Dry type distribution box convenient to maintain

Abstract

The invention, which relates to the technical field of intelligent fault diagnosis and remote maintenance, discloses a convenient-to-maintain dry type distribution box comprising a sensing monitoring module, a main control module, a wireless communication module and a cloud computing module. The sensing monitoring module acquires temperature, humidity, current and voltage data in the distribution box in real time, the master control module processes the data, performs preliminary fault detection and interacts with the cloud computing module through the wireless communication module, and the cloud computing module is responsible for further processing of the data, fault diagnosis, state display and fault report generation; the fault diagnosis method comprises the steps of feature extraction, fault detection, fault positioning, severity evaluation, fault knowledge base construction and maintenance scheme generation. Finally, an operator can output a control instruction according to the fault report, thereby achieving the intelligent maintenance and management of the distribution box. The maintenance difficulty of an existing distribution box is greatly reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent fault diagnosis and remote maintenance, and more specifically to a dry distribution box that is easy to maintain. Background Art

[0002] With the rapid development of social economy and the continuous improvement of science and technology, the reliability and safety of power systems have become increasingly important. In various power facilities, distribution boxes, as key nodes connecting power sources and terminal equipment, undertake important tasks. Especially in industrial production, commercial buildings and residents' lives, the stable operation of distribution boxes directly affects the normal production and life. In recent years, with the development of intelligent and information technology, the functions and performance of distribution boxes have been continuously improved, gradually developing from traditional metal shell distribution boxes to more advanced and reliable dry distribution boxes. These new distribution boxes not only have good moisture resistance, but also have higher safety and longer service life, and are widely used in various environments and scenarios.

[0003] Dry type distribution box is a distribution device designed to prevent moisture intrusion, which is widely used in power systems in outdoor, high humidity and harsh environments. This type of distribution box usually uses high-quality sealing materials and desiccants to keep the interior dry and prevent electrical components from malfunctioning due to moisture. In recent years, with the advancement of materials science and manufacturing technology, the sealing performance and moisture resistance of dry type distribution boxes have been significantly improved. For example, a patent for a waterproof and fog-proof dry type distribution box proposes the use of a permanent magnet switching motor to swap the position of the front molecular sieve slot and the rear molecular sieve slot to complete the moisture absorption cycle and dry the molecular sieve, thereby achieving dry and wet separation, reducing the maintenance personnel's replacement of moisture absorption and drying consumables, and reducing maintenance costs and energy consumption. In addition, the patent also proposes a split component to rationally plan the internal space of the distribution cabinet box, ensure the orderly distribution of electronic components, provide suitable heat dissipation space, and prevent component downtime caused by internal high temperature. It greatly improves the reliability and maintenance convenience of the equipment. However, although the patent solution has been optimized in physical design, it does not involve remote maintenance and diagnostic functions, which means that in some remote areas or places with inconvenient transportation, once the distribution box fails, maintenance personnel still need to spend a lot of time and resources to arrive at the site for troubleshooting and repair. In addition, the internal structure of the distribution box is complex and the electronic components are dense, which makes it difficult to locate the fault. Modern dry distribution boxes integrate a variety of electrical components and control systems. The connections between these components are complex. Once a fault occurs, it is difficult for maintenance personnel to quickly locate the fault point.

[0004] Therefore, the present invention discloses a dry type distribution box which is easy to maintain to solve the above problems. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention discloses a dry-type distribution box that is easy to maintain, thereby improving the maintenance intelligence and maintenance difficulty of the existing distribution box.

[0006] The present invention adopts the following technical solutions:

[0007] A dry distribution box that is easy to maintain, comprising a sensor monitoring module, a main control module, a wireless communication module and a cloud computing module;

[0008] The sensor monitoring module is used to collect real-time data including temperature, humidity, current and voltage inside the distribution box, and convert the real-time collected analog data into digital signals through an analog-to-digital converter, and transmit them to the main control module;

[0009] The main control module is used to process sensor data, perform preliminary fault detection, and output real-time data and fault alarm signals to the wireless communication module; and receive control instructions issued by the cloud computing module, parse and execute diagnostic operations; the main control module includes a data processing submodule, a fault detection submodule, a communication control submodule and an instruction execution submodule; the output end of the data processing submodule is connected to the input end of the fault detection submodule; the output end of the fault detection submodule is connected to the input end of the communication control submodule; the output end of the communication control submodule is connected to the input end of the instruction execution submodule;

[0010] The wireless communication module is used to upload the data processed by the main control module to the cloud computing module and transmit the control instructions of the cloud computing module;

[0011] The cloud computing module is used to receive and process the transmission data of the wireless communication module, display the real-time status data and historical records of the distribution box, generate fault reports, and perform fault diagnosis and processing; the fault report includes the fault type, fault location and repair step suggestions;

[0012] The working method of the cloud computing module for fault diagnosis processing is:

[0013] S1. Extract the real-time operating status of the distribution box through feature engineering, including the fluctuation of current and voltage and the temperature change trend;

[0014] S2, based on the real-time operating status and historical fault data of the distribution box, determine the fault type through a fault detection algorithm based on thresholds and statistical analysis;

[0015] S3. Construct a topological structure model of the distribution box to represent the internal connection relationship of the distribution box, and locate the fault location through a deep learning fusion positioning algorithm based on topological structure and signal propagation;

[0016] S4. According to the fault feature data and historical experience, the severity of the fault is evaluated through the Bayesian network based on the fault impact range, fault development trend vector and the impact index of the fault on the overall operation of the distribution box;

[0017] S5. Based on the fault feature data and historical experience, a fault cloud knowledge base is constructed to store the fault types and the corresponding maintenance step suggestions; the fault cloud knowledge base provides personalized maintenance solutions through an expert system, a maintenance knowledge base and an intelligent reasoning mechanism;

[0018] S6, extracting corresponding maintenance suggestions from the fault cloud knowledge base according to the fault type output by S2, the fault location output by S3 and the severity evaluation output by S4, and outputting a fault report to a computer display platform through a multimodal natural language generation algorithm; the maintenance suggestions include recommended remote operation, maintenance steps, required tools and spare parts information;

[0019] S7. Based on the fault report, the operator outputs control instructions to the main control module, including remote start and stop, load adjustment, and fault isolation instructions.

[0020] As a further technical solution of the present invention, the data processing submodule compares the input data and the reference data bit by bit through a successive approximation ADC, collects sensor data, and stores the data in a ring buffer through a data buffer mechanism; then, the data processing submodule reads data from the ring buffer through a preprocessing unit, the preprocessing unit uses a Kalman filter to smooth the data, remove noise, and converts the data into a standard state distribution through a Z-score standardization method; the preprocessed data is transmitted to a microcontroller for data fusion calculation processing; the microcontroller implements multi-task concurrent processing through a task scheduling and synchronization mechanism; the fault detection submodule includes a threshold detection unit, a statistical analysis unit, and a fault alarm unit; the fault detection submodule sets the upper and lower limit thresholds of temperature, humidity, current and voltage through the threshold detection unit, and compares and determines whether the real-time data exceeds the normal range through a threshold-based fault detection method, and triggers the fault alarm unit if it exceeds the threshold range; the statistical analysis unit analyzes the data trend and abnormal points through a moving average method to further confirm the fault type; when the fault alarm unit is triggered, it generates a fault alarm signal through an interrupt mechanism, and performs a local alarm through an LED and a buzzer; the communication control submodule includes a data encapsulation unit , communication interface unit and instruction receiving unit; the communication control submodule encapsulates the processed data and fault alarm signal into JSON format through the data encapsulation unit, and uses CRC checksum to ensure the integrity of the data; the communication interface unit sends the encapsulated data to the wireless communication module through the serial communication interface, the instruction receiving unit receives the control instructions issued by the cloud computing module through the MQTT protocol, and uses TLS / SSL encryption method to protect the security of data transmission; the instruction execution submodule includes an instruction parsing unit, an operation execution unit and a state feedback unit; the instruction execution submodule receives and parses the control instructions issued by the cloud computing module through the instruction parsing unit, and the instruction parsing unit ensures the accuracy and legality of the instructions through regular expression matching and whitelist mechanism; the operation execution unit realizes remote start and stop through the on and off of the relay control circuit according to the parsed instructions, adjusts the load by pulse width modulation method, and realizes the isolation of faulty equipment through fault isolation mechanism; the state feedback unit monitors the operation results in real time through the state feedback mechanism, and sends the operation results back to the wireless communication module through the serial communication interface, and the state feedback unit also records the operation process and results through the ring log buffer to ensure the transparency and traceability of the operation.

[0021] As a further technical solution of the present invention, the wireless communication module establishes a communication connection with the cloud computing module through a low-power wide area network LPWAN, then encapsulates the data processed by the main control module into messages through the MQTT protocol, and sends the data to the cloud computing module through a publish-subscribe mode; the MQTT protocol ensures the stability of the network connection and the integrity of the data through a heartbeat mechanism and a reconnection mechanism; when the wireless communication module establishes a connection with the cloud computing module, the encryption algorithm and key are negotiated through the TLS / SSL handshake protocol; during data transmission, the wireless communication module adds redundant information to the original data through forward error correction coding to improve the reliability of data transmission; during data transmission, the wireless communication module reduces the amount of data transmission through the compression algorithm LZ77 to improve transmission efficiency; the wireless communication module also ensures the reliability of data transmission and the stability of the system through an automatic retransmission request mechanism and a logging mechanism.

[0022] As a further technical solution of the present invention, the working method of the fault detection algorithm based on threshold and statistical analysis includes the following steps:

[0023] Step u1, extracting feature vectors from real-time data through a sliding window mechanism, including the mean, variance, maximum and minimum values ​​of current, voltage and temperature; and performing dimensionality reduction processing on the extracted feature vectors through principal component analysis;

[0024] Step u2: According to the historical fault data, the threshold is dynamically adjusted through the Kalman filter to adapt to the changes in different time periods and environments; the Kalman filter calculates the dynamic threshold through the error covariance matrix; the calculation formula for dynamically adjusting the threshold through the Kalman filter is:

[0025]

[0026] In formula (1), T k represents the dynamic threshold vector calculated at time k; X k|k-1 is the predicted state vector, which represents the predicted value of the current state before time k; K k represents the Kalman gain, which is used to adjust the difference between the predicted value and the actual observed value; Z k is the observation value vector, which represents the data actually measured at time k; H is the observation matrix, which represents the conversion of the state vector into the representation in the observation space; P k|k-1 is the prediction error covariance matrix, which represents the uncertainty of the current state prediction before time k; Q represents the unit matrix, which is used to maintain the integrity of the matrix operation;

[0027] Step u3: Identify data points that exceed the dynamic threshold through a statistical anomaly detection algorithm, mark them as potential fault points x, and perform time series analysis on potential fault points x through an autocorrelation function to identify the mode and trend of fault occurrence; the formula expression of the autocorrelation function is:

[0028]

[0029] In formula (2), R(c) is the autocorrelation function value; Z t represents the tth data point in the time series, M is the length of the time series, and Z represents the mean of the time series; c is the lag order, which is used to calculate the lag time interval set when calculating the autocorrelation function;

[0030] Step u4: Model the time series analysis results through the Gaussian mixture model and calculate the probability that the data point x belongs to the normal distribution. The calculation formula is:

[0031]

[0032] In formula (3), S is the abnormality detection result, 1 indicates abnormality and 0 indicates normality; ω i is the weight of the i-th Gaussian distribution, which is used to measure the importance of the i-th Gaussian distribution in the mixture model; N(x|μ i , π i ) is the probability density function of the i-th Gaussian distribution, μ i is the mean vector of the i-th Gaussian distribution, used to represent the expected value of the data point x; π i is the covariance matrix of the ith Gaussian distribution, which is used to describe the distribution of the data point x; τ represents the threshold, which is used to determine whether the data point is abnormal; x represents the current data point; T k represents the dynamic threshold vector; It is the inverse of the prediction error covariance matrix, which represents the accuracy of the prediction error and is used to measure the relationship between the data point x and the dynamic threshold T k The deviation between

[0033] Step u5: Based on the abnormal detection result of step u4, the abnormal data point m is classified by a support vector machine to determine the fault type, and the model hyperparameters are optimized by cross-validation to improve the classification accuracy; the formula expression for classifying the data point x by the support vector machine is:

[0034]

[0035] In formula (4), SVM(F) is the classification result of the feature vector F by the support vector machine, a i is the coefficient of the support vector; y i Indicates the label of the support vector; K(Fi ,F) is the kernel function used to calculate the eigenvector F i The similarity between and the feature vector F to be classified, and b represents the bias term.

[0036] As a further technical solution of the present invention, the working steps of the deep learning fusion positioning algorithm based on topological structure and signal propagation include:

[0037] Step t1, construct a distribution box topology matrix G = (V, E); wherein G is a directed graph, representing the topology of the distribution box; V is a node set, representing each electrical component in the distribution box; E is an edge set, representing the connection relationship between components;

[0038] Step t2: Based on the topological structure matrix G of the distribution box, the propagation time of the signal on different paths is calculated by the signal propagation equation. The calculation formula is:

[0039]

[0040] In formula (5), t er represents the propagation time of the signal from node e to node r; L er is the path length from node e to node r; v is the signal propagation speed; X is the attenuation coefficient; R er is the path resistance, which represents the resistance value between node e and node r;

[0041] Step t3: Based on the signal propagation simulation results, construct a signal propagation time matrix T = [t er ], where each element t er represents the propagation time of the signal from node e to node r;

[0042] Step t4: The local and global features of the nodes are fused through the graph convolution function to extract the topological features of the nodes. The formula expression of the graph convolution function is:

[0043]

[0044] In formula (6), H (I) is the node feature matrix of layer I, representing the features of each node in layer I; is the normalized adjacency matrix, which represents the connection relationship between nodes; is the degree matrix, which indicates the degree of each node; σ is the activation function, which is used to introduce nonlinearity; W (I) is the weight matrix, which represents the weight parameters of the Ith layer;

[0045] Step t5: Extract the time series characteristics h of signal propagation through the long short-term memory network t, the LSTM network captures the temporal dependency of signal propagation through the current hidden state and the hidden state of the previous time step;

[0046] Step t6: Use the fully connected layer to transform the topological features H (I+1) and signal propagation characteristics h t The fully connected layer combines the input features to form a comprehensive feature vector F. (I+1) and signal propagation characteristics h t Merge into a feature vector, and perform linear transformation on the concatenated feature vector through the weight matrix and bias vector to obtain the comprehensive feature vector F;

[0047] Step t7: classify the comprehensive feature vector F through the deep learning model, determine the fault location, and optimize the model parameters through the back propagation algorithm; wherein the formula expression for determining the fault location by the deep learning model is:

[0048]

[0049] In formula (7), P(f e |F) is the probability of node e failing, indicating whether node e is a failed node; W f is the model weight matrix, which is used to map the comprehensive feature vector to the probability value; b f Represents the model bias vector, which is used to adjust the output; F r are other feature vectors, representing the features of other nodes.

[0050] As a further technical solution of the present invention, the method for evaluating the severity of a fault using a Bayesian network comprises the following steps:

[0051] Step r1: construct a Bayesian network model. The model construction method is:

[0052] r1.1. According to the structural characteristics of the distribution box and common fault types, the network nodes and node probability distribution are determined using expert knowledge and historical fault data;

[0053] r1.2. Combined with the historical fault data set, the Bayesian network parameters are optimized by the maximum likelihood estimation method;

[0054] r1.3, define the fault feature data, fault impact range, fault development trend vector and fault impact index nodes on the overall operation of the distribution box, and map the conditional probability relationship between the node data through the Bayesian reasoning mechanism;

[0055] Step r2, calculating the posterior probability of the faulty node through a forward reasoning mechanism; the forward reasoning mechanism is based on the node definition and relationship mapping of the Bayesian network, and uses the fault feature data to calculate the posterior probability of the faulty node;

[0056] Step r3: Based on the internal electrical connection diagram of the distribution box, determine the impact of the fault point on other components through the shortest path algorithm in graph theory; and use the weighted average method to comprehensively consider the importance of each component and calculate the overall impact range of the fault;

[0057] Step r4: Model the state data before and after the fault occurs by using the time series analysis ARIMA model to predict the development trend of the fault, and continuously update the fault development trend vector through the sliding window mechanism;

[0058] Step r5, combining the output results of step r3 and step r4, using the AHP hierarchical analysis method to determine the weights of different factors in the evaluation, and according to the weight distribution, using the linear combination method to calculate the impact index of the fault on the overall operation of the distribution box;

[0059] Step r6: Input the indicators outputted from step r3, step r4 and step r5 into the Bayesian network model, calculate the probability distribution of the fault severity through the Bayesian update rule, and determine the fault level according to the probability distribution of the fault severity through the threshold judgment mechanism.

[0060] As a further technical solution of the present invention, the working method of the fault cloud knowledge base is as follows: first, the fault cloud knowledge base extracts data from sensor data, maintenance records, and historical fault reports through a data output ETL process; then, the fault cloud knowledge base constructs fault types and maintenance step suggestions corresponding to the fault types through knowledge representation and data storage methods; the knowledge representation formalizes information including fault types, fault characteristics, and maintenance steps through ontology and semantic web to ensure the structuring and understandability of knowledge; the data storage method uses a relational database and a graph database to store and manage data in the fault cloud knowledge base; then, the fault cloud knowledge base generates personalized maintenance plans through an expert system and an intelligent reasoning mechanism; the expert system converts expert experience and historical data into rules and strategies through a rule engine and a knowledge base management system; the rule engine generates maintenance suggestions by matching rules and input data; the knowledge base management system ensures the update and maintenance of the knowledge base through version control and audit tracking; the intelligent reasoning mechanism identifies fault modes and maintenance strategies through a decision tree training model and generates optimized maintenance suggestions; during the working process of the fault cloud knowledge base, new fault data and maintenance records are received and processed in real time through a data stream processing engine, and the storage information of the cloud knowledge base is updated.

[0061] As a further technical solution of the present invention, the multimodal natural language generation algorithm first comprehensively analyzes the fault type, fault location and severity assessment information through a Bayesian fusion method; the Bayesian fusion method combines information from different data sources through a probability model to generate a comprehensive fault feature description; then, the multimodal natural language generation algorithm generates a natural language text through template matching and content generation methods; the template matching selects the corresponding template according to the fault type and fault location through a predefined template library; the content generation method converts the comprehensive fault feature description into a natural language text through a sequence-to-sequence model; the sequence-to-sequence model encodes the input fault features into an intermediate representation through an encoder-decoder architecture, and then generates a natural language text through a decoder; finally, the multimodal natural language generation algorithm formats the generated natural language text into a structured fault report through a formatting method.

[0062] The positive beneficial effects of the present invention are:

[0063] The present invention can quickly respond to faults and reduce losses caused by faults through the preliminary fault detection and alarm functions of the main control module. When the data processing submodule detects abnormal fluctuations in current or voltage, the fault detection submodule will immediately analyze and determine whether it is a fault. If it is confirmed to be a fault, the main control module will immediately output a fault alarm signal to notify the operator to take action. This rapid response mechanism can not only reduce the losses caused by faults, but also repair problems in a timely manner to ensure the normal operation of the system. In addition, the main control module can also receive control instructions issued by the cloud computing module, parse and execute diagnostic operations, further enhancing the intelligence level of the system.

[0064] More importantly, the solution can accurately locate the fault location through the intelligent diagnosis and fault location functions of the cloud computing module, significantly improving maintenance efficiency. The fault type is determined by a fault detection algorithm based on thresholds and statistical analysis. The fault location is located by a deep learning fusion positioning algorithm based on topology and signal propagation. This series of intelligent diagnosis processes can not only accurately identify the fault type, but also accurately locate the fault location, greatly improving maintenance efficiency.

[0065] In addition, the present invention provides scientific maintenance suggestions through a fault cloud knowledge base and personalized maintenance solutions to guide maintenance personnel to complete tasks efficiently. The fault cloud knowledge base stores various fault types and their corresponding maintenance step suggestions, and provides personalized maintenance solutions through expert systems, maintenance knowledge bases and intelligent reasoning mechanisms. When the system detects a fault, it extracts corresponding maintenance suggestions from the fault cloud knowledge base based on the fault type, location and severity assessment, and outputs a fault report to the computer display platform through a multimodal natural language generation algorithm. The fault report not only contains the fault type, location and severity assessment, but also provides detailed maintenance steps, required tools and spare parts information. This personalized maintenance solution enables maintenance personnel to complete maintenance tasks quickly and accurately, reducing maintenance time and costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 This is a schematic diagram of the overall structure of a dry-type distribution box that is easy to maintain according to the present invention;

[0067] Figure 2 This is a working principle framework diagram of the fault detection submodule of the present invention;

[0068] Figure 3 This is a schematic diagram of the principle framework of the main control module of the present invention;

[0069] Figure 4 A working principle diagram of the method for evaluating the severity of a fault using a Bayesian network according to the present invention;

[0070] Figure 5 This is a working principle framework diagram of the fault cloud knowledge base of the present invention;

[0071] Figure 6 This is a step diagram of the cloud computing module of the present invention. DETAILED DESCRIPTION

[0072] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0073] In the above embodiment, if Figure 1-Figure 6 As shown: A dry distribution box that is easy to maintain includes a sensor monitoring module, a main control module, a wireless communication module and a cloud computing module;

[0074] The sensor monitoring module is used to collect real-time data including temperature, humidity, current and voltage inside the distribution box, and convert the real-time collected analog data into digital signals through an analog-to-digital converter, and transmit them to the main control module;

[0075] The main control module is used to process sensor data, perform preliminary fault detection, and output real-time data and fault alarm signals to the wireless communication module; and receive control instructions issued by the cloud computing module, parse and execute diagnostic operations; the main control module includes a data processing submodule, a fault detection submodule, a communication control submodule and an instruction execution submodule; the output end of the data processing submodule is connected to the input end of the fault detection submodule; the output end of the fault detection submodule is connected to the input end of the communication control submodule; the output end of the communication control submodule is connected to the input end of the instruction execution submodule;

[0076] In the specific implementation, Figure 2 As shown in Figure 1, the working method of the fault detection submodule is as follows: First, the system starts and enters the running state of the fault detection submodule. At the beginning, the system needs to set the thresholds of various monitoring parameters, including temperature, humidity, current and voltage, etc. Each parameter has its corresponding upper and lower thresholds. These thresholds are set based on historical data and expert experience to define the normal range of system operation. After setting the thresholds, the system starts to collect relevant data of the environment and equipment in real time. These data will be used for subsequent fault detection and analysis. Data collection is continuous to ensure that the system can capture any sudden changes. The collected data then enters the threshold detection stage, and the system compares these real-time data with the previously set thresholds. If the data exceeds the set threshold range, it is considered that the system may have a fault, and the system will trigger the fault alarm unit. Once the fault alarm unit is triggered, the system will enter the statistical analysis stage. In this stage, the system uses statistical tools such as the moving average method to analyze data trends and abnormal points to further confirm the fault type. After confirming the fault type, the system generates a fault alarm signal through the interrupt mechanism, and performs local alarms through LEDs and buzzers to ensure that the fault receives timely attention and processing. If the real-time data is within the threshold setting range, that is, the system is operating normally, then the system will not trigger a fault alarm, but continue to monitor new data to ensure the continuity and stability of the system. In this way, the fault detection submodule can effectively detect and respond to potential fault conditions without affecting normal operation.

[0077] In this solution, the wireless communication module is used to upload the data processed by the main control module to the cloud computing module and transmit the control instructions of the cloud computing module;

[0078] In the specific implementation, the wireless communication module first establishes a communication connection with the cloud computing module through the low-power wide area network (LPWAN) technology. LPWAN technology (such as LoRaWAN or NB-IoT) has the characteristics of low power consumption and long-distance transmission, which is suitable for application scenarios such as distribution boxes that require long-term stable operation. Through LPWAN technology, the wireless communication module can realize long-distance data transmission under low power consumption, ensuring the reliability and real-time performance of the data. Next, the wireless communication module realizes efficient data transmission through the MQTT protocol. MQTT is a lightweight message transmission protocol that is particularly suitable for low-bandwidth and high-latency network environments. The wireless communication module uses the MQTT protocol to encapsulate the data processed by the main control module into messages and sends the data to the cloud computing module through the publish / subscribe mode. The MQTT protocol ensures that the connection stability and data integrity can be maintained even when the network is unstable through the heartbeat mechanism and the reconnection mechanism. In order to further improve the reliability of data transmission, the wireless communication module adopts forward error correction (FEC) technology. FEC technology adds redundant information during data transmission, enabling the receiving end to restore the original data when some data is lost. Before sending data, the wireless communication module generates redundant data through the FEC encoding algorithm and sends it together with the original data. After receiving the data, the receiver restores the original data through the FEC decoding algorithm to ensure the integrity and accuracy of the data. In addition, the wireless communication module also uses TLS / SSL encryption technology to ensure the security and privacy of data transmission. TLS / SSL (Transport Layer Security / Secure Sockets Layer) is a widely used network security protocol that encrypts the transmitted data by establishing a secure communication channel to prevent the data from being stolen or tampered with. When the wireless communication module establishes a connection with the cloud computing module, it negotiates the encryption algorithm and key through the TLS / SSL handshake protocol to ensure the security and privacy of the data during transmission. In order to achieve efficient data transmission and management, the wireless communication module uses data compression technology. Data compression technology reduces the amount of data transmission, improves transmission efficiency, and reduces bandwidth occupancy. Before sending data, the wireless communication module compresses the data through a compression algorithm (such as LZ77 or LZ78) to reduce the size of the data. After receiving the compressed data, the receiver restores the original data through the corresponding decompression algorithm to ensure the integrity and accuracy of the data. In addition, the wireless communication module also has an automatic retransmission request (ARQ) mechanism to ensure the reliability of data transmission. The ARQ mechanism ensures the correct transmission of data by establishing a confirmation mechanism between the sender and the receiver. After sending data, the sender waits for the receiver's confirmation reply. If the confirmation reply is not received within the specified time, the sender will automatically retransmit the data until the receiver successfully receives the data and returns a confirmation reply.Through the ARQ mechanism, the wireless communication module can effectively deal with network instability and ensure the reliability and integrity of data transmission. Finally, the wireless communication module ensures the stability and maintainability of the system through real-time monitoring and logging mechanisms. The real-time monitoring mechanism ensures the stability and real-time nature of communication by regularly checking the status of the communication connection. The logging mechanism provides detailed log data by recording various events and error information during the communication process to help maintenance personnel discover and solve problems in a timely manner. The wireless communication module will record relevant log information before and after each data transmission, including transmission time, data size, transmission status, etc., to ensure the transparency and traceability of data transmission.

[0079] In specific implementation, the wireless communication module can efficiently and reliably upload the data processed by the main control module to the cloud computing module to achieve remote monitoring and diagnosis. LPWAN technology ensures long-distance, low-power data transmission, MQTT protocol achieves efficient data transmission, FEC technology improves the reliability of data transmission, TLS / SSL encryption technology ensures the security of data transmission, data compression technology improves transmission efficiency, ARQ mechanism ensures the reliability of data transmission, and real-time monitoring and logging mechanism ensures the stability and maintainability of the system. These technologies together improve the performance and reliability of the wireless communication module in the easy-to-maintain dry distribution box.

[0080] In this solution, the cloud computing module is used to receive and process the transmission data of the wireless communication module, display the real-time status data and historical records of the distribution box, generate fault reports, and perform fault diagnosis and processing; the fault report includes the fault type, fault location and repair step suggestions;

[0081] The working method of the cloud computing module for fault diagnosis processing is as follows: S1. Extract the real-time operating status of the distribution box through feature engineering, including the fluctuation of current and voltage and the temperature change trend; S2. Based on the real-time operating status and historical fault data of the distribution box, determine the fault type through a fault detection algorithm based on threshold and statistical analysis; S3. Construct a topological structure model of the distribution box to represent the internal connection relationship of the distribution box, and locate the fault position through a deep learning fusion positioning algorithm based on topological structure and signal propagation; S4. According to the fault feature data and historical experience, the severity of the fault is evaluated through a Bayesian network based on the fault impact range, the fault development trend vector and the fault impact index on the overall operation of the distribution box; S5. Based on fault feature data and historical experience, a fault cloud knowledge base is constructed to store fault types and maintenance step recommendations corresponding to the fault types; the fault cloud knowledge base provides personalized maintenance solutions through expert systems, maintenance knowledge bases and intelligent reasoning mechanisms; S6, according to the fault type output by S2, the fault location output by S3 and the severity evaluation output by S4, the corresponding maintenance recommendations are extracted from the fault cloud knowledge base, and the fault report is output to the computer display platform through a multimodal natural language generation algorithm; the maintenance recommendations include recommended remote operation, maintenance steps, required tools and spare parts information; S7, based on the fault report, the operator outputs control instructions to the main control module, including remote start and stop, load adjustment, and fault isolation instructions.

[0082] The working method of the fault detection algorithm based on threshold and statistical analysis includes the following steps:

[0083] Step u1, extracting feature vectors from real-time data through a sliding window mechanism, including the mean, variance, maximum value, and minimum value of current, voltage, and temperature; and performing dimensionality reduction processing on the extracted feature vectors through principal component analysis; Step u2, dynamically adjusting the threshold through a Kalman filter based on historical fault data to adapt to changes in different time periods and environments; the Kalman filter calculates the dynamic threshold through an error covariance matrix; the calculation formula for dynamically adjusting the threshold through the Kalman filter is:

[0084]

[0085] In formula (1), T k represents the dynamic threshold vector calculated at time k; X k|k-1 is the predicted state vector, which represents the predicted value of the current state before time k; K k represents the Kalman gain, which is used to adjust the difference between the predicted value and the actual observed value; Z k is the observation value vector, which represents the data actually measured at time k; H is the observation matrix, which represents the conversion of the state vector into the representation in the observation space; P k|k-1is the prediction error covariance matrix, which represents the uncertainty of the current state prediction before time k; Q represents the unit matrix, which is used to maintain the integrity of the matrix operation;

[0086] Step u3: Identify data points that exceed the dynamic threshold through a statistical anomaly detection algorithm, mark them as potential fault points x, and perform time series analysis on potential fault points x through an autocorrelation function to identify the mode and trend of fault occurrence; the formula expression of the autocorrelation function is:

[0087]

[0088] In formula (2), R(c) is the autocorrelation function value; Z t represents the tth data point in the time series, M is the length of the time series, and Z represents the mean of the time series; c is the lag order, which is used to calculate the lag time interval set when calculating the autocorrelation function; step u4, model the time series analysis results through the Gaussian mixture model, and calculate the probability that the data point x belongs to the normal distribution. The calculation formula is:

[0089]

[0090] In formula (3), S is the abnormality detection result, 1 indicates abnormality and 0 indicates normality; ω i is the weight of the i-th Gaussian distribution, which is used to measure the importance of the i-th Gaussian distribution in the mixture model; N(x|μ i , π i ) is the probability density function of the i-th Gaussian distribution, μ i is the mean vector of the i-th Gaussian distribution, used to represent the expected value of the data point x; π i is the covariance matrix of the ith Gaussian distribution, which is used to describe the distribution of the data point x; τ represents the threshold, which is used to determine whether the data point is abnormal; x represents the current data point; T k represents the dynamic threshold vector; It is the inverse of the prediction error covariance matrix, which represents the accuracy of the prediction error and is used to measure the relationship between the data point x and the dynamic threshold T k Deviation between; Step u5, based on the abnormal detection result of step u4, the abnormal data point m is classified by the support vector machine to determine the fault type, and the model hyperparameters are optimized by cross-validation to improve the classification accuracy; the formula expression for the support vector machine to classify the data point x is:

[0091]

[0092] In formula (4), SVM(F) is the classification result of the feature vector F by the support vector machine, a i is the coefficient of the support vector; yi Indicates the label of the support vector; K(F i ,F) is the kernel function used to calculate the eigenvector F i The similarity between and the feature vector F to be classified, and b represents the bias term.

[0093] In specific implementation, the fault detection algorithm based on threshold and statistical analysis can detect abnormal conditions in time by continuously monitoring the real-time operating status data of the distribution box (such as current, voltage, temperature and humidity) and comparing it with the preset threshold. For example, when the current or voltage exceeds the set threshold, the system will immediately trigger an alarm to alert the operator to potential faults. In addition, the algorithm also uses historical fault data for statistical analysis, and by establishing fault models and trend analysis, it can more accurately determine the type of fault. This dual detection mechanism can not only detect instantaneous faults in time, but also predict potential long-term faults, so that preventive measures can be taken in advance. Specifically, when the system detects that the current fluctuation exceeds the normal range, it will not only trigger an immediate alarm, but also combine historical data for trend analysis to determine whether it is an intermittent fault or a gradually worsening fault. This comprehensive analysis method makes fault detection more comprehensive and accurate, reduces the possibility of false alarms and missed alarms, and improves the reliability and stability of the system.

[0094] The working principle of the fault detection algorithm based on threshold and statistical analysis mainly includes two aspects: threshold detection and statistical analysis. First, threshold detection monitors the real-time collected data by setting a series of reasonable thresholds. When a parameter (such as current, voltage, temperature or humidity) exceeds the preset threshold, the system will immediately trigger an alarm to alert the operator to potential faults. The setting of thresholds is usually based on the normal operating range and safety standards of the equipment to ensure that abnormalities can be discovered in time at the early stage of the fault. Secondly, statistical analysis analyzes historical fault data to establish fault models and trend predictions. Specifically, the system collects and stores a large amount of historical data, including normal operating data and known fault data. By performing statistical analysis on this data, the laws and patterns of fault occurrence can be identified and fault models can be established. When the real-time data matches the fault model, the system can accurately determine the type of fault. In addition, statistical analysis can also detect potential long-term faults through trend prediction and take preventive measures in advance. For example, when the system detects that the current fluctuation is gradually increasing, trend analysis can be used to determine whether there is a potential equipment aging or wear problem, so as to carry out maintenance in advance and avoid the occurrence of faults.

[0095] In specific implementation, threshold detection can monitor the operating status of the equipment in real time. When the parameter exceeds the preset threshold, an alarm is triggered immediately to ensure that the fault is discovered at an early stage. Statistical analysis uses historical data to establish a fault model to improve the accuracy of fault detection and reduce the possibility of false alarms and missed alarms. In addition, the threshold detection method is simple and clear, easy to implement, and applicable to various types of distribution boxes. Although statistical analysis involves certain mathematical and statistical knowledge, the development of modern computing technology has made these analysis methods relatively easy to implement and have high reliability. Secondly, the fault detection algorithm based on threshold and statistical analysis combines the advantages of real-time monitoring and historical data analysis, which can fully cover instantaneous faults and long-term faults, and provide more comprehensive and accurate fault diagnosis results. This comprehensive detection method can effectively deal with various complex fault conditions and improve the reliability and stability of the system. At the same time, the algorithm can be flexibly adjusted according to different equipment and application scenarios. For example, for different types of distribution boxes, different thresholds and statistical models can be set according to their specific operating parameters and fault modes to ensure the effectiveness and applicability of the algorithm. Compared with the traditional single threshold detection or pure statistical analysis method, relying solely on threshold detection may lead to false alarms or missed alarms due to unreasonable threshold settings, while pure statistical analysis methods may not be able to detect instantaneous faults in time. The fault detection algorithm based on threshold and statistical analysis can more accurately determine the fault type through a dual detection mechanism, reduce the possibility of false alarms and missed alarms, and improve the reliability and accuracy of fault detection. The algorithm can be flexibly adjusted according to different equipment and application scenarios. For example, for different types of distribution boxes, different thresholds and statistical models can be set according to their specific operating parameters and fault modes to ensure the effectiveness and applicability of the algorithm. This flexibility enables the algorithm to be widely used in various complex industrial environments. In addition, threshold detection can detect instantaneous faults in time, while statistical analysis can predict potential long-term faults. This comprehensive detection method can comprehensively cover various fault types and provide more comprehensive and accurate fault diagnosis results. For example, when the system detects that the current fluctuation is gradually increasing, trend analysis can be used to determine whether there is a potential equipment aging or wear problem, so as to perform maintenance in advance and avoid the occurrence of faults. Secondly, the algorithm can process a large amount of real-time data and historical data, and quickly generate fault reports through efficient calculation and analysis methods. This not only improves the efficiency of fault detection, but also provides rich data support for operation and maintenance personnel, making fault handling more scientific and efficient.

[0096] In order to verify the effectiveness of the fault detection algorithm based on threshold and statistical analysis, Experiment 1 was designed for a comparative experiment. The experiment was divided into two groups. Group A used the fault detection algorithm based on threshold and statistical analysis, and Group B used the traditional single threshold detection algorithm. Both groups of experiments were conducted under the same experimental conditions. Each group conducted five experiments, and the fault detection accuracy, false alarm rate, and missed alarm rate of each experiment were recorded. We selected a typical hardware environment, including the following components:

[0097] Distribution box: Standard dry type distribution box, equipped with temperature, humidity, current and voltage sensors.

[0098] Sensor monitoring module: used to collect temperature, humidity, current and voltage data inside the distribution box in real time.

[0099] Analog-to-digital converter: Converts the collected analog data into digital signals.

[0100] Main control module: processes sensor data, performs preliminary fault detection, and outputs real-time data and fault alarm signals.

[0101] Wireless communication module: uploads the data processed by the main control module to the cloud computing module, and transmits the control instructions of the cloud computing module.

[0102] Cloud computing module: receives and processes data transmitted by the wireless communication module, performs fault diagnosis and processing, and generates fault reports.

[0103] Computer display platform: used to display fault reports and real-time data.

[0104] Based on the above hardware, under the same experimental conditions, the fault detection algorithms of group A and group B were run respectively. Five experiments were conducted for each group, and each experiment lasted for 1 hour. The fault detection accuracy, false alarm rate, and missed alarm rate of each experiment were recorded. The data records are shown in Table 1

[0105] Table 1 Comparative data record table

[0106]

[0107] As shown in Table 1, the average fault detection accuracy of Group A is 97.8%, while the average fault detection accuracy of Group B is 86%. The accuracy of Group A is significantly higher than that of Group B, indicating that the fault detection algorithm based on threshold and statistical analysis can more accurately determine the fault type. The average false alarm rate of Group A is 2%, while the average false alarm rate of Group B is 14%. The false alarm rate of Group A is much lower than that of Group B, indicating that the fault detection algorithm based on threshold and statistical analysis can more effectively reduce false alarms. The false alarm rate of Group A is 0%, while the average false alarm rate of Group B is 4%. There is no false alarm phenomenon in Group A, indicating that the fault detection algorithm based on threshold and statistical analysis can more comprehensively cover various fault types and reduce the possibility of false alarms.

[0108] In summary, the fault detection algorithm based on threshold and statistical analysis is superior to the traditional single threshold detection algorithm in terms of fault detection accuracy, false alarm rate and missed alarm rate. It has higher reliability and accuracy and can effectively improve the maintenance efficiency and safety of distribution boxes.

[0109] In the present invention, the working steps of the deep learning fusion positioning algorithm based on topological structure and signal propagation include:

[0110] Step t1, construct a distribution box topology matrix G = (V, E); wherein G is a directed graph, representing the topology of the distribution box; V is a node set, representing each electrical component in the distribution box; E is an edge set, representing the connection relationship between components;

[0111] Step t2: Based on the topological structure matrix G of the distribution box, the propagation time of the signal on different paths is calculated by the signal propagation equation. The calculation formula is:

[0112]

[0113] In formula (5), t er represents the propagation time of the signal from node e to node r; L er is the path length from node e to node r; v is the signal propagation speed; X is the attenuation coefficient; R er is the path resistance, which represents the resistance value between node e and node r;

[0114] Step t3: Based on the signal propagation simulation results, construct a signal propagation time matrix T = [t er ], where each element t er represents the propagation time of the signal from node e to node r;

[0115] Step t4: The local and global features of the nodes are fused through the graph convolution function to extract the topological features of the nodes. The formula expression of the graph convolution function is:

[0116]

[0117] In formula (6), H (I) is the node feature matrix of layer I, representing the features of each node in layer I; is the normalized adjacency matrix, which represents the connection relationship between nodes; is the degree matrix, which indicates the degree of each node; σ is the activation function, which is used to introduce nonlinearity; W (I) is the weight matrix, which represents the weight parameters of the Ith layer;

[0118] Step t5: Extract the time series characteristics h of signal propagation through the long short-term memory network t , the LSTM network captures the temporal dependency of signal propagation through the current hidden state and the hidden state of the previous time step;

[0119] Step t6: Use the fully connected layer to transform the topological features H (I+1) and signal propagation characteristics h t The fully connected layer combines the input features to form a comprehensive feature vector F. (I+1) and signal propagation characteristics h t Merge into a feature vector, and perform linear transformation on the concatenated feature vector through the weight matrix and bias vector to obtain the comprehensive feature vector F;

[0120] Step t7: classify the comprehensive feature vector F through the deep learning model, determine the fault location, and optimize the model parameters through the back propagation algorithm; wherein the formula expression for determining the fault location by the deep learning model is:

[0121]

[0122] In formula (7), P(f e |F) is the probability of node e failing, indicating whether node e is a failed node; W f is the model weight matrix, which is used to map the comprehensive feature vector to the probability value; b f Represents the model bias vector, which is used to adjust the output; F r are other feature vectors, representing the features of other nodes.

[0123] In the specific implementation, when the system detects that the current or voltage of a certain node is abnormal, the algorithm can quickly locate the specific fault location through the topological structure model and the signal propagation time matrix. This precise fault location not only reduces the troubleshooting time of maintenance personnel, but also improves maintenance efficiency, ensuring that the system can quickly resume normal operation. Specifically, when the current fluctuation of a certain node is abnormal, the algorithm can accurately locate the node where the fault is located by analyzing the propagation path and time of the signal inside the distribution box. Maintenance personnel can go directly to the node for maintenance without having to check all nodes one by one, which greatly saves time and labor costs. In addition, the algorithm can also handle complex distribution box structures, and can accurately locate the fault location even in the case of multiple nodes and multiple paths, improving the reliability and stability of the system.

[0124] The deep learning fusion positioning algorithm based on topology and signal propagation combines information from both topology and signal propagation to more accurately locate faults. By simulating the propagation process of the signal in the distribution box, the algorithm can take into account the attenuation and delay of the signal during propagation, as well as the resistance differences on different paths. This information is crucial for accurately locating faults. In addition, the algorithm also uses deep learning technology to learn the complex relationship between fault characteristics and fault locations through a large amount of training data, thereby improving the accuracy and reliability of positioning.

[0125] The main reasons for choosing the deep learning fusion positioning algorithm based on topology and signal propagation are as follows: Traditional fault location algorithms often rely on single threshold detection or simple statistical analysis, which is easily affected by noise and interference, resulting in false alarms and missed alarms. The deep learning fusion positioning algorithm based on topology and signal propagation can more accurately determine the fault location and reduce the possibility of false alarms and missed alarms through multi-dimensional feature extraction and fusion. Traditional fault location algorithms are prone to misjudgment when dealing with complex distribution box structures. The deep learning fusion positioning algorithm based on topology and signal propagation can capture the propagation characteristics of signals on different paths by constructing a topology model and a signal propagation time matrix, improve the robustness of fault detection, and ensure that the system can operate stably under various complex situations. In addition, traditional fault location algorithms often require a long processing time and cannot meet real-time requirements. The deep learning fusion positioning algorithm based on topology and signal propagation can process a large amount of data in real time, transmit real-time data to the cloud computing module through the wireless communication module, perform rapid fault detection and positioning, and ensure that faults can be discovered and handled at an early stage. At the same time, traditional fault location algorithms are often targeted at specific distribution box structures and application scenarios, and are difficult to adapt to different equipment and environments. The deep learning fusion positioning algorithm based on topology and signal propagation can be flexibly adjusted according to different distribution box structures and application scenarios. By constructing different topology models and signal propagation time matrices, the effectiveness and applicability of the algorithm can be ensured.

[0126] Similarly, comparative experiment 2 was designed to verify the positive and beneficial effects of the deep learning fusion positioning algorithm based on topology and signal propagation. The experiment was divided into two groups. Group A used the deep learning fusion positioning algorithm based on topology and signal propagation, and Group B used the traditional single threshold detection algorithm plus manual positioning. Both groups of experiments were conducted under the same experimental conditions. Each group conducted five experiments, and the fault location accuracy, false alarm rate, missed alarm rate and positioning time of each experiment were recorded. The hardware components of the experiment included:

[0127] High-performance computing server: used to run deep learning models and perform large-scale data processing. The server should have strong computing and storage capabilities to ensure the efficient operation of the algorithm.

[0128] Data acquisition equipment: used to collect status data and signal propagation data of each electrical component inside the distribution box. These devices should have high precision and real-time performance to ensure the accuracy and reliability of the data.

[0129] Communication module: used to transmit the collected data to the computing server for processing. This module should have stable communication performance and high data transmission rate to ensure timely transmission and processing of data.

[0130] Display device: used to display the real-time status data, historical records and generated fault reports of the distribution box. The device should have clear display effects and a friendly user interface so that operation and maintenance personnel can quickly and accurately obtain the required information.

[0131] Based on the above hardware environment, under the same experimental conditions, the fault detection algorithms of group A and group B were run respectively. Five experiments were conducted in each group, and each experiment lasted for 1 hour. The fault location accuracy, false alarm rate, missed alarm rate and location time of each experiment were recorded. The data recording table is shown in Table 2:

[0132] Table 2 Experimental comparison record

[0133]

[0134]

[0135] As shown in Table 2, the average fault location accuracy of Group A is 98.3%, while the average fault location accuracy of Group B is 85.6%. The accuracy of Group A is significantly higher than that of Group B, indicating that the deep learning fusion location algorithm based on topology structure and signal propagation can more accurately determine the fault location. The average false alarm rate of Group A is 1.3%, while the average false alarm rate of Group B is 14.5%. The false alarm rate of Group A is much lower than that of Group B, indicating that the deep learning fusion location algorithm based on topology structure and signal propagation can more effectively reduce false alarms. The average false alarm rate of Group A is 0.3%, while the average false alarm rate of Group B is 4.1%. The false alarm rate of Group A is much lower than that of Group B, indicating that the deep learning fusion location algorithm based on topology structure and signal propagation can more comprehensively cover various fault types and reduce the possibility of false alarms. The average location time of Group A is 1.5 seconds, while the average location time of Group B is 10.1 seconds. The location time of Group A is much shorter than that of Group B, indicating that the deep learning fusion location algorithm based on topology structure and signal propagation can locate the fault location faster and improve the efficiency of fault handling.

[0136] In the present invention, if Figure 4As shown, the method for evaluating the severity of a fault using a Bayesian network includes the following steps: step r1, constructing a Bayesian network model, the model construction method is: r1.1, according to the structural characteristics of the distribution box and common fault types, using expert knowledge and historical fault data to determine the network nodes and node probability distribution; r1.2, combining the historical fault data set, optimizing the Bayesian network parameters by the maximum likelihood estimation method; r1.3, defining fault feature data, fault impact range, fault development trend vector and fault impact index node on the overall operation of the distribution box, and mapping the conditional probability relationship between node data through the Bayesian reasoning mechanism; step r2, calculating the posterior probability of the fault node through the forward reasoning mechanism; the forward reasoning mechanism is based on the node definition and relationship mapping of the Bayesian network, and uses the fault feature data to calculate the posterior probability of the fault node; step r3, based on the internal electrical connection diagram of the distribution box, through The shortest path algorithm in graph theory is used to determine the degree of influence of the fault point on other components; and the weighted average method is used to comprehensively consider the importance of each component and calculate the overall impact range of the fault; step r4, the state data before and after the fault is modeled by the time series analysis ARIMA model, the development trend of the fault is predicted, and the fault development trend vector is continuously updated through the sliding window mechanism; step r5, combined with the output results of step r3 and step r4, the AHP hierarchical analysis method is used to determine the weights of different factors in the evaluation, and according to the weight distribution, the impact index of the fault on the overall operation of the distribution box is calculated by the linear combination method; step r6, the various indicators output by step r3, step r4 and step r5 are input into the Bayesian network model, the probability distribution of the fault severity is calculated by the Bayesian update rule, and the level of the fault is determined according to the probability distribution of the fault severity through the threshold judgment mechanism.

[0137] The Bayesian network is a probabilistic graphical model that represents the dependencies between variables through a directed acyclic graph (DAG) and uses conditional probability distribution to describe these relationships. In the fault diagnosis of the distribution box, the Bayesian network can comprehensively evaluate the impact range, development trend and impact on the overall operation of the distribution box based on the fault feature data and historical experience. In practical applications, the Bayesian network implements the evaluation of the severity of the fault through the following steps: Define nodes: Each node represents a fault feature variable, such as current fluctuation, voltage anomaly, temperature change, etc. Define edges: Edges represent the dependencies between nodes, such as current fluctuations may cause temperature rise. Determine the conditional probability table (CPT): The conditional probability table of each node describes the probability distribution of the node value given the parent node value. The temperature, humidity, current and voltage data inside the distribution box are collected in real time through the sensor monitoring module. The analog data is converted into digital signals using an analog-to-digital converter and transmitted to the main control module for preliminary processing. The main control module transmits the processed data to the cloud computing module through the wireless communication module. The Bayesian network is trained using historical fault data to determine the conditional probability table of each node. The conditional probability distribution of each node is calculated by the maximum likelihood estimation (MLE) or Bayesian estimation (BE) method. When the system detects new fault feature data, the Bayesian network model is used for reasoning to calculate the posterior probability of each fault feature variable. The severity of the fault is comprehensively evaluated by combining the fault impact range, fault development trend vector, and the impact index of the fault on the overall operation of the distribution box.

[0138] Compared with existing fault assessment technologies, Bayesian networks can more accurately describe the dependencies between fault features through probability graph models and conditional probability tables. Through training with historical data, the model can better capture the statistical laws of fault features and improve the accuracy of fault assessment. At the same time, Bayesian networks can handle uncertainty and incomplete data and have strong robustness. Even when some data is missing or the noise is large, the model can still provide reliable assessment results. In addition, Bayesian networks can process large amounts of data in real time and perform rapid fault assessment through cloud computing modules. Real-time performance enables faults to be discovered and handled at an early stage, reducing the losses caused by faults. Secondly, Bayesian networks have good interpretability and can clearly display the dependencies and evaluation process between fault features. Interpretability helps operation and maintenance personnel understand the results of fault assessment and improve the efficiency and accuracy of maintenance work. Finally, Bayesian networks can be dynamically adjusted and optimized according to new fault data and have strong adaptability. Adaptability enables the model to adapt to different distribution box structures and application scenarios, ensuring the effectiveness and applicability of the algorithm.

[0139] In the present invention, the working method of the fault cloud knowledge base is as follows: first, the fault cloud knowledge base extracts data from sensor data, maintenance records, and historical fault reports through a data output ETL process; then, the fault cloud knowledge base constructs fault types and maintenance step suggestions corresponding to the fault types through knowledge representation and data storage methods; the knowledge representation formalizes information including fault types, fault characteristics, and maintenance steps through ontology and semantic web to ensure the structuring and understandability of knowledge; the data storage method uses a relational database and a graph database to store and manage data in the fault cloud knowledge base; then, the fault cloud knowledge base generates personalized maintenance plans through an expert system and an intelligent reasoning mechanism; the expert system converts expert experience and historical data into rules and strategies through a rule engine and a knowledge base management system; the rule engine generates maintenance suggestions by matching rules and input data; the knowledge base management system ensures the update and maintenance of the knowledge base through version control and audit tracking; the intelligent reasoning mechanism identifies fault modes and maintenance strategies through a decision tree training model and generates optimized maintenance suggestions; during the working process of the fault cloud knowledge base, new fault data and maintenance records are received and processed in real time through a data stream processing engine, and the storage information of the cloud knowledge base is updated.

[0140] In specific implementation, Figure 5 As shown in the figure, the fault cloud knowledge base first receives data input from sensor data, maintenance records, and historical fault reports. Then, through the ETL process, useful information is extracted from the input data. Then, ontology and semantic web are used to formally represent the fault type, fault characteristics, and maintenance step information to ensure the structure and understandability of knowledge. At the same time, relational databases and graph databases are used to store and manage fault data and fault relationships. Expert experience and historical data are converted into rules and strategies through the rule engine and knowledge base management system. The rule engine generates maintenance suggestions by matching rules and input data, and the knowledge base management system ensures the update and maintenance of the knowledge base through version control and audit tracking. Then, the output of the rule engine and knowledge base management system is combined to generate maintenance suggestions. And the decision tree training model is used to identify fault modes and maintenance strategies to generate optimized maintenance suggestions. The intelligent reasoning mechanism identifies fault modes. The intelligent reasoning mechanism generates maintenance strategies. New fault data and maintenance records are received and processed in real time, and the storage information of the cloud knowledge base is updated. The data stream processing engine updates the new data to the relational database and graph database in real time to keep the information in the knowledge base up to date.

[0141] In specific implementation, when the current fluctuation of a certain node is abnormal, the system can quickly locate the fault type through the fault cloud knowledge base and provide corresponding repair steps. Maintenance personnel can directly perform repairs according to the suggestions without having to check all nodes one by one, which greatly saves time and labor costs. In addition, the continuous learning and optimization functions of the fault cloud knowledge base enable the system to be continuously improved, improve the accuracy and reliability of fault diagnosis, and ensure that the system can operate stably under different working conditions. Through the application of the fault cloud knowledge base, operation and maintenance personnel can respond to faults more quickly, reduce downtime, and improve the overall operation efficiency and safety of the system.

[0142] The working principle of the fault cloud knowledge base mainly includes the following steps: First, the temperature, humidity, current and voltage data inside the distribution box are collected in real time through the sensor monitoring module. The analog-to-digital converter converts the analog data into digital signals and transmits them to the main control module for preliminary processing. The main control module transmits the processed data to the cloud computing module through the wireless communication module. Secondly, the cloud computing module receives and processes the data transmitted by the wireless communication module and extracts the fault feature data. Feature extraction includes the extraction of key indicators such as current fluctuations, voltage anomalies, and temperature changes. The feature data is further processed using statistical analysis methods such as mean, variance, and standard deviation. Then, a fault cloud knowledge base is constructed to store fault types and their corresponding maintenance step recommendations. Fault types include but are not limited to overload, short circuit, ground fault, high temperature, excessive humidity, etc. The maintenance step recommendations include troubleshooting methods, required tools, spare parts information, etc. Next, the expert system formulates fault diagnosis rules and maintenance strategies based on the knowledge and experience of domain experts. Through the rule engine, the fault feature data is matched with the rules in the fault cloud knowledge base to generate preliminary fault diagnosis results. The maintenance knowledge base stores a large number of maintenance cases and best practices, including historical fault data, maintenance steps, maintenance effects, etc. Through case reasoning, the system can recommend maintenance solutions for similar cases based on the current fault characteristics. Finally, the intelligent reasoning mechanism combines the expert system and the maintenance knowledge base, and uses machine learning and data mining technology to comprehensively evaluate and reason about the fault. Through algorithms such as decision trees, support vector machines (SVMs), and neural networks, personalized maintenance solutions are generated. The entire process realizes real-time data processing and intelligent analysis through the cloud computing module to ensure that faults can be discovered and handled at an early stage.

[0143] The main reason for choosing the fault cloud knowledge base lies in its many advantages and characteristics. First, the fault cloud knowledge base can more accurately judge the type and severity of faults through the combination of expert system and maintenance knowledge base. Through the training of historical data and case reasoning, the model can better capture the statistical laws of fault characteristics and improve the accuracy of fault assessment. Secondly, the fault cloud knowledge base has strong robustness and can handle uncertainty and incomplete data. Even when some data is missing or the noise is large, the model can still provide reliable evaluation results. In addition, the fault cloud knowledge base can process a large amount of data in real time, and perform rapid fault assessment through the cloud computing module to achieve real-time performance, so that faults can be discovered and handled at an early stage, reducing the losses caused by faults. The fault cloud knowledge base also has good interpretability and can clearly show the dependencies and evaluation process between fault characteristics, which helps operation and maintenance personnel understand the results of fault assessment and improve the efficiency and accuracy of maintenance work. Most importantly, the fault cloud knowledge base has strong adaptability and can be dynamically adjusted and optimized according to new fault data to adapt to different distribution box structures and application scenarios, ensuring the effectiveness and applicability of the algorithm. These characteristics give the fault cloud knowledge base significant advantages in practical applications and can significantly improve the efficiency and accuracy of fault diagnosis and repair.

[0144] In the present invention, the multimodal natural language generation algorithm first comprehensively analyzes the fault type, fault location and severity assessment information through a Bayesian fusion method; the Bayesian fusion method combines information from different data sources through a probability model to generate a comprehensive fault feature description; then, the multimodal natural language generation algorithm generates a natural language text through template matching and content generation methods; the template matching selects a corresponding template according to the fault type and fault location through a predefined template library; the content generation method converts the comprehensive fault feature description into a natural language text through a sequence-to-sequence model; the sequence-to-sequence model encodes the input fault features into an intermediate representation through an encoder-decoder architecture, and then generates a natural language text through a decoder; finally, the multimodal natural language generation algorithm formats the generated natural language text into a structured fault report through a formatting method.

[0145] In the specific implementation, when the system detects a fault and completes the evaluation of the fault type, location and severity, the multimodal natural language generation algorithm can automatically generate a detailed fault report and present it to the operation and maintenance personnel in various forms such as text, images and videos. For example, the fault report not only contains a text description, but also can be accompanied by a schematic diagram of the fault location, a video demonstration of the maintenance steps and pictures of the required tools, so that the operation and maintenance personnel can quickly and accurately understand the fault situation and take corresponding maintenance measures. This multimodal fault reporting format not only improves the efficiency of fault handling, but also reduces human errors, ensuring that the system can quickly resume normal operation. Through the application of the multimodal natural language generation algorithm, the operation and maintenance personnel can understand the fault situation more intuitively, improve the accuracy and efficiency of maintenance work, and thus improve the reliability and user satisfaction of the overall system. Specifically, first, information such as the fault type, fault location and severity evaluation is received from the cloud computing module. This information is transmitted to the cloud computing module through the wireless communication module and processed as input data. The input fault type, fault location and severity evaluation information are fused to form a comprehensive fault description. Through the fault cloud knowledge base, the corresponding maintenance suggestions are extracted, including recommended remote operations, maintenance steps, required tools and spare parts information. Then, natural language generation technology is used to convert the fault description and maintenance suggestions into natural language text. Image generation technology is used to generate schematic diagrams of the fault location and pictures of the required tools. Video generation technology is used to generate video demonstrations of the maintenance steps. The generated text, images, and video content are then integrated to form a comprehensive fault report. The fault report is presented to the operation and maintenance personnel in a multimodal form, including text descriptions, schematic diagrams, pictures, and video demonstrations. The generated fault report is presented to the operation and maintenance personnel in a multimodal form through a computer display platform. The operation and maintenance personnel can perform maintenance operations quickly and accurately based on the detailed information in the fault report.

[0146] Compared with the existing fault report generation technology, the multimodal natural language generation algorithm has several significant advantages and features. First, the multimodal natural language generation algorithm can generate fault reports containing multiple forms of information, including text, images and videos, making the fault report more intuitive and easy to understand. Compared with traditional plain text fault reports, multimodal fault reports can provide richer information to help operation and maintenance personnel understand the fault situation more accurately. Secondly, the multimodal natural language generation algorithm can provide detailed maintenance suggestions, including recommended remote operations, maintenance steps, required tools and spare parts information, so that operation and maintenance personnel can perform maintenance operations quickly and accurately, improving maintenance efficiency. This is especially important in industrial environments that require rapid response. Third, the multimodal natural language generation algorithm can generate personalized fault reports based on fault type and severity assessment, improving the pertinence and effectiveness of fault handling. Personalized fault reports can better adapt to different fault situations and provide the most appropriate maintenance solutions. Fourth, the multimodal natural language generation algorithm realizes real-time data processing and intelligent analysis through the cloud computing module to ensure the accuracy and real-time nature of fault reports. This enables operation and maintenance personnel to respond to faults in a timely manner, reduce downtime, and improve the overall operation efficiency and safety of the system. Fifth, the multimodal natural language generation algorithm has high scalability and flexibility, and can be customized according to different application scenarios and requirements to adapt to different distribution box structures and maintenance requirements. Sixth, the multimodal natural language generation algorithm provides a comprehensive fault report through multimodal content integration, allowing operation and maintenance personnel to obtain all necessary information on one platform, simplifying the operation process and improving work efficiency. Finally, the multimodal natural language generation algorithm improves the readability and operability of the fault report through multimodal content generation technology, reduces human errors, and ensures the accuracy and safety of maintenance work. In summary, the multimodal natural language generation algorithm has significant advantages in fault report generation, which can significantly improve the generation quality of fault reports and user experience, and improve the reliability and operation efficiency of the system.

[0147] Although the specific embodiments of the present invention are described above, it should be understood by those skilled in the art that these specific embodiments are only illustrative, and those skilled in the art may omit, replace, and change the details of the above methods and systems in various ways without departing from the principles and essence of the present invention. For example, merging the above method steps so as to perform substantially the same functions in substantially the same manner to achieve substantially the same results is within the scope of the present invention. Therefore, the scope of the present invention is limited only by the appended claims.

Claims

1. A dry type distribution box that is easy to maintain, characterized in that: It includes a sensor monitoring module, a main control module, a wireless communication module and a cloud computing module; The sensor monitoring module is used to collect real-time data including temperature, humidity, current and voltage inside the distribution box, and convert the real-time collected analog data into digital signals through an analog-to-digital converter, and transmit them to the main control module; The main control module is used to process sensor data, perform preliminary fault detection, and output real-time data and fault alarm signals to the wireless communication module; and receive control instructions issued by the cloud computing module, parse and execute diagnostic operations; the main control module includes a data processing submodule, a fault detection submodule, a communication control submodule and an instruction execution submodule; the output end of the data processing submodule is connected to the input end of the fault detection submodule; The output end of the fault detection submodule is connected to the input end of the communication control submodule; The output end of the communication control submodule is connected to the input end of the instruction execution submodule; The wireless communication module is used to upload the data processed by the main control module to the cloud computing module and transmit the control instructions of the cloud computing module; The cloud computing module is used to receive and process the transmission data of the wireless communication module, display the real-time status data and historical records of the distribution box, generate fault reports, and perform fault diagnosis and processing; the fault report includes the fault type, fault location and repair step suggestions; The working method of the cloud computing module for fault diagnosis processing is: S1. Extract the real-time operating status of the distribution box through feature engineering, including the fluctuation of current and voltage and the temperature change trend; S2, based on the real-time operating status and historical fault data of the distribution box, determine the fault type through a fault detection algorithm based on thresholds and statistical analysis; S3. Construct a topological structure model of the distribution box to represent the internal connection relationship of the distribution box, and locate the fault location through a deep learning fusion positioning algorithm based on topological structure and signal propagation; S4. According to the fault feature data and historical experience, the severity of the fault is evaluated through the Bayesian network based on the fault impact range, fault development trend vector and the impact index of the fault on the overall operation of the distribution box; S5. Based on fault feature data and historical experience, a fault cloud knowledge base is constructed to store fault types and corresponding maintenance step suggestions; The fault cloud knowledge base provides personalized maintenance solutions through expert systems, maintenance knowledge bases and intelligent reasoning mechanisms; S6, extracting corresponding maintenance suggestions from the fault cloud knowledge base according to the fault type output by S2, the fault location output by S3 and the severity evaluation output by S4, and outputting a fault report to a computer display platform through a multimodal natural language generation algorithm; the maintenance suggestions include recommended remote operation, maintenance steps, required tools and spare parts information; S7. Based on the fault report, the operator outputs control instructions to the main control module, including remote start and stop, load adjustment, and fault isolation instructions.

2. A dry type distribution box easy to maintain according to claim 1, characterized in that: The data processing submodule compares the input data and the reference data bit by bit through the successive approximation ADC, collects the sensor data, and stores the data in the ring buffer through the data buffer mechanism; then, the data processing submodule reads the data from the ring buffer through the preprocessing unit, and the preprocessing unit uses the Kalman filter to smooth the data, remove the noise, and convert the data into a standard state distribution through the Z-score normalization method; The pre-processed data is transmitted to the microcontroller for data fusion calculation processing; the microcontroller realizes multi-task concurrent processing through task scheduling and synchronization mechanism; The fault detection submodule includes a threshold detection unit, a statistical analysis unit and a fault alarm unit; The fault detection submodule sets the upper and lower thresholds of temperature, humidity, current and voltage through the threshold detection unit, and compares and determines whether the real-time data exceeds the normal range through the threshold-based fault detection method. If it exceeds the threshold range, the fault alarm unit is triggered; the statistical analysis unit analyzes the data trend and abnormal points through the moving average method to further confirm the fault type; When triggered, the fault alarm unit generates a fault alarm signal through an interrupt mechanism, and performs a local alarm through an LED and a buzzer; the communication control submodule includes a data encapsulation unit, a communication interface unit, and an instruction receiving unit; the communication control submodule encapsulates the processed data and the fault alarm signal into a JSON format through the data encapsulation unit, and uses a CRC checksum to ensure the integrity of the data; The communication interface unit sends the encapsulated data to the wireless communication module through the serial communication interface, the instruction receiving unit receives the control instructions issued by the cloud computing module through the MQTT protocol, and uses the TLS / SSL encryption method to protect the security of data transmission; the instruction execution submodule includes an instruction parsing unit, an operation execution unit and a status feedback unit; The instruction execution submodule receives and analyzes the control instructions issued by the cloud computing module through the instruction parsing unit, and the instruction parsing unit ensures the accuracy and legality of the instructions through regular expression matching and whitelist mechanism; The operation execution unit realizes remote start and stop by turning on and off the relay control circuit according to the parsed instructions, adjusts the load by adopting the pulse width modulation method, and isolates the faulty equipment through the fault isolation mechanism; the state feedback unit monitors the operation results in real time through the state feedback mechanism, and sends the operation results back to the wireless communication module through the serial communication interface. The state feedback unit also records the operation process and results through the ring log buffer to ensure the transparency and traceability of the operation.

3. The dry type distribution box easy to maintain according to claim 1 is characterized in that: The wireless communication module establishes a communication connection with the cloud computing module through the low-power wide area network LPWAN, then encapsulates the data processed by the main control module into messages through the MQTT protocol, and sends the data to the cloud computing module through the publish-subscribe mode; the MQTT protocol ensures the stability of the network connection and the integrity of the data through the heartbeat mechanism and the reconnection mechanism; when the wireless communication module establishes a connection with the cloud computing module, the encryption algorithm and key are negotiated through the TLS / SSL handshake protocol; During data transmission, the wireless communication module adds redundant information to the original data through forward error correction coding to improve the reliability of data transmission; during data transmission, the wireless communication module reduces the amount of data transmission through the compression algorithm LZ77 to improve transmission efficiency; the wireless communication module also ensures the reliability of data transmission and the stability of the system through an automatic retransmission request mechanism and a log recording mechanism.

4. The dry type distribution box easy to maintain according to claim 1 is characterized in that: The working method of the fault detection algorithm based on threshold and statistical analysis includes the following steps: Step u1, extracting feature vectors from real-time data through a sliding window mechanism, including the mean, variance, maximum value, and minimum value of current, voltage, and temperature; and performing dimensionality reduction processing on the extracted feature vectors through principal component analysis; Step u2: According to the historical fault data, the threshold is dynamically adjusted through the Kalman filter to adapt to the changes in different time periods and environments; the Kalman filter calculates the dynamic threshold through the error covariance matrix; the calculation formula for dynamically adjusting the threshold through the Kalman filter is: In formula (1), T k represents the dynamic threshold vector calculated at time k; X k|k-1 is the predicted state vector, which represents the predicted value of the current state before time k; K k represents the Kalman gain, which is used to adjust the difference between the predicted value and the actual observed value; Z k is the observation value vector, which represents the data actually measured at time k; H is the observation matrix, which represents the conversion of the state vector into the representation in the observation space; P k|k-1 is the prediction error covariance matrix, which represents the uncertainty of the current state prediction before time k; Q represents the unit matrix, which is used to maintain the integrity of the matrix operation; Step u3: Identify data points that exceed the dynamic threshold through a statistical anomaly detection algorithm, mark them as potential fault points x, and perform time series analysis on potential fault points x through an autocorrelation function to identify the mode and trend of fault occurrence; the formula expression of the autocorrelation function is: In formula (2), R(c) is the autocorrelation function value; Z t represents the tth data point in the time series, M is the length of the time series, and Z represents the mean of the time series; c is the lag order, which is used to calculate the lag time interval set when calculating the autocorrelation function; Step u4: Model the time series analysis results through the Gaussian mixture model and calculate the probability that the data point x belongs to the normal distribution. The calculation formula is: In formula (3), S is the abnormality detection result, 1 indicates abnormality and 0 indicates normality; ω i is the weight of the i-th Gaussian distribution, which is used to measure the importance of the i-th Gaussian distribution in the mixture model; N(x|μ i , π i ) is the probability density function of the i-th Gaussian distribution, μ i is the mean vector of the i-th Gaussian distribution, used to represent the expected value of the data point x; π i is the covariance matrix of the ith Gaussian distribution, which is used to describe the distribution of the data point x; τ represents the threshold, which is used to determine whether the data point is abnormal; x represents the current data point; T k represents the dynamic threshold vector; It is the inverse of the prediction error covariance matrix, which represents the accuracy of the prediction error and is used to measure the relationship between the data point x and the dynamic threshold T k The deviation between Step u5: Based on the abnormal detection result of step u4, the abnormal data point m is classified by a support vector machine to determine the fault type, and the model hyperparameters are optimized by cross-validation to improve the classification accuracy; the formula expression for classifying the data point x by the support vector machine is: In formula (4), SVM(F) is the classification result of the feature vector F by the support vector machine, a i is the coefficient of the support vector; y i Indicates the label of the support vector; K(F i ,F) is the kernel function used to calculate the eigenvector F i The similarity between and the feature vector F to be classified, and b represents the bias term.

5. The dry type distribution box easy to maintain according to claim 1 is characterized in that: The working steps of the deep learning fusion positioning algorithm based on topological structure and signal propagation include: Step t1, construct a distribution box topology matrix G = (V, E); wherein G is a directed graph, representing the topology of the distribution box; V is a node set, representing each electrical component in the distribution box; E is an edge set, representing the connection relationship between components; Step t2: Based on the topological structure matrix G of the distribution box, the propagation time of the signal on different paths is calculated by the signal propagation equation. The calculation formula is: In formula (5), t er represents the propagation time of the signal from node e to node r; L er is the path length from node e to node r; v is the signal propagation speed; X is the attenuation coefficient; R er is the path resistance, which represents the resistance value between node e and node r; Step t3: Based on the signal propagation simulation results, construct a signal propagation time matrix T = [t er ], where each element t er represents the propagation time of the signal from node e to node r; Step t4: The local and global features of the nodes are fused through the graph convolution function to extract the topological features of the nodes. The formula expression of the graph convolution function is: In formula (6), H (I) is the node feature matrix of layer I, representing the features of each node in layer I; is the normalized adjacency matrix, which represents the connection relationship between nodes; is the degree matrix, which indicates the degree of each node; σ is the activation function, which is used to introduce nonlinearity; W (I) is the weight matrix, which represents the weight parameters of the Ith layer; Step t5: Extract the time series characteristics h of signal propagation through the long short-term memory network t , the LSTM network captures the temporal dependency of signal propagation through the current hidden state and the hidden state of the previous time step; Step t6: Use the fully connected layer to transform the topological features H (I+1) and signal propagation characteristics h t The fully connected layer combines the input features to form a comprehensive feature vector F. (I+1) and signal propagation characteristics h t Merge into a feature vector, and perform linear transformation on the concatenated feature vector through the weight matrix and bias vector to obtain the comprehensive feature vector F; Step t7: classify the comprehensive feature vector F through the deep learning model, determine the fault location, and optimize the model parameters through the back propagation algorithm; wherein the formula expression for determining the fault location by the deep learning model is: In formula (7), P(f e |F) is the probability of node e failing, indicating whether node e is a failed node; W f is the model weight matrix, which is used to map the comprehensive feature vector to the probability value; b f Represents the model bias vector, which is used to adjust the output; F r are other feature vectors, representing the features of other nodes.

6. The dry type distribution box easy to maintain according to claim 1 is characterized in that: The Bayesian network method for evaluating the severity of a fault comprises the following steps: Step r1: construct a Bayesian network model. The model construction method is: r1.

1. According to the structural characteristics of the distribution box and common fault types, the network nodes and node probability distribution are determined using expert knowledge and historical fault data; r1.

2. Combined with the historical fault data set, the Bayesian network parameters are optimized by the maximum likelihood estimation method; r1.3, define the fault feature data, fault impact range, fault development trend vector and fault impact index node on the overall operation of the distribution box, and map the conditional probability relationship between the node data through the Bayesian reasoning mechanism; Step r2, calculating the posterior probability of the faulty node through a forward reasoning mechanism; the forward reasoning mechanism is based on the node definition and relationship mapping of the Bayesian network, and uses the fault feature data to calculate the posterior probability of the faulty node; Step r3: Based on the internal electrical connection diagram of the distribution box, determine the impact of the fault point on other components through the shortest path algorithm in graph theory; and use the weighted average method to comprehensively consider the importance of each component and calculate the overall impact range of the fault; Step r4: Model the state data before and after the fault occurs by using the time series analysis ARIMA model to predict the development trend of the fault, and continuously update the fault development trend vector through the sliding window mechanism; Step r5, combining the output results of step r3 and step r4, using the AHP hierarchical analysis method to determine the weights of different factors in the evaluation, and according to the weight distribution, using the linear combination method to calculate the impact index of the fault on the overall operation of the distribution box; Step r6: Input the indicators outputted from step r3, step r4 and step r5 into the Bayesian network model, calculate the probability distribution of the fault severity through the Bayesian update rule, and determine the fault level according to the probability distribution of the fault severity through the threshold judgment mechanism.

7. The dry type distribution box easy to maintain according to claim 1 is characterized in that: The working method of the fault cloud knowledge base is as follows: first, the fault cloud knowledge base extracts data from sensor data, maintenance records, and historical fault reports through a data output ETL process; then, the fault cloud knowledge base constructs fault types and corresponding maintenance step suggestions through knowledge representation and data storage methods; the knowledge representation formalizes information including fault types, fault characteristics, and maintenance steps through ontology and semantic web to ensure the structuring and understandability of knowledge; the data storage method uses relational databases and graph databases to store and manage data in the fault cloud knowledge base; then, the fault cloud knowledge base generates personalized maintenance plans through expert systems and intelligent reasoning mechanisms; the expert system converts expert experience and historical data into rules and strategies through rule engines and knowledge base management systems; The rule engine generates maintenance suggestions by matching rules and input data; The knowledge base management system ensures the updating and maintenance of the knowledge base through version control and audit tracking; The intelligent reasoning mechanism uses a decision tree training model to identify failure modes and maintenance strategies and generate optimized maintenance recommendations; During the operation of the fault cloud knowledge base, new fault data and maintenance records are received and processed in real time through the data stream processing engine, and the storage information of the cloud knowledge base is updated.

8. The dry type distribution box easy to maintain according to claim 1 is characterized in that: The multimodal natural language generation algorithm first uses a Bayesian fusion method to comprehensively analyze the fault type, fault location, and severity assessment information; the Bayesian fusion method combines information from different data sources through a probability model to generate a comprehensive fault feature description; then , the multimodal natural language generation algorithm generates natural language text through template matching and content generation method; the template matching selects the corresponding template according to the fault type and fault location through a predefined template library; The content generation method converts the comprehensive fault feature description into natural language text through a sequence-to-sequence model; the sequence-to-sequence model encodes the input fault features into an intermediate representation through an encoder-decoder architecture, and then generates natural language text through a decoder; finally, the multimodal natural language generation algorithm formats the generated natural language text into a structured fault report through a formatting method.

Images