Inverter power supply protection strategy adaptive adjustment method based on knowledge graph

Abstract

The invention discloses an inverter power supply protection strategy adaptive adjustment method based on a knowledge graph, and relates to the technical field of intelligent power systems, and the method comprises the following steps: S1, collecting operation state data, and carrying out the preprocessing; s2, constructing a knowledge graph of the inverter power supply system; s3, on the basis of a DeepGraphONet model, generating a global feature vector of the inverter power supply system; s4, constructing a fault mode rule base, and generating a fault mode recognition result report; s5, on the basis of an improved AdaptiveGA algorithm, an inverter power supply protection strategy is updated; and S6, generating an inverter power supply protection strategy operation effect report, and performing on-site inspection and supplementary processing by operation and maintenance personnel. According to the method, the limitation that a traditional inverter power supply protection strategy depends on a fixed threshold value and is lagged in response, high in false alarm rate and the like is overcome, and an efficient and accurate solution is provided for intelligent protection and real-time maintenance decision of a power system.

Description

Technical Field

[0001] This invention relates to the field of intelligent power system technology, and in particular to an adaptive adjustment method for inverter power supply protection strategy based on knowledge graphs. Background Technology

[0002] With the continuous development of smart power systems, traditional inverter power supply protection strategies are facing increasingly severe challenges. Most traditional protection strategies are based on preset fixed thresholds and rules, relying primarily on manual judgment and periodic inspections. While these methods can ensure the safe operation of equipment to a certain extent, they often struggle to cope with the complex and ever-changing working environment of power systems and the highly coupled relationships between equipment. Therefore, traditional methods suffer from problems such as delayed response and poor accuracy when dealing with equipment faults.

[0003] The limitations of traditional inverter power supply protection strategies are mainly reflected in the following aspects: First, they lack real-time monitoring and intelligent analysis of equipment status, making it difficult to adapt to the needs of multi-device collaborative operation in power systems; second, protection strategies based on fixed rules cannot fully consider the dynamic changes of equipment under complex operating conditions, resulting in delayed protection response; third, traditional methods are difficult to effectively process and analyze equipment status data with multi-dimensional, multi-source, nonlinear, and highly time-varying characteristics, and cannot accurately judge the potential fault risks of equipment; fourth, they lack analysis of the spatial correlation and functional dependence between equipment, resulting in the inability to fully explore the deep-level correlation between equipment, increasing the risk of false alarms and missed faults.

[0004] Especially in modern power systems, the optimization of condition monitoring and protection strategies for inverters, as one of the core power devices, has become particularly important. Traditional protection strategies based on single devices and simple threshold judgment methods often fail to capture the complex dynamic changes in inverter operation in real time, and cannot perform in-depth analysis and accurate prediction of its multi-dimensional and multi-source data. In addition, traditional methods usually ignore the topological structure and functional dependencies between devices, making it difficult to utilize the spatial correlation between devices, thus failing to effectively improve the overall safety and reliability of the power system.

[0005] Currently, although some methods have incorporated machine learning and deep learning technologies, these methods still fail to fully consider the deep relationship between equipment topology information and equipment status data, making it difficult to achieve accurate and real-time adjustment of inverter power supply protection strategies. Therefore, traditional protection strategies still suffer from insufficient prediction accuracy, delayed response time, and high false alarm rates, seriously affecting the operational safety and stability of power systems.

[0006] Therefore, how to provide an adaptive adjustment method for inverter power supply protection strategies based on knowledge graphs is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] This invention proposes an adaptive adjustment method for inverter power supply protection strategies based on knowledge graphs. By combining an improved DeepGraphONet model with the AdaptiveGA algorithm, the protection strategy of the inverter power supply can be optimized and adjusted more accurately. This method extracts real-time operating status data features of each device in the inverter power supply system using the DeepGraphONet model, and combines this with topological information between devices to improve the comprehensive analysis capability of device status features and topological features. The DeepGraphONet model, through graph convolution operations, fully explores the complex relationships and temporal features of device nodes, achieving accurate extraction of global features of the inverter power supply system and ensuring improved real-time performance and accuracy. The improved AdaptiveGA algorithm uses a Latin hypercube sampling mechanism to initialize the population, ensuring population diversity and avoiding local convergence problems that may occur in traditional algorithms. By introducing a roulette wheel selection mechanism and an elite retention strategy, the protection strategy is adaptively adjusted, accurately optimizing strategy parameters and effectively avoiding overfitting and local optima. The AdaptiveGA algorithm can capture minute changes in equipment operation based on real-time monitoring of equipment status, and automatically adjust protection thresholds and actions according to the operating status, thereby improving the protection accuracy and response speed of the inverter power supply system. This invention overcomes the limitations of traditional inverter power supply protection strategies, such as reliance on fixed thresholds, response lag, and high false alarm rates. It can dynamically optimize protection strategies based on real-time data of the inverter power supply system and the topology information between devices, significantly improving the safety, reliability, and real-time performance of the inverter power supply system, and providing an efficient and accurate solution for intelligent protection and real-time maintenance decision-making in power systems.

[0008] The adaptive adjustment method for inverter power supply protection strategy based on knowledge graph according to an embodiment of the present invention includes the following steps:

[0009] S1. Collect the operating status data of each device in the inverter power system, perform preprocessing operations on the operating status data of each device, and generate the operating status feature vector of each device.

[0010] S2. Construct a knowledge graph of the inverter power supply system based on the operating status feature vectors of each device;

[0011] S3. Input the knowledge graph of the inverter power system into the DeepGraphONet model, extract the node embedding vector of each device, and generate the global feature vector of the inverter power system.

[0012] S4. Construct a fault mode rule base, analyze the equipment operating status based on the global feature vector of the inverter power system, generate a fault mode identification result report, issue an early warning signal, and notify the operation and maintenance personnel to check and maintain.

[0013] S5. Input the global feature vector of the inverter power supply system into the improved AdaptiveGA algorithm, output the optimal parameter combination of the inverter power supply protection strategy, and update the inverter power supply protection strategy.

[0014] S6. Deploy the inverter power supply protection strategy to the actual operating environment, and generate an inverter power supply protection strategy operation effect report by comparing real-time monitoring with historical data. The operation and maintenance personnel conduct on-site inspections and supplementary processing based on the fault mode identification result report.

[0015] Optionally, the operating status data specifically includes voltage, current, frequency, temperature, and load; the preprocessing includes removing missing values, detecting and processing outliers, data smoothing, standardization and normalization, and feature extraction.

[0016] Optionally, S2 specifically includes:

[0017] S21. Set the connection relationships between the various devices in the inverter power supply system, including the connection relationships between devices, transmission paths, and communication relationships;

[0018] S22. Based on the operating state feature vectors and connection relationships of each device in the inverter power system, with each device as a node, the operating state feature vector of the device as the node feature, and the connection relationship of the device as the edge, construct the topology graph of the inverter power system as the knowledge graph of the inverter power system.

[0019] S23. Continuously monitor the inverter power supply system, collect real-time operating status data of each device, and update the knowledge graph of the inverter power supply system.

[0020] Optionally, S3 specifically includes:

[0021] S31. Input the knowledge graph of the inverter power system into the DeepGraphONet model and traverse each node;

[0022] S32. Based on the current node, calculate the state feature vectors of all neighboring nodes. Multiply the device state feature vector of the current node by the modulus of the device state feature vectors of the neighboring nodes to obtain the modulus product. Perform a dot product between the device state feature vector of the current node and the device state feature vectors of the neighboring nodes, and divide by the modulus product to obtain the cosine similarity between the device state feature vector of the current node and the device state feature vectors of the neighboring nodes. Iterate through each neighboring node to obtain the cosine similarity between the device state feature vector of the current node and the device state feature vector of each neighboring node.

[0023] S33. Based on the cosine similarity between the device state feature vector of the current node and the device state feature vector of the neighboring nodes, the attention weights of the neighboring edges of the current node are obtained by processing the ELU activation function and Softmax normalization; each neighboring node is traversed one by one to obtain the attention weights of each neighboring edge.

[0024] S34. Based on the device state feature vector of each neighbor node and the attention weight of each corresponding neighbor edge, the node embedding vector of the current node is obtained by weighted averaging.

[0025] S35. Based on the node embedding vector of the current node, update the node embedding vector through linear transformation of the fully connected layer and ELU activation function;

[0026] S36. After traversing each node of the knowledge graph of the inverter power supply system, the node embedding vector of each device is finally obtained.

[0027] S37. Concatenate the node embedding vectors of each device to generate the global feature vector of the inverter power system.

[0028] Optionally, S4 specifically includes:

[0029] S41. Define fault modes, including overload, overvoltage, undervoltage, short circuit, overtemperature, equipment failure, frequency deviation, electromagnetic interference, communication failure, inverter damage, and abnormal power factor.

[0030] S42. Define fault warning levels, including alert level, warning level, and emergency level;

[0031] S43. Based on historical experience and expert advice, construct a fault mode rule base, including fault type matching rules, fault warning levels, and fault warning signals for each fault mode.

[0032] S44. Based on the global feature vector of the inverter power supply system, analyze the operating status data of each device through the fault mode rule base, match the corresponding fault mode type and fault warning level, generate a fault mode identification result report, and send a fault warning signal through the control unit to notify the operation and maintenance personnel to check and maintain.

[0033] Optionally, the improved AdaptiveGA algorithm specifically includes:

[0034] The parameters for setting the inverter power supply protection strategy include protection action type, maximum output power overload protection threshold, upper limit overvoltage protection threshold for power supply output voltage, lower limit undervoltage protection threshold for power supply output voltage, maximum temperature limit overtemperature protection threshold for power supply, maximum temperature limit overtemperature protection threshold for equipment, upper limit short circuit protection threshold for current, and upper limit short circuit protection threshold for voltage. The protection action type includes disconnecting the power supply, switching the load, adjusting the output power, limiting the voltage output, short circuit isolation, restarting the power supply, activating the cooling system, and activating the backup power supply.

[0035] The reciprocal of the response time is set as the response performance index; the evaluation indexes of the inverter power supply protection strategy are set, including response performance index, protection accuracy, and energy utilization rate; the response performance weight, accuracy weight, and utilization rate weight are set, and the response performance index, protection accuracy, and energy utilization rate are multiplied by weight and summed to obtain the fitness function.

[0036] Based on the Latin hypercube sampling mechanism, each parameter of the inverter power supply protection strategy represents a gene, and each combination of parameters represents an individual. The population capacity is set, and a multidimensional gene space with corresponding dimensions is set according to the number of parameters in the inverter power supply protection strategy. Each gene dimension is divided into several equally spaced gene value intervals. Before each individual is generated, each gene dimension randomly selects a value from each of its own gene value intervals, and the results are aggregated to obtain the candidate gene set for this round. The candidate gene sets for each gene dimension are randomly combined without replacement to form a new individual, which enters the population. The Latin hypercube sampling operation is repeated until the population is full.

[0037] For each individual, the global feature vector of the inverter power supply system is input into the inverter power supply simulation system for simulation testing. Based on the test results, the response performance index, protection accuracy, and energy utilization rate are statistically analyzed. Based on the fitness function, the fitness value of each individual is calculated.

[0038] Based on the roulette wheel selection mechanism, the fitness values ​​of each individual are summed to obtain the population fitness; the fitness of each individual is divided by the population fitness to obtain the selection probability of each individual; each individual is iterated sequentially, and the selection probability of the current individual is added to the selection probabilities of all individuals iterated before to obtain the cumulative probability of the current individual; a random number is generated in the interval [0,1] as a pointer; the breeding pool capacity is set, and the corresponding individual is selected to enter the breeding pool according to the cumulative probability range that the pointer falls into; the roulette wheel selection is repeated until the breeding pool is full.

[0039] Set a crossover ratio, and randomly select several individuals from the breeding pool according to the crossover ratio, with each pair of individuals serving as a parent pair; set up a new population, and according to the crossover mechanism, use each parent pair to generate one offspring individual to enter the new population; the gene value of each gene position of the offspring individual is obtained by randomly selecting one of the two parents; repeat the crossover operation until each parent pair completes the crossover operation to obtain a complete new population;

[0040] Set a mutation ratio, randomly select several individuals from the next generation population according to the mutation ratio, perform mutation operation, and update the new population; the mutation operation refers to randomly selecting a gene position based on the current individual, randomly modifying and replacing the gene value.

[0041] An elite strategy is introduced: an individual is randomly selected from the new population as an elite individual, and the elite individual is retained and directly enters the new population for the next iteration; the new population generated in this iteration is used to replace the original population.

[0042] Based on the adaptive mechanism of crossover ratio and mutation rate, after each iteration, the fitness of the new population is calculated; a population fitness threshold is set. If the fitness of the new population is greater than the population fitness threshold, the population fitness of the new population is subtracted from the population fitness threshold and then divided by the population fitness threshold to obtain the relative increase of the new population; 1 is used to subtract the relative increase of the new population to obtain the crossover ratio correction factor, which is multiplied by the current crossover ratio to obtain the updated crossover ratio; 1 is used to add the relative increase of the new population to obtain the mutation rate correction factor, which is multiplied by the current mutation rate to obtain the updated mutation rate.

[0043] Set a stable period threshold, a fitness threshold, and a fitness change amplitude threshold; the convergence condition of the fitness function is that after a continuous iteration period of the stable period threshold length, the fitness value of the current elite individual is always less than the fitness threshold, and the fitness change amplitude of the current elite individual is always less than the fitness change amplitude threshold.

[0044] The training continues iteratively until the fitness function meets the convergence condition. The current elite individual is then output as the optimal parameter combination for the inverter power supply protection strategy.

[0045] Optionally, S6 specifically includes:

[0046] S61. Update and deploy the inverter power supply protection strategy to the actual working environment, monitor the operating status data in real time, and obtain the operating status data after deployment.

[0047] S62. Compare historical operating status data with post-deployment operating status data, and calculate the corresponding fitness values ​​according to the fitness function to generate an inverter power supply protection strategy operation effect report.

[0048] S63. After the maintenance personnel arrive at the warning site, they shall check the operation effect of the inverter power supply protection strategy, and supplement the handling measures and record the handling strategy in combination with the fault mode identification result report and the actual operation on site.

[0049] The beneficial effects of this invention are:

[0050] (1) This invention significantly improves the accuracy and real-time performance of the adaptive adjustment method for inverter power supply protection strategies by introducing an improved DeepGraphONet model. While traditional graph neural network methods can capture the relationships between device nodes when processing the topology and time-series data of inverter power supply systems, they may still encounter problems such as incomplete information fusion or insufficient feature extraction in scenarios involving complex device state changes and multi-device collaborative management. To address these issues, this invention innovatively introduces the DeepGraphONet model, which adaptively extracts device state features through graph convolution operations and further optimizes the feature fusion effect by combining topological information between devices. This method, by adaptively aggregating the state features of device nodes and the influence of neighboring nodes, can more accurately capture changes in device operating states and fault propagation trends, significantly improving the accuracy and real-time performance of inverter power supply protection strategy adjustment.

[0051] (2) This invention optimizes the parameter adjustment process of the inverter power supply protection strategy by adopting an improved AdaptiveGA algorithm, thereby improving the algorithm's adaptability and search efficiency. Although traditional genetic algorithms have strong global search capabilities, they are often plagued by local optima during parameter optimization and struggle to balance search diversity and convergence speed. To address these issues, this invention innovatively introduces a Latin hypercube sampling mechanism, a roulette wheel selection mechanism, and an elite retention strategy, enabling the population to be evenly distributed in the multidimensional parameter space and maintain efficient diversity, while avoiding early convergence and overfitting problems. Furthermore, the algorithm employs adaptive crossover and mutation strategies, allowing the optimization process to be dynamically adjusted based on the current population performance in each iteration, thereby accelerating the search for the optimal solution and significantly improving the optimization efficiency and global optimality of the inverter power supply protection strategy. Attached Figure Description

[0052] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0053] Figure 1 This is an overall flowchart of the adaptive adjustment method for inverter power supply protection strategy based on knowledge graph proposed in this invention.

[0054] Figure 2This is a flowchart of the improved AdaptiveGA algorithm for the adaptive adjustment method of inverter power supply protection strategy based on knowledge graph proposed in this invention. Detailed Implementation

[0055] The invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0056] refer to Figure 1 and Figure 2 An adaptive adjustment method for inverter power supply protection strategy based on knowledge graphs includes the following steps:

[0057] S1. Collect the operating status data of each device in the inverter power system, perform preprocessing operations on the operating status data of each device, and generate the operating status feature vector of each device.

[0058] S2. Construct a knowledge graph of the inverter power supply system based on the operating status feature vectors of each device;

[0059] S3. Input the knowledge graph of the inverter power system into the DeepGraphONet model, extract the node embedding vector of each device, and generate the global feature vector of the inverter power system.

[0060] S4. Construct a fault mode rule base, analyze the equipment operating status based on the global feature vector of the inverter power system, generate a fault mode identification result report, issue an early warning signal, and notify the operation and maintenance personnel to check and maintain.

[0061] S5. Input the global feature vector of the inverter power supply system into the improved AdaptiveGA algorithm, output the optimal parameter combination of the inverter power supply protection strategy, and update the inverter power supply protection strategy.

[0062] S6. Deploy the inverter power supply protection strategy to the actual operating environment, and generate an inverter power supply protection strategy operation effect report by comparing real-time monitoring with historical data. The operation and maintenance personnel conduct on-site inspections and supplementary processing based on the fault mode identification result report.

[0063] In this embodiment, the operating status data specifically includes voltage, current, frequency, temperature, and load; the preprocessing includes removing missing values, detecting and processing outliers, data smoothing, standardization and normalization, and feature extraction.

[0064] This implementation method comprehensively preprocesses the operating status data of the inverter power supply system, ensuring high data quality and reliability, and improving the accuracy and stability of subsequent model training. Operating status data includes voltage, current, frequency, temperature, and load, serving as key operating parameters of the equipment. Removal of missing values ​​ensures data integrity and avoids training bias caused by missing data. Outlier detection and handling identify and correct abnormal data using statistical methods, ensuring data accuracy and avoiding the negative impact of erroneous data on model performance. Data smoothing removes short-term fluctuations, reduces noise interference, and makes the data exhibit a more stable trend, providing clear signals for subsequent analysis. Standardization and normalization ensure a uniform data scale, eliminating dimensional differences between different feature dimensions and avoiding bias during model training. Feature extraction methods extract key information from the data, enhancing the model's ability to perceive the characteristics of equipment operating status and improving the model's prediction accuracy and generalization ability.

[0065] In this embodiment, S2 specifically includes:

[0066] S21. Set the connection relationships between the various devices in the inverter power supply system, including the connection relationships between devices, transmission paths, and communication relationships;

[0067] S22. Based on the operating state feature vectors and connection relationships of each device in the inverter power system, with each device as a node, the operating state feature vector of the device as the node feature, and the connection relationship of the device as the edge, construct the topology graph of the inverter power system as the knowledge graph of the inverter power system.

[0068] S23. Continuously monitor the inverter power supply system, collect real-time operating status data of each device, and update the knowledge graph of the inverter power supply system.

[0069] This implementation effectively improves the real-time monitoring and fault prediction capabilities of the inverter power system by constructing and continuously updating a knowledge graph of the inverter power system. It defines the connection relationships between the various devices in the inverter power system, including connection relationships, transmission paths, and communication relationships, ensuring that the collaborative information between devices can be accurately modeled, providing rich topological structure information for subsequent analysis. Based on the operating status feature vectors and connection relationships of each device in the inverter power system, using each device as a node, the device's operating status feature vector as node features, and the device's connection relationships as edges, a topology graph of the inverter power system is constructed, serving as the knowledge graph of the inverter power system. This comprehensively represents the interdependencies between devices and their operating states, promoting efficient system modeling and optimization. Continuous monitoring of the inverter power system, real-time collection of operating status data of each device, and updating the knowledge graph of the inverter power system ensure the system's dynamic response capability, guaranteeing timely updates of device status data and topology information, improving the system's real-time performance and accuracy, and providing accurate basis for subsequent decision-making and fault early warning.

[0070] In this embodiment, S3 specifically includes:

[0071] S31. Input the knowledge graph of the inverter power system into the DeepGraphONet model and traverse each node;

[0072] S32. Based on the current node, calculate the state feature vectors of all neighboring nodes. Multiply the device state feature vector of the current node by the modulus of the device state feature vectors of the neighboring nodes to obtain the modulus product. Perform a dot product between the device state feature vector of the current node and the device state feature vectors of the neighboring nodes, and divide by the modulus product to obtain the cosine similarity between the device state feature vector of the current node and the device state feature vectors of the neighboring nodes. Iterate through each neighboring node to obtain the cosine similarity between the device state feature vector of the current node and the device state feature vector of each neighboring node.

[0073] S33. Based on the cosine similarity between the device state feature vector of the current node and the device state feature vector of the neighboring nodes, the attention weights of the neighboring edges of the current node are obtained by processing the ELU activation function and Softmax normalization; each neighboring node is traversed one by one to obtain the attention weights of each neighboring edge.

[0074] S34. Based on the device state feature vector of each neighbor node and the attention weight of each corresponding neighbor edge, the node embedding vector of the current node is obtained by weighted averaging.

[0075] S35. Based on the node embedding vector of the current node, update the node embedding vector through linear transformation of the fully connected layer and ELU activation function;

[0076] S36. After traversing each node of the knowledge graph of the inverter power supply system, the node embedding vector of each device is finally obtained.

[0077] S37. Concatenate the node embedding vectors of each device to generate the global feature vector of the inverter power system.

[0078] This implementation effectively enhances the global analysis capability of device status data by inputting the knowledge graph of the inverter power system into the DeepGraphONet model and using graph neural networks for node feature extraction. The knowledge graph of the inverter power system is input into the DeepGraphONet model, traversing each node to ensure that the status features of each device are effectively captured by the model and participate in subsequent calculations. Based on the current node, the state feature vectors of all neighboring nodes are statistically analyzed. The device status feature vector of the current node is multiplied by the modulus of the device status feature vectors of the neighboring nodes to obtain the modulus product. The dot product of the device status feature vectors of the current node and the device status feature vectors of the neighboring nodes is then calculated and divided by the modulus product to obtain the cosine similarity between the device status feature vectors of the current node and the device status feature vectors of the neighboring nodes. This ensures that the similarity and correlation between devices can be measured, providing a valid basis for subsequent feature fusion. Each neighboring node is traversed one by one to obtain the cosine similarity between the device status feature vectors of the current node and each neighboring node, ensuring that the influence of all neighboring nodes on the features of the current node is considered. Based on the cosine similarity between the device state feature vector of the current node and the device state feature vectors of its neighboring nodes, the attention weights of the neighboring edges of the current node are obtained through ELU activation and Softmax normalization. This further enhances the weight allocation of similarity between nodes, enabling the network to more accurately focus on highly correlated neighboring nodes. Each neighboring node is traversed one by one to obtain the attention weight of each neighboring edge, ensuring that the network handles the relationships between different nodes differently. Based on the device state feature vector of each neighboring node and the corresponding attention weight of each neighboring edge, a weighted average is used to obtain the node embedding vector of the current node. This fully integrates the state information between devices and the influence of neighboring nodes, improving the expressive power of node features. Based on the node embedding vector of the current node, the node embedding vector is updated through linear transformation of the fully connected layer and ELU activation, further optimizing the expression of node features and enhancing the model's understanding of device states. After traversing each node of the knowledge graph of the inverter power supply system, the node embedding vector of each device is finally obtained, ensuring that the state features of all devices in the system can be accurately extracted. The node embedding vectors of each device are concatenated to generate the global feature vector of the inverter power supply system, which integrates the key information of all devices in the system and provides comprehensive feature support for subsequent protection strategy adjustment.

[0079] In this embodiment, S4 specifically includes:

[0080] S41. Define fault modes, including overload, overvoltage, undervoltage, short circuit, overtemperature, equipment failure, frequency deviation, electromagnetic interference, communication failure, inverter damage, and abnormal power factor.

[0081] S42. Define fault warning levels, including alert level, warning level, and emergency level;

[0082] S43. Based on historical experience and expert advice, construct a fault mode rule base, including fault type matching rules, fault warning levels, and fault warning signals for each fault mode.

[0083] S44. Based on the global feature vector of the inverter power supply system, analyze the operating status data of each device through the fault mode rule base, match the corresponding fault mode type and fault warning level, generate a fault mode identification result report, and send a fault warning signal through the control unit to notify the operation and maintenance personnel to check and maintain.

[0084] This implementation method, by constructing a fault mode rule base and analyzing the operating status of the inverter power supply system in real time, can accurately identify equipment faults and provide effective fault warnings and maintenance decision support for operation and maintenance personnel. Fault mode definitions include overload, overvoltage, undervoltage, short circuit, overtemperature, equipment failure, frequency deviation, electromagnetic interference, communication failure, inverter damage, and abnormal power factor, ensuring coverage of common fault types encountered during equipment operation. Fault warning levels are defined as alert, warning, and emergency levels, providing timely response solutions for faults of varying severity. Based on historical experience and expert advice, a fault mode rule base is constructed, including fault type matching rules, fault warning levels, and fault warning signals for each fault mode, thus providing detailed judgment criteria for equipment status analysis. Based on the global feature vector of the inverter power supply system, the operating status data of each device is analyzed through the fault mode rule base to match the corresponding fault mode type and fault warning level, generating a fault mode identification result report to ensure that the fault risk of each device can be accurately assessed. Fault warning signals are issued by the control unit to notify operation and maintenance personnel for inspection and maintenance, improving fault emergency response speed and maintenance efficiency, and effectively ensuring the safe operation of the inverter power supply system.

[0085] In this embodiment, the improved AdaptiveGA algorithm specifically includes:

[0086] The parameters for setting the inverter power supply protection strategy include protection action type, maximum output power overload protection threshold, upper limit overvoltage protection threshold for power supply output voltage, lower limit undervoltage protection threshold for power supply output voltage, maximum temperature limit overtemperature protection threshold for power supply, maximum temperature limit overtemperature protection threshold for equipment, upper limit short circuit protection threshold for current, and upper limit short circuit protection threshold for voltage. The protection action type includes disconnecting the power supply, switching the load, adjusting the output power, limiting the voltage output, short circuit isolation, restarting the power supply, activating the cooling system, and activating the backup power supply.

[0087] The reciprocal of the response time is set as the response performance index; the evaluation indexes of the inverter power supply protection strategy are set, including response performance index, protection accuracy, and energy utilization rate; the response performance weight, accuracy weight, and utilization rate weight are set, and the response performance index, protection accuracy, and energy utilization rate are multiplied by weight and summed to obtain the fitness function.

[0088] Based on the Latin hypercube sampling mechanism, each parameter of the inverter power supply protection strategy represents a gene, and each combination of parameters represents an individual. The population capacity is set, and a multidimensional gene space with corresponding dimensions is set according to the number of parameters in the inverter power supply protection strategy. Each gene dimension is divided into several equally spaced gene value intervals. Before each individual is generated, each gene dimension randomly selects a value from each of its own gene value intervals, and the results are aggregated to obtain the candidate gene set for this round. The candidate gene sets for each gene dimension are randomly combined without replacement to form a new individual, which enters the population. The Latin hypercube sampling operation is repeated until the population is full.

[0089] For each individual, the global feature vector of the inverter power supply system is input into the inverter power supply simulation system for simulation testing. Based on the test results, the response performance index, protection accuracy, and energy utilization rate are statistically analyzed. Based on the fitness function, the fitness value of each individual is calculated.

[0090] Based on the roulette wheel selection mechanism, the fitness values ​​of each individual are summed to obtain the population fitness; the fitness of each individual is divided by the population fitness to obtain the selection probability of each individual; each individual is iterated sequentially, and the selection probability of the current individual is added to the selection probabilities of all individuals iterated before to obtain the cumulative probability of the current individual; a random number is generated in the interval [0,1] as a pointer; the breeding pool capacity is set, and the corresponding individual is selected to enter the breeding pool according to the cumulative probability range that the pointer falls into; the roulette wheel selection is repeated until the breeding pool is full.

[0091] Set a crossover ratio, and randomly select several individuals from the breeding pool according to the crossover ratio, with each pair of individuals serving as a parent pair; set up a new population, and according to the crossover mechanism, use each parent pair to generate one offspring individual to enter the new population; the gene value of each gene position of the offspring individual is obtained by randomly selecting one of the two parents; repeat the crossover operation until each parent pair completes the crossover operation to obtain a complete new population;

[0092] Set a mutation ratio, randomly select several individuals from the next generation population according to the mutation ratio, perform mutation operation, and update the new population; the mutation operation refers to randomly selecting a gene position based on the current individual, randomly modifying and replacing the gene value.

[0093] An elite strategy is introduced: an individual is randomly selected from the new population as an elite individual, and the elite individual is retained and directly enters the new population for the next iteration; the new population generated in this iteration is used to replace the original population.

[0094] Based on the adaptive mechanism of crossover ratio and mutation rate, after each iteration, the fitness of the new population is calculated; a population fitness threshold is set. If the fitness of the new population is greater than the population fitness threshold, the population fitness of the new population is subtracted from the population fitness threshold and then divided by the population fitness threshold to obtain the relative increase of the new population; 1 is used to subtract the relative increase of the new population to obtain the crossover ratio correction factor, which is multiplied by the current crossover ratio to obtain the updated crossover ratio; 1 is used to add the relative increase of the new population to obtain the mutation rate correction factor, which is multiplied by the current mutation rate to obtain the updated mutation rate.

[0095] Set a stable period threshold, a fitness threshold, and a fitness change amplitude threshold; the convergence condition of the fitness function is that after a continuous iteration period of the stable period threshold length, the fitness value of the current elite individual is always less than the fitness threshold, and the fitness change amplitude of the current elite individual is always less than the fitness change amplitude threshold.

[0096] The training continues iteratively until the fitness function meets the convergence condition. The current elite individual is then output as the optimal parameter combination for the inverter power supply protection strategy.

[0097] This implementation method achieves precise optimization and real-time adjustment of the inverter power supply protection strategy through an improved AdaptiveGA algorithm, ensuring the system's high efficiency and stability. Based on the Latin hypercube sampling mechanism, each parameter of the inverter power supply protection strategy is set to represent a gene, and each parameter combination represents an individual. A population size is set, and a multi-dimensional gene space with corresponding dimensions is established according to the number of parameters in the inverter power supply protection strategy, ensuring population diversity and avoiding local convergence problems. Each gene dimension is divided into several equally spaced gene value intervals. Before each individual generation, each gene dimension randomly selects a value from each of its own gene value intervals, summing them to obtain the candidate gene set for this round, ensuring the comprehensiveness of individuals. For each individual, the global feature vector of the inverter power supply system is input into the inverter power supply simulation system for simulation testing. Based on the test results, response performance indicators, protection accuracy, and energy utilization are statistically analyzed. The fitness value of each individual is calculated according to the fitness function, providing a precise basis for individual selection and optimization. Based on the roulette wheel selection mechanism, the fitness values ​​of each individual are summed to obtain the population fitness, ensuring the fairness of population selection. The selection probability of each individual is obtained by dividing the fitness of an individual by the fitness of the population. A random pointer generation mechanism iterates through each individual sequentially, ensuring fair selection. A crossover ratio is set, and several individuals are randomly selected from the breeding pool according to this ratio, with each pair of individuals serving as a parent pair, ensuring the randomness and effectiveness of the crossover operation. A new population is set up, and according to the crossover mechanism, one offspring individual is generated from each parent pair to enter the new population, ensuring the diversity of the new population. A mutation ratio is set, and several individuals are randomly selected from the next generation population according to this ratio, and a mutation operation is performed to update the new population, ensuring the adaptability of the population during the evolutionary process. Based on the adaptive mechanism of crossover ratio and mutation rate, the fitness of the new population is calculated after each iteration, ensuring the adaptability and dynamic optimization of the algorithm. Continuous iterative training continues until the fitness function meets the convergence condition, and the current elite individual is output as the optimal parameter combination for the inverter power supply protection strategy, ensuring that the final output protection strategy is optimal globally.

[0098] In this embodiment, S6 specifically includes:

[0099] S61. Update and deploy the inverter power supply protection strategy to the actual working environment, monitor the operating status data in real time, and obtain the operating status data after deployment.

[0100] S62. Compare historical operating status data with post-deployment operating status data, and calculate the corresponding fitness values ​​according to the fitness function to generate an inverter power supply protection strategy operation effect report.

[0101] S63. After the maintenance personnel arrive at the warning site, they shall check the operation effect of the inverter power supply protection strategy, and supplement the handling measures and record the handling strategy in combination with the fault mode identification result report and the actual operation on site.

[0102] This implementation method ensures the optimization and effectiveness of the inverter power supply protection strategy by real-time monitoring of its operational performance and comparative analysis with historical data. The inverter power supply protection strategy is updated and deployed to the actual operating environment, with real-time monitoring of operational status data to ensure its effective operation in the real-world environment. Historical operational status data is compared with the deployed operational status data, and corresponding fitness values ​​are calculated based on the fitness function to generate an inverter power supply protection strategy performance report, thereby evaluating the strategy's actual performance. After arriving at the warning site, maintenance personnel check the operational performance of the inverter power supply protection strategy, and, based on the fault mode identification results report and the actual on-site operating conditions, supplementary measures are taken and recorded to ensure that on-site problems are resolved promptly and effectively.

[0103] Example 1:

[0104] To verify the feasibility of this invention in optimizing inverter power supply protection strategies, the method was applied to the intelligent protection and regulation system of an inverter power supply system (hereinafter referred to as "Platform X") of a power company. Traditional inverter power supply protection strategies typically rely on fixed threshold rules or strategy adjustment methods based on simple algorithms. These methods not only have poor adaptability in complex operating environments but also struggle to effectively optimize protection strategies, easily leading to false alarms, delayed responses, and low protection efficiency. To address these issues, Platform X decided to adopt the knowledge graph-based adaptive adjustment method for inverter power supply protection strategies proposed in this invention, combining the DeepGraphONet model with the improved AdaptiveGA algorithm.

[0105] During implementation, Platform X extracts real-time status data features of each device in the inverter power system using the DeepGraphONet model and combines this with topological information between devices to improve the fusion capability of device status features and topological features. This method uses graph convolution operations to deeply mine the relationships between device nodes, ensuring accurate and real-time extraction of global system features. The improved AdaptiveGA algorithm, employing a Latin hypercube sampling mechanism, roulette wheel selection mechanism, and elite retention strategy, can adaptively optimize protection strategies, avoiding overfitting and local optima problems. It captures minute changes in device operating status in real time and precisely adjusts the parameters of the protection strategy, thereby significantly improving protection efficiency and response speed.

[0106] Table 1 Comparison of Adaptive Adjustment Methods for Inverter Power Supply Protection Strategies on Platform X

[0107] index Traditional methods Method of the present invention Increase Fault prediction accuracy (%) 78.9 97.2 18.3% False alarm rate (%) 9.5 2.1 7.4% Fault false negative rate (%) 12.8 3.4 9.4% Protection policy response time (seconds) 90 25 72.2% Strategy adjustment time (minutes) 30 10 66.7% Time taken for a single prediction process (seconds) 90 28 68.9% Real-time data processing capacity (data items / second) 190 600 215.8% Average time to locate equipment faults (minutes) 50 15 70.0% Number of unplanned equipment maintenance sessions per month 22 6 72.7% Operation and maintenance costs (ten thousand yuan / month) 48 35 27.1% Improvement in power system stability (%) 83.4 97.3 13.9%

[0108] As shown in Table 1, the inverter power supply protection strategy using the method of this invention exhibits significant improvements in several key performance indicators compared to traditional methods. The fault prediction accuracy increased from 78.9% in the traditional method to 97.2%, an improvement of 18.3%, indicating a significant improvement in the accuracy of fault prediction and facilitating earlier detection of equipment faults. Simultaneously, the false alarm rate and false negative rate decreased from 9.5% and 12.8% to 2.1% and 3.4%, respectively, representing improvements of 7.4% and 9.4%, effectively reducing false alarms and false negatives and improving the reliability and practicality of prediction. Regarding the protection strategy response time, the traditional method has a response time of 90 seconds, while the method of this invention shortens it to 25 seconds, a reduction of 72.2%, ensuring rapid execution of protection actions and improving equipment safety. Furthermore, the strategy adjustment time was reduced from 30 minutes to 10 minutes, a reduction of 66.7%, demonstrating the high efficiency of strategy adjustment. For single prediction processing time, the traditional method requires 90 seconds, while the method of this invention only requires 28 seconds, a reduction of 68.9%, significantly improving the speed of prediction processing and adapting to higher-frequency data processing needs. Simultaneously, real-time data processing capability increases from 190 records / second in the traditional method to 600 records / second, an improvement of 215.8%, enhancing the system's real-time performance and big data processing capabilities. The average time for equipment fault location decreases from 50 minutes to 15 minutes, a reduction of 70.0%, making fault location more efficient and improving maintenance response speed. The number of unplanned equipment maintenance sessions decreases from 22 times per month to 6 times, a reduction of 72.7%, effectively reducing maintenance pressure and improving equipment reliability. Furthermore, maintenance costs decrease from 480,000 yuan per month to 350,000 yuan, a reduction of 27.1%, indicating that the method of this invention performs well in cost reduction. Finally, power system stability improves by 13.9%, from 83.4% in the traditional method to 97.3%, demonstrating a significant improvement in the overall stability and reliability of the system after using the method of this invention. In summary, by adopting the method of this invention, the protection strategy, fault prediction, and maintenance efficiency of the inverter power supply system have been significantly improved, providing strong technical support for the intelligent operation and maintenance and real-time decision-making of the power system.

[0109] The method of this invention successfully realizes intelligent protection and real-time maintenance decision-making for inverter power systems, improving system safety, reliability and real-time performance, reducing operation and maintenance costs, and providing strong technical support for intelligent operation and maintenance of power systems.

[0110] The above-described embodiments are merely preferred embodiments of the present invention, and the scope of protection is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art based on the technical solutions within the scope of the present invention should be included within the scope of protection of the present invention.

Claims

1. An adaptive adjustment method for inverter power supply protection strategy based on knowledge graph, characterized in that, Includes the following steps: S1. Collect the operating status data of each device in the inverter power system, perform preprocessing operations on the operating status data of each device, and generate the operating status feature vector of each device. S2. Construct a knowledge graph of the inverter power supply system based on the operating status feature vectors of each device; S3. Input the knowledge graph of the inverter power system into the DeepGraphONet model, extract the node embedding vector of each device, and generate the global feature vector of the inverter power system. S4. Construct a fault mode rule base, analyze the equipment operating status based on the global feature vector of the inverter power supply system, generate a fault mode identification result report, issue an early warning signal, and notify the operation and maintenance personnel to check and maintain the equipment. S5. Input the global feature vector of the inverter power supply system into the improved AdaptiveGA algorithm, output the optimal parameter combination of the inverter power supply protection strategy, and update the inverter power supply protection strategy. S6. Deploy the inverter power supply protection strategy to the actual operating environment, and generate an inverter power supply protection strategy operation effect report by comparing real-time monitoring with historical data. The operation and maintenance personnel conduct on-site inspections and supplementary processing based on the fault mode identification result report.

2. The adaptive adjustment method for inverter power supply protection strategy based on knowledge graph as described in claim 1, characterized in that, The operating status data specifically includes voltage, current, frequency, temperature, and load; the preprocessing includes removing missing values, detecting and processing outliers, data smoothing, standardization and normalization, and feature extraction.

3. The adaptive adjustment method for inverter power supply protection strategy based on knowledge graphs according to claim 1, characterized in that, S2 specifically includes: S21. Set the connection relationships between the various devices in the inverter power supply system, including the connection relationships between devices, transmission paths, and communication relationships; S22. Based on the operating state feature vectors and connection relationships of each device in the inverter power system, with each device as a node, the operating state feature vector of the device as the node feature, and the connection relationship of the device as the edge, construct the topology graph of the inverter power system as the knowledge graph of the inverter power system. S23. Continuously monitor the inverter power supply system, collect real-time operating status data of each device, and update the knowledge graph of the inverter power supply system.

4. The adaptive adjustment method for inverter power supply protection strategy based on knowledge graph as described in claim 1, characterized in that, S3 specifically includes: S31. Input the knowledge graph of the inverter power system into the DeepGraphONet model and traverse each node; S32. Based on the current node, calculate the state feature vectors of all neighboring nodes. Multiply the device state feature vector of the current node by the modulus of the device state feature vectors of the neighboring nodes to obtain the modulus product. Perform a dot product between the device state feature vector of the current node and the device state feature vectors of the neighboring nodes, and divide by the modulus product to obtain the cosine similarity between the device state feature vector of the current node and the device state feature vectors of the neighboring nodes. Iterate through each neighboring node to obtain the cosine similarity between the device state feature vector of the current node and the device state feature vector of each neighboring node. S33. Based on the cosine similarity between the device state feature vector of the current node and the device state feature vector of the neighboring nodes, the attention weights of the neighboring edges of the current node are obtained by processing the ELU activation function and Softmax normalization; each neighboring node is traversed one by one to obtain the attention weights of each neighboring edge. S34. Based on the device state feature vector of each neighbor node and the attention weight of each corresponding neighbor edge, the node embedding vector of the current node is obtained by weighted averaging. S35. Based on the node embedding vector of the current node, update the node embedding vector through linear transformation of the fully connected layer and ELU activation function; S36. After traversing each node of the knowledge graph of the inverter power supply system, the node embedding vector of each device is finally obtained. S37. Concatenate the node embedding vectors of each device to generate the global feature vector of the inverter power system.

5. The adaptive adjustment method for inverter power supply protection strategy based on knowledge graph as described in claim 1, characterized in that, S4 specifically includes: S41. Define fault modes, including overload, overvoltage, undervoltage, short circuit, overtemperature, equipment failure, frequency deviation, electromagnetic interference, communication failure, inverter damage, and abnormal power factor. S42. Define fault warning levels, including alert level, warning level, and emergency level; S43. Based on historical experience and expert advice, construct a fault mode rule base, including fault type matching rules, fault warning levels, and fault warning signals for each fault mode. S44. Based on the global feature vector of the inverter power supply system, analyze the operating status data of each device through the fault mode rule base, match the corresponding fault mode type and fault warning level, generate a fault mode identification result report, and send a fault warning signal through the control unit to notify the operation and maintenance personnel to check and maintain.

6. The adaptive adjustment method for inverter power supply protection strategy based on knowledge graph as described in claim 1, characterized in that, The improved AdaptiveGA algorithm includes: The parameters for setting the inverter power supply protection strategy include protection action type, maximum output power overload protection threshold, upper limit overvoltage protection threshold for power supply output voltage, lower limit undervoltage protection threshold for power supply output voltage, maximum temperature limit overtemperature protection threshold for power supply, maximum temperature limit overtemperature protection threshold for equipment, upper limit short circuit protection threshold for current, and upper limit short circuit protection threshold for voltage. The protection action type includes disconnecting the power supply, switching the load, adjusting the output power, limiting the voltage output, short circuit isolation, restarting the power supply, activating the cooling system, and activating the backup power supply. The reciprocal of the response time is set as the response performance index; the evaluation indexes of the inverter power supply protection strategy are set, including response performance index, protection accuracy, and energy utilization rate; the response performance weight, accuracy weight, and utilization rate weight are set, and the response performance index, protection accuracy, and energy utilization rate are multiplied by weight and summed to obtain the fitness function. Based on the Latin hypercube sampling mechanism, each parameter of the inverter power supply protection strategy represents a gene, and each combination of parameters represents an individual. The population capacity is set, and a multidimensional gene space with corresponding dimensions is set according to the number of parameters in the inverter power supply protection strategy. Each gene dimension is divided into several equally spaced gene value intervals. Before each individual is generated, each gene dimension randomly selects a value from each of its own gene value intervals, and the results are aggregated to obtain the candidate gene set for this round. The candidate gene sets for each gene dimension are randomly combined without replacement to form a new individual, which enters the population. The Latin hypercube sampling operation is repeated until the population is full. For each individual, the global feature vector of the inverter power supply system is input into the inverter power supply simulation system for simulation testing. Based on the test results, the response performance index, protection accuracy, and energy utilization rate are statistically analyzed. Based on the fitness function, the fitness value of each individual is calculated. Based on the roulette wheel selection mechanism, the fitness values ​​of each individual are summed to obtain the population fitness; the fitness of each individual is divided by the population fitness to obtain the selection probability of each individual; each individual is iterated sequentially, and the selection probability of the current individual is added to the selection probabilities of all individuals iterated before to obtain the cumulative probability of the current individual; a random number is generated in the interval [0,1] as a pointer; the breeding pool capacity is set, and the corresponding individual is selected to enter the breeding pool according to the cumulative probability range that the pointer falls into; the roulette wheel selection is repeated until the breeding pool is full. Set a crossover ratio, and randomly select several individuals from the breeding pool according to the crossover ratio, with each pair of individuals serving as a parent pair; set up a new population, and according to the crossover mechanism, use each parent pair to generate one offspring individual to enter the new population; the gene value of each gene position of the offspring individual is obtained by randomly selecting one of the two parents; repeat the crossover operation until each parent pair completes the crossover operation to obtain a complete new population; Set a mutation ratio, randomly select several individuals from the next generation population according to the mutation ratio, perform mutation operation, and update the new population; the mutation operation refers to randomly selecting a gene position based on the current individual, randomly modifying and replacing the gene value. An elite strategy is introduced: an individual is randomly selected from the new population as an elite individual, and the elite individual is retained and directly enters the new population for the next iteration; the new population generated in this iteration is used to replace the original population. Based on the adaptive mechanism of crossover ratio and mutation rate, after each iteration, the fitness of the new population is calculated; a population fitness threshold is set. If the fitness of the new population is greater than the population fitness threshold, the population fitness of the new population is subtracted from the population fitness threshold and then divided by the population fitness threshold to obtain the relative increase of the new population; 1 is used to subtract the relative increase of the new population to obtain the crossover ratio correction factor, which is multiplied by the current crossover ratio to obtain the updated crossover ratio; 1 is used to add the relative increase of the new population to obtain the mutation rate correction factor, which is multiplied by the current mutation rate to obtain the updated mutation rate. Set a stable period threshold, a fitness threshold, and a fitness change amplitude threshold; the convergence condition of the fitness function is that after a continuous iteration period of the stable period threshold length, the fitness value of the current elite individual is always less than the fitness threshold, and the fitness change amplitude of the current elite individual is always less than the fitness change amplitude threshold. The training continues iteratively until the fitness function meets the convergence condition. The current elite individual is then output as the optimal parameter combination for the inverter power supply protection strategy.

7. The adaptive adjustment method for inverter power supply protection strategy based on knowledge graph as described in claim 1, characterized in that, S6 specifically includes: S61. Update and deploy the inverter power supply protection strategy to the actual working environment, monitor the operating status data in real time, and obtain the operating status data after deployment. S62. Compare historical operating status data with post-deployment operating status data, and calculate the corresponding fitness values ​​according to the fitness function to generate an inverter power supply protection strategy operation effect report. S63. After the maintenance personnel arrive at the warning site, they shall check the operation effect of the inverter power supply protection strategy, and supplement the handling measures and record the handling strategy in combination with the fault mode identification result report and the actual operation on site.

Images