Intelligent power distribution cabinet safety early warning method, system and device and storage medium

Abstract

The invention provides an intelligent power distribution cabinet safety early warning method, system and device and a storage medium, and relates to the field of electric power. The method comprises the steps of obtaining multi-source heterogeneous operation parameters of the intelligent power distribution cabinet, and performing feature extraction on the multi-source heterogeneous operation parameters to obtain operation feature parameters; performing parameter fusion on the operation characteristic parameters to obtain fused characteristic parameters, and inputting the fused characteristic parameters into a preset fault early warning model to obtain fault early warning data; and inputting the fault early warning data into a preset rule matching engine for rule matching to obtain a matching result, and generating an early warning signal according to the matching result. According to the method, the accuracy and timeliness of safety early warning of the intelligent power distribution cabinet are improved.

Description

Technical Field

[0001] The present application relates to the field of electric power, and specifically to a safety early warning method, system, equipment and storage medium for an intelligent power distribution cabinet. Background Art

[0002] Smart distribution cabinets are important equipment in smart grids, and they undertake multiple functions such as power supply, control, and protection. With the rapid development of smart grids and the increasingly complex power environment, the risk of failure faced by distribution cabinets is also increasing. In order to ensure the safety and stability of power supply, it is necessary to monitor and warn the operating status of distribution cabinets in real time to detect and deal with potential failure hazards in a timely manner.

[0003] At present, there have been some studies and applications on safety warnings for power distribution cabinets. A common method is to set a fixed threshold based on a single type of operating parameter, such as current, voltage, temperature, etc., and trigger an alarm when the parameter exceeds the threshold. This method is simple to implement, but it is difficult to adapt to complex and changeable working conditions and is prone to false alarms or missed alarms. Summary of the invention

[0004] The present application provides a smart power distribution cabinet safety warning method, system, device and storage medium, which can improve the accuracy of smart power distribution cabinet safety warning.

[0005] In a first aspect, the present application provides a smart power distribution cabinet safety early warning method, the method comprising: Acquire multi-source heterogeneous operation parameters of an intelligent power distribution cabinet, and perform feature extraction on the multi-source heterogeneous operation parameters to obtain operation feature parameters; the multi-source heterogeneous operation parameters include current parameters, voltage parameters, temperature parameters, humidity parameters, vibration parameters and partial discharge parameters; the feature extraction on the multi-source heterogeneous operation parameters to obtain operation feature parameters includes: performing spectrum analysis on the current parameters to obtain current time-frequency domain features; performing spectrum analysis on the voltage parameters to obtain voltage time-frequency domain features; performing abnormal trend extraction on the temperature parameters to obtain temperature abnormal trend features; performing abnormal trend extraction on the humidity parameters to obtain humidity abnormal trend features; performing time-frequency feature extraction on the vibration parameters to obtain vibration time-frequency domain features; performing feature extraction on the partial discharge parameters to obtain discharge pulse features, discharge mode features and discharge type features, and using the current time-frequency domain features, the voltage time-frequency domain features, the temperature abnormal trend features, the humidity abnormal trend features, the vibration time-frequency domain features, the discharge pulse features, the discharge mode features and the discharge type features as the operation feature parameters; Performing parameter fusion on the operation characteristic parameters to obtain fused characteristic parameters, and inputting the fused characteristic parameters into a preset fault warning model to obtain fault warning data; The fault warning data is input into a preset rule matching engine for rule matching to obtain a matching result, and a warning signal is generated according to the matching result.

[0006] By adopting the above technical solution, the multi-source heterogeneous operating parameters of the intelligent distribution cabinet are obtained, and the operating characteristic parameters are obtained by feature extraction of the multi-source heterogeneous operating parameters, so as to realize the monitoring of the operating status of the distribution cabinet. Since the operating status of the distribution cabinet is affected by many factors, a single type of parameter is difficult to fully reflect the health level of the equipment. Therefore, by comprehensively utilizing parameters of multiple dimensions such as current, voltage, temperature, humidity, vibration, partial discharge, etc., and adopting targeted feature extraction methods, the precursors of faults can be captured from different angles, and the accuracy and reliability of fault warning can be improved.

[0007] After obtaining the operating characteristic parameters, the present invention further performs parameter fusion on the operating characteristic parameters to obtain fused characteristic parameters. This step can remove the redundancy and noise between the characteristic parameters, extract more essential and stable fault indicators, and reduce the complexity and training difficulty of the subsequent warning model. Through parameter fusion, characteristic parameters of different types and scales are mapped to a unified feature space, which is convenient for comprehensive analysis and decision-making.

[0008] By inputting the fusion feature parameters into the preset fault warning model, fault warning data can be obtained. The fault warning model is trained based on a large amount of historical operation data and fault cases. It can automatically learn the mapping relationship between fault features and fault types, and has strong generalization and prediction capabilities. Through the fault warning model, early warning and positioning of faults can be achieved, providing a basis for subsequent maintenance and decision-making. Compared with the traditional threshold comparison method, the warning model based on machine learning can adapt to complex and changeable working conditions and has better robustness and adaptability.

[0009] The fault warning data is input into the preset rule matching engine for rule matching, the matching results are obtained, and the warning signal is generated according to the matching results. The rule matching engine is refined based on expert experience and typical case summaries, and expresses the correspondence between fault characteristics and fault types in the form of rules. Matching the fault warning data with the rules in the rule base can explain the warning results from the perspective of causal logic, enhancing the interpretability and credibility of the warning. At the same time, the results of rule matching can be used as a supplement and verification of the warning model to further improve the accuracy of the warning.

[0010] In a second aspect of the present application, a smart power distribution cabinet safety early warning method system is provided, comprising: A data acquisition module, used to acquire multi-source heterogeneous operation parameters of an intelligent power distribution cabinet, and perform feature extraction on the multi-source heterogeneous operation parameters to obtain operation feature parameters; the multi-source heterogeneous operation parameters include current parameters, voltage parameters, temperature parameters, humidity parameters, vibration parameters and partial discharge parameters; the feature extraction of the multi-source heterogeneous operation parameters to obtain operation feature parameters includes: performing spectrum analysis on the current parameters to obtain current time-frequency domain features; performing spectrum analysis on the voltage parameters to obtain voltage time-frequency domain features; performing abnormal trend extraction on the temperature parameters to obtain temperature abnormal trend features; performing abnormal trend extraction on the humidity parameters to obtain humidity abnormal trend features; performing time-frequency feature extraction on the vibration parameters to obtain vibration time-frequency domain features; performing feature extraction on the partial discharge parameters to obtain discharge pulse features, discharge mode features and discharge type features, and using the current time-frequency domain features, the voltage time-frequency domain features, the temperature abnormal trend features, the humidity abnormal trend features, the vibration time-frequency domain features, the discharge pulse features, the discharge mode features and the discharge type features as the operation feature parameters; A fault warning module is used to perform parameter fusion on the operation characteristic parameters to obtain fused characteristic parameters, and input the fused characteristic parameters into a preset fault warning model to obtain fault warning data; The early warning judgment module is used to input the fault early warning data into a preset rule matching engine for rule matching, obtain a matching result, and generate an early warning signal according to the matching result.

[0011] In a third aspect of the present application, a computer storage medium is provided, wherein the computer storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor and executing the above method steps.

[0012] In the fourth aspect of the present application, an electronic device is provided, including a processor, a memory, a user interface and a network interface, the memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the above method.

[0013] In summary, one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: 1. This application obtains multi-source heterogeneous operating parameters of the intelligent distribution cabinet, and extracts features from the multi-source heterogeneous operating parameters to obtain operating feature parameters, thereby achieving a comprehensive characterization of the operating status of the distribution cabinet. Since the operating status of the distribution cabinet is affected by many factors, a single type of parameter is difficult to fully reflect the health level of the equipment. By comprehensively utilizing parameters in multiple dimensions such as current, voltage, temperature, humidity, vibration, partial discharge, etc., and adopting targeted feature extraction methods, it is possible to capture the precursors of faults from different angles and improve the accuracy and reliability of fault warnings.

[0014] 2. After obtaining the operating characteristic parameters, the present invention further performs parameter fusion on the operating characteristic parameters to obtain fused characteristic parameters. This step can remove the redundancy and noise between the characteristic parameters, extract more essential and stable fault indicators, and reduce the complexity and training difficulty of the subsequent warning model. Through parameter fusion, characteristic parameters of different types and scales are mapped to a unified feature space, which is convenient for comprehensive analysis and decision-making.

[0015] 3. This application inputs the fusion feature parameters into the preset fault warning model to obtain fault warning data. The fault warning model is trained based on a large amount of historical operation data and fault cases. It can automatically learn the mapping relationship between fault characteristics and fault types, and has strong generalization and prediction capabilities. Through the fault warning model, early warning and positioning of faults can be achieved, providing a basis for subsequent maintenance and decision-making. Compared with the traditional threshold comparison method, the warning model based on machine learning can adapt to complex and changeable working conditions and has better robustness and adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of a process flow of a smart power distribution cabinet safety early warning method provided in an embodiment of the present application; Figure 2 An architecture diagram of a smart power distribution cabinet safety warning system provided in an embodiment of the present application; Figure 3 A schematic diagram of the structure of an electronic device provided in this application. DETAILED DESCRIPTION

[0017] In order to enable technicians in this field to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments.

[0018] In the description of the embodiments of the present application, words such as "for example" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "for example" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "for example" or "for example" is intended to present related concepts in a specific way.

[0019] In the description of the embodiments of the present application, the meaning of the term "multiple" refers to two or more. For example, multiple systems refer to two or more systems, and multiple screen terminals refer to two or more screen terminals. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. The terms "include", "comprise", "have" and their variations all mean "including but not limited to", unless otherwise specifically emphasized.

[0020] In order to facilitate understanding of the method and system provided by the embodiments of the present application, before introducing the embodiments of the present application, the background of the embodiments of the present application is first introduced.

[0021] As a key device in the smart grid, the smart distribution cabinet plays an irreplaceable role in power distribution, control and protection. In recent years, with the rapid development of smart grid technology and the increasingly complex power environment, the failure risk faced by the distribution cabinet is also increasing day by day. In order to ensure the safe and stable operation of the power system, there is an urgent need for an efficient and reliable distribution cabinet monitoring and early warning technology that can capture abnormal changes in the operating status of the distribution cabinet in real time, accurately identify and locate potential fault hazards, and thus win precious time for early prevention and rapid disposal of faults.

[0022] At present, domestic and foreign academic and industrial circles have carried out a series of research and practice on the safety warning of power distribution cabinets, and have made certain progress. Among them, a widely used method is the threshold comparison method based on a single operating parameter. This method collects and monitors key parameters such as current, voltage, and temperature of the power distribution cabinet in real time, and sets predetermined upper and lower threshold limits. Once the monitoring parameters exceed the threshold range, the system will automatically trigger an alarm to prompt the staff to take corresponding emergency measures. The advantages of this method are simple principle, easy implementation, and the ability to respond quickly to certain typical faults. However, with the complexity and diversification of the power distribution environment, the single parameter threshold comparison method has gradually exposed some limitations. First, the causes of distribution cabinet failures are complicated, and the signs of many failures are difficult to reflect through the abnormality of a single parameter, which can easily lead to missed reports; second, the operating conditions of the distribution cabinet are ever-changing, and fixed thresholds are difficult to adapt to all situations, especially when the threshold is not selected properly, it is easy to cause false alarms; third, a single parameter lacks horizontal comparison and comprehensive analysis, making it difficult to accurately identify and locate the fault type and location; finally, the threshold comparison method is a post-event passive response, lacks the exploration and prediction of early signs of failure, and is not conducive to the forward-looking prevention of failures.

[0023] After the background introduction of the above content, those skilled in the art can understand the problems existing in the prior art. The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0024] Based on the above background technology, please refer to Figure 1 , Figure 1 A flow chart of a safety early warning method for an intelligent power distribution cabinet provided in an embodiment of the present application. The system can be implemented by a computer program or can be run as an independent tool application. Specifically, in an embodiment of the present application, the method can be applied on a server, but can also be applied to electronic devices such as servers. A safety early warning method for an intelligent power distribution cabinet includes the following steps: S101, acquiring multi-source heterogeneous operating parameters of an intelligent power distribution cabinet, and performing feature extraction on the multi-source heterogeneous operating parameters to obtain operating feature parameters; Specifically, first obtain the multi-source heterogeneous operating parameters of the intelligent distribution cabinet. By installing various types of sensors and collection devices in various key parts of the distribution cabinet, such as the incoming cabinet, outgoing cabinet, bus cabinet, capacitor cabinet, etc., various operating parameters of the distribution cabinet are collected in real time. These operating parameters include but are not limited to: current, voltage, active power, reactive power, temperature, humidity, smoke concentration, SF6 gas pressure, partial discharge, vibration and noise. The frequency of collection can be set according to the characteristics of parameter changes and the degree of importance, generally ranging from once per second to once per minute. By comprehensively collecting the operating parameters of the distribution cabinet, a rich data foundation can be provided for subsequent feature extraction and fault warning.

[0025] After obtaining multi-source heterogeneous operating parameters, it is necessary to extract features from these raw data. This is because the original collected operating parameters usually have characteristics such as large data volume, information redundancy, and noise, which are not conducive to direct use for fault analysis and early warning. Through feature extraction, more essential and sensitive fault indicators can be mined from the raw data, reducing the data dimension and improving the efficiency and accuracy of subsequent analysis.

[0026] In this embodiment, a targeted feature extraction method is adopted according to the physical meaning and statistical characteristics of different operating parameters. For cyclic symmetric quantities such as current and voltage, their mean, peak value, effective value, distortion rate and other indicators are extracted; for trend quantities such as active power and reactive power, their mean, slope, volatility and other indicators are extracted; for slow variables such as temperature and humidity, their mean, standard deviation, slow change rate and other indicators are extracted; for pulse quantities such as partial discharge, their discharge times, discharge amplitude, discharge phase and other indicators are extracted; for waveform quantities such as vibration and noise, their spectrum characteristics, energy characteristics, time-frequency characteristics and other indicators are extracted. By comprehensively using multiple analysis methods such as time domain, frequency domain, time-frequency domain, etc., the operating status of the distribution cabinet is characterized from different dimensions, and a series of operating characteristic parameters are obtained.

[0027] For example, for current parameters, the harmonic components are calculated through Fourier transform to obtain the total harmonic distortion rate of current; the instantaneous characteristics are extracted through wavelet transform to obtain the current mutation index; the current imbalance index is obtained by calculating the root mean square value and peak value, etc. For another example, for partial discharge parameters, the insulation state is judged by the pulse number and amplitude, the discharge type is judged by the pulse phase, and the discharge severity is judged by the pulse spectrum.

[0028] Through the above feature extraction, the original multi-source heterogeneous operating parameters are converted into a set of operating characteristic parameters that reflect the health status of the distribution cabinet. These characteristic parameters are moderate in number and information is concentrated, with stronger fault sensitivity and early warning efficiency. Taking the current distortion rate as an example, its increase often indicates that there are nonlinear loads or harmonic pollution in the system, which is an important precursor to potential faults; for example, the frequent occurrence of partial discharge indicates that the insulation level has decreased, which may cause serious faults such as insulation breakdown. Therefore, the operating characteristic parameters can provide a more reliable and effective basis for the safety assessment and fault warning of the distribution cabinet.

[0029] Based on the above embodiment, as an optional embodiment, the multi-source heterogeneous operation parameters include current parameters, voltage parameters, temperature parameters, humidity parameters, vibration parameters and partial discharge parameters; the feature extraction of the multi-source heterogeneous operation parameters to obtain the operation feature parameters includes: Performing spectrum analysis on the current parameters to obtain current time-frequency domain characteristics; Performing spectrum analysis on the voltage parameters to obtain voltage time-frequency domain characteristics; Extracting abnormal trends of the temperature parameters to obtain abnormal temperature trend characteristics; Extracting abnormal trends of the humidity parameters to obtain abnormal humidity trend characteristics; Extracting time-frequency features of the vibration parameters to obtain vibration time-frequency domain features; Feature extraction is performed on the local discharge parameters to obtain discharge pulse features, discharge mode features and discharge type features, and the current time-frequency domain features, voltage time-frequency domain features, temperature anomaly trend features, humidity anomaly trend features, vibration time-frequency domain features, discharge pulse features, discharge mode features and discharge type features are used as the operating characteristic parameters.

[0030] Specifically, the multi-source heterogeneous operating parameters of the intelligent power distribution cabinet include current parameters, voltage parameters, temperature parameters, humidity parameters, vibration parameters, partial discharge parameters, etc. In order to extract more essential and sensitive fault features from these original parameters, it is necessary to perform targeted feature extraction on each type of parameter.

[0031] For current parameters and voltage parameters, since they are periodic electrical quantities, their time domain waveforms mainly reflect instantaneous characteristics such as amplitude and phase, while the frequency domain spectrum mainly reflects steady-state characteristics such as harmonics and interharmonics. In order to comprehensively characterize the time-varying and frequency-varying laws of current and voltage signals, this embodiment performs spectrum analysis on current parameters and voltage parameters respectively to obtain current time-frequency domain characteristics and voltage time-frequency domain characteristics. Specifically, time-frequency analysis methods such as short-time Fourier transform are used to convert current and voltage signals to a joint time-frequency plane to obtain a spectrum diagram that varies with time, from which indicators such as energy distribution, instantaneous frequency, and instantaneous amplitude of each frequency band are extracted to form a set of time-frequency domain feature vectors that comprehensively reflect the dynamic characteristics of the signal. These features not only include the steady-state components of the signal, but also the transient components, and can sensitively capture weak anomalies such as current and voltage distortion and flicker caused by faults.

[0032] For temperature parameters and humidity parameters, since they are slow variables, their anomalies are mainly manifested as slow trend changes, rather than rapid mutations or fluctuations. In order to promptly discover abnormal trends in temperature and humidity, it is necessary to extract trend features from time series data. This embodiment uses trend extraction algorithms such as sliding average and exponential smoothing to adaptively fit the temperature and humidity change trend curves, and calculates dynamic indicators such as the slope and acceleration of the curve. When the slope or acceleration exceeds the threshold, it is determined that there is an abnormal trend, and the temperature abnormal trend feature and humidity abnormal trend feature are generated. By analyzing the changing trend of the parameters, it is possible to sensitively detect small change signals in the early stage of temperature and humidity anomalies, thereby achieving early warning.

[0033] For vibration parameters, since they contain both periodic mechanical vibration components and random impact vibration components, a single time domain or frequency domain analysis is difficult to fully characterize their characteristics. Therefore, this embodiment extracts time-frequency features of vibration parameters to obtain vibration time-frequency domain features. Specifically, time-frequency analysis methods such as wavelet packet transform and empirical mode decomposition are used to adaptively decompose the vibration signal into multiple narrowband components, extract the time-frequency energy, instantaneous frequency, instantaneous amplitude and other indicators of each component, and then combine the characteristics of each component to obtain the overall time-frequency domain feature vector. Compared with a single time domain statistic or frequency domain indicator, the time-frequency domain features can more comprehensively and dynamically reflect the non-stationary characteristics and transient components of vibration, and more sensitively detect abnormal shocks and resonances caused by faults.

[0034] For partial discharge parameters, since they present typical pulse waveforms and complex discharge patterns, feature extraction is required from both waveform and pattern perspectives. On the one hand, characteristic parameters such as discharge amplitude, duration, rise time, and energy are extracted from the discharge pulse waveform to form discharge pulse features, which are used to characterize the severity of the discharge; on the other hand, pattern parameters such as discharge repetition rate, phase distribution, and energy distribution are extracted from the occurrence law and spatial distribution of the discharge pulse to form discharge pattern features, which are used to characterize the type and nature of the discharge. In addition, the existing discharge recognition model can be used to make a comprehensive judgment based on the characteristics of the discharge pulse and discharge pattern to obtain discharge type characteristics, such as corona discharge, suspended discharge, and surface discharge. By extracting partial discharge features from multiple angles and levels, the time domain and spatial characteristics of the discharge signal can be accurately characterized, and automatic recognition of the discharge type and online evaluation of the severity can be achieved.

[0035] Based on the above embodiment, as an optional embodiment, before obtaining the multi-source heterogeneous operation parameters of the intelligent power distribution cabinet, the following further includes: S201 obtains the topological structure and functional composition of the intelligent power distribution cabinet, and determines key monitoring points based on the topological structure and the functional composition; Specifically, the process of determining key monitoring points mainly includes two steps: topology analysis and function mapping. The first is topology analysis, that is, to fully obtain the internal composition and connection relationship of the intelligent distribution cabinet, and build a complete topology model on this basis. Generally, the graph theory modeling method is adopted to abstract each switch, bus, connection point, etc. as a node in the graph, and the electrical connection between them is abstracted as an edge in the graph, thereby forming a multi-voltage level and multi-branch tree structure. In the modeling process, in addition to describing the static physical connection, the influence of dynamic change factors such as switch state, busbar opening and closing, and backup power supply switching should be fully considered to characterize the topological heterogeneity of the distribution cabinet under different working conditions. At the same time, considering that some special components such as transformers, electric energy meters and other measurement and control units have no explicit electrical connections inside, they undertake indispensable measurement, control and other tasks, which are crucial to the status monitoring of the distribution cabinet. Therefore, when constructing the topology structure, it is also necessary to include it as a logical node to reveal its implicit association with other components. In this way, the internal organizational form of the distribution cabinet can be comprehensively and meticulously portrayed, and the configuration and action mechanism of each component can be intuitively displayed, facilitating further analysis in the future.

[0036] After obtaining the topological structure of the distribution cabinet, it is necessary to further identify and associate the functional attributes of each component, reveal its role in the distribution process, that is, implement the mapping of functional components. Generally, the intelligent distribution cabinet can be divided into different functional partitions such as incoming line, busbar, metering, feeder, capacitor, etc. Each partition undertakes different distribution tasks, and the importance of internal components is also different. For example, the incoming line cabinet is directly connected to the high-voltage power supply and is the total entrance of the power distribution, which has the greatest impact on the reliability of the power supply; the busbar cabinet is the link connecting the two sections of the bus, which is crucial to the flexibility of the power supply; the metering cabinet has built-in metering devices such as electric energy meters and mutual inductors, which are crucial to the metering accuracy and errors; the feeder cabinet is the outlet connecting the user load, which has the greatest impact on the terminal power supply quality. Therefore, it is necessary to fully consider the task attributes of different functional partitions, and assign corresponding functional labels to each node in the topological diagram accordingly, so as to form a logical topology that matches the physical structure and integrates with the business semantics. In this way, the static composition of the distribution system can be clearly portrayed, and the functional changes of each component with the operating status can be dynamically reflected, so that the topological structure is closely integrated with the system behavior, providing a unified semantic basis for state perception.

[0037] Based on the system analysis of the above topological structure and functional composition, this embodiment further adopts a graph-based key node mining algorithm to screen out a group of key monitoring points that have the greatest impact on the safe operation of the distribution cabinet. The basic idea is to evaluate the centrality of each node in the topological graph, that is, its importance in the graph. The higher the centrality, the more critical the role of the node in the network, and the greater the impact of its state change on the entire system. In order to reflect the importance of the node from multiple angles, the algorithm comprehensively considers the following three centrality indicators: one is the connection centrality, which reflects the connectivity of the node in the graph and is proportional to the number of edges associated with it; the second is the betweenness centrality, which reflects the propagation control ability of the node in the graph and is proportional to the number of shortest paths passing through the node; the third is the connectivity centrality, which reflects the contribution of the node to the connectivity of the graph and is proportional to the increase in the number of connected components caused by removing the node. The weighted sum of the above indicators can obtain a comprehensive node importance. On this basis, the top N% of nodes are selected as the key monitoring points for the preliminary selection in order of importance from high to low.

[0038] S202, determining a fault propagation path based on the topological structure and the functional composition, and determining a key monitoring point among the key monitoring points based on the fault propagation path; Specifically, after clarifying the propagation path of the fault, this embodiment further combines the functional composition of the distribution cabinet to carry out the importance ranking of the key nodes in the propagation path, thereby determining the key monitoring points. The reason for introducing functional attributes is that components with different functions have different positions in the power distribution process, and their failure consequences are also different. For example, the incoming switch is the main entrance of the power distribution, and its failure will cause the entire distribution cabinet to lose power, with the largest impact range; while the general circuit breaker or knife switch on a branch line, its abnormality only affects the local load at the end of the line, and the impact is relatively limited. If only the importance of the network is considered, it is impossible to distinguish such differences, and it is easy to have a "fat job" in monitoring. To overcome this problem, this embodiment adopts the hierarchical analysis method, takes the consequences of the fault as the first-level indicator, and takes the topological importance and functional importance as the second-level indicators, and comprehensively scores the nodes in the propagation path from the two perspectives of system failure impact and component action attributes. In order to quantify the importance of the function, two factors, task-based static weight and state-based dynamic weight, are introduced. The former reflects the contribution of components to the overall task of the distribution cabinet, such as the weights of the incoming line cabinet and the busbar cabinet are higher than those of the general feeder cabinet; the latter describes the influence of components on the instantaneous characteristics of the distribution state, such as the busbar cabinet has a greater effect when operating with dual power supplies than when operating with a single power supply, and the power load at the Peak moment is greater than that at the Valley moment. The weighted sum of the two types of weights gives a quantitative representation of the functional properties of the components, which is then multiplied by the topological importance index to obtain the comprehensive score of the components, and thus generate an importance ranking. The nodes with the top M% of the comprehensive score are selected as the top priority of fault risk management, that is, the key monitoring objects for the state perception of the distribution cabinet. For these key objects, the layout of the measurement points needs to be further optimized, covering their key failure modes while taking into account the balance at the system level, and trying to select representative parameters that can reflect the global characteristics, so as to "know the whole from one point".

[0039] S203: Arrange corresponding sensors based on the key monitoring points.

[0040] Specifically, after clarifying the key monitoring points, this embodiment further adopts corresponding sensor configuration measures to implement the monitoring scheme. The reason for carrying out this layout process is that a scientific and complete perception network is the fundamental premise for the intelligent distribution cabinet to achieve comprehensive state perception. As an important link connecting the physical world and the information space, the sensor shoulders the basic mission of signal acquisition and data supply. Its type, layout, parameter setting, etc. directly affect the data quality, and then affect the management and control level of the distribution cabinet. In order to give full play to the indicative role of key monitoring points, it is necessary to ensure the reliability and representativeness of its signal acquisition, otherwise the perfect mathematical model can only be a castle in the air. Therefore, it is necessary to select sensors according to local conditions, not only to meet the measurement point layout requirements of state monitoring, but also to comprehensively consider the installation conditions on site, and strive to be simple, applicable, accurate and reliable. At the same time, the cost-effectiveness of the sensor should be taken into account, and the investment should be reduced as much as possible under the premise of meeting business needs, so as to amplify the data value at the lowest cost and lay a solid foundation for the lean management of the distribution cabinet.

[0041] S102, performing parameter fusion on the operation characteristic parameters to obtain fused characteristic parameters, and inputting the fused characteristic parameters into a preset fault warning model to obtain fault warning data; Specifically, this embodiment uses an autoencoder network based on deep learning to achieve parameter fusion. The autoencoder is an unsupervised learning algorithm that learns the intrinsic structure and representation of the data by encoding the input data into a low-dimensional space and then decoding and reconstructing the original data from the low-dimensional space. In this embodiment, all operating feature parameters are input into the input layer of the autoencoder, and mapped to a low-dimensional feature space through multiple layers of nonlinear transformation to obtain a compressed feature vector, i.e., a fused feature parameter.

[0042] The structure of the autoencoder network is as follows: the number of nodes in the input layer is the number of all running feature parameters, the number of nodes in the output layer is the same as the input layer, and multiple hidden layers are set in the middle. The number of layers and the number of nodes in each layer can be set according to actual conditions. The network training process is divided into two stages: unsupervised pre-training and supervised fine-tuning. In the pre-training stage, the compressed representation of features is learned by minimizing the reconstruction error; in the fine-tuning stage, the network parameters are optimized by minimizing the prediction error. Through these two stages of training, the autoencoder can learn the intrinsic association and mapping rules of the running feature parameters and achieve effective fusion of parameters.

[0043] After obtaining the fused feature parameters, this embodiment inputs them into a preset fault warning model to generate fault warning data. The fault warning model is obtained based on historical operation data and fault case training, and can establish a mapping relationship between fused feature parameters and fault types. When new fused feature parameters are input, the model can automatically determine the fault type and severity to which it belongs, and generate corresponding fault warning data.

[0044] Based on the above embodiment, as an optional embodiment, before inputting the fused feature parameters into a preset fault warning model to obtain fault warning data, the method further includes: S301, collecting historical operation data, the historical operation data including normal operating condition data and various fault data, and preprocessing the historical operation data to obtain preprocessed historical operation data; Specifically, the historical operation data collected in this embodiment includes two categories: normal operating data and various types of fault data. Among them, normal operating data refers to various types of measurement point data collected by the distribution cabinet under normal operating conditions, such as time series records of parameters such as voltage, current, temperature, humidity, vibration and noise, which reflect the health characteristics and behavior patterns of the distribution cabinet. Various types of fault data refer to data collected by the distribution cabinet under various typical fault conditions, such as parameter waveforms and event records when short circuits, grounding, overloads, insulation damage and other faults occur, which reflect the abnormal characteristics and fault modes of the distribution cabinet.

[0045] In the actual collection process, normal operating data can be continuously recorded and archived regularly through the online monitoring system in the daily operation of the distribution cabinet; while fault data can be obtained through manual experiments or simulations. For faults that have not occurred or are difficult to simulate, they can also be supplemented from the equipment defect library, accident case library, etc. In short, it is necessary to collect comprehensive and balanced sample data as much as possible to cover various operating conditions of the distribution cabinet operation and avoid model deviation caused by uneven sample distribution. Ideally, the ratio of normal data to fault data should be close to the actual situation, and generally a ratio of 3:1 to 9:1 can be adopted.

[0046] After collecting enough historical operation data, it is necessary to perform necessary preprocessing on it to improve data quality and prepare for subsequent feature extraction and model training. The main contents of preprocessing include: (1) Data cleaning: Eliminate low-quality data such as errors, omissions, duplications, and inconsistencies to ensure data accuracy and completeness; (2) Data synchronization: aligning data from different sources and time scales in time and space to eliminate errors introduced by inconsistent sampling; (3) Data normalization: Scale data of different dimensions and numerical ranges and map them into the same feature space to avoid dimensional effects; (4) Data noise reduction: Use signal processing technologies such as digital filtering and wavelet denoising to suppress background noise such as power frequency interference and white noise and improve the signal-to-noise ratio; (5) Data enhancement: Use data enhancement techniques such as translation, scaling, rotation, and aliasing to expand the diversity and richness of samples and improve the generalization performance of the model.

[0047] After the above preprocessing, a batch of high-quality historical operation data can be obtained, which is recorded as the preprocessed historical operation data.

[0048] S302, extracting the preprocessed historical operation data to obtain a fault feature data set; Specifically, feature extraction mainly includes two steps: feature selection and feature compression.

[0049] The first is feature selection, which is to identify the feature subset most relevant to the fault from many candidate features and remove redundancy and noise in the data. Since not all features contribute to fault diagnosis, some features may be redundant or even interfere with diagnosis, so it is necessary to use methods such as feature evaluation and feature screening to select a set of optimal feature combinations. Common feature selection methods include variance analysis and chi-square test based on statistics, Pearson coefficient and mutual information based on correlation, and orthogonal matching pursuit and recursive feature elimination based on models. These methods evaluate the relevance of each feature to the fault from different angles, and automatically screen out the key features that contribute most to fault identification based on the set threshold or sorting results.

[0050] Taking variance analysis as an example, it uses one-way variance analysis to compare the significance of the mean difference of each feature under normal and fault conditions. The more significant the difference, the stronger its fault discrimination power. Therefore, the F statistic of each feature can be calculated and compared with the preset confidence threshold. The features above the threshold are selected into the optimal feature set. Similarly, the chi-square test can be used to compare the independence of features and faults, the mutual information can be used to compare the correlation between features and faults, and the recursive feature elimination can be used to compare the diagnostic performance of feature subsets, so as to obtain a set of feature combinations with the smallest dimension and the best diagnostic effect. The second is feature compression, that is, on the basis of the selected optimal feature set, the linear correlation between features is further removed to achieve dimensionality reduction and compression of feature dimensions. Some seemingly different features may actually have strong correlations, carry a lot of repeated information, and cannot provide additional discrimination. In order to eliminate this redundancy, feature transformation methods such as principal component analysis and independent component analysis are needed to map the original high-dimensional features to a low-dimensional feature space and extract a few principal components or independent components that represent the original data as much as possible as new features.

[0051] Taking principal component analysis as an example, it uses orthogonal transformation to recombine multiple linearly related features into a few linearly independent principal components. Each principal component is a linear combination of the original features and carries most of the variance information of the original data. Therefore, the eigenvalues ​​and eigenvectors of the covariance matrix can be calculated, sorted according to the size of the eigenvalues, and the eigenvectors corresponding to the first few largest eigenvalues ​​can be extracted as principal components.

[0052] S303, constructing a fault prediction model according to the fault feature data set, and iteratively training the fault warning model based on the preprocessed historical operation data, and obtaining the preset fault warning model when the fault warning accuracy of the fault warning model reaches a preset fault warning accuracy threshold.

[0053] Specifically, the first step is to select a suitable machine learning algorithm and build the basic framework of the early warning model according to the characteristics of the fault feature data set. Since the fault warning of the distribution cabinet is a typical binary classification problem, common classification algorithms such as support vector machines, logistic regression, and decision trees can be used. Considering that fault data usually presents nonlinear and non-stationary complex characteristics, deep learning algorithms such as convolutional neural networks and long short-term memory networks can also be introduced to extract and learn the deep-level patterns of fault evolution by building a multi-level and multi-time scale network structure. At the same time, different types of algorithms can also be combined to form an integrated learning model to give full play to their respective advantages and achieve complementary enhancement of early warning performance.

[0054] Taking the support vector machine as an example, it separates the normal samples and faulty samples in the fault feature data set by finding an optimal classification hyperplane. This hyperplane can not only maximize the interval between the two types of samples, but also tolerate the classification errors of a small number of samples, so it has good generalization performance. By constructing a support vector machine in a high-dimensional feature space and then implicitly mapping it back to the original space using a kernel function, it is possible to handle complex classification problems that are linearly inseparable. The advantages of the support vector machine are rigorous theoretical foundation, high classification accuracy, and easy solution, but the disadvantage is high computational complexity and is not suitable for large-scale data sets. Secondly, the fault feature data set is used to train the constructed warning model, and the model parameters are adjusted through sample learning so that it can predict the distribution cabinet fault as accurately as possible. The supervised learning paradigm is adopted here, that is, the historical operation data with known fault labels is used as the training set to let the model learn the correspondence between features and labels. Considering that the number of fault samples in the original historical operation data is far less than that of normal samples, in order to balance the proportion of the two types of samples and avoid the model tending to predict the samples as the majority class, it is necessary to use oversampling, undersampling and other techniques during training, or introduce penalty weights, cost-sensitive learning and other strategies. At the same time, in order to enable the model to learn real and reliable warning rules and avoid the risk of overfitting, it is also necessary to adopt methods such as cross-validation and regularization to improve the generalization performance of the model.

[0055] Taking k-fold cross validation as an example, it randomly divides the training set into k mutually exclusive subsets, uses k-1 subsets to train the model each time, and uses the remaining 1 subset to test the model. After k cycles, the average of each test result is taken as the performance indicator of the model. This method can make full use of limited training data, objectively evaluate the model's predictive ability for unknown data, and select the optimal model parameters. During the training process, the performance of the model on the validation set can be monitored in real time. If the validation error is found to continue to rise, the training is stopped in time to prevent the model from overfitting the noise in the training data.

[0056] Finally, the performance of the warning model is evaluated after each round of iterative training. When the model reaches the preset fault warning accuracy threshold, it can be determined as the final preset fault warning model. The performance evaluation indicators here mainly include accuracy, precision, recall, F1 score, area under the ROC curve, etc. They reflect the degree of consistency between the model prediction results and the true labels from different angles. Accuracy measures the proportion of samples predicted correctly by the model to the total samples, precision measures the proportion of real faults in samples predicted by the model as faults, recall measures the proportion of real fault samples predicted by the model, and F1 score is the harmonic average of precision and recall, taking into account the trade-off between the two. The ROC curve describes the changes in the true positive rate and false positive rate of the model at different thresholds. The larger the area under the curve, the better the overall performance of the model.

[0057] Based on the above embodiment, as an optional embodiment, before inputting the fault warning data into a preset rule matching engine for rule matching to obtain a matching result, the method further includes: S401, collecting historical fault cases, and extracting fault features from the historical fault cases based on expert experience knowledge to obtain fault features; Specifically, the construction and training of the fault warning model mainly includes the following steps: The first is model selection, that is, according to the needs of the fault warning task, select a suitable machine learning model as the basis. Since the fault warning of the distribution cabinet is a typical multi-classification problem, multi-class classification models are mainly considered, such as decision trees, random forests, support vector machines, neural networks, etc. Different models have their own characteristics in classification performance, computational efficiency, and interpretability, which need to be weighed comprehensively. Generally speaking, tree models have good interpretability and are suitable for engineering applications; while neural networks have more advantages in classification accuracy and are suitable for complex fault modes. At the same time, the generalization ability and noise resistance of the model should also be considered, and a robust model that is not easy to overfit and has strong robustness to abnormal data should be selected as much as possible.

[0058] Taking random forest as an example, it integrates multiple decision trees and effectively improves the accuracy and stability of classification while retaining the interpretability of a single tree. Therefore, random forest can be selected as the basic classifier. Multiple subsets are randomly extracted from the original training set through the Bootstrap method, and a feature subset is randomly selected from all features. An independent decision tree is trained on each bootstrap sample set and feature subset. Finally, the classification results of all decision trees are weighted averaged to obtain the output of random forest. This idea of ​​ensemble learning effectively reduces the variance of a single model and enhances the adaptability of the early warning model to unknown working conditions and abnormal data. The second is model training, that is, using the fault feature data set to perform parameter learning and performance optimization on the selected machine learning model. The specific training process needs to follow the general process of machine learning, including the division of training set and test set, initialization of model parameters, setting of loss function, selection of optimization algorithm, tuning of hyperparameters, etc. Among them, the fault feature data set is randomly divided into training set and test set according to a certain ratio. The former is used for model training and the latter is used to evaluate model performance. For each training sample, according to its feature input and label output, the loss function is minimized through optimization algorithms such as gradient descent, and the model parameters are iteratively updated to make the model output as close to the true label as possible. At the same time, hyperparameters such as learning rate, regularization term, and number of iterations should be appropriately set to control the convergence speed and stability of the training process. In addition, the classification performance of the model needs to be evaluated, and indicators such as accuracy, precision, and recall need to be calculated to weigh the model's fitting ability and generalization ability, and use methods such as cross-validation to detect overfitting problems.

[0059] S402, constructing a fault association relationship between the fault feature and the fault type based on the historical fault case, and constructing a structured diagnosis rule based on the fault association relationship; Specifically, establishing fault correlation requires the use of a combination of qualitative and quantitative analysis methods.

[0060] Qualitative analysis mainly starts from the physical mechanism and refers to the experience of experts to classify and sort out the causes, manifestation processes, and impact consequences of different fault characteristics, forming a many-to-many association mapping. For example, for the two fault characteristics of vibration and discharge, qualitative analysis found that the two often appear together. The root cause is the looseness caused by poor fixation of the iron core, which then causes the conductor vibration discharge. Therefore, multiple groups of associations such as "loose -> vibration", "loose -> discharge", and "vibration <-> discharge" can be established. For another example, for the two characteristics of heating and cracking, qualitative analysis found that the former may be the result of aging of the iron core insulation, while the latter may be a manifestation of coating failure caused by overheating. The two seem independent, but tracing back to the source, they both point to the decline of insulation performance. Therefore, the associations of "insulation aging -> heating" and "insulation aging -> cracking" can be established.

[0061] On the basis of qualitative analysis, quantitative mining is also needed. A variety of data-driven association analysis algorithms are used to automatically discover the association patterns, association strengths, and association probabilities between different fault features and fault types from a large amount of historical data. Association analysis is generally based on the idea of ​​frequent pattern mining. By calculating the co-occurrence frequency of different features, "strong associations" with a significance higher than the random level are found. Commonly used algorithms include FP-Growth, Apriori, and Eclat. Here, Apriori is mainly used as an example. The algorithm first sets a support threshold to measure the frequency of occurrence of a feature combination; then all features are combined in pairs, the support of each combination is calculated, and the combinations above the threshold are retained to form frequent 1-item sets; on this basis, candidate frequent 2-item sets are generated, and their support is determined by scanning data records, and new frequent 2-item sets are screened out; this is iterated until no higher-order frequent item sets are found. Finally, indicators such as Lift and chi-square are used to evaluate the association strength of each feature in the frequent item set to form a strong association rule.

[0062] Of course, frequent patterns reflect the statistical correlation of feature co-occurrence, which cannot be equated with causal association. In order to further verify the rationality of the association, it is often necessary to combine expert knowledge to eliminate some of the pseudo-associations that are obviously illogical and counterintuitive. At the same time, since the Apriori algorithm enumerates all combinations, the computational complexity increases exponentially with the number of features, and it is less efficient when processing high-dimensional data. Therefore, in engineering practice, the FP-Growth algorithm based on the FP tree or the projection-based Eclat algorithm is generally used to significantly optimize the spatiotemporal performance.

[0063] After the above qualitative and quantitative analysis, a comprehensive fault association network can be constructed to reveal the inherent semantic structure of fault features and form a multi-level mapping between features, components, and faults. This process fragments the originally discrete features into structured knowledge elements, which have good characteristics of being interpretable, computable, and integrable. On this basis, a series of diagnostic rules can be further summarized. Specifically, each fault association is represented as an IF-THEN type rule, in which the antecedent guided by IF is a combination of certain fault features, and the consequent guided by THEN is the corresponding fault mode and faulty component. At the same time, a confidence weight is assigned to each rule according to the strength of association to indicate the degree of credibility of the rule. In addition, in order to enhance the interpretability of the rules, the values ​​of the antecedents need to be discretized, such as dividing the vibration amplitude into three levels: high, medium, and low, and using natural language to describe the rules in text form to generate diagnostic rules such as "IF vibration = high AND discharge = high THEN fault = local overheating". Through this structured representation, the originally obscure empirical knowledge is formalized into a program-readable rule base, with an interface form that can be directly called and matched by computers.

[0064] S403, constructing an initial rule base based on the diagnostic rules, and performing consistency detection on the initial rule base to obtain a target rule base; Specifically, the process of building the initial rule base mainly includes three links: rule collection, rule representation, and rule organization. The first is rule collection, which is to summarize the diagnostic rules formed by association analysis and store them in the rule base. In order to facilitate subsequent management, each rule needs to be given a unique identifier to clarify its scope of application, formulation time, formulation and other meta-information. The second is rule representation. In order to make the rules computable, they need to be uniformly represented as a formal language that can be parsed by the program. Generally, the Prolog language based on Horn clauses is used to represent each rule as a logical expression composed of atomic formulas connected by logical conjunctions such as conjunction and implication. For example, the rule "IF vibration = high AND temperature = high THEN fault = loose" described in natural language is converted into a Prolog clause of the form "fault (loose): - feature (vibration, high), feature (temperature, high)". The last is rule organization. Considering that a single rule base may contain hundreds or thousands of rules, in order to improve matching efficiency, the rules need to be properly indexed and clustered. A common approach is to construct a rule tree, which groups rules with the same antecedent into one category and organizes them into a tree structure, thereby narrowing the matching scope and increasing the search speed. At the same time, in order to support multi-granularity reasoning, a rule tree forest can be further constructed from coarse to fine, with the upper nodes corresponding to abstract and generalized rules and the lower nodes corresponding to specific and specialized rules, thus achieving hierarchical reasoning from the outside to the inside and from cause to effect.

[0065] After the initial rule base is built, it is necessary to further carry out consistency detection, that is, to analyze and evaluate the consistency between the rules in terms of structure, semantics, etc., to discover and eliminate various defects that may exist therein. Common defects mainly include: redundancy, that is, there are multiple repeated rules with exactly the same antecedents or consequents; contradiction, that is, there are multiple opposing rules with the same antecedents but different consequents; incompleteness, that is, there is a lack of corresponding rule descriptions for certain known faults; unsatisfactory, that is, the feature combination corresponding to the rule antecedent cannot appear at the same time in practice; incomprehensible, that is, the rule content is contrary to the existing physical mechanism and does not conform to common sense logic. In this regard, this embodiment adopts different detection strategies for different defects. The first is redundancy detection based on similarity, which identifies repeated rules with a similarity higher than a threshold by calculating the text similarity between the rules. The second is contradiction detection based on SAT, which represents the rule base as a conjunction normal form, uses the SAT solver to determine its satisfiability, and then finds the contradictory rules that lead to unsatisfactory. The third is completeness detection based on matching, which matches the consequent of the rule with the known fault type one by one to find the fault blind spot that has not been covered by the rule. The fourth is the unsatisfiability detection based on sampling. Monte Carlo sampling is used to generate a large number of reasonable feature values ​​to check whether the rule antecedent can be satisfied by these samples. If the satisfaction rate is too low, it is considered an unsatisfiable rule. The fifth is the comprehensibility detection based on ontology. Physical quantities, fault modes, etc. are abstracted into ontology concepts. If the conceptual relationship contained in the rule is contrary to the ontology definition, it is judged as an incomprehensible rule.

[0066] While carrying out consistency detection, this embodiment also proposes corresponding repair measures for the problem rules found. For redundant rules, they can be merged into one by merging similar items; for contradictory rules, their confidence levels need to be further compared, and the one with the highest confidence level is retained; for incomplete faults, new rules can be derived from rules of similar faults through transfer learning; for unsatisfactory and incomprehensible rules, they are directly deleted. At the same time, since the modification of rules may introduce new defects, regression testing is required after the repair to verify the correctness of the repair. This iteration is carried out until all defects are eliminated, forming a self-consistent and complete target rule base.

[0067] S404, constructing a preliminary rule matching engine based on the target rule base, and training the preliminary rule matching engine based on the historical fault cases, and obtaining the preset rule matching engine when the matching accuracy reaches a preset accuracy threshold.

[0068] Specifically, the process of building a preliminary rule matching engine can be divided into two main links: matching algorithm design and reasoning strategy selection. The first is matching algorithm design, that is, designing an appropriate similarity measurement function to calculate the matching degree between the input feature and the rule antecedent. The simplest method is to regard the rule antecedent as a Boolean expression and the input feature as its truth value assignment. If the expression under the assignment is true, it is judged as a match. However, this method is relatively simple and crude, and cannot accurately describe the semantic differences of features. To this end, a semantic-based matching method can be further introduced to measure the proximity between rules through the semantic similarity of ontology concepts. For example, each fault feature is abstracted as an ontology concept, and then the semantic distance between concepts is calculated based on semantic dictionaries such as WordNet. The closer the distance, the higher the similarity. This semantic matching can reveal the deep connection under the word representation and discover more implicit rule matches. Of course, when processing massive monitoring data, the cost of full-text search is high, so it is also necessary to introduce technologies such as inverted indexing to pre-extract the key feature words of each node in the rule tree and establish an inverted mapping from keywords to rules, so as to narrow the matching range to the relevant rule subset and improve the retrieval efficiency.

[0069] The second is the selection of reasoning strategies, that is, when multiple rules are matched, how to weigh the importance of each rule and make the final diagnosis based on it. A common strategy is weighted voting, that is, the consequent of each rule is regarded as a diagnostic hypothesis, whose confidence is equal to the product of the matching degree and the rule weight, and finally the diagnosis with the highest confidence is selected as the output. This soft weighted strategy can comprehensively consider multiple evidences and make basic judgments when the evidence is insufficient, and has strong robustness. However, it may also amplify the influence of minor rules and interfere with the final diagnosis. For this reason, a conflict resolution strategy based on Dempster-Shafer evidence theory can also be adopted. This theory regards rules as different sources of evidence and measures their credibility by the consistency between evidences. When multiple pieces of evidence support a hypothesis at the same time, the confidence of the hypothesis will be significantly improved; conversely, when there is a conflict between the evidences, the confidence of each will be reduced. Finally, the hypothesis with the highest confidence is selected as the diagnostic conclusion. This strategy can adaptively adjust the rule weights, dynamically resolve evidence conflicts, and make the diagnosis more robust and reliable.

[0070] Of course, due to the limitations of training data and expert knowledge, the initially constructed rule matching engine inevitably has some subjective biases and insufficient generalization. It is often difficult to give an accurate diagnosis for some fault modes that have never been encountered. In order to improve its practical application ability, it is necessary to let it continuously learn and improve in the training of massive historical cases, and gradually master the inherent mapping rules from features to faults. To this end, this embodiment adopts an incremental, closed-loop training optimization mechanism. Its basic process is to first use the preliminary engine to carry out batch diagnosis of historical fault cases, compare the diagnosis results with the known root causes, calculate the accuracy, recall rate, F1 score and other evaluation indicators, and update the rule weights and matching thresholds accordingly, so that the weights of rules that are consistent with the actual faults are increased, and the weights of rules that are inconsistent with the actual faults are reduced. At the same time, for missed or misdiagnosed cases, it is necessary to deeply analyze the reasons and dynamically modify the rule base accordingly, such as mining new diagnostic rules for missed cases through association analysis, and deleting related invalid rules for misdiagnosed cases through counterexamples. After the rule base is updated, the new case set is used to re-run the diagnostic test, and the engine parameters are adjusted accordingly to enter the next round of iteration. This is repeated, and as the training continues to deepen, the matching engine will become more and more perfect, and it will be able to make more and more accurate judgments when dealing with various complex faults, until its accuracy rate reaches the preset threshold (such as 95%), which can be considered mature training and become a preset rule matching engine adapted to the current scenario.

[0071] S103, inputting the fault warning data into a preset rule matching engine for rule matching to obtain a matching result, and generating a warning signal according to the matching result.

[0072] Specifically, this application constructs a fault warning knowledge base based on production rules. Each rule in the knowledge base consists of two parts: premise and conclusion, in the form of "IF premises THEN conclusion". The premise describes the fault symptoms and context, and the conclusion gives the corresponding fault type and handling suggestions. These rules are obtained by summarizing and refining a large number of historical fault cases and expert experience, covering various typical operating conditions and fault modes of distribution cabinet operation.

[0073] When matching rules, the fault warning data is parsed into a set of factual assertions, such as "temperature=high, humidity=normal, discharge=frequent", etc., and input into the rule matching engine to start the forward reasoning process. The engine first searches the knowledge base to find all the rules whose premise conditions match the current facts; then determines the final triggered rule through conflict resolution strategies (such as rule priority, time sequence, etc.); finally, according to the conclusion of the triggering rule, it outputs matching results such as fault type, cause explanation, and disposal suggestions.

[0074] For example, there are two rules in the knowledge base: Rule1: IF current_distortion=high AND voltage_unbalance=high THENfault_type="power quality issue", suggestion="check nonlinear load and phasebalance" Rule2: IF temperature=high AND humidity=high AND discharge=frequentTHEN fault_type="insulation aging", suggestion="schedule insulation test andmaintenance" When the input fault warning data is "current_distortion=high, voltage_unbalance=normal, temperature=high, humidity=high, discharge=frequent", the rule matching engine searches the knowledge base and finds that the premise of Rule2 completely matches the current fact, so it triggers Rule2 and outputs the matching result fault_type="insulation aging", suggestion="schedule insulation test andmaintenance", which means that an insulation aging fault is diagnosed and it is recommended to arrange insulation test and maintenance.

[0075] After obtaining the rule matching result, this embodiment generates a corresponding warning signal according to the matching result. The warning signal contains key information such as fault type, fault cause, fault location, severity level, and disposal suggestions, and is presented to the on-duty personnel in an intuitive and summarized manner for their reference and decision-making. At the same time, the warning signal can also link the alarm device (such as sound and light alarm, SMS / email push, etc.), automatically notify the relevant responsible person, and prompt him to take emergency measures in time.

[0076] For example, when the matching result shows that the insulation aging fault occurs in the power distribution cabinet, the generated warning signal is "Warning: Insulation aging detected in Switch Cabinet #3, severity level: HIGH, suggested action: schedule insulation test and maintenance within 24 hours", which means that the insulation aging fault is detected in the switch cabinet #3, the severity level is high, and it is recommended to schedule insulation test and maintenance within 24 hours. At the same time, the warning signal is pushed to the monitoring screen of the main control room and the mobile terminal of the on-duty personnel in the form of a pop-up window, and an audible and visual alarm is triggered to prompt relevant personnel to deal with it in time.

[0077] Based on the above embodiment, as an optional embodiment, after inputting the fault warning data into a preset rule matching engine for rule matching and obtaining a matching result, the method includes: S501, recording the matching result of each rule matching, and forming a rule matching log according to the matching result; Specifically, the process of rule matching recording mainly includes two steps: matching process tracking and result extraction. The first is matching process tracking, that is, in the whole process of the rule matching engine, the input and output data and operation status of the key links are synchronously recorded. Since rule matching usually goes through multiple rounds of iterations, and each round of iteration contains the serial and parallel matching of multiple rules, it is necessary to track the real-time progress of matching layer by layer. For example, for a single rule, it is necessary to record the static properties such as the rule set, priority, matching method, and dynamic information such as input warning data, judgment conditions, thresholds, and output results. In particular, for those composite rules with multiple sub-conditions, it is also necessary to record the actual matching of each sub-condition in detail, including the truth value of each atomic condition, the role of the conjunction, the comprehensive conclusion, etc. For the rule set, it is necessary to record the combination method, execution order, dependency, etc. of each member rule, as well as the final output of the entire rule set.

[0078] S502, performing statistical analysis on the rule matching log, determining the rule performance value of each rule in the preset rule matching engine, and optimizing the rules in the preset rule matching engine according to the rule performance value to obtain an optimized rule matching engine.

[0079] Specifically, the process of rule performance evaluation and optimization mainly includes three steps: statistical indicator design, evaluation result analysis and optimization strategy formulation. The first is statistical indicator design, that is, to build a scientific, reasonable, comprehensive and systematic quantitative evaluation system for rule performance. Generally speaking, rule performance can be characterized from two dimensions: one is diagnostic effectiveness, which reflects the coverage ability of the rule for faults; the other is reasoning efficiency, which reflects the resource consumption of the rule in the matching process. For diagnostic effectiveness, the main statistical indicators are coverage, hit rate, false alarm rate and other indicators of the rule base on various cases. Coverage measures the adaptation range of the rule base to known fault modes, hit rate measures the probability that the fault warning is captured and diagnosed by the rule base, and false alarm rate measures the degree of deviation between the rule diagnosis conclusion and the actual situation. For reasoning efficiency, the main statistical rule tree space complexity (such as the number of nodes, the number of layers, the length of the rule) and the matching time complexity (such as the single reasoning time, the overall response delay) are considered. At the same time, the independent contribution of each rule should also be considered in a targeted manner. For example, the average triggering frequency reflects the intensity of rule use, the independent coverage reflects the irreplaceability of the rule, and the comprehensive confidence reflects the reliability of the rule diagnosis conclusion. Of course, the specific setting of evaluation indicators should be dynamically adjusted according to different application scenarios and business objectives, but the essence remains the same: we must objectively and comprehensively examine all aspects of rule performance, grasp the main contradictions, highlight the core demand of diagnostic effectiveness, and strive to clarify the main issues with the least indicators.

[0080] The second step is to analyze the evaluation results, that is, to substitute the statistical information accumulated during the actual operation of the rule base into the above-mentioned indicator system for quantitative calculation and ranking comparison, and to characterize the advantages and disadvantages of each rule in different indicators. For example, through case data statistics, the fault coverage rate of rule r1 is 85%, the false alarm rate is 3%, the average trigger frequency is 150 times / day, the independent coverage rate is 5%, the average confidence is 75%, and the reasoning time is 10ms; the fault coverage rate of rule r2 is 45%, the false alarm rate is 1%, the average trigger frequency is 30 times / day, the independent coverage rate is 20%, the average confidence is 95%, and the reasoning time is 150ms. It can be seen that rule r1 has a wide range of diagnosis, large volume, and fast speed, but its independent contribution is small and its confidence is not very high; rule r2 has accurate diagnosis and high confidence, but its coverage is narrow and its calculation is slow, which are important clues for rule optimization. After weighted synthesis of various indicators, the rules in the existing rule base can be ranked by advantages and disadvantages, and the problem rules with the highest priority can be identified in combination with specific business impacts. For example, rules with low coverage and high false positive rates are likely to have overly broad or incorrect condition settings; rules with high trigger frequency and low independent coverage may have a large number of duplicates and have room for further merging; rules with low confidence and slow diagnosis have insufficient reasoning chains and need further refinement and improvement. In short, through data, the crux of the problem rules and their quantitative shortcomings will be clear at a glance, and the direction, ideas, and focus of rule base optimization will also become clear.

[0081] The last step is to optimize strategy formulation, that is, to formulate strategies for rule reconstruction, refinement, and expansion based on the results of rule evaluation, and put them into practice.

[0082] Based on the above embodiment, as an optional embodiment, generating a warning signal according to the matching result includes: The credibility of the fault warning data is determined according to the matching result, and when the credibility exceeds a preset credibility threshold, the warning signal is generated.

[0083] Specifically, the process of determining the credibility of fault warning data mainly includes two steps: selection of evaluation factors and quantitative calculation of credibility. First of all, it is necessary to clarify the key factors that affect the credibility of the diagnosis results, which are generally comprehensively evaluated from three perspectives: the rigor of the rule matching process, the authority of the rules themselves, and the typicality of the fault mode. For the rule matching process, the number of rules that trigger the diagnosis result and the sufficiency of evidence are mainly examined. If a fault diagnosis result needs to meet five rules at the same time, and all of them are mandatory rules, the credibility will be relatively higher; on the contrary, if only one or two rules are supported, or all are optional rules, the credibility of the diagnosis will be greatly reduced. For the rules themselves, the verification of them in historical data is mainly examined, such as the theoretical basis for rule formulation and the accuracy of the rules in historical cases.

[0084] After obtaining the credibility score of the warning data, it can be determined whether to trigger the warning signal by comparing it with the preset threshold. Generally speaking, the setting of the credibility threshold needs to balance the sensitivity and accuracy of the diagnosis. It is necessary to strictly control the threshold to reduce the false alarm rate as much as possible, and to avoid the threshold being too high, resulting in untimely warnings and missing the best disposal window. For example, if the warning threshold of a system is set to 9 points, and there are two diagnostic results at this time: A's credibility score is 9.2, and B's credibility score is 8.8, then the system will only issue a warning signal for A, and keep a wait-and-see attitude towards B. Of course, the setting of thresholds should be adapted to the times and local conditions, closely follow the actual business, refer to industry standards and experience, and dynamically float in combination with the current operating status of the system and the diagnostic level. After comprehensively evaluating the timeliness and reliability of the warning, a critical point that takes both into account and has controllable risks is selected as the watershed for starting the warning signal. As a result, high-confidence and high-contribution fault hazards will be focused on, and every second counts under limited human and material resources to avoid and reduce the impact of faults as much as possible.

[0085] See also Figure 2 , Figure 2 An architecture diagram of a smart power distribution cabinet safety warning system provided in an embodiment of the present application, the smart power distribution cabinet safety warning system includes: The data acquisition module 1 is used to acquire multi-source heterogeneous operation parameters of the intelligent power distribution cabinet, and perform feature extraction on the multi-source heterogeneous operation parameters to obtain operation feature parameters; the multi-source heterogeneous operation parameters include current parameters, voltage parameters, temperature parameters, humidity parameters, vibration parameters and partial discharge parameters; the feature extraction on the multi-source heterogeneous operation parameters to obtain operation feature parameters includes: performing spectrum analysis on the current parameters to obtain current time-frequency domain features; performing spectrum analysis on the voltage parameters to obtain voltage time-frequency domain features; performing abnormal trend extraction on the temperature parameters to obtain temperature abnormal trend features; performing abnormal trend extraction on the humidity parameters to obtain humidity abnormal trend features; performing time-frequency feature extraction on the vibration parameters to obtain vibration time-frequency domain features; performing feature extraction on the partial discharge parameters to obtain discharge pulse features, discharge mode features and discharge type features, and using the current time-frequency domain features, the voltage time-frequency domain features, the temperature abnormal trend features, the humidity abnormal trend features, the vibration time-frequency domain features, the discharge pulse features, the discharge mode features and the discharge type features as the operation feature parameters; Fault warning module 2, used for performing parameter fusion on the operation characteristic parameters to obtain fused characteristic parameters, and inputting the fused characteristic parameters into a preset fault warning model to obtain fault warning data; The early warning judgment module 3 is used to input the fault early warning data into a preset rule matching engine for rule matching, obtain a matching result, and generate an early warning signal according to the matching result.

[0086] It should be noted that: when the system provided in the above embodiment realizes its functions, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0087] Please refer to Figure 3 The application also discloses an electronic device. Figure 3 The electronic device 300 may include: at least one processor 301, at least one network interface 304, a user interface 303, a memory 305, and at least one communication bus 302 or end-to-end wireless communication.

[0088] The communication bus 302 is used to realize the connection and communication between these components.

[0089] The user interface 303 may include a display screen (Display) and a camera (Camera). Optionally, the user interface 303 may also include a standard wired interface and a wireless interface.

[0090] The network interface 304 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface).

[0091] Among them, the processor 301 may include one or more processing cores. The processor 301 uses various interfaces and lines to connect various parts in the entire server, and executes various functions of the server and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 305, and calling data stored in the memory 305. Optionally, the processor 301 can be implemented in at least one hardware form of digital signal processing (Digital Signal Processing, DSP), field programmable gate array (Field~Programmable Gate Array, FPGA), and programmable logic array (Programmable Logic Array, PLA). The processor 301 can integrate one or a combination of a central processing unit (Central Processing Unit, CPU), a graphics processing unit (Graphics Processing Unit, GPU) and a modem. Among them, the CPU mainly processes the operating system, user interface and application programs; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor 301, and it can be implemented separately through a chip.

[0092] Among them, the memory 305 may include a random access memory (Random Access Memory, RAM) and may also include a read-only memory (Read~Only Memory). Optionally, the memory 305 includes a non-transitory computer-readable medium. The memory 305 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 305 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store data involved in the above-mentioned various method embodiments, etc. The memory 305 may optionally be at least one storage system located away from the aforementioned processor 301. Refer to Figure 3 , the memory 305 as a computer storage medium may include an operating system, a network communication module, a user interface module and an application program of a smart distribution cabinet safety early warning method.

[0093] exist Figure 3In the electronic device 300 shown, the user interface 303 is mainly used to provide an input interface for the user and obtain the data input by the user; and the processor 301 can be used to call the application program storing the intelligent distribution cabinet safety warning method in the memory 305. When executed by one or more processors 301, the electronic device 300 executes one or more methods in the above-mentioned embodiments. It should be noted that for the aforementioned method embodiments, for the sake of simple description, they are all expressed as a series of action combinations, but those skilled in the art should know that the present application is not limited to the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required for the present application. In the above embodiments, the descriptions of each embodiment have their own emphasis. For the parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0094] In the several implementations provided in this application, it should be understood that the disclosed system can be implemented in other ways. For example, the system embodiments described above are only schematic, such as the division of modules, which is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some service interfaces, indirect coupling or communication connection of systems or modules, which can be electrical or other forms.

[0095] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0096] The present application also provides a computer storage medium that can store multiple instructions, which are suitable for being loaded and executed by a processor as described above. Figure 1 The smart power distribution cabinet safety warning method of the embodiment shown in the figure can be found in the specific execution process. Figure 1 The specific description of the illustrated embodiment will not be repeated here.

[0097] In addition, each functional module in each embodiment of the present application can be integrated into a processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The above integrated modules can be implemented in the form of hardware or software functional modules.

[0098] If the integrated module is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a memory and includes several instructions for a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the various embodiments of the present application. The aforementioned memory includes: various media that can store program codes, such as USB flash drives, mobile hard drives, magnetic disks or optical disks.

[0099] The above are only exemplary embodiments of the present disclosure and cannot be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure. After considering the disclosure of the specification and practice, those skilled in the art will easily think of other embodiments of the present disclosure.

[0100] This application is intended to cover any variation, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary technical means in the art not described in the present disclosure. The description and examples are to be regarded as exemplary only, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A safety early warning method for an intelligent power distribution cabinet, characterized in that: The method comprises: Acquire multi-source heterogeneous operation parameters of an intelligent power distribution cabinet, and perform feature extraction on the multi-source heterogeneous operation parameters to obtain operation feature parameters; the multi-source heterogeneous operation parameters include current parameters, voltage parameters, temperature parameters, humidity parameters, vibration parameters and partial discharge parameters; the feature extraction on the multi-source heterogeneous operation parameters to obtain operation feature parameters includes: performing spectrum analysis on the current parameters to obtain current time-frequency domain features; performing spectrum analysis on the voltage parameters to obtain voltage time-frequency domain features; performing abnormal trend extraction on the temperature parameters to obtain temperature abnormal trend features; performing abnormal trend extraction on the humidity parameters to obtain humidity abnormal trend features; performing time-frequency feature extraction on the vibration parameters to obtain vibration time-frequency domain features; performing feature extraction on the partial discharge parameters to obtain discharge pulse features, discharge mode features and discharge type features, and using the current time-frequency domain features, the voltage time-frequency domain features, the temperature abnormal trend features, the humidity abnormal trend features, the vibration time-frequency domain features, the discharge pulse features, the discharge mode features and the discharge type features as the operation feature parameters; Performing parameter fusion on the operation characteristic parameters to obtain fused characteristic parameters, and inputting the fused characteristic parameters into a preset fault warning model to obtain fault warning data; The fault warning data is input into a preset rule matching engine for rule matching to obtain a matching result, and a warning signal is generated according to the matching result.

2. The method according to claim 1, characterized in that: Before inputting the fault warning data into a preset rule matching engine for rule matching to obtain a matching result, the method further includes: Collect historical fault cases, and extract fault features from the historical fault cases based on expert experience knowledge to obtain fault features; Building a fault association relationship between the fault characteristics and the fault type based on the historical fault cases, and building a structured diagnosis rule based on the fault association relationship; Building an initial rule base based on the diagnostic rules, and performing consistency detection on the initial rule base to obtain a target rule base; A preliminary rule matching engine is constructed based on the target rule base, and the preliminary rule matching engine is trained based on the historical fault cases. When the matching accuracy reaches a preset accuracy threshold, the preset rule matching engine is obtained.

3. The method according to claim 1, characterized in that Before inputting the fused feature parameters into a preset fault warning model to obtain fault warning data, the method further includes: Collecting historical operation data, the historical operation data including normal operating condition data and various fault data, and preprocessing the historical operation data to obtain preprocessed historical operation data; Extracting the preprocessed historical operation data to obtain a fault feature data set; A fault prediction model is constructed according to the fault feature data set, and the fault warning model is iteratively trained based on the preprocessed historical operation data. When the fault warning accuracy of the fault warning model reaches a preset fault warning accuracy threshold, the preset fault warning model is obtained.

4. The method according to claim 1, characterized in that Before obtaining the multi-source heterogeneous operation parameters of the intelligent power distribution cabinet, the method further includes: Obtaining the topological structure and functional composition of the intelligent power distribution cabinet, and determining key monitoring points based on the topological structure and the functional composition; Determine a fault propagation path based on the topological structure and the functional composition, and determine a key monitoring point among the key monitoring points based on the fault propagation path; Corresponding sensors are arranged based on the key monitoring points.

5. The method according to claim 1, characterized in that After the fault warning data is input into a preset rule matching engine for rule matching and a matching result is obtained, the method further comprises: Record the matching results of each rule match and form a rule matching log based on the matching results; The rule matching log is statistically analyzed to determine the rule performance value of each rule in the preset rule matching engine, and the rules in the preset rule matching engine are optimized according to the rule performance value to obtain an optimized rule matching engine.

6. The method according to claim 1, characterized in that Generating a warning signal according to the matching result includes: The credibility of the fault warning data is determined according to the matching result, and when the credibility exceeds a preset credibility threshold, the warning signal is generated.

7. An intelligent power distribution cabinet safety early warning system, characterized in that: The system comprises: A data acquisition module, used to acquire multi-source heterogeneous operation parameters of an intelligent power distribution cabinet, and perform feature extraction on the multi-source heterogeneous operation parameters to obtain operation feature parameters; the multi-source heterogeneous operation parameters include current parameters, voltage parameters, temperature parameters, humidity parameters, vibration parameters and partial discharge parameters; the feature extraction of the multi-source heterogeneous operation parameters to obtain operation feature parameters includes: performing spectrum analysis on the current parameters to obtain current time-frequency domain features; performing spectrum analysis on the voltage parameters to obtain voltage time-frequency domain features; performing abnormal trend extraction on the temperature parameters to obtain temperature abnormal trend features; performing abnormal trend extraction on the humidity parameters to obtain humidity abnormal trend features; performing time-frequency feature extraction on the vibration parameters to obtain vibration time-frequency domain features; performing feature extraction on the partial discharge parameters to obtain discharge pulse features, discharge mode features and discharge type features, and using the current time-frequency domain features, the voltage time-frequency domain features, the temperature abnormal trend features, the humidity abnormal trend features, the vibration time-frequency domain features, the discharge pulse features, the discharge mode features and the discharge type features as the operation feature parameters; A fault warning module is used to perform parameter fusion on the operation characteristic parameters to obtain fused characteristic parameters, and input the fused characteristic parameters into a preset fault warning model to obtain fault warning data; The early warning judgment module is used to input the fault early warning data into a preset rule matching engine for rule matching, obtain a matching result, and generate an early warning signal according to the matching result.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor and executing the method according to any one of claims 1 to 6.

9. An electronic device, characterized in that: It includes a processor, a memory and a transceiver, the memory is used to store instructions, the transceiver is used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the method according to any one of claims 1 to 6.

Images