A multifunctional remote power management diagnosis method, system and device

Abstract

The present invention discloses a multifunctional remote power management diagnosis method, system and device, which belongs to the field of data analysis technology. The present invention accurately captures subtle changes in power supply operation through state indicator identification, improves monitoring accuracy; deeply explores fault trends, generates forward-looking fault characterization knowledge, and realizes fault prediction; adopts feature adversarial processing, constructs authentication and adversarial trend characterization knowledge, enhances robustness and adapts to complex environments, and reduces maintenance costs; generates parameter-focused characterization knowledge chains, realizes effective association and traceability, and improves fault handling efficiency; finally, based on authentication and adversarial discrimination probabilities, accurately determines fault diagnosis viewpoints, improves accuracy and realizes intelligent early warning. The present invention realizes the intelligence, foresight and efficiency of multifunctional remote power fault diagnosis through innovative technical means.

Description

Technical Field

[0001] The present invention belongs to the technical field of data analysis, and in particular relates to a multifunctional remote power management diagnosis method, system and device. Background Art

[0002] In today's highly information-based and intelligent era, the multifunctional remote power supply is the heart of various electronic devices, and its stable operation is crucial to ensure the normal operation of electronic devices. However, due to the complex and changeable working environment, the multifunctional remote power supply often faces the risk of various potential failures, and most of the existing fault diagnosis technologies are threshold-based fault diagnosis technologies. This technology monitors the operating data of the power supply equipment in real time by presetting a series of thresholds of key parameters (such as voltage, current, temperature, etc.), and compares it with the preset thresholds. Once a parameter is detected to be out of the threshold range, it is determined to be a fault state. However, this technology has the following three disadvantages:

[0003] (1) Lack of flexibility: Threshold settings are often based on experience or fixed standards, which are difficult to adapt to complex and changing working environments. The optimal thresholds for different devices or the same device under different working conditions may be different, resulting in false positives or false negatives;

[0004] (2) Lack of foresight: This technology can only identify faults after they occur, but cannot predict fault trends in advance, and cannot effectively avoid the occurrence of faults;

[0005] (3) Dependence on manual intervention: Threshold adjustment and optimization usually require manual participation, which increases maintenance costs and complexity.

[0006] It can be seen that how to efficiently and accurately monitor the power supply status and predict fault trends in advance has become a problem that needs to be solved urgently. Summary of the invention

[0007] The present invention provides a multifunctional remote power management diagnosis method, system and device, which can solve or partially solve the technical problems involved in the above-mentioned background technology.

[0008] The present invention provides a multifunctional remote power management and diagnosis method, which is applied to a management and diagnosis system. The method comprises: performing state indicator recognition on state monitoring information of a multifunctional remote power supply to obtain a state indicator recognition result; performing fault trend mining on X preset fault diagnosis viewpoints to obtain corresponding X fault trend representation knowledge; performing feature confrontation processing on the X fault trend representation knowledge to obtain X authentication trend representation knowledge corresponding to each fault trend representation knowledge in the X fault trend representation knowledge, and X confrontation trend representation knowledge corresponding to each fault trend representation knowledge in the X fault trend representation knowledge, where X is a positive integer, each authentication trend representation knowledge comprises a power supply device state record for indicating that its corresponding preset fault diagnosis viewpoint is compatible with the state monitoring information, and each confrontation trend representation knowledge comprises a power supply device state record for indicating that its corresponding preset fault diagnosis viewpoint is not compatible with the state monitoring information. The invention discloses a method for generating a discriminant representation knowledge chain based on the X authentication trend representation knowledge chains and the X confrontation trend representation knowledge chains, and performing parameter focusing processing on the state indicator identification result through the discriminant representation knowledge chain to obtain a parameter focusing representation knowledge chain; the parameter focusing representation knowledge chain is used to represent a correlation feature map between the discriminant representation knowledge chain and the state indicator identification result; based on the parameter focusing representation knowledge chain, X authentication discriminant probabilities corresponding to one authentication trend representation knowledge chain among the X authentication trend representation knowledge chains and X confrontation discriminant probabilities corresponding to one confrontation trend representation knowledge chain among the X confrontation trend representation knowledge chains are generated; based on the X authentication discriminant probabilities and the X confrontation discriminant probabilities, a target fault diagnosis viewpoint that is involved with the state monitoring information is determined from the X preset fault diagnosis viewpoints; the authentication discriminant probability corresponding to the target fault diagnosis viewpoint is greater than the confrontation discriminant probability corresponding to the target fault diagnosis viewpoint.

[0009] The present invention provides a management diagnosis system, comprising at least one processor and a memory; the memory stores computer execution instructions; the at least one processor executes the computer execution instructions stored in the memory, so that the at least one processor executes the above method.

[0010] The present invention provides a readable storage medium, on which a program or instruction is stored, and the program or instruction implements the steps of the above method when executed by a processor.

[0011] The present invention provides a multifunctional remote power management and diagnosis device, which is used for:

[0012] Performing status indicator recognition on status monitoring information of the multifunctional remote power supply to obtain status indicator recognition results;

[0013] Perform fault trend mining on X preset fault diagnosis viewpoints respectively to obtain corresponding X fault trend representation knowledge;

[0014] Performing feature confrontation processing on the X fault trend characterization knowledge respectively, obtaining X authentication trend characterization knowledge respectively corresponding to each fault trend characterization knowledge in the X fault trend characterization knowledge, and X confrontation trend characterization knowledge respectively corresponding to each fault trend characterization knowledge in the X fault trend characterization knowledge, X is a positive integer, each authentication trend characterization knowledge includes a power supply device state record used to indicate that its corresponding preset fault diagnosis viewpoint is adapted to the state monitoring information, and each confrontation trend characterization knowledge includes a power supply device state record used to indicate that its corresponding preset fault diagnosis viewpoint is not adapted to the state monitoring information;

[0015] Generate a discriminant representation knowledge chain according to the X authentication trend representation knowledge and the X confrontation trend representation knowledge, perform parameter focusing processing on the state indicator identification result through the discriminant representation knowledge chain, and obtain a parameter focusing representation knowledge chain; the parameter focusing representation knowledge chain is used to represent a correlation feature graph between the discriminant representation knowledge chain and the state indicator identification result;

[0016] Generate, based on the parameter focused representation knowledge chain, X authentication discrimination probabilities corresponding to one authentication trend representation knowledge among the X authentication trend representation knowledge, and X adversarial discrimination probabilities corresponding to one adversarial trend representation knowledge among the X adversarial trend representation knowledge;

[0017] According to the X authentication discrimination probabilities and the X adversarial discrimination probabilities, a target fault diagnosis viewpoint related to the state monitoring information is determined from the X preset fault diagnosis viewpoints; the authentication discrimination probability corresponding to the target fault diagnosis viewpoint is greater than the adversarial discrimination probability corresponding to the target fault diagnosis viewpoint.

[0018] The present invention proposes an innovative solution to the problems of insufficient flexibility, lack of foresight and reliance on manual labor in multifunctional remote power supply fault diagnosis. Through state indicator identification, subtle changes in power supply operation can be accurately captured to improve monitoring accuracy; fault trends can be deeply explored to generate forward-looking fault characterization knowledge to achieve fault prediction; feature adversarial processing can be used to build authentication and adversarial trend characterization knowledge to enhance robustness and adapt to complex environments and reduce maintenance costs; parameter-focused characterization knowledge chains can be generated to achieve effective association and traceability and improve fault handling efficiency; finally, based on authentication and adversarial discrimination probabilities, fault diagnosis viewpoints can be accurately determined to improve accuracy and achieve intelligent early warning. The present invention realizes the intelligence, foresight and efficiency of multifunctional remote power supply fault diagnosis through innovative technical means. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The present invention provides a flowchart of a multifunctional remote power management diagnosis method.

[0020] Figure 2 This is a structural schematic diagram of a management and diagnosis system provided by the present invention. DETAILED DESCRIPTION

[0021] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] The terms "first", "second", etc. in the present invention are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present invention can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second", etc. are generally of the same type, and the number of objects is not limited. For example, the first object can be one or more. In addition, in the present invention, "and / or" means at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0023] Figure 1 A multifunctional remote power management diagnosis method is shown, which is applied to a management diagnosis system. The method includes S101-S105.

[0024] S101, performing status indicator recognition on status monitoring information of a multifunctional remote power supply to obtain a status indicator recognition result.

[0025] The first step of this application is to identify the status indicators of the status monitoring information of the multifunctional remote power supply. This process is the basis for all subsequent analyses, and its core lies in extracting key indicators that can directly reflect the current status of the power supply from massive and complex data. The management and diagnosis system can automatically screen and parse status monitoring data such as voltage, current, temperature, power factor, etc. through a preset algorithm model (such as the target power management diagnosis algorithm), and convert these raw data into status indicator identification results that are easy to understand and analyze. This step not only simplifies the amount of data for subsequent processing, but also ensures the accuracy and pertinence of the analysis object, providing a solid foundation for subsequent fault diagnosis.

[0026] S102, respectively perform fault trend mining on X preset fault diagnosis viewpoints to obtain corresponding X fault trend representation knowledge; respectively perform feature adversarial processing on the X fault trend representation knowledge to obtain X authentication trend representation knowledge corresponding to each fault trend representation knowledge in the X fault trend representation knowledge, and X adversarial trend representation knowledge corresponding to each fault trend representation knowledge in the X fault trend representation knowledge.

[0027] Wherein, X is a positive integer, each authentication trend characterization knowledge includes a power supply equipment status record used to indicate that its corresponding preset fault diagnosis viewpoint is compatible with the state monitoring information, and each counter-trend characterization knowledge includes a power supply equipment status record used to indicate that its corresponding preset fault diagnosis viewpoint is incompatible with the state monitoring information.

[0028] After obtaining the preliminary status indicators, this application enters a more in-depth analysis phase: fault trend mining. This link conducts in-depth mining of fault trends for X preset fault diagnosis viewpoints (X is a positive integer representing different fault settings or modes). The management diagnosis system uses advanced technologies such as machine learning and data mining to learn and extract fault trend representation knowledge corresponding to each fault diagnosis viewpoint from historical data. These fault trend representation knowledge is a bridge between status monitoring information and fault diagnosis, and can reveal the failure risks that power supply equipment may face under specific conditions.

[0029] Next, in order to enhance the robustness and accuracy of fault diagnosis, this application introduces a feature adversarial processing mechanism. For each fault trend characterization knowledge, the system generates X certified trend characterization knowledge and X adversarial trend characterization knowledge. The certified trend characterization knowledge contains power supply equipment status records that are compatible with the status monitoring information, that is, those data that support the current fault diagnosis point of view; while the adversarial trend characterization knowledge, on the contrary, contains power supply equipment status records that are incompatible with the status monitoring information, that is, those data that may refute or weaken the current fault diagnosis point of view. This step provides a comprehensive perspective for subsequent discriminant analysis by constructing both positive and negative arguments.

[0030] S103: Generate a discriminant representation knowledge chain according to the X authentication trend representation knowledge and the X confrontation trend representation knowledge, and perform parameter focusing processing on the state indicator recognition result through the discriminant representation knowledge chain to obtain a parameter focusing representation knowledge chain.

[0031] The parameter-focused representation knowledge chain is used to represent a correlation feature graph between the discriminant representation knowledge chain and the state indicator recognition result.

[0032] After obtaining rich knowledge of fault trend representation and its confrontation information, this application further integrates this knowledge into a discriminant representation knowledge chain. This step is the key link between the state indicator identification results and the fault diagnosis viewpoint. It establishes a set of systematic knowledge frameworks to logically link each independent representation knowledge to form a complex network that can comprehensively reflect the power supply status and fault trends.

[0033] Subsequently, the management diagnosis system uses the discriminant representation knowledge chain to perform parameter focusing on the state indicator identification results. This process is equivalent to finding the most relevant fault trend path for each state indicator in a huge knowledge network, thereby generating a parameter focusing representation knowledge chain. This parameter focusing representation knowledge chain not only shows the direct connection between the state indicator and fault diagnosis, but also intuitively presents the strength and directionality of this connection in the form of a correlation feature graph, providing accurate input for subsequent fault probability calculations.

[0034] S104: Generate, based on the parameter focused representation knowledge chain, X authentication discrimination probabilities corresponding to one authentication trend representation knowledge among the X authentication trend representation knowledge, and X adversarial discrimination probabilities corresponding to one adversarial trend representation knowledge among the X adversarial trend representation knowledge.

[0035] Based on the parameter-focused characterization knowledge chain, the present application then generates discrimination probabilities corresponding to X authentication trend characterization knowledge and X confrontation trend characterization knowledge. These discrimination probability values ​​are quantitative assessments of the likelihood of each fault diagnosis view being established, where the authentication discrimination probability reflects the strength of evidence supporting the view, and the confrontation discrimination probability reflects the strength of evidence refuting the view. Through this step, the management diagnosis system can transform complex fault diagnosis problems into simple probability comparison problems, providing an intuitive and quantitative basis for subsequent decision support.

[0036] S105. Determine, based on the X authentication discrimination probabilities and the X adversarial discrimination probabilities, a target fault diagnosis viewpoint related to the state monitoring information from among the X preset fault diagnosis viewpoints; the authentication discrimination probability corresponding to the target fault diagnosis viewpoint is greater than the adversarial discrimination probability corresponding to the target fault diagnosis viewpoint.

[0037] Finally, this application determines the target fault diagnosis viewpoint that is involved with the status monitoring information from the X preset fault diagnosis viewpoints based on the generated X authentication discrimination probabilities and X adversarial discrimination probabilities. The basic principle followed in this decision-making process is that the authentication discrimination probability corresponding to the target fault diagnosis viewpoint must be greater than its corresponding adversarial discrimination probability. In other words, after comprehensively considering all supporting and opposing evidence, this fault diagnosis viewpoint is considered to be the most likely explanation for the current power supply status. Through this step, the management diagnosis system not only realizes the accurate prediction of power supply failures, but also provides clear guidance for subsequent repair and maintenance work.

[0038] The multifunctional remote power supply status monitoring and fault diagnosis technical solution under the management diagnosis system proposed in this application realizes comprehensive monitoring of the power supply status and accurate prediction of fault trends through a series of refined information processing processes. From status indicator identification to fault trend mining, to feature adversarial processing, parameter focus processing, and finally to the final fault diagnosis point of view determination, each link is closely linked and together constitutes an efficient and accurate fault diagnosis system. The application of this application can not only significantly improve the operating reliability of power supply equipment, but also effectively reduce maintenance costs, which is of great significance to promoting the intelligent development of the power industry.

[0039] In order to more specifically illustrate the technical solution proposed in this application, an example will be further explained below.

[0040] 1. Algorithm model example: Fault trend mining and discrimination based on decision tree

[0041] For example, there is a multifunctional remote power supply whose status monitoring information includes voltage, current, temperature and power factor. Decision tree is selected as the algorithm model for fault trend mining and discrimination.

[0042] 2. Status indicator identification:

[0043] Condition monitoring data is collected over a period of time and preprocessed, such as normalization. Then, the data is classified using a decision tree algorithm to identify different condition indicators. For example, a decision tree may classify the status into "normal", "overload", and "underload" based on the voltage and current levels.

[0044] 3. Status indicator identification results:

[0045] Normal state: voltage = 0.9 (normalized value), current = 0.8;

[0046] Overload state: voltage = 1.2, current = 1.5;

[0047] Underload state: voltage = 0.7, current = 0.5.

[0048] 4. Fault trend mining:

[0049] For each preset fault diagnosis viewpoint (such as "overheating fault", "capacitor aging", etc.), the corresponding fault trend representation knowledge is mined using historical data and decision tree algorithms. This representation knowledge can be expressed in the form of feature vectors.

[0050] 5. Knowledge of fault trend characterization (taking overheating fault as an example):

[0051] Eigenvector: [voltage = 1.1, current = 1.3, temperature = 0.95, power factor = 0.85].

[0052] 6. Feature adversarial processing:

[0053] For each fault trend characterization knowledge, corresponding authentication trend characterization knowledge and adversarial trend characterization knowledge are generated.

[0054] Certified trend characterization knowledge (overheat fault example): [voltage = 1.1, current = 1.3, temperature = 0.95, power factor = 0.85] (records adapted to condition monitoring information);

[0055] Knowledge of counter-trend characterization (overheating fault example): [voltage = 0.9, current = 0.8, temperature = 0.7, power factor = 0.9] (records that do not match condition monitoring information).

[0056] 7. Discriminative representation knowledge chain generation and parameter focusing processing:

[0057] All the authentication trend representation knowledge and confrontation trend representation knowledge are integrated into a discriminant representation knowledge chain. Then, this knowledge chain is used to perform parameter focusing on the state indicator identification results to obtain a parameter focused representation knowledge chain. This process can be regarded as finding the fault trend closest to the current state in the feature space.

[0058] 8. Parameter Focused Characterization The knowledge chain is an ordered set of feature vectors, representing the mapping path from state indicators to fault diagnosis.

[0059] 9. The correlation feature graph can be a two-dimensional graph, where the horizontal axis represents the state indicator and the vertical axis represents the fault diagnosis point of view. The points on the graph represent the strength of the correlation between the state indicator and the fault diagnosis. For example, if a state indicator has a high correlation with the overheating fault, the value at that point will be large.

[0060] 10. Generation of judgment probability and determination of fault diagnosis viewpoints:

[0061] Based on the parameter focusing characterization knowledge chain, the authentication discrimination probability and adversarial discrimination probability of each fault diagnosis viewpoint are calculated. Then, the two probability values ​​are compared to determine the target fault diagnosis viewpoint that is involved with the condition monitoring information.

[0062] 11. Discriminant probability:

[0063] Overheating fault authentication probability: 0.8 (indicates that there is 80% evidence supporting overheating fault);

[0064] Probability of adversarial discrimination of overheating fault: 0.2 (indicates that there is 20% evidence to refute overheating fault);

[0065] Since the authentication discrimination probability is greater than the adversarial discrimination probability, it is determined that the most likely explanation for the current state is an overheating fault.

[0066] It is worth mentioning that the parameter-focused representation knowledge chain, as a bridge between status indicators and fault diagnosis, is based on a series of ordered feature vectors to show the mapping path from raw status data to specific fault diagnosis results. The following is a specific example to further illustrate the concept of the parameter-focused representation knowledge chain.

[0067] For example, there is a multifunctional remote power supply, and three key status indicators have been identified through status indicators: voltage (V), current (I) and temperature (T). At the same time, two fault diagnosis viewpoints are preset: overheat fault (OF) and capacitor aging (CE).

[0068] 1. Status indicator identification results (expressed in normalized values):

[0069] State 1: [V=0.9, I=0.8, T=0.75];

[0070] State 2: [V=1.1, I=1.2, T=0.9];

[0071] ...(other status).

[0072] 2. Knowledge of fault trend characterization:

[0073] Overtemperature fault (OF) trend: [V_high=1.05, I_high=1.15, T_high=0.85];

[0074] Capacitor aging (CE) trend: [V_low=0.85, I_normal=1.0, T_normal=0.7].

[0075] 3. Parameter Focus Characterization Knowledge Chain Construction:

[0076] For state 1: [V=0.9, I=0.8, T=0.75], the system may first recognize that this is close to the capacitor aging (CE) trend of voltage low (V_low=0.85) and temperature normal (T_normal=0.7), but also notices that the current is slightly low. Therefore, the initial mapping path may be biased towards CE, but with uncertainty.

[0077] Next, the system considers state changes or more state information (such as state 2) and combines other feature vectors in the discriminant representation knowledge chain to gradually focus on the most likely fault diagnosis.

[0078] After further analysis, the system found that both the voltage and current in State 2 increased, and the temperature was also close to the overtemperature threshold, which more strongly indicated the possibility of an overtemperature fault (OF).

[0079] 4. Example of parameter-focused characterization knowledge chain:

[0080] Initial mapping (preliminary path from state 1 to fault diagnosis):

[0081] [V=0.9→CE_V_low=0.85, I=0.8→CE_I_normal=1.0 (but low), T=0.75→CE_T_normal=0.7];

[0082] Focus update (combined with state 2 information):

[0083] [State 1+State 2→OF_V_high=1.05 (close to State 2’s V=1.1), OF_I_high=1.15 (close to State 2’s I=1.2), OF_T_high=0.85 (close to State 2’s T=0.9)].

[0084] The final parameter focusing characterization knowledge chain can be a feature vector sequence that integrates multiple state indicators and fault diagnosis trends, which reflects the gradual focusing process from the initial state to the final fault diagnosis.

[0085] 5. In the correlation feature graph, each state indicator (such as voltage, current, temperature) has a correlation strength value with the fault diagnosis point of view (such as overheating fault, capacitor aging). These values ​​can be obtained by calculating the similarity or distance between the state indicator and the fault diagnosis trend feature vector.

[0086] For example, in the above example, the voltage, current, and temperature of state 2 are closer to the characteristic vector of the overheating fault trend, so the correlation strength between state 2 and the overheating fault will be higher on the correlation feature graph.

[0087] In summary, the present invention proposes an innovative solution to the key problems existing in the current multifunctional remote power fault diagnosis technology, such as insufficient flexibility, lack of foresight and reliance on manual intervention, which has significant beneficial effects, which are specifically reflected in the following aspects:

[0088] First, by implementing status indicator recognition, the present invention can accurately capture subtle changes in the operation of the multifunctional remote power supply, providing a solid data foundation for subsequent fault diagnosis. This step effectively solves the problem of false alarms or missed alarms caused by improper threshold settings in traditional threshold-based technologies, and improves the accuracy and reliability of monitoring;

[0089] Secondly, the present invention conducts in-depth fault trend mining on the preset fault diagnosis viewpoints and generates forward-looking fault trend characterization knowledge. This innovation enables the system to not only identify faults that have already occurred, but more importantly, to predict the development trend of faults in advance, making it possible to take preventive measures in time to avoid faults.

[0090] Furthermore, through feature adversarial processing, the present invention constructs authentication trend characterization knowledge and adversarial trend characterization knowledge, laying a solid foundation for the generation of discriminant characterization knowledge chains. This process not only enhances the robustness of fault diagnosis, but also enables the system to adapt to more complex and changeable working environments without relying on frequent manual adjustment of thresholds, significantly reducing maintenance costs and complexity;

[0091] In addition, the generation of parameter-focused characterization knowledge chains realizes the effective association between the discriminant characterization knowledge chains and the state indicator identification results, forming an intuitive correlation feature graph. This feature makes the fault diagnosis process more transparent and traceable, making it easier for technicians to quickly locate the root cause of the problem and improving the efficiency of fault handling;

[0092] Finally, based on the authentication discrimination probability and adversarial discrimination probability generated by the parameter-focused characterization knowledge chain, the present invention can accurately determine the target fault diagnosis viewpoints that are involved in the state monitoring information. This result not only improves the accuracy of fault diagnosis, but also provides strong technical support for the realization of intelligent and automated fault warning and processing.

[0093] In some optional embodiments, the performing feature confrontation processing on the X fault trend characterization knowledge respectively to obtain X authentication trend characterization knowledge respectively corresponding to each fault trend characterization knowledge in the X fault trend characterization knowledge, and X confrontation trend characterization knowledge respectively corresponding to each fault trend characterization knowledge in the X fault trend characterization knowledge, includes:

[0094] 1) inputting the X fault trend characterization knowledge into a target power management diagnosis algorithm; the target power management diagnosis algorithm includes a feature adversarial processing branch;

[0095] 2) obtaining the authentication feature variable and the confrontation feature variable in the feature confrontation processing branch; the authentication feature variable refers to the algorithm weight used to characterize the diagnosis hit result, the diagnosis hit result refers to the view feature used to characterize the adaptation of the fault diagnosis view to the state monitoring information, and the confrontation feature variable refers to the algorithm weight used to characterize the diagnosis disturbance result, and the diagnosis disturbance result refers to the view feature used to characterize the incompatibility of the fault diagnosis view to the state monitoring information;

[0096] 3) generating X authentication trend characterization knowledge corresponding to each of the X fault trend characterization knowledge respectively through the authentication feature variable and the X fault trend characterization knowledge, and generating X adversarial trend characterization knowledge corresponding to each of the X fault trend characterization knowledge respectively through the adversarial feature variable and the X fault trend characterization knowledge.

[0097] It can be immediately seen that the core of this embodiment is to perform feature adversarial processing on X fault trend characterization knowledge (hereinafter referred to as "fault knowledge") to generate two sets of new characterization knowledge: one set is the authentication trend characterization knowledge, which is used to verify the consistency between the fault diagnosis viewpoint and the actual state monitoring information; the other set is the adversarial trend characterization knowledge, which is used to simulate and identify disturbance factors in fault diagnosis. The entire processing flow relies on a specially designed target power management diagnosis algorithm, which has a built-in feature adversarial processing branch to achieve in-depth analysis and processing of fault knowledge.

[0098] First, X fault trend characterization knowledge is input into the target power management diagnosis algorithm. These fault knowledge are extracted based on historical data, real-time monitoring information and expert experience, reflecting the various fault modes and development trends that may occur in the power system. Each fault knowledge can be regarded as a feature vector, which contains multi-dimensional information describing the fault characteristics, such as voltage fluctuation, current abnormality, temperature rise, etc.

[0099] The target power management diagnosis algorithm contains a feature adversarial processing branch, which is the core module for implementing feature adversarial processing. It is responsible for receiving input fault knowledge and generating authentication feature variables and adversarial feature variables through a series of algorithm operations.

[0100] Certified feature variables: These variables represent the weight parameters in the algorithm that can correctly reflect the matching of fault diagnosis viewpoints with condition monitoring information. Their role is to strengthen those fault diagnosis features that are highly consistent with the actual situation, thereby improving the accuracy of diagnosis.

[0101] Adversarial feature variables: These variables represent the weight parameters in the algorithm that can identify and respond to disturbance factors in fault diagnosis. Their role is to introduce "adversarial" thinking in the fault diagnosis process, that is, to consider factors that may lead to misdiagnosis or missed diagnosis, thereby enhancing the robustness of the algorithm.

[0102] In the feature adversarial processing branch, authentication feature variables and adversarial feature variables are extracted and optimized through specific algorithm logic (such as adversarial training in deep learning, kernel function optimization in support vector machines, etc.). This process can involve multiple iterative calculations to continuously adjust the variable values ​​until the predetermined optimization goal is achieved, such as minimizing the diagnostic error, maximizing the diagnostic stability, etc.

[0103] The extracted authentication feature variables are combined with the input X fault trend representation knowledge, and X authentication trend representation knowledge corresponding to each fault knowledge is generated through operations such as feature fusion and weight adjustment. These authentication knowledge not only retains the core features of the original fault knowledge, but also incorporates the verification information of the algorithm on the fault diagnosis viewpoint, making the diagnosis result more reliable.

[0104] For example, a feature fusion method based on a deep neural network can be used to use the authentication feature variable as part of the network weight, and forward propagate with the fault knowledge, and obtain the authentication trend representation knowledge through nonlinear transformation and feature learning. In this process, the network continuously adjusts the weight through the back-propagation algorithm to minimize the loss function, thereby improving the accuracy of authentication.

[0105] Similarly, the extracted adversarial feature variables are combined with the input X fault trend representation knowledge to generate X adversarial trend representation knowledge corresponding to each fault knowledge. These adversarial knowledge simulates the disturbances that may occur in fault diagnosis, such as noise interference, data deviation, etc., to help the algorithm better identify and deal with these adverse factors during the diagnosis process. Exemplarily, the idea of ​​generative adversarial network (GAN) can be used to design a generator network and a discriminator network. The generator network receives fault knowledge and adversarial feature variables to generate adversarial trend representation knowledge; the discriminator network attempts to distinguish the generated adversarial knowledge from the real fault knowledge. Through the adversarial training of the two, the deception ability of the generator and the recognition ability of the discriminator are continuously improved, and finally high-quality adversarial trend representation knowledge is obtained.

[0106] Through the above-mentioned feature confrontation processing, the present invention achieves the following technical effects:

[0107] 1) Improve diagnostic accuracy: The generation of knowledge for certification trend representation strengthens the consistency between fault diagnosis viewpoints and condition monitoring information, reducing the possibility of misdiagnosis and missed diagnosis;

[0108] 2) Enhance algorithm robustness: The introduction of adversarial trend characterization knowledge enables the algorithm to identify and respond to disturbance factors in fault diagnosis, improving the diagnostic stability in complex environments;

[0109] 3) Optimizing algorithm performance: The design of the feature adversarial processing branch continuously improves the overall performance of the algorithm, including diagnostic speed, accuracy, and generalization ability, by iteratively optimizing the authentication feature variables and adversarial feature variables.

[0110] In some other possible embodiments, generating a discriminant representation knowledge chain based on the X authentication trend representation knowledge and the X confrontation trend representation knowledge, performing parameter focusing processing on the state indicator identification result through the discriminant representation knowledge chain to obtain a parameter focusing representation knowledge chain includes:

[0111] (1) inputting the X authentication trend characterization knowledge, the X countermeasure trend characterization knowledge and the state indicator identification result into a target power management diagnosis algorithm; the target power management diagnosis algorithm includes a parameter focusing branch;

[0112] (2) according to the parameter focusing branch, generating a discriminant representation knowledge chain of the inter-class feature enhanced convolution kernel through the X authentication trend representation knowledge and the X adversarial trend representation knowledge, generating an identification feature chain of the inter-class feature enhanced convolution kernel according to the state indicator recognition result, and generating an abnormal tracing feature chain of the inter-class feature enhanced convolution kernel according to the state indicator recognition result;

[0113] (3) Generate a basic parameter focused representation knowledge chain based on the discrimination representation knowledge chain, the identification feature chain, the abnormality tracing feature chain, and the feature recognition degree corresponding to the status indicator identification result, and calculate the basic parameter focused representation knowledge chain layer by layer to obtain the parameter focused representation knowledge chain corresponding to the status indicator identification result.

[0114] In a complex power management system, accurately identifying state indicators and effectively processing them is the key to ensuring stable operation of the system. This embodiment aims to improve the accuracy and robustness of state indicator identification by introducing authentication trend characterization knowledge, adversarial trend characterization knowledge, and parameter focus processing mechanism. Specifically, this embodiment uses the target power management diagnostic algorithm, combined with the parameter focus branch, to generate a convolution kernel with inter-class feature enhancement capabilities, thereby forming a discriminant representation knowledge chain, and finally obtains a parameter focus representation knowledge chain by performing parameter focus processing on the state indicator identification results.

[0115] In detail, the specific implementation steps of the above embodiment are as follows.

[0116] 1. Input data and algorithm initialization

[0117] Input data: The X authentication trend representation knowledge, the X confrontation trend representation knowledge and the state indicator identification result are used as input data. These input data respectively represent the authentication state, confrontation state and current state indicator identification of the system in different dimensions, and are the basis for subsequent processing.

[0118] Algorithm initialization: Initialize the target power management diagnostic algorithm, which contains a parameter focusing branch for performing subsequent parameter focusing processing. The parameter focusing branch is designed to achieve precise focusing on key parameters through in-depth analysis and feature extraction of input data, thereby improving the accuracy and efficiency of status indicator identification.

[0119] 2. Generate inter-class feature enhancement convolution kernel

[0120] Generation of discriminative representation knowledge chain: In the parameter focusing branch, the X authentication trend representation knowledge and the X adversarial trend representation knowledge are first used to generate a discriminative representation knowledge chain of the inter-class feature enhanced convolution kernel through a specific convolution operation and feature fusion strategy. This process aims to explore the intrinsic relationship between authentication trends and adversarial trends, as well as their impact on the identification of state indicators, so as to construct a feature representation with strong discriminative ability.

[0121] Convolution operation: Use multi-scale convolution kernels to perform convolution operations on the input data to capture feature information at different scales. The weight of each convolution kernel is optimized through the back propagation algorithm to maximize its performance in distinguishing different state indicators.

[0122] Feature fusion: The feature maps output by different convolution kernels are fused by using weighted summation, maximum pooling or concatenation to integrate multi-scale and multi-level feature information and enhance the feature representation capability.

[0123] Inter-class feature enhancement: Based on feature fusion, an inter-class feature enhancement mechanism is further introduced to highlight the distinguishing features between classes and suppress the common features within classes by calculating the feature differences between different classes and performing weighted enhancement. This helps to improve the robustness and generalization ability of the discriminative representation knowledge chain in complex scenarios.

[0124] 3. Generate identification feature chain and abnormal tracing feature chain

[0125] Identification feature chain generation: Based on the state indicator recognition results, an identification feature chain of the inter-class feature enhancement convolution kernel is generated. The identification feature chain is used to represent the unique features of the state indicator in a normal or abnormal state, and is an important basis for subsequent parameter focusing processing.

[0126] Feature extraction: Feature extraction is performed on the state indicator recognition results, and the most representative feature vectors are extracted using methods such as principal component analysis (PCA) and linear discriminant analysis (LDA).

[0127] Feature encoding: Encode the extracted feature vectors to form an identification feature chain. The encoding method can use sparse coding, local linear embedding (LLE), etc. to retain the main information of the feature and reduce the dimension.

[0128] Abnormal tracing feature chain generation: Based on the state indicator recognition results, an abnormal tracing feature chain of the inter-class feature enhanced convolution kernel is generated. The abnormal tracing feature chain is used to track the change trajectory of the state indicator under abnormal conditions, providing strong support for fault location and diagnosis.

[0129] Anomaly detection: Use statistical anomaly detection methods (such as the 3σ principle, box plots, etc.) or machine learning-based anomaly detection methods (such as support vector machines, random forests, etc.) to perform anomaly detection on the status indicator identification results.

[0130] Trajectory tracking: Track the detected abnormal points, record their changes over time, and form an abnormal tracing feature chain. The tracking method can use algorithms such as time series analysis and dynamic programming.

[0131] 4. Generate parameter-focused representation knowledge chain

[0132] Basic parameter focused representation knowledge chain generation: Based on the discrimination representation knowledge chain, identification feature chain, abnormal tracing feature chain and the feature recognition degree corresponding to the state indicator recognition result, the basic parameter focused representation knowledge chain is generated. This process aims to comprehensively analyze and integrate various feature information to form a comprehensive and accurate representation of the state indicator recognition result.

[0133] Feature fusion: weighted summation, feature splicing and other methods are used to fuse the discriminant representation knowledge chain, identification feature chain, anomaly tracing feature chain and feature recognition to form a basic parameter focused representation knowledge chain.

[0134] Feature selection: Perform feature selection on the fused features, using feature selection methods based on statistics (such as chi-square test, F-value test, etc.) or feature selection methods based on machine learning (such as recursive feature elimination, tree model-based feature selection, etc.) to remove redundant features and retain the most discriminative features.

[0135] Layer-by-layer calculation and parameter focusing: The basic parameter-focused representation knowledge chain is calculated layer by layer, and high-level feature information is gradually extracted through layer-by-layer convolution, pooling, full connection and other operations, and key parameters are focused. This process aims to further refine feature information and highlight the impact of key parameters on state indicator identification, thereby obtaining the final parameter-focused representation knowledge chain.

[0136] Layer-by-layer convolution: A multi-layer convolutional neural network (CNN) is used to perform layer-by-layer convolution operations on the basic parameter focus representation knowledge chain to extract feature information at different levels. The weight of each layer of convolution kernel is optimized through the back propagation algorithm.

[0137] Pooling operation: After each convolution layer, operations such as maximum pooling or average pooling are used to reduce the dimension of the feature map to reduce the amount of calculation and retain key features.

[0138] Fully connected layer: After the convolution and pooling operations, the fully connected layer (FC) is introduced to map the convolution features to the sample tag space to obtain the final parameter-focused representation knowledge chain. The weights of the fully connected layer are also optimized through the back-propagation algorithm.

[0139] It can be seen that this embodiment achieves accurate focusing and efficient processing of the state indicator identification results by introducing authentication trend characterization knowledge, confrontation trend characterization knowledge and parameter focus processing mechanism. The specific advantages of this embodiment are as follows: by exploring the intrinsic relationship between authentication trends and confrontation trends, as well as their influence on state indicator identification, a discriminant characterization knowledge chain with strong discriminant ability is constructed, which improves the accuracy of state indicator identification; by introducing the inter-class feature reinforcement mechanism, the inter-class distinguishing features are highlighted, the common features within the class are suppressed, and the robustness and generalization ability of the discriminant characterization knowledge chain in complex scenarios are enhanced; by generating an abnormal tracing feature chain, the changing trajectory of the state indicator under abnormal conditions can be tracked, which provides strong support for fault location and diagnosis; through layer-by-layer calculation and parameter focus processing, the feature information is gradually refined and the influence of key parameters is highlighted, which improves the processing efficiency and real-time performance of state indicator identification.

[0140] In the next step, the discriminant representation knowledge chain of the inter-class feature enhancement convolution kernel is generated according to the X authentication trend representation knowledge and the X adversarial trend representation knowledge, the identification feature chain of the inter-class feature enhancement convolution kernel is generated according to the state indicator recognition result, and the abnormal tracing feature chain of the inter-class feature enhancement convolution kernel is generated according to the state indicator recognition result, including: weighting the discriminant feature value list in the parameter focusing branch with the X authentication trend representation knowledge and the X adversarial trend representation knowledge to obtain the discriminant representation knowledge chain of the inter-class feature enhancement convolution kernel; weighting the identification feature value list in the parameter focusing branch with the state indicator recognition result to obtain the identification feature chain of the inter-class feature enhancement convolution kernel; weighting the abnormal tracing feature value list in the parameter focusing branch with the state indicator recognition result to obtain the abnormal tracing feature chain of the inter-class feature enhancement convolution kernel; the discriminant feature value list, the identification feature value list and the abnormal tracing feature value list are all lists formed by soft thresholds in the parameter focusing branch.

[0141] Furthermore, the basic parameter focused representation knowledge chain is generated based on the feature recognition degree corresponding to the discriminant representation knowledge chain, the identification feature chain, the abnormality tracing feature chain and the status indicator identification result, including: updating the abnormality tracing feature chain to obtain a first risk mapping feature chain, determining the weighted processing result of the discriminant representation knowledge chain and the first risk mapping feature chain as a second risk mapping feature chain, generating a third risk mapping feature chain based on the feature recognition degree corresponding to the second risk mapping feature chain and the status indicator identification result, quantitatively correcting the third risk mapping feature chain to obtain a quantitatively corrected array, and determining the weighted processing result of the quantitatively corrected array and the abnormality tracing feature as the basic parameter focused representation knowledge chain.

[0142] In the following steps, this embodiment elaborates on an innovative technical solution, which aims to generate a feature chain with high representation ability and discrimination through a series of sophisticated processing steps, and then provide solid support for the construction of the basic parameter focus representation knowledge chain. The core of this technical solution is how to effectively use X authentication trend representation knowledge and X adversarial trend representation knowledge, as well as the state indicator recognition results, to generate the discriminative representation knowledge chain, identification feature chain and abnormal tracing feature chain of the inter-class feature enhanced convolution kernel, and finally generate the basic parameter focus representation knowledge chain based on these feature chains and feature recognition.

[0143] 1. Feature chain generation of inter-class feature enhancement convolution kernel

[0144] ① Generation of discriminative representation knowledge chain

[0145] When constructing the discriminative representation knowledge chain of the inter-class feature enhanced convolution kernel, the discriminative feature value list in the parameter focusing branch is first processed in detail. This list is carefully extracted from the original data using advanced soft thresholding technology, and it contains the key features necessary to distinguish different categories.

[0146] In order to closely combine these features with authentication trends and adversarial trends, a weighted processing strategy is adopted. Specifically, each feature value in the discriminant feature value list is weighted and fused with X authentication trend representation knowledge and X adversarial trend representation knowledge. The cleverness of this step is that it can not only highlight the unique differences between categories, but also effectively resist adversarial attacks, thereby enhancing the robustness and generalization ability of the model.

[0147] Through weighted processing, a feature chain rich in discriminative information is obtained, that is, the discriminative representation knowledge chain of the inter-class feature enhanced convolution kernel. This feature chain will play a vital role in subsequent model training and classification tasks.

[0148] ② Generation of identification feature chain

[0149] When generating the identification feature chain of the inter-class feature enhancement convolution kernel, we also focus on the identification feature value list in the parameter focusing branch. This list is also extracted by the soft thresholding technique, but it pays more attention to the unique features of the state indicators in normal or abnormal states.

[0150] In order to closely combine these features with the state indicator recognition results, a weighted processing strategy is used again. Each feature value in the identification feature value list is weighted and fused with the state indicator recognition result. The purpose of this step is to enable the model to more accurately identify and classify different states by strengthening the feature representation of the state indicator.

[0151] Through weighted processing, a feature chain rich in identification information is obtained, that is, the identification feature chain of the inter-class feature enhanced convolution kernel. This feature chain will show excellent performance in state recognition and classification tasks.

[0152] ③Generation of abnormal tracing feature chain

[0153] When generating the anomaly tracing feature chain of the inter-class feature enhancement convolution kernel, we focus on the anomaly tracing feature value list in the parameter focusing branch. This list is also extracted by the soft thresholding technique, but it is specifically used to track the change trajectory of the state indicator under abnormal conditions.

[0154] In order to closely combine these features with the status indicator recognition results and highlight the characteristics of abnormal states, a weighted processing strategy is used again. Each feature value in the abnormal tracing feature value list is weighted and fused with the status indicator recognition results. The purpose of this step is to enable the model to more accurately detect and trace the occurrence and development of abnormal states by strengthening the feature representation of abnormal states.

[0155] Through weighted processing, a feature chain rich in anomaly tracing information is obtained, namely, the anomaly tracing feature chain of the inter-class feature enhanced convolution kernel. This feature chain will play an irreplaceable role in anomaly detection and tracing tasks.

[0156] 2. Generation of knowledge chain focusing on basic parameters

[0157] After successfully generating the discriminant representation knowledge chain, identification feature chain, and abnormal tracing feature chain, these feature chains are further combined with the feature recognition corresponding to the state indicator recognition result to generate the basic parameter focused representation knowledge chain. This step is the essence of the entire technical solution.

[0158] ① Update of abnormal tracing feature chain

[0159] First, the abnormality tracing feature chain is updated to reflect the latest abnormal status information. This update process may include smoothing, filtering or other advanced processing techniques to remove noise and redundant information and retain key abnormal features. Through this update, a more accurate and reliable abnormality tracing feature chain is obtained, namely the first risk mapping feature chain.

[0160] ② Generation of the second risk mapping feature chain

[0161] Next, the discriminant representation knowledge chain is weighted with the first risk mapping feature chain to fuse the discriminant features and the abnormal features. The purpose of this step is to enable the model to have both strong discriminant and abnormal detection capabilities by organically combining two different types of features. The weighted processing process may involve complex algorithms and models to ensure effective fusion and representation of features.

[0162] Through weighted processing, a feature chain rich in discriminative and abnormal information is obtained, namely the second risk mapping feature chain. This feature chain will show stronger robustness and generalization ability in subsequent model training and classification tasks.

[0163] ③ Generation of the third risk mapping feature chain

[0164] Then, based on the feature recognition degree corresponding to the second risk mapping feature chain and the state indicator identification result, the third risk mapping feature chain is generated. Feature recognition degree is an important indicator that reflects the reliability and confidence of the state indicator identification result. In order to make full use of this information, each feature value in the second risk mapping feature chain is multiplied by the corresponding feature recognition degree to obtain a weighted feature value. Then, these weighted feature values ​​are combined to form the third risk mapping feature chain.

[0165] The purpose of this step is to further adjust and optimize the representation ability of the feature chain by introducing feature recognition information, so that the model can more accurately identify and classify different states when facing complex and changing data.

[0166] ④Quantitative correction and basic parameter focus characterization knowledge chain generation

[0167] Finally, the third risk mapping feature chain is quantitatively modified to obtain a more accurate feature representation. Quantization modification is an important step that can map continuous feature values ​​into a limited discrete space, thereby facilitating subsequent model processing and calculation. The process of quantitative modification may involve quantitative coding, quantitative compression or other advanced quantitative processing techniques.

[0168] Through quantitative correction, a quantized feature array, namely the quantitative correction array, is obtained. Then, the quantitative correction array is fused with the weighted processing result of the abnormal tracing feature to generate the final basic parameter focused characterization knowledge chain. This fusion process may involve simple splicing, weighted sum or other complex fusion methods to ensure the integrity and characterization capability of the feature chain.

[0169] Thus, this embodiment successfully generates a feature chain with high representation and discrimination capabilities through a series of sophisticated processing steps, and provides solid support for the construction of a basic parameter focused representation knowledge chain. In this way, it can not only effectively improve the classification and recognition capabilities of the model, but also enhance the robustness and generalization capabilities of the model, laying a solid foundation for subsequent model application and deployment.

[0170] In order to implement the above technical solution clearly and completely, the following exemplary formula is used to introduce the generation of feature chains of inter-class feature enhancement convolution kernels and the generation of basic parameter focus representation knowledge chains.

[0171] 1. Feature chain generation of inter-class feature enhancement convolution kernel

[0172] ① Generation of discriminative representation knowledge chain

[0173] Input: discriminant feature value list D=[d1, d2, ..., dn], authentication trend representation knowledge C=[c1, c2, ..., cx], adversarial trend representation knowledge A=[a1, a2, ..., ax].

[0174] For each di, calculate the weighted fusion eigenvalue:

[0175] di'=(w1i*c1*di)+(w2i*c2*di)+...+(wxi*cx*di)+(w1i'*a1*di)+(w2i'*a2*di)+...+(wxi'*ax*di)

[0176] Where w1i, w2i, ..., wxi and w1i', w2i', ..., wxi' are the corresponding weighting coefficients.

[0177] Output: discriminant representation knowledge chain D'=[d1', d2', ..., dn'].

[0178] ② Generation of identification feature chain

[0179] Input: identification feature value list I=[i1, i2, ..., im], state indicator recognition result R=[r1, r2, ..., rm].

[0180] For each il, calculate the weighted fusion eigenvalue: il'=wl*rl*il

[0181] Where wl is the corresponding weighting coefficient.

[0182] Output: Identity feature chain I'=[i1', i2', ..., im'].

[0183] ③Generation of abnormal tracing feature chain

[0184] Input: abnormal tracing feature value list T=[t1, t2, ..., tp], state indicator recognition result R=[r1, r2, ..., rp] (set the dimension of R to be consistent with T).

[0185] For each tq, calculate the weighted fusion eigenvalue tq'=wq*rq*tq

[0186] Where wq is the corresponding weighting coefficient.

[0187] Output: abnormal tracing feature chain T'=[t1', t2', ..., tp'].

[0188] 2. Generation of knowledge chain focusing on basic parameters

[0189] ① Update of abnormal tracing feature chain (first risk mapping feature chain)

[0190] Input: Exception tracing feature chain T'.

[0191] Processing: Apply smoothing, filtering and other processing methods to obtain the updated abnormal tracing feature chain T'' (i.e., the first risk mapping feature chain).

[0192] Output: The first risk mapping feature chain T''.

[0193] ② Generation of the second risk mapping feature chain

[0194] Input: discriminant representation knowledge chain D', first risk mapping feature chain T''.

[0195] For each corresponding eigenvalue, calculate the weighted fused eigenvalue fs=α*ds'+β*ts''

[0196] Where α and β are weighting coefficients, ds' is an element in D', and ts'' is an element in T''.

[0197] Output: The second risk mapping feature chain F = [f1, f2, ..., fn] (assuming that the lengths of D' and T'' are consistent or processed).

[0198] ③ Generation of the third risk mapping feature chain

[0199] Input: The second risk mapping feature chain F, the feature recognition degree G=[g1, g2, ..., gn] corresponding to the status indicator recognition result.

[0200] For each fs, calculate the weighted eigenvalue fs'=gs*fs

[0201] where gs is an element in G.

[0202] Output: The third risk mapping feature chain F'=[f1', f2', ..., fn'].

[0203] ④Quantitative correction and basic parameter focus characterization knowledge chain generation

[0204] Input: Third risk mapping feature chain F'.

[0205] Processing: Apply the quantization correction technology to map the continuous eigenvalues ​​in F' to the discrete space and obtain the quantization correction array Q.

[0206] Fusion:

[0207] The quantitative correction array Q is merged with the weighted processing result of the abnormal tracing feature (T'' can be used if necessary), for example, by simple splicing, weighted sum, etc., to obtain the basic parameter focusing characterization knowledge chain K.

[0208] Output: Basic parameter focused representation knowledge chain K.

[0209] Those skilled in the art can and should understand that the weighting coefficients in the above formula (such as w1i, w2i, ..., wxi, w1i', w2i', ..., wxi', wl, wq, α, β) and the specific quantization correction and fusion methods need to be adjusted and optimized according to the actual application scenarios and data.

[0210] In some further optional embodiments, generating a discriminant representation knowledge chain based on the X authentication trend representation knowledge and the X confrontation trend representation knowledge, performing parameter focusing processing on the state indicator identification result through the discriminant representation knowledge chain to obtain a parameter focusing representation knowledge chain includes:

[0211] (1) inputting the X authentication trend characterization knowledge, the X confrontation trend characterization knowledge and the state indicator identification result into a target power management diagnosis algorithm; the target power management diagnosis algorithm includes a parameter focusing branch, and the parameter focusing branch includes Y feature operation subnets; Y is a positive integer;

[0212] (2) combining the X authentication trend representation knowledge and the X confrontation trend representation knowledge into a discriminant representation knowledge chain through the Y feature operation subnetworks, performing parameter focusing processing on the discriminant representation knowledge chain and the state indicator identification result, and obtaining local parameter focusing representation knowledge corresponding to the Y feature operation subnetworks;

[0213] (3) Performing knowledge interaction on Y local parameter-focused representation knowledge to obtain a parameter-interaction knowledge chain, performing knowledge collision on the parameter-interaction knowledge chain and the parameter-interaction knowledge quantification relationship network to obtain a basic parameter-focused representation knowledge chain, and performing layer-by-layer calculation on the basic parameter-focused representation knowledge chain to obtain a parameter-focused representation knowledge chain corresponding to the state indicator recognition result; the parameter-interaction knowledge quantification relationship network is the soft threshold in the parameter-focused branch.

[0214] In the actual application process, this technical scheme proposes an innovative solution to the problem of how to generate a discriminant representation knowledge chain based on the given X authentication trend representation knowledge and X confrontation trend representation knowledge, and further perform parameter focusing processing on the state indicator recognition result through the discriminant representation knowledge chain to obtain a parameter focused representation knowledge chain. This scheme makes full use of advanced algorithm architecture and knowledge processing strategy to improve the accuracy and efficiency of power management diagnosis.

[0215] First, we will describe in detail how to input X authentication trend representation knowledge, X adversarial trend representation knowledge, and status indicator identification results into a specially designed target power management diagnosis algorithm. This algorithm is a highly integrated and powerful system, the core of which is a module called parameter focus branch. The parameter focus branch contains Y feature operation subnets, where Y is a positive integer representing the multi-dimensional analysis capability of the algorithm when processing complex information. These feature operation subnets can not only work independently, but also collaborate when necessary to achieve deep mining and efficient processing of input data.

[0216] Next, we enter the core stage of algorithm processing. Y feature operation subnetworks begin to play their key roles, each combining and integrating X authentication trend representation knowledge and X adversarial trend representation knowledge in a unique way. This process is not a simple splicing or weighting, but an advanced feature extraction and fusion technology based on deep learning algorithms, aiming to generate discriminative representation knowledge chains with strong discriminative power. These knowledge chains not only contain the core information of the original data, but are also optimized and enhanced through the internal mechanism of the algorithm, so that they can better serve the subsequent parameter focusing processing.

[0217] After successfully generating the discriminant representation knowledge chain, the algorithm enters the next key step: using these knowledge chains to perform parameter focusing on the status indicator identification results. Parameter focusing is a crucial part of power management diagnosis. It requires the algorithm to accurately capture the key parameters closely related to the power status from a large amount of data, and to effectively integrate and analyze them. In this embodiment, this task is undertaken by Y feature operation subnets, each of which performs in-depth parameter focusing on the status indicator identification results based on the information in the discriminant representation knowledge chain. In this process, the algorithm will make full use of the authentication trend and confrontation trend information in the discriminant representation knowledge chain to ensure the accuracy and reliability of parameter focusing.

[0218] After parameter focusing, the Y feature operation subnetworks will output their corresponding local parameter focusing representation knowledge. These local knowledge are the intermediate results generated by the algorithm during the processing. Although they are independent, they together constitute a comprehensive description of the state indicator recognition results. In order to further improve the accuracy and efficiency of diagnosis, the algorithm will perform knowledge interaction processing on these local parameter focusing representation knowledge.

[0219] Knowledge interaction is a complex and delicate process, which requires the algorithm to achieve effective communication and integration between local knowledge while maintaining the independence of local knowledge. In this embodiment, this task is achieved by building a parameter interaction knowledge chain. The parameter interaction knowledge chain is a dynamically updated data structure that can automatically adjust its structure and content according to the changes in the local parameter focus representation knowledge. Through knowledge interaction processing, the algorithm can make full use of the complementarity between local knowledge, thereby obtaining a more comprehensive and accurate parameter focus result.

[0220] After obtaining the parameter interaction knowledge chain, the algorithm will perform knowledge collision with the parameter interaction knowledge quantification relationship network. The parameter interaction knowledge quantification relationship network is an important soft threshold in the parameter focus branch, which defines the rules and conditions for knowledge collision. Through knowledge collision processing, the algorithm can further refine and optimize the information in the parameter interaction knowledge chain, thereby obtaining the basic parameter focus representation knowledge chain. This step is a key turning point in the algorithm processing process, which marks the algorithm's transition from the integration and interaction of local knowledge to the refinement and optimization of global knowledge.

[0221] Finally, the algorithm will calculate the basic parameter focused characterization knowledge chain layer by layer. This step is the final stage of the algorithm processing and is also the key step to obtain the parameter focused characterization knowledge chain corresponding to the state indicator identification result. Through layer-by-layer calculation processing, the algorithm can make full use of the information in the basic parameter focused characterization knowledge chain to obtain a comprehensive, accurate and reliable description of the state indicator identification result. This result not only contains detailed information about the power supply status, but also reflects all the optimizations and adjustments made by the algorithm during the processing process. It is an indispensable and important basis for power management diagnosis.

[0222] In some preferred embodiments, the generating, based on the parameter focused representation knowledge chain, X authentication discrimination probabilities corresponding to one authentication trend representation knowledge among the X authentication trend representation knowledge, and X adversarial discrimination probabilities corresponding to one adversarial trend representation knowledge among the X adversarial trend representation knowledge, respectively, comprises:

[0223] (1) Determining the focus weights corresponding to the X authentication trend representation knowledge in the parameter focus representation knowledge chain as X authentication trend description values, and weighting the X authentication trend description values ​​with preset discrimination coefficients to obtain authentication discrimination probabilities corresponding to the X authentication trend representation knowledge respectively;

[0224] (2) Determine the focus weights corresponding to the X confrontation trend representation knowledge in the parameter focus representation knowledge chain as X confrontation trend description values, and perform weighted processing on the X confrontation trend description values ​​and the preset discrimination coefficients respectively to obtain confrontation discrimination probabilities corresponding to the X confrontation trend representation knowledge.

[0225] Furthermore, this technical solution proposes a sophisticated and efficient solution to the problem of how to generate the discrimination probabilities corresponding to X authentication trend representation knowledge and X confrontation trend representation knowledge based on the parameter focus representation knowledge chain. This solution aims to accurately calculate the discrimination probability corresponding to each authentication trend representation knowledge and confrontation trend representation knowledge by deeply analyzing the information in the parameter focus representation knowledge chain, thereby providing a strong basis for subsequent decision support.

[0226] First, it is necessary to clarify the core position of the parameter focus characterization knowledge chain. This knowledge chain is obtained after a series of complex processing, and it contains rich information about power status, authentication trends, and confrontation trends. In order to make full use of this information, this embodiment extracts the focus weights related to X authentication trend characterization knowledge in the parameter focus characterization knowledge chain as X authentication trend description values. These description values ​​not only reflect the importance of each authentication trend in the whole, but also reflect their close connection with the power status.

[0227] Next, in order to obtain the authentication discrimination probability corresponding to each authentication trend representation knowledge, this embodiment introduces a preset discrimination coefficient. This coefficient is preset based on actual conditions and experience, and is used to perform weighted processing on the authentication trend description value. Through weighted processing, the authentication trend description value can be converted into a more intuitive and comparable authentication discrimination probability. This process not only enhances the accuracy of the discrimination result, but also improves its interpretability in practical applications.

[0228] Similarly, for X antagonistic trend characterization knowledge, this embodiment also adopts the same processing strategy. First, the focus weights related to the antagonistic trend characterization knowledge are extracted from the parameter focus characterization knowledge chain as X antagonistic trend description values. These description values ​​also reflect the importance of each antagonistic trend in the whole and its relationship with the power state. Then, the antagonistic trend description values ​​are weighted using the preset discrimination coefficient to obtain the antagonistic discrimination probability corresponding to each antagonistic trend characterization knowledge.

[0229] It is worth noting that the selection of the preset discriminant coefficient has an important impact on the accuracy of the discriminant probability. In practical applications, this coefficient needs to be reasonably set according to specific circumstances and needs. For example, factors such as the actual needs of power management, the distribution characteristics of data, and the performance of the algorithm can be considered. By carefully selecting and adjusting the preset discriminant coefficient, the discriminant probability can be made more in line with the actual situation, thereby improving the accuracy and reliability of decision support.

[0230] In addition, when dealing with power management problems of different scales, we can adapt to different application scenarios and needs by adjusting the generation method of parameter focus characterization knowledge chain, optimizing the setting of preset discrimination coefficients, and other strategies. This makes this technical solution more widely applicable and more practical in practical applications.

[0231] In order to further improve the accuracy and reliability of the discrimination probability, this embodiment can also be combined with other advanced technologies for joint processing. For example, a machine learning algorithm can be used to deeply mine and analyze the parameter focus characterization knowledge chain to discover the hidden laws and patterns therein; or big data analysis technology can be used to efficiently process and analyze massive data to extract more comprehensive and accurate information. The combination of these technologies can make the generation process of the discrimination probability more intelligent and automated, thereby improving the efficiency and accuracy of power management diagnosis.

[0232] In the specific implementation process, this embodiment also focuses on the stability and robustness of the algorithm. By adopting advanced numerical calculation methods and optimization algorithms, it can be ensured that there will be no problems such as numerical instability or overflow when calculating the discrimination probability. At the same time, by introducing fault-tolerant mechanisms and exception handling strategies, the algorithm can maintain stable and reliable performance when facing abnormal situations.

[0233] In some exemplary formulas, a parameter-focused characterization knowledge chain is defined as K, which includes information related to power status, authentication trend, and confrontation trend.

[0234] For X authentication trend representation knowledge, the corresponding focus weight is extracted from K and recorded as the authentication trend description value D_auth_i (where i represents the i-th authentication trend and the value range of i is 1 to X).

[0235] Similarly, for X adversarial trend representation knowledge, the corresponding focus weights are extracted from K and recorded as adversarial trend description values ​​D_adv_j (where j represents the jth adversarial trend and the value range of j is 1 to X).

[0236] Next, the preset discriminant coefficients R1 (for authentication trend) and R2 (for countering trend) are introduced. These coefficients are pre-set based on actual conditions and experience.

[0237] Using these coefficients, we can calculate the authentication discrimination probability P_auth_i corresponding to each authentication trend representation knowledge and the adversarial discrimination probability P_adv_j corresponding to each adversarial trend representation knowledge. The specific calculation formula is as follows:

[0238] P_auth_i=R1*D_auth_i

[0239] P_adv_j=R2*D_adv_j

[0240] Among them, the selection of R1 and R2 is crucial to the accuracy of the judgment probability. They need to be reasonably set according to factors such as the actual needs of power management, the distribution characteristics of data, and the performance of the algorithm.

[0241] Through the above formula, the discrimination probability corresponding to each authentication trend representation knowledge and confrontation trend representation knowledge can be accurately calculated, providing a strong basis for subsequent decision support.

[0242] In some alternative embodiments, the X preset fault diagnosis viewpoints include historical fault diagnosis viewpoints, the X authentication discrimination probabilities include historical authentication discrimination probabilities corresponding to the historical fault diagnosis viewpoints, and the X adversarial discrimination probabilities include historical adversarial discrimination probabilities corresponding to the historical fault diagnosis viewpoints; and determining, based on the X authentication discrimination probabilities and the X adversarial discrimination probabilities, a target fault diagnosis viewpoint that is related to the state monitoring information from the X preset fault diagnosis viewpoints comprises: if the historical authentication discrimination probability is greater than the historical adversarial discrimination probability, determining the historical fault diagnosis viewpoint as a target fault diagnosis viewpoint that is related to the state monitoring information; and if the historical authentication discrimination probability is not greater than the historical adversarial discrimination probability, determining that the historical fault diagnosis viewpoint is not related to the state monitoring information.

[0243] In this embodiment, for the problem of how to accurately identify the target fault diagnosis viewpoint that is directly related to the current state monitoring information from the pre-set X fault diagnosis viewpoints based on the calculated X authentication discrimination probabilities and X adversarial discrimination probabilities, the present technical scheme proposes a detailed and systematic solution: by comparing and analyzing the authentication discrimination probability and adversarial discrimination probability corresponding to each fault diagnosis viewpoint, it is determined whether the viewpoint is involved in the state monitoring information, thereby providing accurate guidance for subsequent fault diagnosis and maintenance decisions.

[0244] First, the X preset fault diagnosis viewpoints in this embodiment are a set of multiple possible fault causes, and these viewpoints may be constructed based on historical data, expert experience or theoretical models. Among them, it is particularly worth mentioning that the historical fault diagnosis viewpoints represent the fault modes that have appeared in the past and have been confirmed, and are the key reference objects in this diagnosis. Correspondingly, the X authentication discrimination probabilities and the X adversarial discrimination probabilities respectively reflect the authentication degree and adversarial degree of these fault diagnosis viewpoints under the current state monitoring information, which are important bases for judging whether a fault exists and its type.

[0245] In the specific implementation process, this embodiment first analyzes each historical fault diagnosis viewpoint one by one. For any historical fault diagnosis viewpoint, its corresponding historical authentication discrimination probability and historical confrontation discrimination probability are obtained through the aforementioned parameter focus representation knowledge chain extraction, focus weight calculation, and preset discrimination coefficient weighting steps. These two probability values ​​respectively represent the possibility of the fault diagnosis viewpoint being authenticated as correct and the possibility of being confronted (i.e., denied) in the current state.

[0246] Next, this embodiment adopts a direct and effective comparison strategy to determine the relevance of historical fault diagnosis viewpoints and state monitoring information. Specifically, if the historical authentication discrimination probability corresponding to a historical fault diagnosis viewpoint is greater than its historical adversarial discrimination probability, then it can be considered that the viewpoint has a high credibility in the current state, that is, it is directly involved with the state monitoring information. Therefore, the historical fault diagnosis viewpoint is determined as the target fault diagnosis viewpoint and is the focus of subsequent fault diagnosis and maintenance work.

[0247] On the contrary, if the historical authentication discrimination probability is not greater than the historical adversarial discrimination probability, it means that the credibility of the historical fault diagnosis viewpoint is low in the current state, and it may not be directly involved in the state monitoring information, or its corresponding fault mode has not occurred in the current situation. In this case, this embodiment will exclude the historical fault diagnosis viewpoint from the candidate list of the target fault diagnosis viewpoint to avoid misleading subsequent work.

[0248] It is worth noting that this embodiment does not simply adopt a single threshold or fixed judgment standard when judging the relevance of historical fault diagnosis viewpoints and status monitoring information, but performs dynamic analysis based on the specific circumstances of each fault diagnosis viewpoint. This flexible processing method not only improves the accuracy of diagnosis, but also enhances the adaptability and robustness of the solution.

[0249] In addition, in order to further improve the efficiency and accuracy of fault diagnosis, this embodiment can also be combined with other relevant information for comprehensive analysis. For example, factors such as the timeliness, completeness, and correlation of the status monitoring information with other data can be considered to assist in determining the relevance of the fault diagnosis viewpoint. At the same time, advanced technologies such as machine learning and data mining can also be used to conduct in-depth analysis of the parameter focus characterization knowledge chain to discover the hidden laws and patterns therein, providing more comprehensive support for fault diagnosis.

[0250] In practical applications, the technical solution proposed in this embodiment has significant advantages and effects. First, by comparing and analyzing the authentication discrimination probability and the adversarial discrimination probability, the target fault diagnosis viewpoints that are involved in the state monitoring information can be accurately identified, avoiding the misjudgment and missed judgment problems that may occur in traditional methods. Secondly, the solution adopts the strategy of dynamic analysis and comprehensive judgment to improve the accuracy and adaptability of fault diagnosis. Finally, combined with other relevant information and the support of advanced technologies, the efficiency and accuracy of fault diagnosis can be further improved, providing strong guarantees for the stable operation and timely maintenance of equipment.

[0251] In order to more specifically illustrate the technical solution of this embodiment, an actual power management scenario is used as an example for explanation below. It is assumed that in a certain power management system, multiple fault diagnosis viewpoints are pre-set, including overcurrent fault, overvoltage fault, short circuit fault, etc. Through the extraction and processing of parameter-focused characterization knowledge chains, the authentication discrimination probability and adversarial discrimination probability corresponding to each fault diagnosis viewpoint are obtained. Next, these probability values ​​are analyzed one by one to determine the relevance of each fault diagnosis viewpoint to the current state monitoring information.

[0252] Taking overcurrent fault as an example, if its corresponding authentication discrimination probability is greater than the adversarial discrimination probability, then it can be considered that overcurrent fault is one of the most likely fault causes in the current state. Therefore, overcurrent fault is determined as the target fault diagnosis point of view, and corresponding measures are taken to troubleshoot and repair it. Similarly, for other fault diagnosis points of view, the same method can be used for analysis and judgment.

[0253] In summary, by comparing and analyzing the authentication discrimination probability and the adversarial discrimination probability, the target fault diagnosis viewpoints that are involved in the condition monitoring information are accurately identified in the pre-set fault diagnosis viewpoints. This scheme not only improves the accuracy and efficiency of fault diagnosis, but also provides strong support for the stable operation and timely maintenance of equipment.

[0254] In other optional embodiments, the state indicator identification of the state monitoring information of the multifunctional remote power supply to obtain the state indicator identification result includes: acquiring the state monitoring information of the multifunctional remote power supply, and inputting the state monitoring information of the multifunctional remote power supply into the target power management diagnosis algorithm; the target power management diagnosis algorithm includes an output state identification model, a frequency state identification model and an operating load identification model; mining the power output data in the state monitoring information through the output state identification model to obtain output state linear knowledge; mining the frequency record data in the state monitoring information through the frequency state identification model to obtain frequency state linear knowledge; mining the operating load data in the state monitoring information through the operating load identification model to obtain operating load linear knowledge; generating the state indicator identification result based on the output state linear knowledge, the frequency state linear knowledge and the operating load linear knowledge.

[0255] It can be understood that this technical solution proposes a comprehensive solution for efficient and accurate status indicator identification of the status monitoring information of the multifunctional remote power supply. The core of this solution is to respectively mine different dimensional data in the status monitoring information by constructing and applying three specialized neural network models: output status identification model, frequency status identification model and operating load identification model, and then generate detailed status indicator identification results. The following is a detailed description of the technical solution.

[0256] 1. This embodiment first obtains the status monitoring information of the multifunctional remote power supply, which usually includes power supply output data, frequency recording data, and operating load data. Subsequently, this information is input into a designed target power management diagnosis algorithm. The algorithm consists of three key models, which are responsible for processing different types of data and extracting valuable linear knowledge (i.e., status indicator features) from them. Finally, based on these linear knowledge, a comprehensive status indicator recognition result is generated to provide strong support for the maintenance and management of the power supply.

[0257] 2. Neural Network Model and Linear Knowledge Extraction

[0258] ① Output state recognition model and output state linear knowledge

[0259] Neural network algorithm example: The output state recognition model can use structures such as convolutional neural network (CNN) or recurrent neural network (RNN). Considering that the power output data may contain time series characteristics, RNN (or its variants such as LSTM, GRU) may be more appropriate. These networks can capture the time dependency in the data, thereby more accurately identifying the output state.

[0260] Output state linear knowledge: Output state linear knowledge mainly refers to the linear features contained in the power supply output data that are directly related to the power supply working state. These features may include the stability of the output voltage, the change trend of the output current, and the power factor. Through the mining of the output state recognition model, the inherent connection between these linear features and the power supply working state can be revealed, providing an important basis for state monitoring and fault diagnosis.

[0261] ②Frequency state recognition model and frequency state linear knowledge

[0262] Neural network algorithm example: The frequency state recognition model can also use RNN or its variants to process the time series information in the frequency record data. In addition, considering the periodic characteristics of frequency data, you can also try to use Fourier transform and other signal processing methods to assist model training and improve recognition accuracy.

[0263] Frequency state linear knowledge: Frequency state linear knowledge focuses on the stability, fluctuation range, and possible abnormal frequency components of the power supply output frequency. These linear characteristics are crucial for evaluating the stability and reliability of the power supply. Through in-depth mining of the frequency state recognition model, frequency anomalies can be discovered in a timely manner, providing strong support for the maintenance and debugging of the power supply.

[0264] ③ Operation load identification model and operation load linear knowledge

[0265] Neural network algorithm example: The operating load identification model can use structures such as deep neural network (DNN) or convolutional neural network (CNN). Considering that the operating load data may contain multi-dimensional features (such as load type, load size, load change trend, etc.), these networks can effectively capture and process these complex features.

[0266] Operation load linearity knowledge: Operation load linearity knowledge mainly involves the load conditions that the power supply bears during operation, including the type, size, change trend, and possible load mutations of the load. These linear characteristics are of great significance for evaluating the load capacity of the power supply, optimizing the power supply configuration, and predicting future load demand. Through the accurate identification of the operation load identification model, a scientific basis can be provided for the scheduling and management of the power supply.

[0267] To facilitate understanding of the above technical solution, exemplary details of three types of linear knowledge are listed below.

[0268] 1) Output state linear knowledge mainly focuses on the output characteristics of the power supply, which can be represented by a series of numerical feature vectors.

[0269] Output voltage stability characteristic vector: average output voltage (V_avg); output voltage standard deviation (V_std); maximum output voltage (V_max); minimum output voltage (V_min); output voltage fluctuation range (V_range=V_max-V_min).

[0270] Output current change trend characteristic vector: average output current (I_avg); output current standard deviation (I_std); output current change rate (dI / dt, which can be obtained by time series analysis of current data); output current peak value (I_peak); output current valley value (I_valley).

[0271] Power factor characteristic vector: average power factor (PF_avg); power factor standard deviation (PF_std); minimum power factor (PF_min, reflecting the efficiency of the power supply under the worst conditions)

[0272] 2) Frequency state linearity knowledge mainly focuses on the stability and fluctuation of the power supply output frequency.

[0273] Output frequency stability characteristic vector: average output frequency (f_avg); output frequency standard deviation (f_std); maximum output frequency (f_max); minimum output frequency (f_min); output frequency fluctuation range (f_range=f_max-f_min).

[0274] Frequency fluctuation characteristic vector: frequency fluctuation period (T_fluctuation, reflecting the periodicity of frequency fluctuation); frequency fluctuation amplitude (A_fluctuation, reflecting the amplitude of frequency fluctuation); frequency jitter number (N_jitter, reflecting the number of times the frequency changes rapidly in a short period of time).

[0275] 3) Operating load linearity knowledge mainly focuses on the load conditions that the power supply bears during operation.

[0276] Load type characteristic vector: resistive load ratio (R_ratio, reflecting the ratio of resistive components in the load); inductive load ratio (L_ratio, reflecting the ratio of inductive components in the load); capacitive load ratio (C_ratio, reflecting the ratio of capacitive components in the load); non-linear load ratio (NL_ratio, reflecting the ratio of non-linear components in the load).

[0277] Load size feature vector: average load power (P_avg); load power standard deviation (P_std); load power peak (P_peak); load power valley (P_valley).

[0278] Load change trend feature vector: load change rate (dP / dt, which can be obtained by time series analysis of load data); number of load surges (N_surge, reflecting the number of times the load changes rapidly in a short period of time); load duration (T_duration, reflecting the duration of the load at a certain level)

[0279] 3. Generation of state indicator identification results

[0280] After extracting the output state linear knowledge, frequency state linear knowledge and operating load linear knowledge through the above three models, this technical solution integrates and analyzes these linear knowledge. By comparing, correlating and fusing linear features of different dimensions, more comprehensive and accurate state indicator identification results can be generated. These results can not only help operation and maintenance personnel to understand the working status of the power supply in a timely manner, but also provide strong support for power supply fault diagnosis, preventive maintenance and performance optimization.

[0281] By constructing and applying the output state recognition model, frequency state recognition model and operating load recognition model, the comprehensive mining and accurate recognition of multifunctional remote power supply state monitoring information is realized. By extracting different types of linear knowledge, detailed state indicator recognition results are generated, providing strong support for the maintenance and management of power supply.

[0282] In some other optional embodiments, the mining of fault trends for X preset fault diagnosis viewpoints to obtain corresponding X fault trend representation knowledge includes: acquiring X preset fault diagnosis viewpoints, and inputting the X preset fault diagnosis viewpoints into a target power management diagnosis algorithm; the target power management diagnosis algorithm includes a fault hidden danger identification model; performing diagnostic viewpoint annotation on the X preset fault diagnosis viewpoints to obtain corresponding X diagnostic annotation feature chains; a diagnostic annotation feature in the diagnostic annotation feature chain refers to a field representation obtained after a preset fault diagnosis viewpoint is labeled; based on the fault hidden danger identification model, performing feature deduction on the X diagnostic annotation feature chains to obtain abnormal state development information corresponding to the X diagnostic annotation feature chains, and mining the X abnormal state development information to obtain fault trend representation knowledge corresponding to the X preset fault diagnosis viewpoints.

[0283] In the technical field of fault diagnosis and prediction of complex power management systems, accurately identifying and predicting fault trends is the key to improving system stability and maintenance efficiency. To achieve this goal, this embodiment proposes an innovative technical solution that conducts in-depth analysis of X preset fault diagnosis viewpoints to mine their potential fault trends, thereby generating corresponding X fault trend characterization knowledge. The implementation of this process not only relies on efficient data processing and algorithm design, but also combines advanced fault potential identification models to achieve accurate prediction of the fault development trend of the power management system.

[0284] First, the starting point of this embodiment is to obtain X preset fault diagnosis viewpoints. These viewpoints are usually derived from historical fault data, expert experience or theoretical analysis, and are preliminary judgments on the types of faults that may occur in the power management system. In order to ensure the comprehensiveness and accuracy of these viewpoints, they need to be strictly screened and verified. For example, the specific value of X is 10, that is, there are 10 preset diagnostic viewpoints based on different fault modes, which cover a variety of possible fault conditions from voltage anomalies, current overloads to battery aging.

[0285] Next, these 10 preset fault diagnosis viewpoints are input into the target power management diagnosis algorithm. This algorithm is the core of the embodiment. It integrates a fault potential identification model, which is built based on machine learning or deep learning technology. It can identify early signs and potential development trends of faults through in-depth analysis of input data. The design of the target power management diagnosis algorithm takes into account the complexity and dynamics of the power management system, and can process a large amount of real-time monitoring data and extract useful information for fault diagnosis and prediction.

[0286] When processing these 10 preset fault diagnosis viewpoints, the algorithm first performs the diagnostic viewpoint labeling operation. The purpose of this step is to convert each diagnostic viewpoint into a series of features with clear meanings, which can reflect different aspects of the fault, such as fault type, frequency of occurrence, impact, etc. Through the labeling operation, each preset fault diagnosis viewpoint is converted into a diagnostic labeling feature chain, and each diagnostic labeling feature in the chain is a specific description of a certain aspect of the original viewpoint. For example, for a diagnostic viewpoint about voltage anomaly, its diagnostic labeling feature chain may include features such as "voltage fluctuation range", "abnormal duration" and "historical occurrence times".

[0287] After completing the annotation of diagnostic viewpoints, the next step is to perform feature deduction on these 10 diagnostic annotation feature chains based on the fault potential identification model. Feature deduction is a complex data processing process that uses the algorithmic logic within the model to perform in-depth analysis and calculation on the input feature chain to reveal the fault development trend hidden behind the data. In this process, the model considers multiple factors, such as the relationship between features, the pattern of historical data, and the current state of the system, to generate abnormal state development information corresponding to each diagnostic annotation feature chain. This information describes the possible path of the fault from the initial state to the final occurrence, as well as the key nodes and turning points that may appear in the process.

[0288] Taking the diagnostic annotation feature chain of voltage anomalies as an example, the fault potential identification model may analyze the historical data of the voltage fluctuation range, identify the trend of gradually increasing fluctuation amplitude, and combine the extension of the abnormal duration to predict the possible voltage out-of-control events in the future. At the same time, the model will also consider the impact of the historical occurrence frequency on the fault probability, as well as the changes in the current system load and environmental conditions, so as to obtain a comprehensive abnormal state development information.

[0289] Finally, the 10 abnormal state development information obtained is mined to extract the fault trend characterization knowledge corresponding to each fault diagnosis viewpoint. This step is a summary and improvement of all the previous work. It generates fault trend prediction results with practical guiding significance through in-depth analysis and interpretation of abnormal state development information. These results not only include the possibility and time range of fault occurrence, but also provide the possible impact of the fault and suggestions for countermeasures. For example, for the diagnosis viewpoint of voltage anomaly, the fault trend characterization knowledge may indicate that the risk of voltage out of control will gradually increase in the next three months, and it is recommended to take measures to strengthen monitoring and preventive maintenance to avoid the occurrence of faults.

[0290] Throughout the implementation process, data accuracy and algorithm reliability are critical. To ensure the accuracy of the results, the input diagnostic viewpoints need to be strictly verified and screened, and the fault potential identification model needs to be continuously optimized and updated to adapt to the continuous changes and upgrades of the power management system. In addition, an effective feedback mechanism needs to be established to promptly discover and correct possible problems in actual applications.

[0291] It can be seen that this embodiment obtains X preset fault diagnosis viewpoints and inputs them into the target power management diagnosis algorithm, uses the fault potential identification model to perform feature deduction on the diagnostic annotation feature chain, and finally mines the fault trend characterization knowledge corresponding to each fault diagnosis viewpoint. This process not only realizes the accurate prediction of the fault trend of the power management system, but also provides strong support for the maintenance and optimization of the system.

[0292] In some independent embodiments, the step of training the target power management diagnostic algorithm includes:

[0293] (1) obtaining condition monitoring training data information and prior training annotations, inputting the condition monitoring training data information into a basic power management diagnosis algorithm, and performing condition indicator recognition on the condition monitoring training data information through a condition indicator recognition branch of the basic power management diagnosis algorithm to obtain condition indicator training data; the basic power management diagnosis algorithm further includes a fault hidden danger recognition model, a feature adversarial processing branch, and a parameter focusing branch;

[0294] (2) obtaining a set of fault diagnosis viewpoints, mining the fault diagnosis viewpoints in the set of fault diagnosis viewpoints according to the fault potential identification model, and obtaining a description of the fault diagnosis viewpoints;

[0295] (3) performing feature confrontation processing on the fault diagnosis viewpoint description according to the feature confrontation processing branch to obtain a certification trend training feature corresponding to the fault diagnosis viewpoint description and a confrontation trend training feature corresponding to the fault diagnosis viewpoint description; the certification trend training feature includes a power supply device status record used to indicate that its corresponding fault diagnosis viewpoint is compatible with the state monitoring training data information, and the confrontation trend training feature includes a power supply device status record used to indicate that its corresponding fault diagnosis viewpoint is incompatible with the state monitoring training data information;

[0296] (4) Based on the parameter focusing branch, the authentication trend training feature and the adversarial trend training feature are combined into a discriminant representation knowledge chain, and the state indicator training data is subjected to parameter focusing processing through the discriminant representation knowledge chain to obtain a parameter focusing training feature chain, and the authentication training probability corresponding to the authentication trend training feature and the adversarial training probability corresponding to the adversarial trend training feature are generated based on the parameter focusing training feature chain; the parameter focusing training feature chain is used to represent a correlation feature graph between the discriminant representation knowledge chain and the state indicator training data;

[0297] (5) Generate a fault diagnosis opinion confidence based on the certification training probability and the adversarial training probability, generate an algorithm training error through the prior training annotations and the fault diagnosis opinion confidence, and optimize the algorithm weights of the feature adversarial processing branch and the parameter focusing branch of the basic power management diagnosis algorithm based on the algorithm training error, until the basic power management diagnosis algorithm meets the algorithm quality assessment requirements, and then determine the state indicator identification branch, the fault potential identification model, the optimized feature adversarial processing branch and the optimized parameter focusing branch as the target power management diagnosis algorithm.

[0298] In detail, the method of generating the fault diagnosis opinion confidence based on the certification training probability and the adversarial training probability includes: obtaining a learning rate, generating a first diagnostic opinion activation weight based on the certification training probability and the learning rate, and generating a second diagnostic opinion activation weight based on the adversarial training probability and the learning rate; the learning rate is a training basis indicating the timeliness of stability changes of the basic power management diagnostic algorithm; and generating the fault diagnosis opinion confidence based on the first diagnostic opinion activation weight and the second diagnostic opinion activation weight.

[0299] Furthermore, the method of generating an algorithm training error through the prior training annotations and the fault diagnosis viewpoint confidence includes: determining the fault diagnosis training viewpoint corresponding to the state monitoring training data information based on the certification training probability and the adversarial training probability; if the fault diagnosis training viewpoint matches the prior training annotation, determining the fault diagnosis viewpoint confidence as a first error quantization value, and generating an algorithm training error based on an error enhancement factor and the first error quantization value; the error enhancement factor is determined based on the number of certification trend training data, and the certification trend training data is the state monitoring training data information that carries the prior training annotation in the state monitoring training data information set; if the fault diagnosis training viewpoint does not match the prior training annotation, generating a second error quantization value based on the fault diagnosis viewpoint confidence, and generating an algorithm training error based on the error enhancement factor and the second error quantization value.

[0300] In the field of fault diagnosis of power management systems, it is crucial to build an efficient and accurate target power management diagnosis algorithm. The training process of this algorithm is a complex and delicate task, which involves the collaborative work of multiple branches and the in-depth analysis and processing of a large amount of data. The following will elaborate on the training steps of this target power management diagnosis algorithm to ensure that it can achieve accurate monitoring of the power supply device status and effective prediction of faults.

[0301] First, the training process begins with obtaining condition monitoring training data information and prior training annotations. Condition monitoring training data information is the basis for algorithm learning, which contains detailed records of power supply equipment under various operating conditions. This information may include data in multiple dimensions such as voltage, current, temperature, and their changing trends over time. Prior training annotations are prior annotations of these state data, which indicate which states are normal and which states may indicate the occurrence of faults.

[0302] Inputting these state monitoring training data information into the basic power management diagnosis algorithm is the first step in the training process. The basic power management diagnosis algorithm is a preliminary constructed model that contains multiple branches for processing different types of data and tasks. Among them, the state indicator identification branch is responsible for extracting key state indicators from the input state monitoring training data information. These indicators may be the fluctuation range of voltage, abnormal peak value of current, etc., which can directly reflect the operating status of the power supply equipment. Through the processing of the state indicator identification branch, the state indicator training data is obtained, which is an important input for the subsequent training steps.

[0303] In addition to the state indicator recognition branch, the basic power management diagnosis algorithm also includes a fault potential identification model, a feature adversarial processing branch, and a parameter focusing branch. The fault potential identification model is used to mine potential fault information. It may be based on machine learning or deep learning technology and can identify abnormal patterns in the data. The feature adversarial processing branch is responsible for feature adversarial processing of the input fault diagnosis viewpoints to generate certification trend training features and adversarial trend training features. These features represent the power supply equipment status records that are compatible and incompatible with the state monitoring training data information, respectively, and they provide positive and negative samples for the training of the algorithm.

[0304] After obtaining the state indicator training data and fault diagnosis view description, the next step is to perform feature adversarial processing on the fault diagnosis view description based on the feature adversarial processing branch. This process generates certification trend training features and adversarial trend training features by comparing and analyzing the differences and similarities between the fault diagnosis view and the state monitoring training data. These features not only reflect the degree of correlation between the fault diagnosis view and the actual state, but also provide an important basis for subsequent parameter focus processing.

[0305] Subsequently, based on the parameter focusing branch, the authentication trend training features and the adversarial trend training features are combined into a discriminative representation knowledge chain. The discriminative representation knowledge chain is a comprehensive feature representation that integrates both positive and negative information and can more comprehensively reflect the status of the power supply equipment. The parameter focusing training feature chain is obtained by performing parameter focusing on the state indicator training data through the discriminative representation knowledge chain. This feature chain represents the correlation feature graph between the discriminative representation knowledge chain and the state indicator training data, which provides a more accurate and comprehensive input for the training of the algorithm.

[0306] After obtaining the parameter-focused training feature chain, the next step is to generate the certification training probability corresponding to the certification trend training feature and the adversarial training probability corresponding to the adversarial trend training feature based on the feature chain. The certification training probability and adversarial training probability respectively reflect the degree of fit and misfit between the fault diagnosis viewpoint and the condition monitoring training data information. They are obtained by further analyzing and calculating the parameter-focused training feature chain, providing a key probability basis for the algorithm training.

[0307] In order to generate the confidence of the fault diagnosis viewpoint, it is necessary to obtain the learning rate, and generate the first diagnosis viewpoint activation weight based on the certification training probability and the learning rate, and generate the second diagnosis viewpoint activation weight based on the adversarial training probability and the learning rate. The learning rate is an important parameter in the algorithm training process, which controls the learning speed and stability of the algorithm. By combining with the learning rate, the certification training probability and the adversarial training probability can be converted into corresponding activation weights, which reflect the importance and credibility of the fault diagnosis viewpoint in the algorithm.

[0308] According to the activation weight of the first diagnostic viewpoint and the activation weight of the second diagnostic viewpoint, the confidence of the fault diagnosis viewpoint can be generated. Confidence is a comprehensive evaluation index that comprehensively considers both positive and negative information, as well as the impact of the learning rate on the stability of the algorithm. Through the calculation of confidence, the reliability of the fault diagnosis viewpoint can be quantitatively evaluated, providing a basis for subsequent algorithm optimization and decision-making.

[0309] After obtaining the confidence of the fault diagnosis viewpoint, the next step is to generate the algorithm training error through the prior training annotations and the fault diagnosis viewpoint confidence. This process is achieved by comparing and analyzing the differences and similarities between the prior training annotations and the fault diagnosis viewpoint confidence. If the fault diagnosis training viewpoint matches the prior training annotation, the fault diagnosis viewpoint confidence is determined as the first error quantization value, and the algorithm training error is generated based on the error enhancement factor and the first error quantization value. The error enhancement factor is a coefficient used to amplify or reduce the error, which is determined based on the number of certified trend training data. This ensures that the algorithm pays more attention to those fault diagnosis views that match the prior knowledge during the training process.

[0310] If the fault diagnosis training viewpoint does not match the prior training annotation, a second error quantization value is generated based on the fault diagnosis viewpoint confidence, and the algorithm training error is generated based on the error enhancement factor and the second error quantization value. The purpose of this step is to penalize those fault diagnosis views that do not match the prior knowledge, so as to encourage the algorithm to handle such views more cautiously in the subsequent training process.

[0311] Finally, the algorithm weights of the feature adversarial processing branch and parameter focusing branch of the basic power management diagnosis algorithm are optimized based on the generated algorithm training error. The optimization process is to adjust the parameters and weights in the algorithm to make the algorithm perform better on the training data. This step is iterated repeatedly until the basic power management diagnosis algorithm meets the algorithm quality assessment requirements, and then the state indicator identification branch, fault potential identification model, optimized feature adversarial processing branch and optimized parameter focusing branch are determined as the target power management diagnosis algorithm.

[0312] Throughout the training process, the stability and accuracy of the algorithm are crucial. To ensure this, the input data needs to be strictly screened and preprocessed to eliminate the influence of noise and outliers. At the same time, the algorithm needs to be continuously monitored and debugged to promptly discover and solve possible problems. In addition, in order to improve the performance and efficiency of the algorithm, some advanced optimization techniques and parallel computing strategies can also be used to accelerate the training and reasoning process of the algorithm.

[0313] To facilitate understanding of the above training process, an exemplary implementation is provided below for introduction.

[0314] ①Data preparation

[0315] Get the condition monitoring training data information Mx and prior training annotations My.

[0316] Mx is a matrix containing state features, and My is the class label vector corresponding to Mx.

[0317] ②Basic algorithm processing

[0318] Mx is input into the basic power management diagnostic algorithm.

[0319] The algorithm contains multiple branches: status indicator identification, fault potential identification, feature adversarial processing and parameter focusing.

[0320] ③ Status indicator identification

[0321] The key status indicator is extracted from Mx through the status indicator identification branch and output as Mx_indicator.

[0322] ④ Fault diagnosis viewpoint mining

[0323] Get the fault diagnosis viewpoint set V.

[0324] The fault potential identification model H_hidden is used to mine V, and the output is the description of fault diagnosis viewpoints.

[0325] ⑤ Feature adversarial processing

[0326] The feature adversarial processing branch is used to perform feature adversarial processing on the fault diagnosis view description and Mx.

[0327] The output is the authentication trend training feature F_auth and the adversarial trend training feature F_adv.

[0328] An exemplary formula is expressed as: (F_auth, F_adv) = f_adversarial(H_hidden(V), Mx)

[0329] ⑥Parameter focus processing

[0330] A parameter-focused branch is used to combine F_auth and F_adv into a discriminative representation knowledge chain.

[0331] Perform parameter focusing processing on Mx_indicator and output it as parameter focusing training feature chain C_focus.

[0332] An exemplary formula is expressed as: C_focus = f_focus (F_auth, F_adv, Mx_indicator)

[0333] ⑦Generate training probability

[0334] Based on C_focus, the authentication training probability P_auth corresponding to F_auth and the adversarial training probability P_adv corresponding to F_adv are generated.

[0335] The exemplary formula is: P_auth = g_auth (C_focus), P_adv = g_adv (C_focus)

[0336] ⑧ Generate fault diagnosis opinion confidence

[0337] Get the learning rate η.

[0338] The first diagnostic viewpoint activation weight W_auth is generated using P_auth and η.

[0339] The second diagnostic viewpoint activation weight W_adv is generated using P_adv and η.

[0340] The exemplary formula is: W_auth=η*P_auth, W_adv=η*P_adv

[0341] Generate the fault diagnosis opinion confidence Conf based on W_auth and W_adv.

[0342] The exemplary formula is: Conf = W_auth / (W_auth + W_adv)

[0343] ⑨Generate algorithm training error

[0344] Generate algorithm training error E using My and Conf.

[0345] If the fault diagnosis training viewpoint matches My, then E = λ*(1-Conf).

[0346] If there is no match, then E=λ*Conf.

[0347] λ is the error enhancement factor, which is determined according to the number of certification trend training data.

[0348] ⑩Algorithm optimization

[0349] E is used to optimize the algorithm weights of the feature adversarial processing branch and the parameter focusing branch of the basic power management diagnosis algorithm.

[0350] The optimization steps are repeated repeatedly until the algorithm meets the quality assessment requirements.

[0351] Through the above steps, an optimized target power management diagnosis algorithm can be obtained. Please note that in actual implementation, the specific form of each function (such as f_adversarial, f_focus, g_auth, g_adv) needs to be designed and adjusted according to the specific application scenario and data characteristics.

[0352] Further, Figure 2 FIG. 2 is a schematic diagram of a management and diagnosis system 200 provided by the present invention. Figure 2 The management and diagnosis system 200 shown includes a processor 210 , which can call and run a computer program from a memory to implement the method of the present invention.

[0353] Alternatively, if Figure 2 As shown, the management and diagnosis system 200 may further include a memory 230. The processor 210 may call and run a computer program from the memory 230 to implement the method of the present invention.

[0354] The memory 230 may be a separate device independent of the processor 210 , or may be integrated into the processor 210 .

[0355] Alternatively, if Figure 2 As shown, the management and diagnosis system 200 may further include a transceiver 220 , and the processor 210 may control the transceiver 220 to interact with other devices, specifically, to send information or data to other devices, or to receive information or data sent by other devices.

[0356] Optionally, the management and diagnosis system 200 may implement the corresponding processes corresponding to the storage engine or components (such as processing modules) in the storage engine or devices deployed with the storage engine in the various methods of the present invention, which will not be described in detail here for the sake of brevity.

[0357] It should be understood that the processor of the present invention may be an integrated circuit chip having signal processing capabilities.

[0358] It is understood that the memory in the present invention can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memory. It should be noted that the memory of the systems and methods described herein is intended to include but is not limited to suitable types of memory.

[0359] Based on the above, a readable storage medium is provided, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps of the above method are implemented.

[0360] Based on the same inventive concept, the present invention provides a multifunctional remote power management and diagnosis device, which is used for:

[0361] Performing status indicator recognition on status monitoring information of the multifunctional remote power supply to obtain status indicator recognition results;

[0362] Perform fault trend mining on X preset fault diagnosis viewpoints respectively to obtain corresponding X fault trend representation knowledge;

[0363] Performing feature confrontation processing on the X fault trend characterization knowledge respectively, obtaining X authentication trend characterization knowledge respectively corresponding to each fault trend characterization knowledge in the X fault trend characterization knowledge, and X confrontation trend characterization knowledge respectively corresponding to each fault trend characterization knowledge in the X fault trend characterization knowledge, X is a positive integer, each authentication trend characterization knowledge includes a power supply device state record used to indicate that its corresponding preset fault diagnosis viewpoint is adapted to the state monitoring information, and each confrontation trend characterization knowledge includes a power supply device state record used to indicate that its corresponding preset fault diagnosis viewpoint is not adapted to the state monitoring information;

[0364] Generate a discriminant representation knowledge chain according to the X authentication trend representation knowledge and the X confrontation trend representation knowledge, perform parameter focusing processing on the state indicator identification result through the discriminant representation knowledge chain, and obtain a parameter focusing representation knowledge chain; the parameter focusing representation knowledge chain is used to represent a correlation feature graph between the discriminant representation knowledge chain and the state indicator identification result;

[0365] Generate, based on the parameter focused representation knowledge chain, X authentication discrimination probabilities corresponding to one authentication trend representation knowledge among the X authentication trend representation knowledge, and X adversarial discrimination probabilities corresponding to one adversarial trend representation knowledge among the X adversarial trend representation knowledge;

[0366] According to the X authentication discrimination probabilities and the X adversarial discrimination probabilities, a target fault diagnosis viewpoint related to the state monitoring information is determined from the X preset fault diagnosis viewpoints; the authentication discrimination probability corresponding to the target fault diagnosis viewpoint is greater than the adversarial discrimination probability corresponding to the target fault diagnosis viewpoint.

[0367] It should be noted that, in this article, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the method and device in the embodiment of the present invention is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0368] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for enabling a terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0369] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the purpose of the present invention and the scope of protection of the present invention, all of which are within the protection of the present invention.

Claims

1. A multifunctional remote power management diagnosis method, characterized in that: The multifunctional remote power management diagnosis method is performed by a management diagnosis system, and the method includes: Performing status indicator recognition on status monitoring information of the multifunctional remote power supply to obtain status indicator recognition results; Perform fault trend mining on X preset fault diagnosis viewpoints respectively to obtain corresponding X fault trend representation knowledge; Performing feature confrontation processing on the X fault trend characterization knowledge respectively, obtaining X authentication trend characterization knowledge respectively corresponding to each fault trend characterization knowledge in the X fault trend characterization knowledge, and X confrontation trend characterization knowledge respectively corresponding to each fault trend characterization knowledge in the X fault trend characterization knowledge, X is a positive integer, each authentication trend characterization knowledge includes a power supply device state record used to indicate that its corresponding preset fault diagnosis viewpoint is adapted to the state monitoring information, and each confrontation trend characterization knowledge includes a power supply device state record used to indicate that its corresponding preset fault diagnosis viewpoint is not adapted to the state monitoring information; Generate a discriminant representation knowledge chain according to the X authentication trend representation knowledge and the X confrontation trend representation knowledge, perform parameter focusing processing on the state indicator identification result through the discriminant representation knowledge chain, and obtain a parameter focusing representation knowledge chain; the parameter focusing representation knowledge chain is used to represent a correlation feature graph between the discriminant representation knowledge chain and the state indicator identification result; Generate, based on the parameter focused representation knowledge chain, X authentication discrimination probabilities corresponding to one authentication trend representation knowledge among the X authentication trend representation knowledge, and X adversarial discrimination probabilities corresponding to one adversarial trend representation knowledge among the X adversarial trend representation knowledge; According to the X authentication discrimination probabilities and the X adversarial discrimination probabilities, a target fault diagnosis viewpoint related to the state monitoring information is determined from the X preset fault diagnosis viewpoints; the authentication discrimination probability corresponding to the target fault diagnosis viewpoint is greater than the adversarial discrimination probability corresponding to the target fault diagnosis viewpoint.

2. The method according to claim 1, characterized in that The performing feature confrontation processing on the X fault trend characterization knowledge respectively to obtain X authentication trend characterization knowledge respectively corresponding to each fault trend characterization knowledge in the X fault trend characterization knowledge and X confrontation trend characterization knowledge respectively corresponding to each fault trend characterization knowledge in the X fault trend characterization knowledge include: Inputting the X fault trend characterization knowledge into a target power management diagnosis algorithm; the target power management diagnosis algorithm includes a feature adversarial processing branch; Acquire the authentication feature variable and the confrontation feature variable in the feature confrontation processing branch; the authentication feature variable refers to the algorithm weight used to characterize the diagnosis hit result, the diagnosis hit result refers to the view feature used to characterize the adaptation of the fault diagnosis view to the state monitoring information, and the confrontation feature variable refers to the algorithm weight used to characterize the diagnosis disturbance result, and the diagnosis disturbance result refers to the view feature used to characterize the incompatibility of the fault diagnosis view to the state monitoring information; X authentication trend characterization knowledge corresponding to each of the X fault trend characterization knowledge are generated by using the authentication feature variable and the X fault trend characterization knowledge, and X adversarial trend characterization knowledge corresponding to each of the X fault trend characterization knowledge are generated by using the adversarial feature variable and the X fault trend characterization knowledge.

3. The method according to claim 1, characterized in that The generating of a discriminant representation knowledge chain based on the X authentication trend representation knowledge and the X confrontation trend representation knowledge, and performing parameter focusing processing on the state indicator identification result through the discriminant representation knowledge chain to obtain a parameter focusing representation knowledge chain includes: Inputting the X authentication trend characterization knowledge, the X confrontation trend characterization knowledge and the state indicator identification result into a target power management diagnosis algorithm; the target power management diagnosis algorithm includes a parameter focusing branch; According to the parameter focusing branch, a discriminant representation knowledge chain of an inter-class feature enhanced convolution kernel is generated through the X authentication trend representation knowledge and the X adversarial trend representation knowledge, an identification feature chain of the inter-class feature enhanced convolution kernel is generated according to the state indicator recognition result, and an abnormal tracing feature chain of the inter-class feature enhanced convolution kernel is generated according to the state indicator recognition result; Generate a basic parameter focused representation knowledge chain based on the discrimination representation knowledge chain, the identification feature chain, the abnormal tracing feature chain, and the feature recognition degree corresponding to the state indicator identification result, and calculate the basic parameter focused representation knowledge chain layer by layer to obtain a parameter focused representation knowledge chain corresponding to the state indicator identification result; The step of generating a discriminant characterization knowledge chain of an inter-class feature enhancement convolution kernel based on the X authentication trend characterization knowledge and the X adversarial trend characterization knowledge, generating an identification feature chain of the inter-class feature enhancement convolution kernel based on the status indicator recognition result, and generating an abnormal tracing feature chain of the inter-class feature enhancement convolution kernel based on the status indicator recognition result includes: Performing weighted processing on the discriminant feature value list in the parameter focusing branch, the X authentication trend representation knowledge, and the X adversarial trend representation knowledge to obtain a discriminant representation knowledge chain of the inter-class feature enhanced convolution kernel; Performing weighted processing on the identification feature value list in the parameter focusing branch and the state indicator recognition result to obtain the identification feature chain of the inter-class feature enhancement convolution kernel; The abnormal tracing feature value list in the parameter focusing branch and the state indicator recognition result are weighted to obtain the abnormal tracing feature chain of the inter-class feature enhanced convolution kernel; the discriminant feature value list, the identification feature value list and the abnormal tracing feature value list are all lists formed by the soft threshold in the parameter focusing branch; The generating of a basic parameter focused representation knowledge chain based on the discrimination representation knowledge chain, the identification feature chain, the abnormality tracing feature chain and the feature recognition degree corresponding to the state indicator recognition result includes: The abnormality tracing feature chain is updated to obtain a first risk mapping feature chain, a weighted processing result of the discrimination representation knowledge chain and the first risk mapping feature chain is determined as a second risk mapping feature chain, a third risk mapping feature chain is generated according to the feature recognition degree corresponding to the second risk mapping feature chain and the status indicator identification result, the third risk mapping feature chain is quantitatively corrected to obtain a quantitative correction array, and the weighted processing result of the quantitative correction array and the abnormality tracing feature is determined as a basic parameter focused representation knowledge chain.

4. The method according to claim 1, characterized in that The generating of a discriminant representation knowledge chain based on the X authentication trend representation knowledge and the X confrontation trend representation knowledge, and performing parameter focusing processing on the state indicator identification result through the discriminant representation knowledge chain to obtain a parameter focusing representation knowledge chain includes: Inputting the X authentication trend characterization knowledge, the X confrontation trend characterization knowledge and the state indicator identification result into a target power management diagnosis algorithm; the target power management diagnosis algorithm includes a parameter focusing branch, and the parameter focusing branch includes Y feature operation subnets; Y is a positive integer; The X authentication trend representation knowledge and the X confrontation trend representation knowledge are respectively combined into a discriminant representation knowledge chain through the Y feature operation subnets, and parameter focusing processing is performed on the discriminant representation knowledge chain and the state indicator identification result to obtain local parameter focusing representation knowledge corresponding to the Y feature operation subnets respectively; Knowledge interaction is performed on Y local parameter-focused representation knowledge to obtain a parameter-interaction knowledge chain, knowledge collision is performed on the parameter-interaction knowledge chain and the parameter-interaction knowledge quantification relationship network to obtain a basic parameter-focused representation knowledge chain, and layer-by-layer calculation is performed on the basic parameter-focused representation knowledge chain to obtain a parameter-focused representation knowledge chain corresponding to the state indicator identification result; the parameter-interaction knowledge quantification relationship network is a soft threshold in the parameter-focused branch.

5. The method according to claim 1, characterized in that The generating, based on the parameter focused representation knowledge chain, X authentication discrimination probabilities corresponding to one authentication trend representation knowledge among the X authentication trend representation knowledge, and X adversarial discrimination probabilities corresponding to one adversarial trend representation knowledge among the X adversarial trend representation knowledge, respectively, comprises: Determine the focus weights corresponding to the X authentication trend representation knowledge in the parameter focus representation knowledge chain as X authentication trend description values, and perform weighted processing on the X authentication trend description values ​​and preset discrimination coefficients respectively to obtain authentication discrimination probabilities corresponding to the X authentication trend representation knowledge respectively; The focus weights corresponding to the X confrontation trend characterization knowledge in the parameter focus characterization knowledge chain are determined as X confrontation trend description values, and the X confrontation trend description values ​​are weighted with the preset discrimination coefficients to obtain confrontation discrimination probabilities corresponding to the X confrontation trend characterization knowledge.

6. The method according to claim 1, characterized in that The X preset fault diagnosis viewpoints include historical fault diagnosis viewpoints, the X authentication discrimination probabilities include historical authentication discrimination probabilities corresponding to the historical fault diagnosis viewpoints, and the X adversarial discrimination probabilities include historical adversarial discrimination probabilities corresponding to the historical fault diagnosis viewpoints; and determining, based on the X authentication discrimination probabilities and the X adversarial discrimination probabilities, a target fault diagnosis viewpoint that is involved in the state monitoring information from among the X preset fault diagnosis viewpoints includes: If the historical authentication discrimination probability is greater than the historical adversarial discrimination probability, determining the historical fault diagnosis viewpoint as a target fault diagnosis viewpoint that is involved with the state monitoring information; If the historical authentication discrimination probability is not greater than the historical adversarial discrimination probability, it is determined that there is no involvement between the historical fault diagnosis viewpoint and the state monitoring information.

7. The method according to claim 1, characterized in that The state indicator identification is performed on the state monitoring information of the multifunctional remote power supply to obtain the state indicator identification result, including: Acquire status monitoring information of the multifunctional remote power supply, and input the status monitoring information of the multifunctional remote power supply into a target power supply management diagnosis algorithm; the target power supply management diagnosis algorithm includes an output state identification model, a frequency state identification model, and an operating load identification model; Mining the power supply output data in the state monitoring information through the output state recognition model to obtain output state linear knowledge; Mining the frequency record data in the state monitoring information through the frequency state recognition model to obtain frequency state linear knowledge; Mining the operating load data in the state monitoring information through the operating load identification model to obtain operating load linear knowledge; A state indicator identification result is generated according to the output state linear knowledge, the frequency state linear knowledge and the operation load linear knowledge.

8. The method according to claim 1, characterized in that The fault trend mining is performed on the X preset fault diagnosis viewpoints respectively to obtain corresponding X fault trend representation knowledge, including: Obtaining X preset fault diagnosis viewpoints, and inputting the X preset fault diagnosis viewpoints into a target power management diagnosis algorithm; the target power management diagnosis algorithm includes a fault hidden danger identification model; The X preset fault diagnosis viewpoints are respectively annotated with diagnostic viewpoints to obtain corresponding X diagnostic annotation feature chains; a diagnostic annotation feature in the diagnostic annotation feature chain refers to a field representation obtained after an annotation operation is performed on a preset fault diagnosis viewpoint; According to the fault hidden danger identification model, feature deduction is performed on the X diagnosis annotation feature chains to obtain abnormal state development information corresponding to the X diagnosis annotation feature chains respectively, and the X abnormal state development information is mined to obtain fault trend representation knowledge corresponding to the X preset fault diagnosis viewpoints respectively.

9. A management diagnosis system, characterized in that: The method comprises at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the method according to any one of claims 1 to 8.

10. A multifunctional remote power management and diagnostic device, characterized in that: The device is used for: Performing status indicator recognition on status monitoring information of the multifunctional remote power supply to obtain status indicator recognition results; Perform fault trend mining on X preset fault diagnosis viewpoints respectively to obtain corresponding X fault trend representation knowledge; Performing feature confrontation processing on the X fault trend characterization knowledge respectively, obtaining X authentication trend characterization knowledge respectively corresponding to each fault trend characterization knowledge in the X fault trend characterization knowledge, and X confrontation trend characterization knowledge respectively corresponding to each fault trend characterization knowledge in the X fault trend characterization knowledge, X is a positive integer, each authentication trend characterization knowledge includes a power supply device state record used to indicate that its corresponding preset fault diagnosis viewpoint is adapted to the state monitoring information, and each confrontation trend characterization knowledge includes a power supply device state record used to indicate that its corresponding preset fault diagnosis viewpoint is not adapted to the state monitoring information; Generate a discriminant representation knowledge chain according to the X authentication trend representation knowledge and the X confrontation trend representation knowledge, and perform parameter focusing processing on the state indicator identification result through the discriminant representation knowledge chain to obtain a parameter focusing representation knowledge chain; The parameter-focused representation knowledge chain is used to represent a correlation feature graph between the discriminant representation knowledge chain and the state indicator recognition result; Generate, based on the parameter focused representation knowledge chain, X authentication discrimination probabilities corresponding to one authentication trend representation knowledge among the X authentication trend representation knowledge, and X adversarial discrimination probabilities corresponding to one adversarial trend representation knowledge among the X adversarial trend representation knowledge; According to the X authentication discrimination probabilities and the X adversarial discrimination probabilities, a target fault diagnosis viewpoint related to the state monitoring information is determined from the X preset fault diagnosis viewpoints; the authentication discrimination probability corresponding to the target fault diagnosis viewpoint is greater than the adversarial discrimination probability corresponding to the target fault diagnosis viewpoint.

Images