducing computational complexity; and the sparsity characteristics of the hydropower system network are fully utilized, with a compressed storage format used to optimize computation. These improvements enable the algorithm to achieve approximately linear time complexity when calculating the shortest path in a large-scale hydropower monitoring system.

[0045] The initial step size was set to 0.5, and then dynamically adjusted based on the convergence of each iteration. When a rapid convergence trend was detected, the step size was increased to 0.8; when oscillations occurred, the step size was decreased to 0.3. Incomplete Cholesky decomposition was used as a preprocessor to accelerate the convergence of the conjugate gradient method. After 7 iterations, the algorithm converged on the subproblem, yielding the shortest path from the excitation system EXC01 to the generator protection device PROT01 as EXC01→AVR01→GEN01→PROT01, with a total path weight of 2.45. This path was output as part of the fault propagation path data, indicating that an excitation system fault may affect the generator protection device through this path.

[0046] S2: As Figure 3 As shown, a submodular objective function is constructed using fault propagation path data, and Matroid constraints are designed. An approximate submod Matroid constraint partitioning method is then used to generate optimized equipment monitoring grouping results. These results include the equipment group hierarchy and the inter-group correlation strength matrix. The equipment includes power generation equipment, control and protection equipment, and auxiliary equipment in a hydropower station.

[0047] S201: Based on the fault propagation path data, extract the equipment type matching degree, functional parameter similarity and control logic correlation degree, and construct a functional correlation function.

[0048] In this embodiment, a functional correlation function is designed to evaluate the degree of functional correlation of devices within a group, taking into account the matching degree of device type, similarity of functional parameters and correlation of control logic, and encouraging devices with similar functions to be grouped together to facilitate professional diagnosis.

[0049] Specifically, based on fault propagation path data, the device type matching degree is 1.0 for GEN01 and GEN02; the functional parameter similarity is 0.6 for GEN01 and EXC01; and the control logic correlation degree is 0.85 for GEN01 and PROT01. The functional correlation function is obtained by weighted combination of device type matching degree, functional parameter similarity, and control logic correlation degree.

[0050] S202: Based on the functional correlation function, calculate the degree of concentration of physical locations of equipment and construct the location proximity function.

[0051] Specifically, a proximity function is designed to assess the concentration of physical locations of devices within a group, encouraging devices that are physically close to each other to be grouped together, thus reducing the spatial span required for diagnosis.

[0052] Based on the physical coordinates of the equipment, the concentration of equipment in the same factory building is calculated. GEN01 and TRB01 are 5 meters apart, with a similarity score of 0.9.

[0053] S203: Based on the location proximity function and fault propagation path data, evaluate the effect of grouping on fault isolation, construct the fault propagation correlation function, and form a submodular objective function through weighted combination.

[0054] The calculation of equipment type matching degree is based on a predefined equipment classification system and is achieved through type tree hierarchical analysis and functional compatibility assessment.

[0055] For any two device nodes u and v, the expression for type matching degree is: ; In the formula, LCA(u,v) represents the Lowest Common Ancestor level. The maximum depth of the type tree. The adjustment coefficient is empirically set at 0.6-0.8; C(u,v) is derived from the functional compatibility score of the expert knowledge base. This represents the topological distance in the type tree.

[0056] Build a feature vector for each device It includes key parameters such as voltage deviation rate, load rate, temperature change rate, and vibration amplitude.

[0057] The weighted cosine similarity is used to calculate the similarity of operating states between devices, and the expression is: ; The feature distribution is continuously updated using a sliding time window mechanism, and KL divergence is introduced for dynamic correction. The expression is: ; In the formula, and These represent the probability distributions corresponding to the two objects. This represents the KL divergence, which measures the difference between two probability distributions.

[0058] The evaluation of control logic correlation requires the integration of multi-source data such as SCADA control charts, protection interlock tables, and communication matrices. The expression for establishing a three-layer correlation model is as follows: ; In the formula, This represents a Boolean function used to determine direct control relationships.

[0059] Using the above formula, the correlation between the generator and the excitation system can reach 0.93, while the correlation with the cooling system may be 0.15.

[0060] The fault propagation correlation function is constructed based on path analysis of historical fault data, taking into account both path frequency and intensity. The calculation formula is as follows: ; In the formula, This represents the actual number of times path p occurs. The number of possible occurrences is given by |p|, which represents the path length.

[0061] The resulting submodular objective function is a weighted combination of three parts, calculated as follows: ; In the formula, S represents the set currently being evaluated, |S| represents the number of elements in set S, Match(u,v) represents the matching degree or correlation between u and v, and F(S) represents the comprehensive scoring function of set S.

[0062] The fault propagation correlation function is designed based on the fault propagation path to evaluate the effectiveness of grouping for fault isolation. It encourages each group to cover the complete fault propagation path as much as possible and penalizes cases where a single fault propagation path is split into multiple groups. The three sub-functions are combined using weighted sums to form the final objective function, with weight coefficients used to balance the importance of each aspect.

[0063] S204: Based on the Submodular objective function, set the range of the number of devices in each group and construct a balanced partitioning Matroid constraint.

[0064] The balanced partitioning design of Matroid ensures that the number of devices in each group is within a reasonable range, avoiding overly large or undersized groups. In practical applications, partitioning can be done according to device type, setting an appropriate capacity limit for each type of device.

[0065] Specifically, the following constraints apply: balanced zoning: each group should have no more than 3 generating units and no more than 8 control and protection units; redundancy guarantee: each group should contain at least 2 independent protection units; and diagnostic efficiency: equipment in the same generating unit should be grouped together, while equipment in different plants should be grouped in different groups, following the plant-unit-equipment hierarchical structure.

[0066] S205: Based on the balanced partitioning Matroid constraint and the redundancy backup requirements of critical functional devices, construct a redundant protection Matroid constraint.

[0067] Specifically, the design redundancy guarantee of Matroid ensures that critical functional devices have sufficient redundancy backups to improve reliability. This constraint ensures that each group contains a sufficient number of critical functional devices to meet redundancy backup requirements.

[0068] S206: Based on the redundancy guarantee Matroid constraint, and based on the plant-unit-equipment hierarchical structure of the hydropower system, construct the diagnostic efficiency Matroid constraint to obtain the Matroid constraint conditions.

[0069] Specifically, the design of the diagnostic efficiency Matroid ensures that the grouping structure meets the hierarchical requirements of the diagnostic process, thereby improving diagnostic efficiency. This constraint ensures that the grouping structure follows the natural hierarchy of the hydropower system, such as the plant-unit-equipment hierarchy. Combining these three types of Matroid constraints forms the final constraint set.

[0070] S207: Based on the Submodular objective function and Matroid constraints, an approximate algorithm combining a greedy strategy and continuous relaxation is used to solve the Submod Matroid constrained partitioning problem and obtain an approximately optimal device grouping scheme.

[0071] S208: Based on the device grouping scheme, construct a hierarchical structure of device groups to form a multi-level diagnostic view from the system level to the device level, and calculate the correlation strength matrix between groups to obtain the device monitoring grouping results.

[0072] Based on the grouping scheme, feature vectors are extracted for each equipment group, including functional features, topological features, fault features, and diagnostic features. Based on the group feature vectors, a hierarchical clustering algorithm is applied to construct the hierarchical structure of the groups. A similarity matrix between groups is calculated, and a hierarchical tree is constructed. Based on the hierarchical tree, the equipment groups are divided into multiple hierarchical levels, from the highest level of the entire hydropower system, to the intermediate level of functional subsystems and functional units, and then to the lowest level of basic equipment groups. The correlation strength matrix between groups within the same level is calculated to guide correlation analysis during the diagnostic process.

[0073] Specifically, the grouping results are as follows: Group 1 {GEN01, TRB01, EXC01, SPD01}, Group 2 {GEN02, TRB02, EXC02, SPD02}, and Group 3 {TRF01, TRF02, TRF03, Plant Transformer}. Hierarchical clustering algorithm is applied to calculate the similarity matrix between groups. The similarity between Group 1 and Group 2 is 0.85, and the similarity between Group 1 and Group 3 is 0.4. Single-link clustering method is used to progressively merge the most similar groups, forming a hierarchical tree. Finally, a three-layer structure is constructed: System Level - Functional Subsystem Level - Equipment Group Level. The System Level includes all equipment; the Functional Subsystem Level includes the power generation system and auxiliary systems; and the Equipment Group Level includes basic equipment groups. The association strength matrix between groups is calculated, and the association strength between the power generation system and the auxiliary system is 0.5.

[0074] S3: Based on the equipment monitoring grouping results, perform expert rule diagnosis, equipment mechanism model analysis, signal feature analysis and event sequence analysis in parallel, and construct the above analysis results into a multi-objective optimization problem. Then, use the improved interior point method based on the logarithmic barrier function to fuse them and output a comprehensive diagnostic conclusion and confidence score.

[0075] S301: Based on the device monitoring grouping results, identify potentially abnormal device groups and generate a list of abnormal device groups. Anomaly detection is based on simple threshold detection or rate of change detection, aiming to quickly narrow down the scope requiring in-depth diagnosis. For each potentially abnormal device group, create a diagnostic task.

[0076] Specifically, the equipment group of Unit 1 in the first group was identified as a potential anomaly group. The assigned diagnostic task allocated 4 computing cores and 8GB of memory to the first group to generate a list of parallel diagnostic tasks.

[0077] S302: Based on the list of abnormal device groups, create diagnostic tasks and allocate computing resources to generate a list of parallel diagnostic tasks. The priority of diagnostic tasks is determined based on the following factors: the importance level of the group, the severity of the anomaly, and the dependencies between groups.

[0078] S303: Based on the parallel diagnostic task list, a subset of expert rules is introduced and logical reasoning is performed through the forward inference engine to output the rule diagnostic results.

[0079] Load a subset of expert rules related to the current device group, and use the forward inference engine to match the monitoring data with the rule conditions to trigger the corresponding diagnostic conclusions.

[0080] Specifically, a subset of expert rules related to Unit 1 was loaded, including 152 rules such as generator overheating rules, excitation anomaly rules, and shaft vibration rules. The forward inference engine matched the current monitoring data with the rule conditions, triggering 23 rules. The preliminary rule diagnosis result was output: the automatic voltage regulator (AVR) of Unit 1 may be faulty, with a confidence level of 0.82.

[0081] S304: Based on the rule-based diagnostic results, establish a device mechanism model and calculate the theoretical value under the current operating conditions, and output the deviation analysis results.

[0082] Based on the operating principles of the equipment group, a theoretical model is established to calculate the theoretical values ​​under current operating conditions. These values ​​are then compared with actual measurements to identify significant deviations. Deviation analysis is particularly suitable for detecting performance degradation and early signs of failure.

[0083] Specifically, a mechanism model for the excitation of Unit 1 was established, with the current operating conditions input as follows: active power 80MW, reactive power 15Mvar. The theoretical excitation voltage was calculated to be 220V±5V, the excitation current should be 1050A±30A, and the AVR output should be 4.2V±0.2V. Comparison with actual measurements: the actual measured excitation voltage was 185V-240V, the actual measured excitation current was 1030A-1120A, and the actual measured AVR output was 3.8V~4.5V. Output deviation analysis showed abnormal fluctuations in the AVR output signal, leading to instability in the excitation voltage and current, with a deviation of 27%, exceeding the normal operating range.

[0084] S305: Based on the deviation analysis results, perform time-domain and frequency-domain analysis on the high-frequency acquired data and output signal feature analysis results.

[0085] For high-frequency acquired data, such as vibration, pressure pulsation, and current waveforms, time-domain and frequency-domain analyses are performed to extract characteristic quantities that reflect the health status of the equipment. In particular, for spectrum analysis, attention is paid to anomalies in specific frequency components, including rotational frequencies, harmonics, and bearing fault characteristic frequencies.

[0086] Specifically, FFT analysis of the 2kHz rotor current waveform detected 5Hz and 10Hz oscillation characteristics with a confidence level of 0.88.

[0087] S306: Based on the signal feature analysis results, SOE records are used to perform event time window clustering, and the event sequence analysis results are output to obtain the results of the multi-dimensional diagnostic method.

[0088] By utilizing the high-precision timestamps recorded by SOE, time window clustering is performed on events before and after the fault occurs, and the initial triggering event and chain reaction path are identified based on a predefined causal relationship matrix.

[0089] Specifically, SOE events within the 30 minutes prior to the failure were collected, totaling 27 records, including operational events, alarm events, and state change events. Key event sequences were identified through time window clustering. 1) The load of Unit 1 increased from 60MW to 80MW; 2) The excitation of Unit 1 is switched to automatic mode; 3) The AVR reference voltage is adjusted to 110%; 4) Excitation voltage fluctuation alarm; 5) Generator stator temperature rise alarm; Final output event sequence analysis results: Switching the excitation mode and adjusting the reference voltage after the load increases may have triggered AVR instability, with a confidence level of 0.75. It is recommended to check the AVR parameter settings and feedback loop.

[0090] S307: Collect the results of multi-dimensional diagnostic methods, map the heterogeneous diagnostic results to a unified diagnostic hypothesis space, and construct a set of diagnostic hypotheses.

[0091] Specifically, the results of the above diagnostic methods were collected and mapped to a unified diagnostic hypothesis space. The identified diagnostic hypotheses were: H1 = AVR control parameter mismatch, H2 = AVR feedback loop fault, H3 = excitation transformer fault, and H4 = generator overload. Supporting evidence and confidence levels were recorded for each hypothesis: the initial confidence level for H1 was 0.65, for H2 it was 0.82, for H3 it was 0.30, and for H4 it was 0.15.

[0092] S308: Based on the set of diagnostic hypotheses, comprehensively evaluate the credibility of each diagnostic hypothesis and establish a multi-objective optimization problem.

[0093] For each diagnostic hypothesis, its credibility is evaluated from multiple dimensions: the strength of supporting evidence, the consistency of evidence, the reliability of the diagnostic method, and the completeness of the diagnostic logic. The fusion problem is represented as an optimization problem, with the goal of finding the set of diagnostic hypotheses that best explains the current observation data, while considering the relationships between the hypotheses.

[0094] Specifically, for H1, supporting evidence includes the time-series relationship of parameter adjustments after load changes, but direct parameter measurement evidence is lacking; the time-series relationship strength is 0.7, and the AVR oscillation characteristic strength is 0.6. For H2, supporting evidence includes characteristic frequency analysis results (0.9), voltage fluctuation patterns (0.8), and theoretical model deviation (0.85), showing high consistency. For H3, only indirect evidence supports the hypothesis, and it does not perfectly match the observed phenomena. For H4, there is direct evidence to refute the hypothesis. Considering the historical reliability of each diagnostic method: signal feature analysis has an accuracy rate of 92% in similar faults, mechanistic model analysis has an accuracy rate of 88%, rule-based diagnosis has an accuracy rate of 80%, and event sequence analysis has an accuracy rate of 75%. These factors are combined into a multi-objective optimization problem, with the goal of maximizing the weighted sum of evidence support, evidence consistency, and diagnostic method reliability.

[0095] S309: Based on a multi-objective optimization problem, a logarithmic barrier function is introduced to construct a fusion model based on an improved interior-point method. The improved interior-point method is applied to transform the problem into an optimization problem with a barrier function. By introducing the logarithmic barrier function, factors such as hypothesis confidence scores, reliability weights of diagnostic methods, and constraints on the relationships between hypotheses are unified into a single objective function.

[0096] Based on the multi-objective optimization problem, a logarithmic barrier function is introduced to construct a fusion model based on the improved interior point method. The objective function used is as follows: ; In the formula, x represents the confidence vector. ; Represents the i-th diagnostic hypothesis Confidence level; This represents the weight of the i-th diagnostic method. This indicates support for the hypothesis. The strength of evidence, with a value ranging from [0,1], This indicates the degree of influence of the control barrier function.

[0097] S310: Based on the improved interior-point method, a fusion model is established, and constraints are set between hypotheses to obtain a constrained optimization model. The following constraints are added: H1 and H2 are not completely mutually exclusive but have a competing relationship, expressed as 0.7·x1+x2≤1.5; H3 and H2 have a certain degree of mutual exclusivity, expressed as x2+x3≤1.2; the sum of the confidence levels of all hypotheses should be close to but not necessarily equal to 1, expressed as 0.8≤∑x i ≤ 1.2.

[0098] S311: Based on the constrained optimization model, the confidence level of each diagnostic hypothesis is adjusted by an iterative solution method. The main fault hypothesis, supporting evidence list and uncertainties are output to obtain the comprehensive diagnostic conclusion and confidence score.

[0099] The algorithm iteratively solves the problem, gradually adjusting the confidence levels of each diagnostic hypothesis until the optimal solution is reached. This process automatically balances the contributions of different diagnostic methods while considering the structured relationships between the diagnostic hypotheses. Finally, it outputs the fused diagnostic conclusions, including the main fault hypotheses and their combined confidence scores, a list of key evidence supporting each hypothesis, possible alternative explanations and their confidence levels, uncertainties, and matters requiring further confirmation.

[0100] Specifically, starting from the initial point x0 = (0.65, 0.82, 0.30, 0.15), the interior-point method iteratively is executed. After the first iteration, x1 = (0.58, 0.86, 0.25, 0.12); after the fifth iteration, x5 = (0.42, 0.91, 0.18, 0.08); after the tenth iteration, the algorithm converges to x. 10 = (0.38, 0.92, 0.15, 0.05). The final comprehensive diagnostic conclusion is: the excitation AVR feedback loop of Unit 1 is faulty, with a confidence level of 0.92. It is recommended that maintenance personnel prioritize checking the voltage feedback loop of the AVR.

[0101] S4: Based on the comprehensive diagnostic conclusions and confidence scores, calculate the diagnostic accuracy, recall, and timeliness, and monitor changes in system topology, equipment parameters, and operating modes to generate diagnostic effectiveness evaluation results and system status change detection results.

[0102] S401: Collect comparative data of comprehensive diagnostic conclusions and confidence scores with actual fault conditions, calculate diagnostic accuracy, recall, precision and F1 score, and generate a set of performance indicators.

[0103] Establish a diagnostic result tracking mechanism to record the complete process and results of each diagnostic activity. Once the fault is finally confirmed and resolved, the operations and maintenance personnel will fill out a feedback form to record the actual fault situation. The diagnostic results will be automatically compared with the actual situation to calculate a series of performance indicators, including diagnostic accuracy, recall, precision, F1 score, and average diagnostic time.

[0104] Specifically, the calculated performance metrics are: diagnostic accuracy of 92%, recall rate of 95%, and F1 score of 0.93. Adjustments to the AVR parameters of Unit 1 were detected, with the impact assessed as moderate, and status change detection results were generated.

[0105] S402: Based on the set of performance indicators, decompose and analyze according to fault type, equipment type and fault severity to construct a classification performance analysis result.

[0106] The evaluation results are decomposed into different dimensions, including fault type, equipment type, and fault severity. The fault type dimension includes mechanical faults, electrical faults, and control system faults, while the equipment type dimension includes generators, turbines, and transformers. The fault severity dimension includes severe, moderate, and minor. This decomposition analysis helps to identify the strengths and weaknesses of the diagnostic model in specific areas.

[0107] S403: Based on the classification performance analysis results, identify misdiagnosed and missed cases and establish a failure case knowledge base. For misdiagnosed and missed cases, conduct in-depth failure reason analysis. Failure reasons may include: sensor data quality issues, incomplete rule coverage, inaccurate model parameters, and unidentified fault modes. Each failure case is categorized and recorded, forming a continuously improving knowledge base.

[0108] S404: Based on a failure case knowledge base, categorize failure causes and analyze fault handling time intervals to generate timeliness analysis results. Analyze diagnostic timeliness, assessing the time intervals from fault occurrence to detection, from detection to preliminary diagnosis, and from preliminary diagnosis to final confirmation. For critical equipment requiring rapid response, timeliness is a particularly important performance indicator.

[0109] S405: Based on the timeliness analysis results, key indicators are displayed in a visual format, and the diagnostic effect evaluation results are output.

[0110] In this embodiment, a comprehensive diagnostic performance report is generated based on all evaluation data, including overall performance indicators, categorized performance analysis, typical failure case studies, timeliness analysis, and improvement recommendations. The report is presented in a visual format, using dashboards, trend charts, and comparison charts to display key indicators.

[0111] S406: Monitor the topology, equipment parameters, and operating modes of hydropower systems, compare the current state with the baseline state, and identify the type of change.

[0112] In this embodiment, three key types of changes are continuously monitored: topology changes, equipment parameter changes, and operating mode changes. Topology changes include the addition, decommissioning, or alteration of equipment connections; equipment parameter changes include adjustments to rated values ​​and modifications to protection settings; and operating mode changes include switching from power generation to phase regulation and from full-load operation to partial-load operation.

[0113] S407: Based on the type of change, analyze the changed topology, equipment parameters and operating modes to obtain three key types of changes.

[0114] S408: Based on three types of key changes, assess the criticality and connectivity of the changed elements to determine the importance of the topological changes.

[0115] In this embodiment, for changes in the network topology, the addition or removal of nodes and edges is automatically identified by comparing the current network topology graph with the baseline topology graph. When a change is detected, the importance of the change is assessed, and a decision is made on whether to trigger a model update based on the criticality of the changed element and its connectivity with other devices.

[0116] S409: Based on the importance of topology changes, set multi-level change thresholds to obtain the degree of impact of device parameter changes.

[0117] In this embodiment, for changes in equipment parameters, historical baseline values ​​of the equipment parameters are maintained, and multi-level change thresholds are set. Minor changes may only be logged, moderate changes may trigger local model updates, while major changes will trigger a complete model reconstruction. The importance of parameter changes depends not only on the magnitude of the change, but also on the degree of impact of the parameter on system operation and fault diagnosis.

[0118] S410: Based on the degree of impact of changes in equipment parameters, identify the current operating mode and detect mode transitions, and generate a record of operating mode changes.

[0119] In this embodiment, for changes in operating mode, the system identifies its current operating mode based on a predefined mode feature library. When a mode transition is detected, a set of diagnostic parameters suitable for the new operating mode is automatically loaded to ensure that the diagnostic logic matches the current operating state.

[0120] S411: Based on the record of changes in operating mode, a multi-confirmation mechanism is used to perform cross-validation and evaluate the cumulative effect, and output the system state change detection results.

[0121] In this embodiment, a multiple verification mechanism is employed to improve the reliability of change detection. For significant changes detected, cross-validation is performed using multiple information sources to avoid misjudgments due to data anomalies. The system can also assess the cumulative effect of changes. Some seemingly minor individual changes may significantly impact the diagnostic model after accumulating to a certain extent. By tracking the cumulative change over a period of time, the system determines whether the threshold for updating the model has been reached.

[0122] Specifically, the adjustment of AVR parameters of Unit 1 was detected, the impact was assessed as moderate, and a status change detection result was generated.

[0123] S5: Based on the diagnostic effect evaluation results and system state change detection results, adaptively adjust and improve the internal point method parameters and equipment partitioning scheme.

[0124] Step S501: Based on the diagnostic effect evaluation results, identify the performance bottleneck under the current parameter settings, and adjust the key parameters of the interior point method by adopting a rule-based parameter adjustment strategy combined with an adaptive search mechanism.

[0125] In this embodiment, the key parameters of the interior-point method include: the barrier function coefficient μ, the convergence threshold ε, the step size control parameter τ, the preprocessor selection, and the initial point selection strategy. Based on the diagnostic effect evaluation results, performance bottlenecks under the current parameter settings are identified. A rule-based parameter adjustment strategy combined with an adaptive search mechanism is adopted. The rule base defines the basic principles of parameter adjustment.

[0126] S502: Based on the system state change detection results, adjust the initial solution construction method according to the scale and location of the change, update the coefficients in the weight calculation formula, and switch to a parameter set optimized for a specific mode.

[0127] In this embodiment, parameter adjustments follow different strategies in response to system state changes. Changes in topology primarily affect the initial point selection strategy of the interior-point method, adjusting the initial solution construction method according to the scale and location of the change. Changes in device parameters mainly affect the weight calculation of relevant edges, updating the coefficients in the weight calculation formula. Changes in operating mode require switching to a parameter set optimized for that specific mode. Multiple preset parameter sets are maintained, corresponding to different operating scenarios and diagnostic task types. When a specific scenario is detected, the system can quickly switch to the corresponding parameter set and then fine-tune it, avoiding parameter searches from scratch and improving adaptation speed.

[0128] S503: Based on the diagnostic effect evaluation results and system state change detection results, evaluate the effectiveness of the current partitioning scheme, identify partitions that need optimization, and re-execute the Submod Matroid constrained partitioning algorithm.

[0129] In this embodiment, the dynamic optimization of the device partitioning scheme is divided into two modes: periodic optimization and triggered optimization. Periodic optimization is performed at preset time intervals to comprehensively evaluate the effectiveness of the current partitioning scheme. Triggered optimization is initiated when significant system changes are detected or diagnostic performance deteriorates significantly. The optimization process first evaluates the effectiveness of the current partitioning scheme. Evaluation indicators include the functional relevance of devices within the partition, physical location concentration, fault propagation isolation effect, and computational efficiency and accuracy during actual diagnostics. A weighted composite score of these indicators is calculated as the benchmark score for the current scheme. Specific partitions requiring optimization are identified. Not all partitions need to be optimized simultaneously; partitions with lower scores or those more affected by system changes are prioritized for adjustment. This selective optimization strategy reduces computational resource consumption and minimizes the impact on normal diagnostic operations.

[0130] S504: Based on the results of the re-execution of the Submod Matroid constrained partitioning algorithm, change cost constraints are introduced. The performance of the optimized partitioning scheme is evaluated through simulation, and a phased deployment strategy is adopted to update the device partitioning scheme.

[0131] In this embodiment, to maintain the stability of the partitioning scheme, a change cost constraint is introduced during the optimization process. Each adjustment to the current partitioning scheme calculates a change cost, and the adjustment is only executed when the optimization benefits significantly outweigh the change cost. This avoids frequent changes to the partitioning scheme due to minor benefits, which could lead to system instability. The optimized partitioning scheme's performance is evaluated through simulation. Using historical diagnostic cases or generated fault scenarios, the diagnostic performance under the old and new partitioning schemes is compared to confirm that the optimization has indeed brought substantial improvements. Only schemes that pass simulation evaluation are actually deployed. To ensure a smooth transition, a phased deployment strategy is implemented. The new scheme is first deployed in non-critical areas or offline environments to verify its stability before being gradually expanded to more critical areas. This gradual deployment reduces potential risks.

[0132] like Figure 4 As shown, a diagnostic method for a hydropower centralized control and dispatch monitoring system includes a diagnostic system comprising: The data acquisition module is used to acquire monitoring data from the hydropower centralized control and dispatch monitoring system and convert the monitoring data into a weighted directed graph structure. The path analysis module is used to extract fault propagation path data based on the improved interior point method. The equipment grouping module is used to generate optimized equipment monitoring grouping results based on fault propagation path data and an approximate Submod Matroid constrained partitioning method; the equipment includes power generation equipment, control and protection equipment, and auxiliary equipment of hydropower stations; The diagnostic fusion module is used to execute multi-dimensional diagnostic methods in parallel based on the device monitoring grouping results, fuse the diagnostic results, and output a comprehensive diagnostic conclusion and confidence score. The effectiveness evaluation module is used to calculate diagnostic performance indicators and monitor system state changes, and generate diagnostic effectiveness evaluation results and system state change detection results. The adaptive optimization module is used to adaptively adjust and improve the interior point method parameters and equipment partitioning scheme based on the diagnostic effect evaluation results and system state change detection results.

[0133] It should be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0134] The above provides a detailed description of a diagnostic method and apparatus for a hydropower centralized control and electrical dispatch monitoring system provided by embodiments of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and core ideas of the present invention. It should be noted that those skilled in the art can make various improvements and modifications to the present invention without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

Claims

1. A diagnostic method for a hydropower centralized control and dispatch monitoring system, characterized in that, Includes the following steps: S1: Acquire monitoring data from the hydropower centralized control and dispatch monitoring system, convert the monitoring data into a weighted directed graph structure, and then extract fault propagation path data based on the improved interior point method; the monitoring data includes analog quantities, digital quantities, and SOE events; The fault propagation path data includes the fault source node, propagation path, and affected terminal nodes. S2: Construct a Submodular objective function and design Matroid constraints using fault propagation path data, and use an approximate Submod Matroid constraint partitioning method to generate optimized device monitoring grouping results; S3: Based on the equipment monitoring grouping results, expert rule diagnosis, equipment mechanism model analysis, signal feature analysis and event sequence analysis are performed in parallel. The above analysis results are constructed into a multi-objective optimization problem, and then the improved interior point method based on the logarithmic barrier function is used for fusion to output a comprehensive diagnostic conclusion and confidence score. S4: Based on the comprehensive diagnostic conclusions and confidence scores, calculate the diagnostic accuracy, recall, and timeliness, and monitor changes in system topology, equipment parameters, and operating modes to generate diagnostic effectiveness evaluation results and system status change detection results; S5: Based on the diagnostic effect evaluation results and system state change detection results, adaptively adjust and improve the internal point method parameters and equipment partitioning scheme.

2. The diagnostic method for a hydropower centralized control and dispatch monitoring system according to claim 1, characterized in that, In step S1, converting the monitoring data into a weighted directed graph structure includes the following sub-steps: 1) Extract the device nodes corresponding to physical devices, the measurement point nodes corresponding to sensors, and the logic unit nodes corresponding to control and protection function units based on monitoring data, and assign a unique identifier and type attribute to each node to construct a node set; 2) Based on the node set, identify the physical connection relationships, functional association relationships, and fault propagation relationships between nodes, and establish an edge set; 3) Based on the edge set, calculate the physical connection strength, logical association strength and historical correlation of each edge to generate an edge weight matrix and obtain a weighted directed graph structure.

3. The diagnostic method for a hydropower centralized control and dispatch monitoring system according to claim 1, characterized in that, In step S1, the extraction of fault propagation path data based on the improved interior point method includes the following sub-steps: 1) Based on the weighted directed graph structure, construct the linear programming form of the standard interior point method, introduce slack variables and logarithmic barrier functions, and establish an improved mathematical model of the interior point method; 2) Detect local changes in the network topology or edge weights of a weighted directed graph structure, determine the affected region, and construct an incremental computation subproblem; 3) Based on the incremental computation subproblem, an adaptive step size selection and preprocessing conjugate gradient algorithm is used to solve the problem and output the fault propagation path data.

4. The diagnostic method for a hydropower centralized control and dispatch monitoring system according to claim 1, characterized in that, In step S2, constructing the submodular objective function using fault propagation path data includes the following sub-steps: 1) Based on the fault propagation path data, extract the equipment type matching degree, functional parameter similarity, and control logic correlation degree, and construct a functional correlation function; 2) Based on the functional correlation function, calculate the degree of concentration of physical locations of equipment and construct a location proximity function; 3) Based on the location proximity function and the fault propagation path data, evaluate the effect of grouping on fault isolation, construct the fault propagation correlation function, and form a submodular objective function through weighted combination.

5. The diagnostic method for a hydropower centralized control and dispatch monitoring system according to claim 1, characterized in that, In step S2, the design of the Matroid constraints includes: 1) Based on the Submodular objective function, set the range of the number of devices in each group and construct a balanced partitioning Matroid constraint; 2) Based on the balanced partitioning Matroid constraint and the redundancy backup requirements of critical functional devices, construct a redundancy protection Matroid constraint; 3) Based on the redundancy guarantee Matroid constraint, and based on the plant-unit-equipment hierarchical structure of the hydropower system, a diagnostic efficiency Matroid constraint is constructed to obtain the Matroid constraint conditions.

6. The diagnostic method for a hydropower centralized control and dispatch monitoring system according to claim 1, characterized in that, Step S3, which involves calculating diagnostic accuracy, recall, and timeliness, and monitoring changes in system topology, device parameters, and operating modes, includes the following sub-steps: 1) Based on the device monitoring grouping results, identify potentially abnormal device groups and generate a list of abnormal device groups; 2) Based on the list of abnormal device groups, create diagnostic tasks and allocate computing resources to generate a list of parallel diagnostic tasks; 3) Based on the parallel diagnostic task list, a subset of expert rules is introduced and logical reasoning is performed through the forward inference engine to output the rule diagnostic results; 4) Based on the rule-based diagnostic results, establish a device mechanism model and calculate the theoretical values ​​under the current operating conditions, and output the deviation analysis results; 5) Based on the deviation analysis results, perform time-domain and frequency-domain analysis on the high-frequency acquired data, and output the signal characteristic analysis results; 6) Based on the signal feature analysis results, use SOE records to perform event time window clustering, output event sequence analysis results, and obtain the results of the multi-dimensional diagnostic method.

7. The diagnostic method for a hydropower centralized control and dispatch monitoring system according to claim 1, characterized in that, In step S3, the above analysis results are constructed into a multi-objective optimization problem, and then fused using an improved interior-point method based on a logarithmic barrier function, including the following sub-steps: 1) Collect the results of multi-dimensional diagnostic methods, map the heterogeneous diagnostic results to a unified diagnostic hypothesis space, and construct a set of diagnostic hypotheses; 2) Based on the set of diagnostic hypotheses, comprehensively evaluate the credibility of each diagnostic hypothesis and establish a multi-objective optimization problem; 3) Based on the multi-objective optimization problem, a logarithmic barrier function is introduced to construct a fusion model based on the improved interior point method; 4) Based on the improved interior point method, the fusion model is used to set the constraint relationships between assumptions and obtain the constrained optimization model; 5) Based on the constrained optimization model, the confidence level of each diagnostic hypothesis is adjusted by an iterative solution method, and the main fault hypothesis, supporting evidence list and uncertainty factors are output to obtain the comprehensive diagnostic conclusion and confidence score.

8. The diagnostic method for a hydropower centralized control and dispatch monitoring system according to claim 1, characterized in that, In step S4, the diagnostic accuracy, recall, and timeliness are calculated based on the comprehensive diagnostic conclusion and confidence score, including the following sub-steps: 1) Collect comparative data of comprehensive diagnostic conclusions and confidence scores with actual fault conditions, calculate diagnostic accuracy, recall, precision and F1 score, and generate a set of performance indicators; 2) Based on the set of performance indicators, decompose and analyze them according to fault type, equipment type and fault severity to construct a classification performance analysis result; 3) Based on the classification performance analysis results, identify misdiagnosed and missed cases and establish a knowledge base of failure cases; 4) Based on the failure case knowledge base, classify the causes of failure and analyze the time interval of fault handling to generate timeliness analysis results; 5) Based on the timeliness analysis results, key indicators are displayed in a visual format, and the diagnostic effect evaluation results are output.

9. The diagnostic method for a hydropower centralized control and dispatch monitoring system according to claim 1, characterized in that, Step S4, the changes in the monitoring system topology, equipment parameters, and operating mode, includes the following sub-steps: 1) Monitor the topology, equipment parameters, and operating modes of the hydropower system, compare the current state with the baseline state, and identify the types of changes; 2) Based on the type of change, the topology, equipment parameters, and operating modes of the changes are analyzed to identify three key types of changes; 3) Based on the three types of key changes, assess the criticality and connectivity of the changed elements to determine the importance of the topological changes; 4) Based on the importance of topology changes, set multiple levels of change thresholds to obtain the degree of impact of equipment parameter changes; 5) Based on the degree of impact of changes in equipment parameters, identify the current operating mode and detect mode transitions, generating a record of operating mode changes; 6) Based on the record of changes in operating mode, a multi-confirmation mechanism is used to perform cross-validation and evaluate the cumulative effect, and the system state change detection results are output.

10. A diagnostic system for a hydropower centralized control and dispatch monitoring system as described in claims 1-9, characterized in that, include: The data acquisition module is used to acquire monitoring data from the hydropower centralized control and dispatch monitoring system and convert the monitoring data into a weighted directed graph structure. The path analysis module is used to extract fault propagation path data based on the improved interior point method. The equipment grouping module is used to generate optimized equipment monitoring grouping results based on fault propagation path data and an approximate Submod Matroid constrained partitioning method; the equipment includes power generation equipment, control and protection equipment, and auxiliary equipment of hydropower stations; The diagnostic fusion module is used to execute multi-dimensional diagnostic methods in parallel based on the device monitoring grouping results, fuse the diagnostic results, and output a comprehensive diagnostic conclusion and confidence score. The effectiveness evaluation module is used to calculate diagnostic performance indicators and monitor system state changes, and generate diagnostic effectiveness evaluation results and system state change detection results. An adaptive optimization module is used to adaptively adjust and improve the interior point method parameters and equipment partitioning scheme based on the diagnostic effect evaluation results and system state change detection results.

Images