High-risk loop fault identification method, system, medium, equipment and program

Abstract

The invention relates to the technical field of fault detection, and provides a high-risk loop fault identification method and system, a medium, equipment and a program. The high-risk loop fault identification method comprises the following steps: carrying out feature selection on preprocessed high-risk loop data, and carrying out feature fusion on selected features to obtain fused features; inputting the fused features into an encoder of a deep learning model to extract local features; inputting the extracted local features into a one-dimensional convolutional layer to obtain refined features; based on the refined features, a decoder is adopted to obtain a fault identification result of the high-risk loop; in the training process of the deep learning model, hyper-parameters of the model are optimized by adopting a hunting algorithm for adaptive parameter adjustment, and a learning strategy of the model is adjusted according to data changes by adopting an incremental learning method for category balance and data selection. According to the method, high-dimensional time series data can be effectively processed, and the identification precision of a high-risk loop is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault detection, and particularly to a high-risk loop fault identification method, system, medium, device and program. Background Art

[0002] The statements in this part merely provide background technical information related to the present invention and do not necessarily constitute prior art.

[0003] In modern industrial systems, the identification of high-risk loops is crucial, especially in fields such as power, chemical engineering, and manufacturing. Traditional loop identification methods usually rely on expert experience and rule-based algorithms, which are less efficient and vulnerable to human factors when dealing with complex systems and large amounts of data.

[0004] With the rapid development of deep learning, its capabilities in pattern recognition and data processing have been significantly improved, and it can effectively handle complex non-linear relationships and high-dimensional data. Deep learning models perform well in image, time series, and sensor data analysis and are suitable for high-risk loop identification.

[0005] Chinese Patent with the application number "CN2023115703653" discloses an intelligent diagnosis method and system for faults and potential hazards of power production equipment based on a knowledge graph. The method includes: 1) constructing a knowledge graph of faults of power production equipment; 2) identifying the severity level of faults for the obtained fault alarm information and sorting them according to the order of fault treatment; 3) preprocessing the fault text to be processed using a vector space model to obtain feature terms and performing vectorization representation; 4) weighting the vectorized feature terms, combining the constructed knowledge graph, and using a similarity calculation method to obtain the fault diagnosis result. However, the current description of potential electricity safety hazards contains a large amount of structured and unstructured data (such as safety hazard inspection records, etc.), and this invention only targets structured data, which affects the comprehensiveness of safety hazard diagnosis. In addition, relying on the similarity calculation to judge the fault level model cannot accurately diagnose low-similarity fault information and requires manual assistance for judgment, which limits the practicality of this invention, increases labor costs and time costs, and reduces the efficiency of electricity safety hazard diagnosis.

[0006] The Chinese patent with the application number "CN2024105417861" discloses a smart grid fault monitoring method and system applying artificial intelligence, including: Step 1, obtaining the to-be-processed power IoT sensing monitoring data, and determining a first fault discrimination decision model and a second fault discrimination decision model for fault identification of the to-be-processed power IoT sensing monitoring data, wherein the second fault discrimination decision model has learned the fault location sensing monitoring data; Step 2, based on the first fault discrimination decision model, performing first power grid operation state mining on the to-be-processed power IoT sensing monitoring data to obtain first power grid operation state vectors of the to-be-processed power IoT sensing monitoring data at at least one semantic fine-grained level; Step 3, based on the second fault discrimination decision model and according to the past fault location guidance vectors, performing second power grid operation state mining on the to-be-processed power IoT sensing monitoring data to obtain second power grid operation state vectors of the to-be-processed power IoT sensing monitoring data at the at least one semantic fine-grained level; Step 4, performing state consistency analysis on the first power grid operation state vectors and the second power grid operation state vectors at the same semantic fine-grained level to obtain state similarity and difference analysis viewpoints; Step 5, based on the state similarity and difference analysis viewpoints, performing fault point location warning on the to-be-processed power IoT sensing monitoring data. However, the model adopted by this invention has too high computational complexity, and for small and medium-sized power enterprises, the cost is greater than the benefit. Moreover, the large amount of on-site sensor data required by this model is applicable to application scenarios with a large number of sensors deployed. When the number of sensors in the application scenario itself is insufficient, the model has insufficient learning data, which may limit the accuracy and comprehensiveness of annotation, and cannot cover entity and relationship types with different characteristics, thus resulting in deviation in model learning. Summary of the Invention

[0007] To solve the technical problems existing in the above-mentioned background technology, the present invention provides a high-risk loop fault identification method, system, medium, device and program. The deep learning model of the present invention that integrates the cheetah algorithm with adaptive parameter adjustment and the incremental learning method of class balance and data selection can effectively process high-dimensional time series data and improve the identification accuracy of high-risk loops. Through the self-attention mechanism, the model dynamically captures the complex time dependencies in the power system, and at the same time, the incremental learning method mechanism of class balance and data selection enables the model to update parameters in real time when new data is received, further reducing missed reports and false alarms and improving the reliability of decision-making.

[0008] To achieve the above object, the present invention adopts the following technical solutions: The first aspect of the present invention provides a high-risk loop fault identification method.

[0009] A high-risk loop fault identification method includes: Identify high-risk circuits, obtain high-risk circuit data, and perform preprocessing; Perform feature selection on the preprocessed high-risk circuit data, fuse the selected features, and obtain fused features; Input the fused features into the encoder of the deep learning model to extract local features; input the extracted local features into a one-dimensional convolutional layer to obtain refined features; based on the refined features, use a decoder to obtain the fault identification results of high-risk circuits; Among them, during the training process of the deep learning model, the cheetah algorithm with adaptive parameter adjustment is used to optimize the hyperparameters of the model, and the incremental learning method of class balance and data selection is used to automatically adjust the learning strategy of the model according to the changes in the data.

[0010] Further, the high-risk circuits include: power system circuits, control system circuits, power system circuits, high-temperature and high-pressure system circuits, data acquisition and monitoring system circuits, electrical start and stop circuits, fire protection circuits, and toxic substance discharge circuits.

[0011] Further, after identifying the high-risk circuits, it also includes: obtaining key equipment in the high-risk circuits, the operating status of the equipment, the location of the electrical system, voltage levels, line distribution, tag items, and data sets. The key equipment includes transformers, circuit breakers, disconnectors, busbars, capacitors, and reactors.

[0012] Further, the high-risk circuit data includes: high-risk circuit and its load operation data, high-risk circuit historical operation data, protection equipment data, sensing equipment data, historical fault data, user behavior data, equipment or cable line maintenance record data, environmental data, and network topology data.

[0013] Further, the preprocessing includes: cleaning and standardizing the high-risk circuit data, and screening data useful for risk identification.

[0014] Further, the method for performing feature selection on the preprocessed high-risk circuit data includes: dividing the high-risk circuit data into equipment-side data and environmental data. For environmental data, first perform variational mode decomposition on the environmental data, and for each decomposed mode, use the aggregate statistical method to extract relevant features for feature selection; based on the equipment-side data and relevant features, use the chi-square test method to measure the deviation between the observed data and the hypothesized data, and calculate the probability value; if the probability value is less than the significance level, retain the corresponding data, otherwise, delete it.

[0015] Further, the method for performing feature fusion on the selected features includes: performing normalization processing on the selected features, forming a high-dimensional feature matrix with the normalized features, and using the principal component analysis method to reduce the dimension of the high-dimensional feature matrix, mapping the high-dimensional feature matrix to a low-dimensional feature matrix to obtain the fused features.

[0016] Further, the method for inputting the extracted local features into a one-dimensional convolutional layer to obtain refined features includes: inputting the local features into a one-dimensional convolutional layer for a convolutional operation of weighted summation and adding a bias, performing batch normalization on the result of the convolutional operation, inputting the normalized features into an activation function, and inputting the features output by the activation function into a max pooling layer for downsampling to obtain refined features.

[0017] Further, during the training process of the deep learning model, the cheetah algorithm with adaptive parameter adjustment is used to optimize the hyperparameters of the model. The method includes: generating a random cheetah group, where each cheetah individual represents a potential solution; calculating the optimal fitness and the corresponding optimal position of each cheetah, and the cheetah moves quickly based on the current position and the global optimal solution, selects search, wait, and attack strategies, explores a new solution space, updates the cheetah position, updates the optimal fitness and the optimal position, and determines whether the termination condition is reached. If so, the optimal solution is output to optimize the learning rate, the number of attention heads, and the hidden layer dimension hyperparameters of the deep learning model.

[0018] Further, during the training process of the deep learning model, the incremental learning method of class balance and data selection is used to automatically adjust the learning strategy of the model according to the change of data. The method includes: loading and training the model, freezing the pre-trained model, adding a new layer, training the new layer, fine-tuning, and saving and loading the new layer.

[0019] Further, the fault identification result includes the fault type and the risk level.

[0020] The second aspect of the present invention provides a high-risk circuit fault identification system.

[0021] A high-risk circuit fault identification system includes: A data acquisition module configured to: determine a high-risk circuit, acquire high-risk circuit data, and perform preprocessing; A feature fusion module configured to: perform feature selection on the preprocessed high-risk circuit data, and perform feature fusion on the selected features to obtain fused features; An output module configured to: input the fused features into the encoder of the Transformer model to extract local features; input the extracted local features into a one-dimensional convolutional layer to obtain refined features; and based on the refined features, use a decoder to obtain the fault identification result of the high-risk circuit. Among them, during the training process of the Transformer model, the cheetah algorithm with adaptive parameter adjustment is used to optimize the hyperparameters of the model, and the incremental learning method of class balance and data selection is used to automatically adjust the learning strategy of the model according to the changes in the data.

[0022] Furthermore, the preprocessing includes: cleaning and standardizing the high-risk loop data.

[0023] The third aspect of the present invention provides a computer-readable storage medium.

[0024] A computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the steps in the high-risk loop fault identification method described in the first aspect above.

[0025] The fourth aspect of the present invention provides a computer device.

[0026] A computer device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps in the high-risk loop fault identification method described in the first aspect above.

[0027] The fifth aspect of the present invention provides a computer program product or a computer program.

[0028] The present invention provides a computer program product or a computer program. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the steps in the high-risk loop fault identification method described in the first aspect above.

[0029] Compared with the prior art, the beneficial effects of the present invention are: The present invention aims to solve the problem of automatic identification of high-risk loops through deep learning technology to address the limitations of traditional methods that rely on manual experience, have low efficiency, and are prone to errors. Deep learning can effectively process complex, high-dimensional, and non-linear data, thereby extracting key features and significantly improving the accuracy and efficiency of risk identification. In addition, the present invention strengthens the real-time monitoring and early warning capabilities of the system, can identify potential risks in a timely manner, take preventive measures, and significantly reduce the incidence of mechanical failures and overload events. Through this innovation, the overall safety and reliability of the industrial system are improved, which not only ensures the stable operation of the equipment but also provides a scientific basis for safety management and promotes the development of intelligent monitoring technology.

[0030] The deep learning model of the present invention, which integrates the cheetah algorithm with adaptive parameter adjustment and the incremental learning method of class balance and data selection, can effectively process high-dimensional time series data and improve the recognition accuracy of high-risk circuits. Through the self-attention mechanism, the model dynamically captures the complex time dependencies in the power system, while the incremental learning mechanism enables the model to update parameters in real time when new data is received, further reducing false negatives and false positives and improving the reliability of decision-making.

[0031] The cheetah algorithm with adaptive parameter adjustment adopted by the present invention demonstrates excellent capabilities in hyperparameter optimization, significantly shortening the model training time and computational resource consumption. Combined with the incremental learning method of class balance and data selection, the model can adaptively adjust the learning rate and other key parameters during the continuous learning process, thus quickly finding the optimal solution, improving the convergence speed, and accelerating the model deployment time.

[0032] The present invention has strong real-time data processing capabilities and can quickly analyze and feedback the operating status of the system. The combination of dynamic monitoring capabilities and incremental learning enables the system to give early warnings before risks occur, effectively preventing potential failures and power outages and ensuring the continuity of the normal operation of the system.

[0033] By identifying high-risk circuits in advance and using the incremental learning method of class balance and data selection to optimize the decision-making process, the present invention significantly reduces the occurrence frequency of unexpected failures, reduces maintenance and repair costs. The optimized maintenance strategy is more scientific and reasonable, improves the resource allocation efficiency, and reduces unnecessary expenses.

[0034] Under the technical route of the present invention, the single-loop identification model may need to be improved in terms of identification ability and adaptability. The dual-model scheme can combine their respective advantages to provide stronger risk identification capabilities and adapt to different application requirements and scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The accompanying drawings forming a part of this specification are used to provide a further understanding of the present invention. The schematic embodiments and descriptions thereof of the present invention are used to explain the present invention and do not constitute an improper limitation of the present invention.

[0036] Figure 1 It is a flowchart of the high-risk circuit fault identification method shown in the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0037] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0038] It should be noted that the following detailed descriptions are all illustrative and are intended to provide further explanations of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0039] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they specify the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0040] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of methods and systems according to various embodiments of the present disclosure. It should be noted that each block in the flowchart or block diagram may represent a module, a program segment, or a part of code, and the module, program segment, or part of code may include one or more executable instructions for implementing the logical functions specified in each embodiment. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, or they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the flowchart and / or block diagram, and the combinations of blocks in the flowchart and / or block diagram, may be implemented using a dedicated hardware-based system for performing the specified functions or operations, or may be implemented using a combination of dedicated hardware and computer instructions.

[0041] Embodiment 1 As Figure 1 shown, this embodiment provides a method for identifying high-risk loop failures. This embodiment takes the application of this method to a server as an example for illustration. It can be understood that this method can also be applied to a terminal, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network servers, cloud communications, middleware services, domain name services, security services CDN, and big data and artificial intelligence platforms. The terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited thereto. The terminal and the server can be directly or indirectly connected through wired or wireless communication methods, and this application does not make any restrictions here. In this embodiment, the method includes the following steps: Step 1: Classify and sort out (which high-risk loops in industrial enterprises).

[0042] Power system circuit: Involving main transformers, circuit breakers, relays, etc., it is prone to cause large-scale system failures or electrical fires.

[0043] Control system circuit: Such as PLC and DCS systems, if a failure occurs, it may lead to out-of-control production processes.

[0044] Power system circuit: Including motors and drive equipment, its potential risks mainly come from mechanical failures or overload situations.

[0045] High-temperature and high-pressure system circuit: Involving boilers and pressure vessels, improper operation may lead to explosions or leaks.

[0046] Data acquisition and monitoring system circuit: A system used to monitor production parameters in real time. If there are problems with data acquisition, it may affect production efficiency or safety. The communication circuit between the control system and the monitoring system. If there are network or interface failures, it may affect the normal operation of the system.

[0047] Electrical start and stop circuit: Control devices such as buttons, switches, and relays. If there is a failure, the equipment may not be able to start. The circuit used to stop the equipment emergently or normally. If it malfunctions, the equipment may not be able to be stopped when needed.

[0048] Fire protection circuit: Mainly composed of fire detectors, sprinkler systems, and fire extinguishers, etc. The risk is that failure to detect and extinguish the fire source in time may cause significant losses.

[0049] Toxic substance emission circuit: The toxic substance emission circuit involves the storage, treatment, and emission systems of toxic substances. Its main risk is that leakage and emission may pose threats to the environment and personal safety.

[0050] Step 2: The objects to be identified (what key equipment is included in the circuit, what is the operating state of the equipment, where it is in the electrical system, what voltage level, under which transformer to which line), tag items, and data sets.

[0051] (1) Key equipment: Transformers: Main transformers, standby transformers.

[0052] Circuit breakers: Switching devices used to protect and control circuits.

[0053] Isolating switches: Switches used for equipment maintenance and repair.

[0054] Busbars: Power transmission channels in high-voltage transmission lines.

[0055] Capacitors and reactors: Used for power factor compensation and system stability.

[0056] (2) Operating state of the equipment Normal operation: The device operates within the normal operating range.

[0057] Standby: The device is in a standby state but not in use.

[0058] Fault: The device has a fault and cannot operate normally.

[0059] Under maintenance: The device is under maintenance or repair and is temporarily unavailable.

[0060] Shutdown: The device has stopped due to a fault or planned outage and has not resumed operation.

[0061] (3) Location of the electrical system Main substation: The main power conversion and distribution center.

[0062] Distribution substation: A secondary substation used to distribute power to specific areas.

[0063] Feeder: A power line branched out from the substation to supply power to each load point.

[0064] Branch circuit: A more subdivided line branched out from the feeder to supply power to specific equipment.

[0065] (4) Voltage levels Low voltage: Usually refers to 220V, 380V, supplied to terminal equipment.

[0066] Medium voltage: Usually refers to 1kV to 35kV, used for regional power distribution.

[0067] High voltage: Usually refers to above 35kV, used for long-distance power transmission and main substations.

[0068] (5) Line distribution Transmission line: A high-voltage power transmission line from the power plant to the substation.

[0069] Distribution line: A line that distributes power from the substation to the end users.

[0070] (6) Label items Equipment number: A number that uniquely identifies the equipment.

[0071] Equipment type: Such as transformer, circuit breaker, busbar, etc.

[0072] Voltage level: The voltage level of the equipment.

[0073] Location: The location of the equipment in the power grid (such as substation location, line number).

[0074] Status: The operating status of the equipment (normal, faulty, under maintenance).

[0075] Maintenance record: The maintenance and repair records of the equipment.

[0076] Operation record: The historical record of equipment operation.

[0077] (7) Dataset Equipment list: includes detailed information of the equipment, such as model, specification, and installation location.

[0078] Operation data: real-time parameters of the equipment such as current, voltage, power, etc.

[0079] Fault record: equipment fault information, including fault type and handling results.

[0080] Maintenance records: Historical maintenance and repair records of equipment, including repair time and content.

[0081] Operation record: The operation history of the equipment, including start, stop and other operation records.

[0082] Step 3: Feature extraction (what kind of feature should be collected to identify different risk loops)

[0083] High-risk circuits and their load operating data: the circuit where the user's production load is located and the operating power of the main electrical equipment in the circuit, the circuit where the user's security load is located and the operating power of the main electrical equipment.

[0084] Historical operation data of high-risk circuits: historical load data of important circuits where users’ production loads are located (including load type, load amount, load balance), power quality data (including voltage, current, voltage fluctuation and flicker, power factor and harmonics). The data provided needs to include accidental power outage data.

[0085] Protection equipment data: protection equipment configuration data, including configuration information of circuit breakers, fuses, relay protection devices, etc.; protection equipment action record data, including the number and time of protection equipment actions.

[0086] Sensor equipment data: vibration sensor data, including equipment vibration conditions; sound sensor data, including abnormal sound detection conditions; temperature sensor data, including temperature data of cables or electrical equipment.

[0087] Historical fault data: Detailed fault information of the faulty cable or equipment, including fault type (short circuit, overload, ground fault, etc.), fault cause and fault frequency.

[0088] User behavior data: electricity consumption pattern data of important industrial high-voltage users (users’ electricity consumption habits and patterns), production plan data (users’ production plans and load forecasts), and operation record data (users’ operation records on the power system).

[0089] Device or cable line maintenance record data: Maintenance time data, including the time and frequency of regular maintenance; Maintenance content data, including specific maintenance items and measures; Maintenance result data, including the post-maintenance status and effectiveness evaluation.

[0090] Environmental data: Temperature data: including ambient temperature and equipment temperature; Meteorological data, including weather conditions such as rainfall, lightning, etc.; Humidity data.

[0091] Network topology data: Topological structure diagram of the power grid: including the layout of lines, substations, circuit breakers, etc. The connection methods and relationships between each loop help identify potential risk points and the location of faulty equipment.

[0092] Step 4: Algorithm design (what kind of algorithm is used to complete loop identification, key point, patent requirement) First, it is necessary to divide the risk levels. According to system reliability data, equipment status, environmental factors, emergency response capabilities, and economic impacts, it is divided into no, low, medium, and high risks.

[0093] No risk: There are no significant risks in the loop, almost no possibility of failure, the system operates stably, and it will not have any impact on equipment and the environment.

[0094] Low risk: The impact on equipment operation is limited, it will not cause major problems, the probability of equipment failure is low, and it usually only affects a single device or a small area of power supply, such as local power fluctuations or short-term power outages. In addition, when the fire protection loop is at this level, it may only cause a minor fire risk, and the leakage of the toxic substance discharge loop has a small impact and usually does not pose a threat to personal safety.

[0095] Medium risk: It may cause certain interference to the normal operation of equipment, but it will not lead to a complete system paralysis. The probability of a fault or event occurring is medium, and it may affect a larger range of equipment. For example, a substation equipment failure may cause a large-scale power outage and require a certain amount of time to recover. At the same time, the fire protection loop may face certain fire hazards at this level, and the toxic substance discharge loop may have a small-scale leakage, resulting in environmental impacts.

[0096] High risk: It has a serious impact on equipment operation and may lead to large-scale power outages or system crashes. High-probability events such as extreme weather conditions and major equipment failures may cause large-scale power outages, long-term power supply interruptions, or have a significant impact on the economy and society. At this risk level, the fire protection loop faces a greater fire risk, and the toxic substance discharge loop may have a serious leakage, endangering personal safety and environmental health.

[0097] Based on the rich feature data provided above and the classification of risk levels, a method for identifying high-risk circuits can be designed, combining deep learning and data analysis techniques to identify potential high-risk circuits and optimize the reliability of the power system. The following is a detailed step design: Step 4.1: Data collection and preprocessing Step 4.1.1: Data collection: High-risk circuit and its load operation data: including the operating power of the main electrical equipment in the production load circuit and the security load circuit.

[0098] Historical operation data: including load data, power quality data, and accident power outage data.

[0099] Protection equipment data: including configuration data and action record data.

[0100] Sensing equipment data: including vibration, sound, and temperature sensor data.

[0101] Historical fault data: fault information and types.

[0102] User behavior data: power consumption pattern data, production plan data, and operation record data.

[0103] Maintenance record data: maintenance time, content, and result data.

[0104] Environmental data: environmental temperature, meteorological data, and humidity data.

[0105] Network topology data: power grid topology structure diagram.

[0106] Step 4.1.2: Data preprocessing The initial data may be disturbed by missing values, outliers, and noise. First, the data needs to be cleaned.

[0107] Since the scales of different features may be different, the data needs to be standardized to reach the same scale and improve the training effect of the model.

[0108] Select features useful for risk identification, such as load balance degree, power factor, etc.

[0109] Step 4.2: Feature fusion and selection.

[0110] Step 4.2.1: Feature selection Feature selection is a very important step in the machine learning modeling process, which selects the most useful features from a large number of original features. This not only helps to improve the performance of the model, but also reduces the training time, avoids overfitting, and improves the interpretability of the model. Since the types of features are different, feature selection is performed on different types of data separately, which can be roughly divided into device-side data (load data, power quality data, etc.) and environmental data (temperature, humidity, etc.). For environmental data, variational mode decomposition (VMD) is first performed on it, and for each mode obtained after decomposition, aggregation statistical methods (such as mean, standard deviation, kurtosis, skewness, etc.) are used to extract features such as trends and periodicity in the time series data for feature selection.

[0111] For feature selection, the Chi-square Test method is used, and the p-value (probability value) is used to measure the deviation between the observed data and the hypothesis. In the context of feature selection, the p-value represents the significance level of the independence between the feature and the target variable. Specifically, a low p-value indicates a significant association between the feature and the target variable, while a high p-value indicates independence between the feature and the target variable, with no significant association. Here, the p-value is compared with the significance level (α). The significance level is set to 0.05. A result with a p-value less than 0.05 indicates a significant correlation between the feature and the target variable, and this feature can be retained.

[0112] p-value < 0.05: It indicates that there is a significant correlation between the feature and the target variable with a 95% confidence level, and it is considered that this feature should be retained.

[0113] p-value ≥ 0.05: It indicates that there is no significant correlation between the feature and the target variable, and it is considered that this feature should be excluded.

[0114] Step 4.2.2: Feature fusion In the machine learning modeling process, feature fusion is an important step, especially when dealing with different types of data. The goal of feature fusion is to effectively combine the data features from different sources or different types so that their advantages can be utilized simultaneously to improve the performance of the model. For the features selected in the feature selection method, fusion is to convert these features into a unified format or combine them into more meaningful features for training the model.

[0115] Feature fusion adopts the principal component analysis (PCA) method. First, the selected features are standardized, and then all the feature data to be fused are combined into a high-dimensional feature matrix. Subsequently, PCA is used to reduce the dimension of the matrix. By extracting the main components in the data, PCA maps the high-dimensional features to a lower-dimensional space while preserving as much important information in the data as possible, and finally generates a new set of features that are more representative and informative, thus providing a more effective input for the model.

[0116] Step 4.3: Model Design and Evaluation The present invention proposes an intelligent prediction and detection framework that combines a deep learning model (Transformer), an adaptive parameter adjustment cheetah algorithm (CAPA), and an incremental learning method strategy for class balance and data selection, aiming to solve the problems of accuracy and adaptability in industrial load prediction and fault detection. Through the innovative integration of time series data processing capabilities, global optimization algorithms, and dynamic learning mechanisms, this designed model not only improves the prediction accuracy of the model but also enhances the adaptive ability of the model under changing data conditions. Specifically, the model consists of three main parts: the Transformer model is used to process time series data, the adaptive parameter adjustment cheetah algorithm (CAPA) is used for hyperparameter optimization, and the incremental learning method strategy for class balance and data selection ensures that the model can be updated in real time and continuously learn when new data arrives.

[0117] In the present invention, the Transformer model, as the core component, undertakes multiple key tasks from the input features to the output results. First, the input time series data are preprocessed, and positional encoding is added to the features at each time step to inject temporal order information. Since the Transformer model itself does not have the ability to process temporal order, positional encoding enables the model to understand the relative order between time steps by adding it to the original input features, thereby processing the temporal information in the time series data. Next, the core mechanism of the Transformer model, the self-attention mechanism, comes into play. The self-attention mechanism allows the model to automatically learn the dependencies between each time step when processing the input sequence. By calculating the correlation between each time step and other time steps, the model assigns a weight to each time step, and these weights reflect the degree of influence of different time steps on the current time step. In this way, the model can dynamically adjust the contribution value of each feature, thus more accurately capturing the long-term dependencies in the sequence. This mechanism greatly improves the ability of traditional time series models to capture long-term dependencies.

[0118] To further enhance the model's expressive power, Transformer introduces the multi-head attention mechanism. The key to the multi-head attention mechanism lies in dividing the input feature space into multiple sub-spaces and independently performing self-attention calculations in each sub-space. By calculating multiple attention heads in parallel, the model can capture diverse dependency patterns in time-series data from different perspectives, thereby better understanding the complex relationships in the data. For example, short-term changes and long-term periodic patterns may have different dependency structures, and the multi-head attention mechanism can handle these different levels of dependencies simultaneously.

[0119] Looking further, the structure of Transformer includes two main parts: the encoder and the decoder. The encoder is responsible for processing the input data and extracting features. It consists of multiple self-attention layers and feed-forward neural network layers. Each encoder layer can capture the relationships between different time steps according to the self-attention mechanism. Through multi-layer stacking, the encoder can gradually perform feature transformation and information compression on the input data, extracting deep temporal patterns.

[0120] The decoder part is responsible for generating the final prediction results from the features output by the encoder. The decoder learns to decode temporal information through the self-attention mechanism and simultaneously uses the encoder-decoder attention mechanism to focus on the global information of the input sequence. This dual attention mechanism enables the decoder to consider both the global information provided by the encoder and the previously generated results when generating the output, thereby generating more accurate predictions. In the decoder part of the model, decoder 1 and decoder 6 each have different roles. The decoder is usually used to convert the encoded feature information into specific outputs. The results output by the decoder will undergo a linear transformation and finally be converted into specific prediction values for the target task, such as load prediction or failure probability.

[0121] However, traditional Transformer still faces the problem of limited local feature mining ability when dealing with complex data. Although the self-attention mechanism performs well in modeling long-term dependencies, due to its calculation depending on the entire sequence, there are certain limitations in capturing local dependency relationships, which may affect the prediction accuracy. To solve this problem, a convolutional neural network is introduced to extract local features after the Transformer encoder. This method enables the model to not only effectively model long-range dependencies but also better capture the local associations between adjacent time steps in the short term, thereby improving the prediction accuracy.

[0122] First, the input data X is fed into a one-dimensional convolutional layer (Conv). This layer performs weighted summation of the features and adds a bias. The formula is expressed as: (1) where, is the weight matrix of the convolutional layer, is the bias term, and C represents the extracted local features. The main purpose of this step is to extract local features from the data and further supplement the global information from the Transformer encoder.

[0123] To improve the generalization ability of the model, batch normalization (BatchNorm1d) is performed after the convolutional operation to accelerate training and reduce the dependence on the initialized weights: (2) where D represents the features after normalization processing.

[0124] Next, the exponential linear unit (ELU) is used as the activation function. ELU can provide a smooth non-linear transformation, help alleviate the vanishing gradient problem, and better capture complex data features: (3) where E represents the features output by the activation function.

[0125] Finally, a max pooling layer (MaxPool1d) is added to downsample the features. The pooling layer extracts the important features of the local area by summarizing the information of adjacent data points, reducing the data dimension and the computational complexity: (4) where M represents the features output by the max pooling layer.

[0126] This process not only makes the model less sensitive to local positions but also enhances the translational invariance of the features, which is very important for prediction. Through this design, local and global information can be effectively combined to extract more refined feature representations from the output of the Transformer encoder.

[0127] The present invention introduces an improved cheetah optimization algorithm, the Cheetah Algorithm with Adaptive Parameter Adjustment (CAPA). This is a heuristic optimization algorithm based on the hunting behavior of cheetahs, aiming to optimize the hyperparameters in the Transformer model. The cheetah algorithm can quickly find the optimal hyperparameter combination by simulating the acceleration and deceleration strategies of cheetahs during the hunting process and combining the balance mechanism of global search and local refinement, thus significantly improving the training efficiency and prediction accuracy of the model. The Cheetah Algorithm with Adaptive Parameter Adjustment (CAPA) is a heuristic optimization algorithm based on the hunting behavior of cheetahs, aiming to solve optimization problems by simulating the characteristics of cheetahs such as speed changes and strategy adjustments during the pursuit of prey. Traditional cheetah algorithms usually use fixed parameters for optimization, but this method may perform poorly in complex high-dimensional problems. To solve this problem, an adaptive parameter adjustment mechanism is introduced to dynamically adjust key parameters such as speed, step size, inertia weight, and temperature, thereby improving the adaptability and efficiency of the algorithm. The speed and step size are larger in the initial stage of the algorithm for extensive global search and gradually decrease in the later stage of the search for more refined local search; the adaptive adjustment of the inertia weight and temperature helps to balance global exploration and local exploitation, avoiding premature convergence to local optimal solutions. Through these dynamic adjustments, the algorithm can flexibly respond to changes in the search environment at different stages, improving the convergence speed and the quality of the solution, and avoiding the problem of getting stuck in local optimal solutions. In this application, CAPA mainly optimizes hyperparameters such as the learning rate, the number of attention heads, and the hidden layer dimension, and these adjustments effectively improve the convergence speed and overall performance of the model. Especially when dealing with large-scale time series data, CAPA demonstrates excellent optimization capabilities, can quickly adapt to complex data patterns while ensuring accuracy, and improve the generalization ability and stability of the model. Its main steps are as follows: initialize parameters, calculate the fitness function value, adaptively adjust parameters, update the optimal solution, judge whether the termination condition is reached. If so, output; otherwise, continue iterative update. Through the cheetah algorithm with adaptive parameter adjustment, the model can not only obtain a faster training process but also maintain efficient learning ability when facing a constantly changing data environment.

[0128] To ensure that the model can continuously adapt to new data in a dynamically changing industrial environment, the present invention adopts an incremental learning strategy. Incremental learning is a learning method that can continuously receive new data and gradually update on the basis of the existing model without having to retrain the entire model each time. This not only significantly improves computational efficiency but also reduces the storage and processing requirements for a large amount of historical data. In the traditional training mode, as the amount of data continuously increases, the model often needs to be retrained on all historical data, which consumes a large amount of computational resources and time. However, through incremental updates, incremental learning enables the model to flexibly adapt to new data, avoiding the high cost of training from scratch. Its main steps are: initializing the model, obtaining new data, data selection, handling class imbalance problems, updating the model, model evaluation and verification, and model output.

[0129] In traditional machine learning methods, when the model is trained on new data, the knowledge of the old data may be covered by the new data, resulting in the model performing well in new tasks but having poor performance in old tasks. Through designing appropriate algorithms, incremental learning can retain old knowledge while quickly adapting to new data when updating the model, which is particularly important for long-running industrial systems as these systems often face challenges of long-term evolution and data distribution changes.

[0130] The incremental learning method of class balance and data selection has significant advantages over traditional incremental learning methods in solving problems such as data distribution changes and class imbalance. Traditional incremental learning methods usually rely on a single data update mechanism. As new data is continuously added, the model is easily affected by catastrophic forgetting and gradually loses its memory of old tasks. Moreover, since it does not consider the imbalance between classes, it often performs poorly on minority class samples, leading to a decline in overall performance. In contrast, the incremental learning method of class balance and data selection can effectively ensure the sufficiency of training for each class, especially minority class samples, while learning new tasks by introducing strategies such as dynamic resampling, sample selection, and data weighting, avoiding the bias of the model towards imbalanced data. In addition, these methods usually also combine memory replay or gradient-based regularization techniques. By storing and replaying historical data, they help the model maintain its memory of old tasks when learning new tasks, thereby preventing catastrophic forgetting and enhancing the stability of the model in multi-task learning. Through appropriate sample selection and data processing, incremental learning can continuously optimize learning performance when facing large-scale and dynamically changing data, ensuring good generalization for all classes. Therefore, the incremental learning method of class balance and data selection not only enhances the adaptability and long-term learning ability of the model but also makes its performance in complex task environments more balanced and efficient, overcoming the deficiencies of traditional incremental learning methods in practical applications.

[0131] In industrial load forecasting, the patterns of electricity demand and energy consumption are often non-linear and affected by many factors, such as weather, holidays, equipment operating status, etc. Traditional models may need to be updated frequently. Incremental learning, by making progressive updates in real-time data streams, can ensure that in a changing data environment, the model can always reflect the latest trends, avoid the influence of outdated data, and thus improve the accuracy and reliability of load forecasting.

[0132] For equipment fault detection, incremental learning also plays an important role. Industrial equipment may exhibit different fault patterns during long-term operation. Traditional models often rely on historical fault data for training, and changes in fault patterns may lead to model failure. Incremental learning can process newly occurring fault cases in real-time and incorporate these new patterns into consideration when updating the model, enabling the system to adapt to the evolving fault patterns, detect potential problems in a timely manner, and reduce equipment downtime and maintenance costs.

[0133] The model described in the present invention integrates three major technologies: the Transformer model, the Cheetah Algorithm with Adaptive Parameter Adjustment (CAPA), and the incremental learning method for class balance and data selection. The input of the model first passes through a data preprocessing module, and the processed time-series data enters the Transformer model for prediction tasks. The incremental learning module adjusts the model parameters according to newly arriving real-time data, while the Cheetah Algorithm with Adaptive Parameter Adjustment optimizes the hyperparameters during each training to ensure that the model is always in an optimal state. The entire system can be dynamically updated in a continuous industrial data stream, ensuring the real-time and accuracy of load forecasting and fault detection tasks.

[0134] Experimental results on multiple industrial datasets show that the designed model of the present invention can significantly improve the prediction accuracy and training efficiency compared with traditional methods. In the load forecasting task, the Root Mean Square Error (RMSE) and F1 score of the model are better than those of the existing technologies. At the same time, the incremental learning mechanism enables the model to be updated in real-time as the data changes, avoiding the accuracy degradation caused by fixed models in traditional methods. In addition, the introduction of the Cheetah Algorithm with Adaptive Parameter Adjustment further improves the training efficiency of the model on large-scale datasets, especially in the processing of complex time-series data, demonstrating good generalization ability.

[0135] Step 4.4: High-risk loop identification and analysis Use the trained model to perform risk scoring and classification on the loops.

[0136] Risk ranking: Rank the loops according to the risk scores to identify high-risk loops.

[0137] Fault prediction: Analyze historical fault data to predict possible fault types and occurrence probabilities.

[0138] Step 4.5: Result Application and Feedback Deploy the model for real-time monitoring, automatically identify high-risk circuits and issue early warnings.

[0139] The present invention introduces a convolutional neural network into Transformer. This method enables the model to not only effectively model long-range dependencies but also better capture local correlations between adjacent time steps in the short term.

[0140] The present invention introduces an incremental learning method for class balance and data selection, which can automatically adjust the learning strategy of the model according to data changes. By designing a suitable algorithm, it can retain old knowledge when updating the model and quickly adapt to new data.

[0141] The present invention introduces a cheetah algorithm with adaptive parameter adjustment, aiming to optimize the hyperparameters in the Transformer model. It can quickly adapt to complex data patterns while ensuring accuracy, improving the generalization ability and stability of the model.

[0142] Step 5: What is the identification result (taking a certain industry as an example)?

[0143] Objective: To detect abnormal states or faults in the power grid in a timely manner, reduce power outage time and improve power grid reliability.

[0144] Identification result: Fault detection: Identify possible fault points or abnormal patterns in the power grid, such as abnormal fluctuations in transmission lines.

[0145] Fault prediction: Based on historical data and real-time monitoring data, predict possible faults so as to take measures in advance.

[0146] Fault classification: Distinguish different types of faults (such as short circuits, overloads) and risk levels (low, medium, high), helping maintenance personnel quickly locate problems.

[0147] Embodiment 2 This embodiment provides a high-risk circuit fault identification system.

[0148] A high-risk circuit fault identification system includes: A data acquisition module, which is configured to: determine high-risk circuits, acquire high-risk circuit data, judge whether there is a data imbalance problem, and perform preprocessing; A feature fusion module, which is configured to: perform feature selection on the preprocessed high-risk circuit data using the chi-square test method, and perform feature fusion on the selected features using the principal component analysis method to obtain fusion features; An output module, configured to: input the fused features into an encoder of a deep learning model to extract local features; input the extracted local features into a one-dimensional convolutional layer to obtain refined features; and based on the refined features, use a decoder to obtain a fault identification result of a high-risk loop. Wherein, during the training process of the deep learning model, a cheetah algorithm with adaptive parameter adjustment is used to optimize the hyperparameters of the model, and an incremental learning method of class balance and data selection is used to automatically adjust the learning strategy of the model according to the change of data.

[0149] In some embodiments, the high-risk loops include: power system loops, control system loops, power system loops, high-temperature and high-pressure system loops, data acquisition and monitoring system loops, electrical start and stop loops, fire protection loops, and toxic substance emission loops.

[0150] In some embodiments, after determining the high-risk loops, it further includes: obtaining key equipment, operating status of the equipment, location of the electrical system, voltage level, line distribution, tag items, and data sets in the high-risk loops, and the key equipment includes transformers, circuit breakers, disconnectors, busbars, capacitors, and reactors.

[0151] In some embodiments, the high-risk loop data includes: high-risk loops and their load operation data, high-risk loop historical operation data, protection equipment data, sensing equipment data, historical fault data, user behavior data, equipment or cable line maintenance record data, environmental data, and network topology data.

[0152] In some embodiments, the preprocessing includes: cleaning and standardizing the high-risk loop data, and screening the data useful for risk identification.

[0153] In some embodiments, the feature fusion module is further configured to: divide the high-risk loop data into equipment-side data and environmental data. For the environmental data, first perform variational mode decomposition on the environmental data, and for each decomposed mode, use an aggregation statistical method to extract relevant features for feature selection; based on the equipment-side data and the relevant features, use a chi-square test method to measure the deviation degree between the observed data and the hypothesized data, and calculate the probability value; if the probability value is less than the significance level, retain the corresponding data, otherwise, delete it.

[0154] In some embodiments, the feature fusion module is further configured to: perform standardization processing on the selected features, form a high-dimensional feature matrix with the standardized features, and use the principal component analysis method to reduce the dimension of the high-dimensional feature matrix, and map the high-dimensional feature matrix to a low-dimensional feature matrix to obtain fused features.

[0155] In some embodiments, the output module is further configured to: perform a convolution operation of weighted summation and adding a bias on the local features by inputting them into a one-dimensional convolutional layer, perform batch normalization on the result after the convolution operation, input the normalized features into an activation function, and input the features output by the activation function into a max pooling layer for downsampling to obtain refined features.

[0156] In some embodiments, the cheetah optimization algorithm with adaptive parameter adjustment plays a key role. It is configured to: generate a random cheetah group, where each cheetah individual represents a potential solution; calculate the optimal fitness and the corresponding optimal position of each cheetah. The cheetah moves quickly based on the current position and the global optimal solution, selects search, wait, and attack strategies, explores the new solution space, updates the cheetah position, updates the optimal fitness and the optimal position, and determines whether the termination condition is reached. If so, output the optimal solution and optimize the hyperparameters of the learning rate, the number of attention heads, and the hidden layer dimension of the deep learning model.

[0157] In some embodiments, the incremental learning module for class balance and data selection plays a key role. This module first initializes the parameters, initializes the model by operations such as obtaining the function value of the new data through an adaptive adjustment algorithm and updating the optimal solution, then obtains the new data, and then performs data selection and processes the class imbalance problem to ensure the data quality. During the model training process, the model is continuously optimized through steps such as hyperparameter optimization, model evaluation, and performance verification. In each iteration process, it is determined whether the termination condition is reached. If not, continue to update the model for learning. If so, output the model result. Incremental learning is reflected in this process as continuously obtaining new data, processing the data imbalance problem, and optimizing and updating the model according to the new data and model evaluation results, enabling the model to continuously adapt to the new data and environment. The method includes: initializing the model, obtaining new data, data selection, processing the class imbalance problem, updating the model, model evaluation and performance verification, and model output.

[0158] In some embodiments, the deep learning model is the core component. First, the input time series data is preprocessed, and positional encoding is added to the features of each time step to inject chronological information. Next, the core mechanism of the deep learning model, the self-attention mechanism, starts to play a role. The self-attention mechanism allows the model to automatically learn the dependencies between each time step when processing the input sequence. By calculating the correlation between each time step and other time steps, the model assigns a weight to each time step, and these weights reflect the influence degree of different time steps on the current time step. In this way, the model can dynamically adjust the contribution value of each feature, thereby more accurately capturing the long-term dependencies in the sequence.

[0159] The decoder part is responsible for generating the final prediction results from the features output by the encoder. The decoder learns to decode the temporal information through the self-attention mechanism.

[0160] Meanwhile, the encoder-decoder attention mechanism is used to focus on the global information of the input sequence. This dual attention mechanism enables the decoder to consider both the global information provided by the encoder and the previously generated results when generating the output, thereby generating more accurate predictions. The results output by the decoder will undergo a linear transformation and finally be converted into specific prediction values for the target task, such as load prediction or fault occurrence probability.

[0161] In some embodiments, the fault identification result includes the fault type and the risk level.

[0162] Embodiment III This embodiment provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the steps in the high-risk loop fault identification method described in Embodiment I above.

[0163] Embodiment IV This embodiment provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps in the high-risk loop fault identification method described in Embodiment I above.

[0164] Embodiment V This embodiment provides a computer program product or a computer program. The computer program product or the computer program includes computer instructions, which are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the steps in the high-risk loop fault identification method described in Embodiment I above.

[0165] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a hardware embodiment, a software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories and optical memories, etc.) containing computer-usable program code.

[0166] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate a means for implementing the functions specified in one or more flows and / or blocks. Figure 1 in one or more flows and / or blocks Figure 1 in one or more blocks.

[0167] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including an instruction means that implements the functions specified in one or more flows and / or blocks. Figure 1 in one or more flows and / or blocks Figure 1 in one or more blocks.

[0168] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more flows and / or blocks. Figure 1 in one or more flows and / or blocks Figure 1 in one or more blocks.

[0169] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM), etc.

[0170] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A high-risk circuit fault identification method, characterized in that: include: Identify high-risk loops, obtain high-risk loop data, and perform preprocessing; Perform feature selection on the preprocessed high-risk loop data, perform feature fusion on the selected features, and obtain fusion features; Input the fused features into the encoder of the deep learning model to extract local features; The extracted local features are input into the one-dimensional convolution layer to obtain refined features; Based on the refined features, a decoder is used to obtain the fault identification results of high-risk circuits; During the training process, the deep learning model uses the cheetah algorithm with adaptive parameter adjustment to optimize the model's hyperparameters, and uses an incremental learning method with category balance and data selection to adjust the model's learning strategy according to data changes.

2. The high-risk circuit fault identification method according to claim 1, characterized in that: The high-risk circuits include: power system circuits, control system circuits, power system circuits, high-temperature and high-pressure system circuits, data acquisition and monitoring system circuits, electrical start and stop circuits, fire protection circuits, and toxic substance emission circuits.

3. The high-risk circuit fault identification method according to claim 1, characterized in that: After determining the high-risk circuit, it also includes: obtaining the key equipment in the high-risk circuit, the operating status of the equipment, the location of the electrical system, the voltage level, line distribution, label items and data sets. The key equipment includes transformers, circuit breakers, disconnectors, busbars, capacitors and reactors.

4. The high-risk circuit fault identification method according to claim 1, characterized in that: The high-risk circuit data includes: high-risk circuit and its load operation data, high-risk circuit historical operation data, protection equipment data, sensor equipment data, historical fault data, user behavior data, equipment or cable line maintenance record data, environmental data and network topology data.

5. The high-risk circuit fault identification method according to claim 1, characterized in that: The preprocessing includes: cleaning and standardizing the high-risk loop data.

6. The high-risk circuit fault identification method according to claim 1, characterized in that: The method for performing feature selection on the preprocessed high-risk loop data includes: dividing the high-risk loop data into device-side data and environmental data, and for the environmental data, first performing variational mode decomposition on the environmental data, and using an aggregate statistical method for each mode after decomposition to extract relevant features for feature selection; based on the device-side data and relevant features, using a chi-square test method to measure the degree of deviation between the observed data and the hypothesized data, and calculating the probability value; if the probability value is less than the significance level, retaining the corresponding data, otherwise, deleting it.

7. The high-risk circuit fault identification method according to claim 1, characterized in that: The method for fusing the selected features includes: standardizing the selected features, forming a high-dimensional feature matrix with the standardized features, using a principal component analysis method to reduce the dimension of the high-dimensional feature matrix, mapping the high-dimensional feature matrix to a low-dimensional feature matrix, and obtaining fused features.

8. The high-risk circuit fault identification method according to claim 1, characterized in that: The extracted local features are input into a one-dimensional convolutional layer to obtain refined features; The method includes: inputting local features into a one-dimensional convolution layer for weighted summation and adding a biased convolution operation, performing batch normalization on the results of the convolution operation, inputting the normalized features into an activation function, and inputting the features output by the activation function into a maximum pooling layer for downsampling to obtain refined features.

9. A high-risk circuit fault identification system, characterized in that: include: A data acquisition module, which is configured to: determine high-risk circuits, acquire high-risk circuit data, and perform preprocessing; A feature fusion module is configured to: perform feature selection on the preprocessed high-risk loop data, perform feature fusion on the selected features, and obtain fusion features; An output module is configured to: input the fused features into an encoder of a deep learning model to extract local features; The extracted local features are input into the one-dimensional convolution layer to obtain refined features; Based on the refined features, a decoder is used to obtain the fault identification results of high-risk circuits; During the training process, the deep learning model uses the cheetah algorithm with adaptive parameter adjustment to optimize the model's hyperparameters, and uses an incremental learning method with category balance and data selection to automatically adjust the model's learning strategy according to data changes.

10. The high-risk circuit fault identification system according to claim 9, characterized in that: The preprocessing includes: cleaning and standardizing the high-risk loop data.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the high-risk circuit fault identification method as described in any one of claims 1 to 8 are implemented.

12. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the high-risk circuit fault identification method according to any one of claims 1 to 8 are implemented.

13. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps in the high-risk circuit fault identification method according to any one of claims 1 to 8 are implemented.

Images