Intelligent measurement and control system and control method of withdrawable switch cabinet

Abstract

The present invention discloses an intelligent measurement and control system for a withdrawable switch cabinet and a control method thereof, which relate to the technical field of automatic control of electric power systems, and solve the problems of slow response speed, inflexible control strategy and difficulty in coping with extreme weather conditions in existing intelligent measurement and control systems when responding to changes in power grids; real-time data acquisition and analysis are realized by integrating an edge processing module, and state prediction and anomaly detection are performed in combination with a long short-term memory network; an adaptive control module is used to optimize the control strategy according to dynamic monitoring results; at the same time, the emergency response module monitors environmental variables through a SCADA system, and an intelligent scheduling algorithm adjusts the switch cabinet operation mode and switches standby equipment; the present invention greatly improves the safety and reliability of the switch cabinet, significantly enhances the rapid response capability and adaptive control capability of the power system to changes in the power grid, and ensures the stability and safety of the power supply.

Description

Technical Field

[0001] The present invention relates to the technical field of power system automation control, and more specifically to an intelligent measurement and control system of a withdrawable switch cabinet and a control method thereof. Background Art

[0002] With the rapid development of smart grids, the power system has increasingly higher requirements for the intelligence and automation of switchgear. As an important part of the power system, the intelligent measurement and control system of withdrawable switchgear has become the key to improving the efficiency of power grid operation and ensuring power supply safety. Driven by information technology, Internet of Things technology and artificial intelligence technology, the intelligent measurement and control system of withdrawable switchgear is gradually developing in a more efficient, smarter and more reliable direction to meet the needs of modern power systems for real-time monitoring of equipment status, fault warning and remote control.

[0003] For example, patent CN221151023U discloses a switch cabinet integrated measurement and control device, which integrates multiple functional modules such as human body sensing lighting unit, intelligent locking unit, protection unit, voice prompt unit, power module, measurement module, and communication module. The measurement module monitors the operating status of the switch cabinet in real time, and coordinates with the protection unit, voice prompt unit, and intelligent locking unit to realize centralized monitoring of the switch cabinet. The protection unit provides protection for multiple faults such as leakage, high temperature, and arc light. The communication module can perform remote communication and data transmission to realize intelligent remote measurement and control. However, in actual applications, since the control method mainly relies on the collection and analysis of operation data, the intelligent measurement and control system needs to respond quickly to grid changes to ensure the safety and stability of power supply. The delay in operation data analysis leads to the inability to respond in time in emergency situations such as power distribution short circuit and overload, which affects the safety and reliability of the switch cabinet. In addition, traditional control algorithms are often based on fixed rules or preset modes, which are difficult to adapt to the dynamic changes in the power system. For example, in the face of sudden surges or decreases in power demand, existing control methods may not be able to quickly adjust the output, resulting in waste of power resources or insufficient power supply. In addition, due to the lack of sufficient flexibility, the existing system may not provide the best solution when encountering special situations such as emergencies under extreme weather conditions, thereby increasing the possibility of accidents. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention discloses an intelligent measurement and control system for a withdrawable switch cabinet and a control method thereof, aiming to solve the problems in the background technology.

[0005] In order to achieve the above technical effects, the present invention adopts the following technical solutions:

[0006] An intelligent measurement and control system for a withdrawable switch cabinet, comprising:

[0007] The edge processing module is used to collect the operating status information of the switch cabinet in real time, including voltage, current, temperature, switch status, and protection action information, and to perform real-time analysis on the voltage, current, and temperature of the switch cabinet. When abnormal data is detected, the edge processing module is also used to issue an early warning notification through an early warning mechanism;

[0008] A dynamic monitoring and analysis module is used to perform state prediction and anomaly detection through a long short-term memory network based on the real-time data collected by the edge processing module. If a potential short circuit or overload risk is detected, a warning signal is generated, and the optimal adjustment strategy is calculated through a switch cabinet power failure mapping model;

[0009] An adaptive control module, used to adjust the output power and protection settings of the switch cabinet according to the output results of the dynamic monitoring and analysis module, and optimize the control strategy through a reinforcement learning algorithm during the adjustment process;

[0010] Emergency response module, which is used to monitor environmental variables through the SCADA system. If an extreme weather warning is detected, the switch cabinet operation mode is adjusted through the intelligent scheduling algorithm, and backup equipment, including UPS uninterruptible power supply and diesel generators, is switched when data anomalies are detected;

[0011] The remote diagnosis module is used to provide a remote access interface, allowing operation and maintenance personnel to monitor the status of the switch cabinet in real time, receive early warning information, and perform remote fault diagnosis.

[0012] As a further technical solution of the present invention, in the edge processing module, the working method for detecting data anomalies is: when collecting each set of voltage, current, and temperature data, a consistency check is performed through a data cross-validation algorithm, and similar data collected by different sensors at the same time are compared; if the voltage data difference, the current data difference, or the temperature data difference does not exceed the preset error range, continue to monitor and update the data; if any of the voltage data difference, the current data difference, or the temperature data difference exceeds the preset error range, send real-time detection data and historical detection data within a preset time period to the dynamic monitoring and analysis module, and send early warning information in real time to the central monitoring platform or the mobile device of the operation and maintenance personnel, and the information content includes the abnormality type, occurrence time, and abnormal value.

[0013] As a further technical solution of the present invention, in the dynamic monitoring and analysis module, the method for state prediction and anomaly detection by the long short-term memory network is as follows: based on data timestamp information and historical operation data, through the forget gate mechanism of the long short-term memory network, a dynamic weight distribution algorithm is used to perform a weight distribution operation on the hidden state information, screen key information and output the optimized hidden state information to the hidden layer; the hidden layer is used to perform a nonlinear transformation operation through a tanh activation function based on the input data of the input gate and the optimized hidden state information output by the forget gate, and output the time series feature data to the standard feature template, perform a deviation value calculation operation, and obtain the deviation value data between the current data and the standard feature; the standard feature template is generated by historical data training and an expert system; if the deviation value exceeds a preset threshold, a risk judgment operation is performed through an anomaly detection algorithm based on probability distribution, and if the output risk judgment result is yes, a warning signal generation mechanism is triggered, and the output includes the fault probability and the affected range information to the switch cabinet power failure prediction model; if the output risk judgment result is no, the state prediction and anomaly detection operation is terminated; the anomaly detection algorithm based on probability distribution fuses historical fault data and current abnormal features through a Bayesian network to perform fault judgment.

[0014] As a further technical solution of the present invention, the switch cabinet power failure mapping model includes a state assessment layer, an RL environment simulation layer, a multi-objective optimization layer, a decision selection layer, an instruction generation layer and a self-correction and feedback layer; the working method of the switch cabinet power failure mapping model is:

[0015] The state evaluation layer performs real-time evaluation on the current power grid state; the state evaluation layer constructs a dynamic state-action value function through a deep Q network algorithm, quantifies the long-term benefits of different control actions, and outputs the data of the state-action value function to the RL environment simulation layer;

[0016] The RL environment simulation layer simulates the system response under different control strategies; the RL environment simulation layer generates operation scenarios through simulation modules, including fast action of circuit breakers, load redistribution, and generator output adjustment; and outputs simulation results and evaluation data of each scenario to the multi-objective optimization layer;

[0017] The multi-objective optimization layer finds a balance point between multiple conflicting objectives; the multi-objective optimization layer maps multiple control strategies into a multi-dimensional objective space through a non-dominated sorting genetic algorithm, calculates the fitness value of each strategy, and selects the optimal solution on the Pareto front as a recommended solution; after the solution is selected, the optimized control strategy list is output to the decision selection layer;

[0018] The decision selection layer selects the optimal control strategy according to the optimization result; the decision selection layer selects the control strategy and outputs it to the instruction generation layer through a rule-based decision tree combined with real-time monitoring data and historical experience;

[0019] Generate adjustment instructions through the instruction generation layer and record them in the log system at the same time;

[0020] The self-correction and feedback layer monitors the actual feedback and performs self-correction. When the self-correction and feedback layer detects that the actual feedback deviates from the expected result, it identifies the error factor through the causal relationship diagram and the Bayesian network, and uses the gradient descent method to adjust the model parameters and feed back to the state evaluation layer.

[0021] As a further technical solution of the present invention, the instruction generation layer fuzzifies the input state variables through membership functions, converts the numerical values ​​into language variables, and performs reasoning operations through fuzzy reasoning algorithms to calculate preliminary fuzzy control outputs; then, the control quantity value is calculated based on the preliminary fuzzy control output through the center of gravity method to determine at least the adjustment position of the transformer tap in the switch cabinet and the switching capacity parameters of the reactive compensation device.

[0022] As a further technical solution of the present invention, the method for the adaptive control module to optimize the control strategy through a reinforcement learning algorithm is: first, at least voltage, current, and temperature variables are converted into discrete state representations through a state encoder, and then a state-action value function is constructed to quantify the long-term benefits of taking different control actions under each state; when a new state change is detected, the reinforcement learning algorithm selects the optimal action according to the latest state-action value function and executes the corresponding control instructions; during the execution process, the adaptive control module continuously monitors the actual feedback of the system and adjusts the parameters of the state-action value function through a self-correction mechanism.

[0023] As a further technical solution of the present invention, the intelligent scheduling algorithm first integrates the weather forecast data of the meteorological department, the geographic information data of the area where the switch cabinet is located, and the real-time operation data of the current power system into a unified format through a data format conversion and cleaning method, and filters out erroneous or redundant information; the real-time operation data of the current power system includes the load conditions of different lines and the temperature status of each switch cabinet; then, the potential risk scenarios corresponding to the current extreme weather are located in the knowledge base through a pattern matching algorithm and an expert system; based on the matched risks, the power outage scope, overload risk and backup power use time constraints are comprehensively considered through an optimization strategy based on mixed integer linear programming, and a scheduling plan including line switching and load allocation is generated through a linear programming solver.

[0024] As a further technical solution of the present invention, the remote diagnosis module is used to build a remote access channel through the WebRTC protocol, receive the output data of the edge processing module, encrypt the transmitted data through an asymmetric encryption algorithm, and verify the identity of the operation and maintenance personnel through multi-factor authentication; if an abnormal situation is detected, the early warning mechanism is triggered, and the abnormal alarm information is transmitted to the operation and maintenance terminal through the RESTful API interface. The remote diagnosis module stores the timestamp, type and detailed description of each abnormal event through a distributed log system; the remote diagnosis module is also used to allow the operation and maintenance personnel to transmit control instructions to the physical layer equipment for execution through the industrial communication protocol EtherCAT and the interactive interface.

[0025] As a further technical solution of the present invention, an intelligent measurement and control method for a withdrawable switch cabinet comprises the following steps:

[0026] a. In each preset measurement and control cycle, the edge processing module uses the sensor network to perform ultra-high frequency sampling of the electrical parameters in the withdrawable switch cabinet, and performs real-time filtering and noise reduction processing on the sampled data through an adaptive filtering algorithm to obtain pre-processed electrical parameter data;

[0027] b. Determine the operating status of the switch cabinet by using a data cross-validation algorithm; if it is determined to be in a normal operating state, execute step c; otherwise, execute step d;

[0028] c. Continue data collection and monitoring operations, store current data in the normal operation data history library, and exit the measurement and control process of the current measurement and control cycle;

[0029] d. Perform trend analysis on abnormal state data and historical data within a certain period of time through long short-term memory network;

[0030] e. According to the fault prediction result, the fault level is judged according to the fault severity classification standard; if the fault prediction time is judged to be less than the preset emergency warning time threshold, step f is executed; otherwise, step g is executed;

[0031] f. Immediately trigger the emergency warning signal sending operation, and send an emergency warning message containing the fault type, expected occurrence time, and switch cabinet location information to the remote monitoring center through the wireless communication module. At the same time, start the emergency protection plan in the local switch cabinet, including cutting off the switch equipment operation of the fault line, and then exit the measurement and control process of the current measurement and control cycle;

[0032] g. Send general warning information to remind operation and maintenance personnel to check and handle at the appropriate time, and continue to monitor the development trend of the fault. If the trend worsens, re-enter step d;

[0033] h. Generate a control action combination through the switch cabinet power failure prediction model according to the current operating state of the switch cabinet, including the electrical parameters collected in real time, the operating state information output in step b, and the future state information output in step d;

[0034] i. For each generated control action combination, the stability index, reliability index and economic index of the power system are calculated through the reinforcement learning model to obtain the comprehensive evaluation score of each control action combination;

[0035] j. Based on the evaluation scores, the particle swarm optimization algorithm is used to search for the optimal control action combination;

[0036] k. Apply the optimized control strategy to the control operation of the withdrawable switchgear, and use the fuzzy control strategy to output control signals to the corresponding switchgear and compensation device to achieve dynamic adjustment of the switchgear operating parameters. During the adjustment process, the control strategy is optimized through the reinforcement learning algorithm;

[0037] l. When an extreme situation is detected, the intelligent scheduling algorithm will be used to match and compare the monitored data with the various extreme situation feature data in the extreme situation knowledge base, adjust the switch cabinet operation mode, and switch to backup equipment, including UPS uninterruptible power supply and diesel generators, when abnormal data is detected;

[0038] m. During the emergency handling process, continuously determine whether the extreme situation has been resolved; if it has been resolved, execute step n; otherwise, continue to execute step l;

[0039] n. Gradually restore the normal operation of the switch cabinet, including reclosing the previously disconnected switch and adjusting the equipment operating parameters to normal status. At the same time, store the data and operation records of the emergency handling process in the emergency handling history library.

[0040] Based on the above technical solutions, the positive and beneficial effects of the present invention are:

[0041] 1. Through real-time data collection and analysis of the edge processing module, combined with the early warning mechanism, the present invention can immediately issue an early warning notification when data anomalies are detected, significantly shortening the response time. The application of long short-term memory network (LSTM) makes state prediction and anomaly detection more accurate, and can generate early warning signals before potential short circuit or overload risks occur, thereby effectively avoiding the problem of delayed emergency response caused by delayed operation data analysis; it solves the problem of delayed operation data analysis leading to the inability to respond in time in emergency situations such as power distribution short circuit and overload, and improves the safety and reliability of the switch cabinet.

[0042] 2. The adaptive control module continuously optimizes the control strategy based on the output results of the dynamic monitoring and analysis module through the reinforcement learning algorithm, and realizes the flexible adjustment of the switch cabinet output power and protection settings. This adaptive control mechanism can cope with the dynamic changes of the power system, especially in the face of sudden changes in power demand, and can quickly adjust the output to avoid waste of power resources or insufficient power supply; it solves the problem that traditional control algorithms are difficult to adapt to the dynamic changes in the power system, and improves the flexibility and efficiency of power supply.

[0043] 3. The emergency response module monitors environmental variables through the SCADA system. When an extreme weather warning is detected, it can adjust the switch cabinet operation mode through an intelligent scheduling algorithm to ensure the stability and safety of power supply. At the same time, when data anomalies are detected, it can automatically switch to backup equipment (such as UPS uninterruptible power supply and diesel generators), further enhancing the robustness of the system; solving the problem of lack of sufficient flexibility to deal with emergencies under extreme weather conditions and reducing the possibility of accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work, among which:

[0045] Figure 1 This is a structural diagram of an intelligent measurement and control system for a withdrawable switch cabinet of the present invention;

[0046] Figure 2 This is a working principle diagram of the edge processing module of the present invention;

[0047] Figure 3 It is a working principle diagram of the dynamic monitoring and analysis module of the present invention;

[0048] Figure 4 A diagram showing the working method of the switch cabinet power failure mapping model of the present invention;

[0049] Figure 5 The present invention is a flowchart of the steps of the intelligent measurement and control method of the withdrawable switch cabinet. DETAILED DESCRIPTION

[0050] 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.

[0051] In a specific embodiment, an intelligent measurement and control system for a withdrawable switch cabinet is provided. Figure 1 As shown, the output end of the edge processing module is connected to the input end of the dynamic monitoring and analysis module, the emergency response module and the remote diagnosis module; the output end of the emergency response module, the remote diagnosis module and the dynamic monitoring and analysis module is connected to the input end of the adaptive control module; the output end of the dynamic monitoring and analysis module is connected to the input end of the remote diagnosis module. The system includes:

[0052] The edge processing module is used to collect the operating status information of the switch cabinet in real time, including voltage, current, temperature, switch status, and protection action information, and to perform real-time analysis on the voltage, current, and temperature of the switch cabinet. When abnormal data is detected, the edge processing module is also used to issue an early warning notification through an early warning mechanism; Figure 2 As shown, in the edge processing module, the working method for detecting data anomalies is: when collecting each set of voltage, current, and temperature data, a consistency check is performed through a data cross-validation algorithm to compare similar data collected by different sensors at the same time; if the voltage data difference, the current data difference, or the temperature data difference does not exceed the preset error range, continue to monitor and update the data; if any of the voltage data difference, the current data difference, or the temperature data difference exceeds the preset error range, send the real-time detection data and the historical detection data within the preset time period to the dynamic monitoring and analysis module, and send the early warning information in real time to the central monitoring platform or the mobile device of the operation and maintenance personnel, and the information content includes the abnormality type, occurrence time, and abnormal value.

[0053] In the specific implementation, the edge processing module uses a composite sensor array to collect the operating status information of the switch cabinet in real time. These sensors are based on micro-electromechanical system (MEMS) technology and have the characteristics of high precision and high sensitivity. For voltage acquisition, the principle of capacitor voltage division is used to convert the high voltage into a measurable weak current signal in proportion through a precisely designed capacitor network, and then converted into a digital signal through a high-precision analog-to-digital converter (ADC). The sampling frequency can reach the kilohertz level to ensure that the instantaneous fluctuation of the voltage can be captured. The current measurement uses the Hall effect sensor. When the current passes through the current-carrying conductor, a magnetic field is generated around the conductor. The Hall element is placed in the magnetic field to generate a Hall voltage proportional to the current intensity, which is amplified and filtered by the signal conditioning circuit and converted by the ADC. The temperature sensor uses a thermistor, whose resistance value shows a specific nonlinear relationship with temperature. The resistance value is measured and converted into a temperature value using a pre-calibrated lookup table. This multi-parameter synchronous acquisition method provides basic data for the subsequent comprehensive analysis of the switch cabinet status. When collecting switch status and protection action information, an optocoupler isolation circuit is used. When the switch is activated or the protection device is triggered, the light-emitting diode inside the optocoupler emits light, causing the phototransistor to conduct, resulting in high and low level changes that are captured by the digital input port. This method effectively isolates strong electrical interference and ensures the accuracy of signal acquisition.

[0054] The core of the data cross-validation algorithm is to compare the same type of data collected by different sensors at the same time. Taking voltage as an example, suppose there are three voltage sensors collecting phase A voltage at the same time. After they are independently signal conditioned and converted by ADC, the data enters the algorithm comparison link. First, the absolute value of the difference between the sensor data, namely |V1-V2|, |V1-V3|, |V2-V3|, is calculated and compared with the preset voltage error range. This preset error range is determined based on the accuracy index of the sensor, the voltage fluctuation range of the normal operation of the switch cabinet, and a large amount of experimental data, and is generally within ±0.5% of the rated voltage. For current and temperature data, the same logic is followed. The current error range is set according to the load characteristics and the accuracy of the transformer, and the temperature error range considers the temperature measurement accuracy of the sensor and the allowable temperature fluctuation in the switch cabinet. If all differences do not exceed the preset error range, it means that the data is reliable, and the system continues to monitor and update the data; if any difference exceeds the range, the subsequent exception handling process is immediately triggered.

[0055] Once data anomalies are detected, the edge processing module quickly starts the early warning mechanism. The early warning mechanism first integrates and packages the real-time detection data with the historical detection data within a preset time period (such as the past 10 minutes, which can be adjusted according to the operating characteristics of the switch cabinet and the requirements of fault analysis). For historical data, a circular buffer is used for storage, and new data overwrites old data to ensure the timeliness of the data. Then, these data are sent to the dynamic monitoring and analysis module through a wireless communication module (such as a communication chip that supports 4G / 5G or industrial Ethernet), providing a comprehensive basis for the upper-level module to further diagnose faults. At the same time, the SMS gateway or a dedicated operation and maintenance APP push service is used to send the early warning information to the central monitoring platform or the mobile device of the operation and maintenance personnel in real time. The early warning information includes the type of abnormality (such as excessive voltage, unbalanced current, overheated temperature, etc.), the time of occurrence (accurate to milliseconds, provided by the system clock), and the abnormal value (specific voltage, current, and temperature values ​​that exceed the normal range), so that the operation and maintenance personnel can quickly understand the fault situation and take corresponding measures.

[0056] When the system is initialized, the MCU performs self-checks on the sensor, signal conditioning circuit, storage unit, wireless communication module, etc. to ensure that all components are working properly. Then, the data acquisition process is started according to the preset sampling period (such as 1 second), and the sensor is controlled to collect voltage, current, temperature and switch status information, which are sent to the MCU after signal conditioning. The MCU runs the data cross-validation algorithm to perform real-time verification on the collected data. During normal operation, if the data is normal, the MCU updates the new data to the historical data buffer in the storage unit, and regularly sends heartbeat packets to the remote monitoring system through the wireless communication module to indicate that its working status is normal. Once data anomalies are detected, the early warning process is immediately executed. As mentioned above, data is sent and early warning information is pushed. After that, the abnormal situation is continuously monitored. If the abnormality continues to deteriorate or new abnormalities appear, the early warning information is updated in time.

[0057] Compared with the traditional switch cabinet measurement and control that only uses a single sensor or a simple combination of sensors, the composite sensor array of this edge processing module provides more comprehensive and accurate data. Sensors based on various principles verify each other, greatly improving the reliability of the data and reducing the risk of misjudgment. For example, in voltage measurement, multiple sensors are compared and verified to avoid erroneous data collection caused by the failure of a single sensor. In addition, the data cross-validation algorithm has strong real-time performance and performs verification at the moment of data collection. Compared with traditional post-data analysis, it can detect potential problems at the first time. Moreover, the algorithm flexibly sets the error range based on the actual operating characteristics of the switch cabinet to adapt to different working conditions. Secondly, the timeliness and comprehensiveness of the early warning mechanism are outstanding. Not only can it quickly push abnormal information to the operation and maintenance personnel, but it also combines real-time and historical data and sends it to the upper-level analysis module to provide in-depth information for fault diagnosis, which is unmatched by the existing simple alarm system.

[0058] It also includes a dynamic monitoring and analysis module, which is used to perform state prediction and anomaly detection through a long short-term memory network based on the real-time data collected by the edge processing module. If a potential short circuit or overload risk is detected, a warning signal is generated, and the optimal adjustment strategy is calculated through the switch cabinet power failure mapping model; Figure 3 As shown, in the dynamic monitoring and analysis module, the method for state prediction and anomaly detection by the long short-term memory network is as follows: based on the data timestamp information and the historical operation data, through the forget gate mechanism of the long short-term memory network, a dynamic weight distribution algorithm is used to perform a weight distribution operation on the hidden state information, screen the key information and output the optimized hidden state information to the hidden layer; the hidden layer is used to perform a nonlinear transformation operation based on the input data of the input gate and the optimized hidden state information output by the forget gate through the tanh activation function, and output the time series feature data to the standard feature template, perform a deviation value calculation operation, and obtain the deviation value data between the current data and the standard feature; the standard feature template is generated by historical data training and expert system; if the deviation value exceeds the preset threshold, a risk judgment operation is performed through an anomaly detection algorithm based on probability distribution, and if the output risk judgment result is yes, the warning signal generation mechanism is triggered, and the output includes the fault probability and the affected range information to the switch cabinet power failure prediction model; if the output risk judgment result is no, the state prediction and anomaly detection operation is terminated; the anomaly detection algorithm based on probability distribution fuses the historical fault data and the current abnormal features through the Bayesian network to perform fault judgment.

[0059] In a specific embodiment, the forget gate is based on the sigmoid function. It receives the current input data and the hidden state information of the previous moment, and uses a dynamic weight allocation algorithm to allocate weights to each element in the hidden state according to the timeliness information contained in the data timestamp and the rules summarized from the historical operation data. For example, for the hidden state part corresponding to the data with a newer timestamp, because it reflects the latest operation trend of the switch cabinet at the moment, the algorithm assigns a higher weight; while the weight corresponding to the old data is reduced accordingly. The mathematical model of the algorithm continuously optimizes the weight allocation strategy through complex matrix operations and combined with the back propagation training of the neural network, so that the network can accurately focus on key information, effectively avoiding the gradient disappearance or gradient explosion problems of traditional neural networks when processing time series data, thereby screening out the most valuable key information for current state prediction and anomaly detection, and outputting the optimized hidden state information to the hidden layer. In the hidden layer, the tanh function has an S-shaped curve feature, which can map the input linear combination data to a fixed interval (usually -1 to 1), and mine the deep time series association between data through nonlinear transformation. For example, for subtle fluctuations in voltage and current data over time, the tanh function can keenly capture their changing trends, amplify or suppress these trend features, convert them into nonlinear feature representations that are easier to process later, and output data rich in time series features to standard feature templates.

[0060] The standard feature template is constructed by training a large amount of historical data and combining it with the knowledge of the expert system. The historical data covers the multi-dimensional operating parameters of the switchgear during normal operation, critical states of various faults, and when faults occur. The expert system incorporates the in-depth understanding of the operating characteristics and fault manifestations of the switchgear by experts in the power field. Through machine learning algorithms, such as cluster analysis and principal component analysis, typical feature patterns are extracted from historical data, and they are corrected and improved in combination with expert experience to form a standard feature template. When the time series feature data output by the hidden layer arrives, the deviation value is calculated, that is, the current data is compared element by element with the data at the corresponding position of the standard feature template through mathematical methods such as Euclidean distance and cosine similarity, and accurate deviation value data is obtained to intuitively reflect the degree of deviation between the current operating state and the normal or expected state.

[0061] In the specific implementation, the dynamic monitoring and analysis module uses the long short-term memory network (LSTM) to perform state prediction and anomaly detection. Specifically, this method is based on data timestamp information and historical operation data, uses the forget gate mechanism of LSTM, and adopts a dynamic weight distribution algorithm to perform weight distribution operations on hidden state information, screen key information and output optimized hidden state information to the hidden layer. The hidden layer is based on the input data of the input gate and the optimized hidden state information output by the forget gate, and performs nonlinear transformation operations through the tanh activation function, and outputs the time series feature data to the standard feature template, performs deviation value calculation operations, and obtains the deviation value data between the current data and the standard feature.

[0062] Compared with the existing technology, the forget gate mechanism of LSTM combined with the dynamic weight distribution algorithm enables the model to more effectively screen out key information and avoid the interference of redundant data. This mechanism improves the accuracy and reliability of state prediction, and can detect potential risks at an early stage and take preventive measures in advance. In addition, through the nonlinear transformation operation of the tanh activation function, LSTM can quickly process a large amount of time series data and generate accurate time series features. Once an abnormal situation is detected, the early warning signal generation mechanism is immediately triggered, and the fault probability and scope information are sent to the switch cabinet power failure prediction model to ensure that the operation and maintenance personnel can take action in the shortest time to avoid the expansion of the accident. Secondly, the anomaly detection algorithm based on probability distribution fuses historical fault data with current abnormal features through the Bayesian network to construct a more comprehensive and accurate fault judgment model. This method not only improves the accuracy of anomaly detection, but also provides detailed fault probability and scope information, helping operation and maintenance personnel to better understand the nature of the fault and formulate response strategies. The standard feature template is generated through historical data training and expert system, and can be continuously updated to adapt to new working conditions. This adaptive learning ability enables the system to always maintain optimal performance and improve the stability and reliability of long-term operation. In addition, the present invention adopts modular design, and each component is connected through a standard interface, which is convenient for maintenance and upgrading. At the same time, it supports multiple communication protocols and can flexibly access different network environments to meet the needs of diverse application scenarios.

[0063] Furthermore, the switch cabinet power failure mapping model includes a state assessment layer, an RL environment simulation layer, a multi-objective optimization layer, a decision selection layer, an instruction generation layer, and a self-correction and feedback layer; Figure 4 As shown, the working method of the switch cabinet power failure mapping model is:

[0064] The state evaluation layer performs real-time evaluation on the current power grid state; the state evaluation layer constructs a dynamic state-action value function through a deep Q network algorithm, quantifies the long-term benefits of different control actions, and outputs the data of the state-action value function to the RL environment simulation layer;

[0065] The RL environment simulation layer simulates the system response under different control strategies; the RL environment simulation layer generates operation scenarios through simulation modules, including fast action of circuit breakers, load redistribution, and generator output adjustment; and outputs simulation results and evaluation data of each scenario to the multi-objective optimization layer;

[0066] The multi-objective optimization layer finds a balance point between multiple conflicting objectives; the multi-objective optimization layer maps multiple control strategies into a multi-dimensional objective space through a non-dominated sorting genetic algorithm, calculates the fitness value of each strategy, and selects the optimal solution on the Pareto front as a recommended solution; after the solution is selected, the optimized control strategy list is output to the decision selection layer;

[0067] The decision selection layer selects the optimal control strategy according to the optimization result; the decision selection layer selects the control strategy and outputs it to the instruction generation layer through a rule-based decision tree combined with real-time monitoring data and historical experience;

[0068] The instruction generation layer generates adjustment instructions and records them in the log system at the same time; the instruction generation layer performs fuzzification processing on the input state variables through the membership function, converts the numerical values ​​into language variables, and performs reasoning operations through the fuzzy reasoning algorithm to calculate the preliminary fuzzy control output; then the center of gravity method is used to calculate the control quantity value according to the preliminary fuzzy control output to determine at least the adjustment position of the transformer tap in the switch cabinet and the switching capacity parameters of the reactive compensation device.

[0069] The self-correction and feedback layer monitors the actual feedback and performs self-correction. When the self-correction and feedback layer detects that the actual feedback deviates from the expected result, it identifies the error factor through the causal relationship diagram and the Bayesian network, and uses the gradient descent method to adjust the model parameters and feed back to the state evaluation layer.

[0070] In a specific embodiment, the deep Q network (DQN) algorithm uses a neural network to approximate the state-action value function. The input is the current grid state information of the switch cabinet. This information covers multi-dimensional data such as voltage, current, and load. After being processed by multiple layers of neurons, the long-term benefit estimates corresponding to different control actions are output. During the training process, by continuously allowing the agent to take actions in the environment and updating the parameters of the neural network based on reward feedback, the value function becomes more accurate. For example, for the control action of adjusting the transformer tap, DQN can comprehensively consider the subsequent long-term impact on voltage stability, power loss, and other aspects, output its corresponding value, and provide a quantitative basis for subsequent decision-making. The output state-action value function data is passed to the RL environment simulation layer.

[0071] The simulation module of the RL environment simulation layer generates various operation scenarios based on the physical model and operation rules of the actual power system, such as the rapid action of the circuit breaker to simulate the rapid cut-off protection when a short-circuit fault occurs, the load redistribution to simulate the adjustment under the sudden change of power demand, and the generator output adjustment to simulate the response of the power generation end. Through complex power system simulation algorithms, considering multiple physical processes such as electromagnetic transients and electromechanical transients, the changes of various system parameters under different control strategies, such as voltage fluctuation curves and current impact peaks, are accurately calculated, and combined with preset evaluation indicators, the evaluation data of each scenario is given and output to the multi-objective optimization layer, providing rich references for strategy selection.

[0072] The non-dominated sorting genetic algorithm (NSGA-II) treats multiple conflicting objectives in the operation of switchgear, such as improving power supply reliability, reducing operating costs, and reducing equipment losses. It regards multiple control strategies as population individuals and maps them to a multidimensional target space. Based on the concept of Pareto optimality, the population is iteratively evolved through the selection, crossover, and mutation operations of the genetic algorithm. The fitness value of each strategy is calculated, and the optimal solution on the Pareto front is screened out, that is, the strategy set that achieves the optimal goal without sacrificing too much other goals. It is output as a recommended solution to the decision selection layer to achieve balanced optimization of multiple objectives.

[0073] The rule-based decision tree of the decision selection layer integrates real-time monitoring data and historical experience. The decision tree nodes are branched according to different grid status characteristics (such as voltage range, load severity, etc.), and each branch corresponds to different control strategy rules. When real-time monitoring data is input, the appropriate control strategy is quickly matched along the branch path of the tree. For example, when the voltage is low and the load is heavy, the strategy of increasing reactive compensation and appropriately adjusting the transformer tap is selected and output to the instruction generation layer to make efficient and targeted decisions.

[0074] The instruction generation layer uses membership functions to fuzzify the input state variables and converts precise values ​​into language variables such as "high voltage" and "low current" to better match the fuzzy reasoning logic. The fuzzy reasoning algorithm is based on a fuzzy rule base built based on expert experience, such as "if the voltage is high and the load is light, then the reactive compensation should be appropriately reduced", to infer the fuzzified variables and calculate the preliminary fuzzy control output. Then, through the centroid method, based on the membership function distribution of the fuzzy output, the control quantity value is accurately calculated, and specific parameters such as the transformer tap adjustment position and the reactive compensation device switching capacity are determined to generate executable adjustment instructions.

[0075] The self-correction and feedback layer uses the causal relationship diagram to intuitively present the causal relationship between each variable, and combines the probabilistic reasoning ability of the Bayesian network to quickly locate the error factor when the actual feedback is detected to be different from the expected result. For example, if it is found that the voltage adjustment does not meet expectations, the causal relationship diagram can be traced back to the possible load mutation or reactive compensation calculation error. The Bayesian network gives the probability of each factor based on historical data, uses the gradient descent method to adjust the model parameters in a targeted manner, and feeds back to the state evaluation layer to continuously optimize the model performance.

[0076] During implementation, the switch cabinet power fault mapping model is based on a high-performance server, equipped with a multi-core CPU for coordinating the operation of each module and conventional data processing, and equipped with a GPU to accelerate the training and reasoning of deep learning algorithms such as deep Q network to ensure fast computing. Large-capacity memory stores real-time power grid data, simulation scenario data, model parameters, etc. Equipped with a high-speed storage hard disk to record operation logs. At the same time, it is equipped with a professional power system simulation card with a built-in high-precision power model to assist the simulation of the RL environment simulation layer. In terms of network interface, it has a high-speed Ethernet interface to communicate with the edge processing module and other systems to ensure real-time data interaction. The CPU, GPU and memory are connected through a high-speed bus to achieve high-speed data reading and writing. The power simulation card is connected to the server through the PCIe interface to ensure smooth collaboration with the main system. The server's network interface is connected to the edge processing module, remote monitoring terminal, etc. through a switch, following communication protocols such as TCP / IP to stably transmit data. The storage hard disk is connected to the server motherboard through the SATA interface to ensure reliable log recording.

[0077] Traditional switch cabinet fault handling is mostly based on fixed rules or single-objective optimization. This model uses deep Q networks, multi-objective optimization and other algorithms, which can adapt to complex and changeable grid conditions, comprehensively consider multi-objective balance, and make accurate decisions. For example, unlike the past, it only focuses on voltage stability, but also takes into account cost and equipment life. Fuzzy reasoning combined with membership functions when generating instructions combines the advantages of precision and fuzzy control, which not only utilizes expert experience but also adapts to uncertainty and improves control flexibility. Secondly, the self-correction and feedback mechanism optimizes the model in real time, dynamically adjusts according to actual operation, maintains high adaptability of the model, and overcomes the limitations of traditional static models.

[0078] It also includes an adaptive control module, which is used to adjust the output power and protection settings of the switch cabinet according to the output results of the dynamic monitoring and analysis module. During the adjustment process, the control strategy is optimized by a reinforcement learning algorithm; the method by which the adaptive control module optimizes the control strategy by the reinforcement learning algorithm is: first, at least voltage, current, and temperature variables are converted into discrete state representations through a state encoder, and then a state-action value function is constructed to quantify the long-term benefits brought about by taking different control actions under each state; when a new state change is detected, the reinforcement learning algorithm selects the optimal action according to the latest state-action value function and executes the corresponding control instructions; during the execution process, the adaptive control module continuously monitors the actual feedback of the system and adjusts the parameters of the state-action value function through a self-correction mechanism.

[0079] In specific implementation, the state encoder is a key component that converts continuous physical variables into discrete state representations. For key variables such as voltage, current, and temperature, it is divided according to pre-set quantization intervals. For example, according to the voltage range of normal operation of the switchgear, it is subdivided into multiple sub-intervals, such as low-voltage interval, rated voltage interval, high-voltage warning interval, etc., and the same is true for current and temperature. Through this quantization method, complex continuous signals are simplified into a finite number of discrete states, which is convenient for subsequent algorithm processing. This process utilizes the principles of threshold judgment and data mapping. When the real-time data collected falls into a certain interval, it corresponds to a specific discrete state encoding, laying the foundation for building an accurate state-action value function, so that the system can quickly identify the current working conditions.

[0080] The state-action value function aims to quantify the long-term benefits of different control actions in various states. It takes the discrete state output by the state encoder as input and combines the neural network architecture to approximate this complex function. The hidden layer of the neural network extracts and combines the features of the input state to learn the weights of the influence of various control actions (such as adjusting the transformer tap position, changing the switching state of the reactive compensation device, etc.) on system stability, economy and other aspects under different states. In the training stage, through a large amount of simulation or actual operation data, according to the reward mechanism, when a certain control action makes the system perform well in terms of voltage stability, power loss reduction and other goals, positive rewards are given to promote the continuous optimization of the function, accurately reflect the value of the optimal control action in each state, and provide a basis for real-time decision-making.

[0081] When a new state change is detected, the reinforcement learning algorithm quickly selects the optimal action based on the latest state-action value function. It uses exploration and utilization balance mechanisms such as the ε-greedy strategy. On the one hand, it selects the optimal action under the current value function evaluation with a high probability (1-ε) to ensure the efficient operation of the system; on the other hand, it randomly explores new actions with a small probability of ε to avoid falling into local optimality. For example, in the common state where the voltage is slightly low, it is highly likely to choose the action of increasing reactive compensation based on the value function, but occasionally other parameters will be adjusted by trial and error. By continuously collecting system feedback after these attempts, the decision strategy is continuously optimized, the control instructions of the selected action are executed, and the switch cabinet is driven to respond. In the process of executing the control instructions, the adaptive control module collects the actual feedback of the system in real time, such as the actual change value of voltage and current, the actual operating temperature of the equipment, etc. By comparing the expected results after executing the action with the actual feedback, the parameters of the state-action value function are adjusted in reverse using the gradient descent method or a similar optimization algorithm. If it is found that a certain control action does not achieve the expected voltage stabilization effect in the current state, the function weights related to the action and state are fine-tuned in a targeted manner, so that subsequent decisions are more accurate, adapt to the dynamic changes of the system, and always maintain the effectiveness of the control strategy.

[0082] In the specific implementation, after the system is powered on and initialized, the pre-trained state-action value function model and related parameters are loaded from the storage medium. The data acquisition card continuously collects the voltage, current, and temperature data of the switch cabinet, sends them to the state encoder to convert them into discrete states, and inputs them into the constructed value function module. Once the state change is detected, the reinforcement learning algorithm makes a quick decision and sends a control command to the switch cabinet for execution. At the same time, the self-correction mechanism is activated, and the value function parameters are adjusted in real time according to the feedback. This cycle is repeated to ensure the adaptive optimization of the switch cabinet operation control.

[0083] Compared with the existing technology, this module uses a reinforcement learning algorithm, and the system can dynamically adjust the control strategy according to real-time data and historical experience to adapt to different working conditions and failure modes. In addition, the application of deep neural networks improves the estimation accuracy of the state-action value function. At the same time, the self-correction mechanism enhances the robustness of the system and reduces misoperation and failure rates. Secondly, it also realizes closed-loop management from data collection, state monitoring to control decision-making, improves the operating efficiency and safety of the switch cabinet, and reduces operation and maintenance costs.

[0084] It also includes an emergency response module, which is used to monitor environmental variables through the SCADA system. If an extreme weather warning is detected, the switch cabinet operation mode is adjusted through an intelligent scheduling algorithm, and backup equipment, including UPS uninterruptible power supply and diesel generator, is switched when data abnormality is detected; the intelligent scheduling algorithm first integrates the weather forecast data of the meteorological department, the geographic information data of the area where the switch cabinet is located, and the real-time operation data of the current power system into a unified format through data format conversion and cleaning methods, and screens out erroneous or redundant information; the real-time operation data of the current power system includes the load conditions of different lines and the temperature status of each switch cabinet; then, the potential risk scenarios corresponding to the current extreme weather are located in the knowledge base through the pattern matching algorithm and the expert system; based on the matched risks, the power outage scope, overload risk and backup power use time constraints are comprehensively considered through the optimization strategy based on mixed integer linear programming, and a scheduling plan including line switching and load allocation is generated through a linear programming solver.

[0085] In the specific implementation, the emergency response module uses data adapters and parsers to deal with weather forecast data from the meteorological department (usually in a specific meteorological format, such as GRIB, etc.), geographic information data of the area where the switchgear is located (possibly based on the vector or raster format of the GIS system), and real-time operation data of the power system (mostly derived from the private protocol data format of various power monitoring equipment). The adapter maps the data into a structured format that can be processed internally by writing corresponding conversion scripts based on the format specifications of different data sources. For example, key information such as temperature, wind speed, and precipitation probability in meteorological data is extracted and organized into a table form. The cleaning process uses algorithms based on statistics and rules, such as identifying temperature extremes that exceed the normal range (based on the statistical limits of historical meteorological data) and abnormal spikes in power loads (compared with past load curve characteristics), using data interpolation to repair a small number of missing values, and filtering out erroneous or redundant information to ensure that the data input into the intelligent scheduling algorithm is accurate and reliable, laying a solid foundation for subsequent risk assessment.

[0086] The pattern matching algorithm is based on a pre-built library of extreme weather feature patterns, and uses string matching, feature vector similarity calculation and other technologies to quickly scan key features in the input data. For example, when a combination of features such as a sudden increase in wind speed and frequent lightning activity is detected, it is quickly associated with extreme weather patterns of strong winds and rainstorms accompanied by lightning strikes. On this basis, the expert system incorporates the years of experience accumulated by senior engineers in the power field and stores it in the form of a knowledge rule base, such as "In strong wind weather, overhead lines are prone to the risk of dancing short circuits, and special attention should be paid to changes in line tension." The two work together to accurately locate potential risk scenarios corresponding to current extreme weather in the knowledge base, covering a wide range of hidden dangers from short circuits caused by moisture in electrical equipment, damage to lines due to external forces to water ingress to substations, and provide directions for the generation of targeted strategies.

[0087] The optimization strategy based on mixed integer linear programming (MILP) is the core algorithm to ensure the stable operation of switchgear under extreme conditions. It takes minimizing the scope of power outage, minimizing the risk of overload, and rationalizing the use time of backup power as the objective function, defines the operational decision variables such as line switching and load allocation in the power system as integers or continuous variables, and constructs linear inequality group constraints based on the physical constraints of the power network (such as line capacity, equipment rated power) and operating rules (power flow balance equation). Linear programming solvers (such as commonly used commercial software or open source solvers such as CPLEX and Gurobi) use simplex methods, branch and bound algorithms, etc. to efficiently search for the optimal solution in the feasible solution space that meets the constraints, and generate a scheduling plan that includes accurate line switching paths and load distribution ratio adjustments. For example, before the arrival of a hurricane, it is determined that non-critical load lines should be cut off first, important loads should be transferred to backup circuits, and the timing and load distribution of UPS and diesel generators should be reasonably arranged to ensure uninterrupted critical power supply.

[0088] In a specific embodiment, the emergency response module is composed of a series of hardware, including a high-performance microcontroller (such as ARMCortex-M series), an FPGA acceleration board, a communication module (such as an Ethernet interface, a Wi-Fi module), and a storage device (such as an SSD hard disk). These hardware components are connected together through a standard industrial bus (such as Modbus RTU or CAN bus) to form an efficient and reliable real-time monitoring and scheduling platform.

[0089] After the system is initialized, each hardware component is ready for self-check. The network communication module continuously receives real-time meteorological, geographical and power data. After the data enters the server, it is first processed by the data format conversion and cleaning module, converted to a standard format and abnormal data is removed, and then sent to the pattern matching and expert system module. Once an extreme weather warning is detected, the risk scenario is quickly located, the MILP optimization strategy is triggered, and the solver generates a scheduling plan. On the one hand, the control command is sent to the switch cabinet through the communication module to execute line switching and load allocation. On the other hand, the intelligent power switching controller is on standby in real time. Once the power is abnormal, the backup power supply is switched immediately. The whole process is dynamically monitored to ensure timely and efficient power emergency response.

[0090] In actual applications, the emergency response module first starts the data format conversion and cleaning method to integrate the data from the meteorological department, geographic information system and power system into a unified format and filter out erroneous or redundant information. Next, the potential risk scenarios corresponding to the current extreme weather are located in the knowledge base through pattern matching algorithms and expert systems. Based on the matched risks, the optimal scheduling plan is generated through an optimization strategy based on mixed integer linear programming and sent to the physical layer device for execution through Profinet or EtherCAT industrial communication protocols. During the execution process, the emergency response module continuously monitors the actual feedback of the system and adjusts the optimization strategy through a self-correction mechanism to ensure the stability and reliability of the system.

[0091] Compared with the prior art, the emergency response module proposed in the present invention ensures the quality and reliability of data through ETL tools and outlier detection algorithms, and improves the accuracy of subsequent analysis. The combination of pattern matching algorithm and expert system makes risk assessment more comprehensive and accurate, and can detect potential risks at an early stage and take preventive measures in advance. In addition, the MILP model comprehensively considers the scope of power outage, overload risk and backup power supply usage time constraints to ensure that the scheduling scheme is both scientific and highly adaptable. The linear programming solver can quickly generate the optimal scheduling scheme, which improves the scheduling efficiency and decision-making quality. The generated scheduling scheme is sent to the physical layer device for execution through the Profinet or EtherCAT industrial communication protocol to ensure that the control action can be implemented quickly and reliably. The actual feedback of the system is continuously monitored, and the optimization strategy is adjusted through the self-correction mechanism to ensure the stability and reliability of the system. Secondly, a modular design is adopted, and each component is connected through a standard interface for easy maintenance and upgrading. It supports multiple communication protocols and can flexibly access different network environments to meet the needs of diverse application scenarios. At the same time, by continuously optimizing the scheduling scheme, the emergency response module can quickly adjust the output in the face of sudden power demand fluctuations to avoid waste of power resources or insufficient power supply. This not only improves the safety and stability of power supply, but also effectively reduces operating costs and maximizes economic benefits.

[0092] It also includes a remote diagnosis module for providing a remote access interface, allowing operation and maintenance personnel to monitor the status of the switch cabinet in real time, receive warning information, and perform remote fault diagnosis. The remote diagnosis module is used to build a remote access channel through the WebRTC protocol, receive the output data of the edge processing module, and encrypt the transmitted data based on an asymmetric encryption algorithm, and verify the identity of the operation and maintenance personnel through multi-factor authentication; if an abnormal situation is detected, the early warning mechanism is triggered, and the abnormal alarm information is transmitted to the operation and maintenance terminal through the RESTful API interface. The remote diagnosis module stores the timestamp, type and detailed description of each abnormal event through a distributed log system; the remote diagnosis module is also used to allow operation and maintenance personnel to transmit control instructions to the physical layer equipment for execution through the industrial communication protocol EtherCAT and the interactive interface.

[0093] In a specific embodiment, the WebRTC (Web Real-Time Communication) protocol is the key technical support for achieving real-time remote interaction. It uses the ICE (Interactive Connectivity Establishment) framework to achieve network penetration, and detects and traverses the restrictions of network address translation (NAT) devices and firewalls by collecting the IP address, port and other information of the local network interface, combined with the STUN (Session Traversal Utilities for NAT) server and the TURN (Traversal Using Relay NAT) server. In a complex network environment, whether it is a switch cabinet in a private network or a different network area where the operation and maintenance personnel are located, a point-to-point direct communication link can be quickly established, and the operating status data collected by the edge processing module of the switch cabinet, such as voltage, current, temperature, and switch status, can be transmitted to the operation and maintenance personnel terminal in real time in a low-latency, high-bandwidth manner, ensuring the real-time and smoothness of monitoring, just as convenient as direct local operation.

[0094] Asymmetric encryption algorithms use a key pair of public and private keys. Before data transmission, the remote diagnosis module uses the pre-configured public key to encrypt the switch cabinet data to be sent. The encryption process is based on complex mathematical operations to convert plain text data into ciphertext, so that even if the data is illegally intercepted during network transmission, the attacker cannot decrypt and obtain the real information because he does not have the corresponding private key. Only the authorized operation and maintenance personnel terminal with the private key can perform decryption operations after receiving the data and restore the original switch cabinet operation data, effectively preventing the risk of data leakage and ensuring the confidentiality and security of power data.

[0095] The distributed log system uses a distributed file system (HDFS) based on Hadoop. When an abnormality occurs in the switch cabinet, the remote diagnosis module stores the timestamp (accuracy is guaranteed by a high-precision clock synchronization mechanism), type (detailed record according to the fault classification standard) and detailed description (including relevant parameter change trends, equipment operation history, etc.) of each abnormal event on multiple nodes through a distributed storage protocol. This storage method not only improves the reliability of data and prevents log loss due to single point failures, but also facilitates subsequent data analysis and fault tracing. By mining massive log data with big data analysis tools, potential fault patterns can be discovered, providing a basis for preventive maintenance.

[0096] In implementation, the working method of the remote diagnosis module is as follows: after the system is started, the server initializes each hardware component and loads encryption keys, network configuration and other parameters. The network interface card continuously monitors WebRTC connection requests. Once the operation and maintenance personnel initiate access, a communication link is established through the ICE framework, and the edge processing module data is received, which is encrypted and transmitted using an asymmetric encryption algorithm. At the same time, the multi-factor authentication module verifies the identity of the operation and maintenance personnel in real time. During operation, if an abnormality is detected, an alarm message is pushed to the operation and maintenance terminal through the RESTfulAPI interface, and detailed information on the abnormal event is stored in the distributed log system. According to the monitoring situation, the operation and maintenance personnel issue control instructions through the interactive interface, which are transmitted to the physical layer equipment of the switch cabinet via the EtherCAT protocol for execution. The whole process is repeated to ensure the efficient operation of remote diagnosis and control.

[0097] Compared with the existing technology, the remote diagnosis module ensures the security of remote access and data transmission through asymmetric encryption algorithms and multi-factor authentication technology. In addition, the WebRTC protocol and EtherCAT protocol are used to achieve real-time and accurate remote monitoring and control. Secondly, through the collaborative work of the edge processing module and the remote diagnosis module, real-time monitoring and abnormal warning of the switch cabinet status are achieved, improving the operation and maintenance efficiency and the accuracy of fault diagnosis. At the same time, the use of a distributed log system and a RESTful API interface ensures the scalability and maintainability of the system, which facilitates subsequent functional upgrades and operation and maintenance management.

[0098] In the above embodiment, if Figure 5 As shown, an intelligent measurement and control method for a withdrawable switch cabinet includes the following steps:

[0099] a. In each preset measurement and control cycle, the edge processing module uses the sensor network to perform ultra-high frequency sampling of the electrical parameters in the withdrawable switch cabinet, and performs real-time filtering and noise reduction processing on the sampled data through an adaptive filtering algorithm to obtain pre-processed electrical parameter data;

[0100] b. Determine the operating status of the switch cabinet by using a data cross-validation algorithm; if it is determined to be in a normal operating state, execute step c; otherwise, execute step d;

[0101] c. Continue data collection and monitoring operations, store current data in the normal operation data history library, and exit the measurement and control process of the current measurement and control cycle;

[0102] d. Perform trend analysis on abnormal state data and historical data within a certain period of time through long short-term memory network;

[0103] e. According to the fault prediction result, the fault level is judged according to the fault severity classification standard; if the fault prediction time is judged to be less than the preset emergency warning time threshold, step f is executed; otherwise, step g is executed;

[0104] f. Immediately trigger the emergency warning signal sending operation, and send an emergency warning message containing the fault type, expected occurrence time, and switch cabinet location information to the remote monitoring center through the wireless communication module. At the same time, start the emergency protection plan in the local switch cabinet, including cutting off the switch equipment operation of the fault line, and then exit the measurement and control process of the current measurement and control cycle;

[0105] g. Send general warning information to remind operation and maintenance personnel to check and handle at the appropriate time, and continue to monitor the development trend of the fault. If the trend worsens, re-enter step d;

[0106] h. Generate a control action combination through the switch cabinet power failure prediction model according to the current operating state of the switch cabinet, including the electrical parameters collected in real time, the operating state information output in step b, and the future state information output in step d;

[0107] i. For each generated control action combination, the stability index, reliability index and economic index of the power system are calculated through the reinforcement learning model to obtain the comprehensive evaluation score of each control action combination;

[0108] j. Based on the evaluation scores, the particle swarm optimization algorithm is used to search for the optimal control action combination;

[0109] k. Apply the optimized control strategy to the control operation of the withdrawable switchgear, and use the fuzzy control strategy to output control signals to the corresponding switchgear and compensation device to achieve dynamic adjustment of the switchgear operating parameters. During the adjustment process, the control strategy is optimized through the reinforcement learning algorithm;

[0110] l. When an extreme situation is detected, the intelligent scheduling algorithm will be used to match and compare the monitored data with the various extreme situation feature data in the extreme situation knowledge base, adjust the switch cabinet operation mode, and switch to backup equipment, including UPS uninterruptible power supply and diesel generators, when abnormal data is detected;

[0111] m. During the emergency handling process, continuously determine whether the extreme situation has been resolved; if it has been resolved, execute step n; otherwise, continue to execute step l;

[0112] n. Gradually restore the normal operation of the switch cabinet, including reclosing the previously disconnected switch and adjusting the equipment operating parameters to normal status. At the same time, store the data and operation records of the emergency handling process in the emergency handling history library.

[0113] In its specific implementation, this solution, with its ultra-high frequency sampling and adaptive filtering algorithms, can capture the instantaneous changes of electrical parameters with extremely high precision and effectively eliminate the noise interference mixed in them, making the initial monitoring data of the switch cabinet operating status accurate and reliable, greatly improving the early detection capability of hidden fault hazards, strangling potential risks in the bud, and laying a solid foundation for the safe and stable operation of the power system from the source.

[0114] The long and short-term memory network works in coordination with the fault classification and early warning mechanisms, not only accurately grasping the trend of fault development, but also implementing differentiated processing strategies based on the severity of the fault. This means that for emergency faults, high-intensity response measures can be quickly initiated to avoid serious consequences such as large-scale power outages caused by fault deterioration. For minor faults, operation and maintenance resources can be reasonably arranged to avoid unnecessary losses caused by overreaction, thereby comprehensively improving the scientificity and accuracy of emergency response and ensuring the continuity and stability of power supply.

[0115] Reinforcement learning, particle swarm optimization and fuzzy control are integrated to break the rigid mode of traditional fixed control strategies and dynamically optimize the control action combination according to the real-time operation status of the power system. Whether it is to cope with daily load fluctuations or complex and changeable working conditions, the operating parameters of the switch cabinet can be flexibly adjusted to accurately achieve the multi-objective balance of power system stability, reliability and economy, ensure the high quality of power supply, and meet the stringent requirements of various power consumption scenarios.

[0116] The intelligent dispatching algorithm is very effective in extreme situations. Its knowledge base, which integrates knowledge from multiple fields, is like an all-knowing and omnipotent escort. Once extreme conditions such as heavy rain, lightning strikes, and earthquakes are detected, it can quickly match the corresponding risk scenarios, adjust the switch cabinet operation mode in time, and decisively switch to backup equipment at critical moments to protect the power supply and ensure that even in harsh environments, key power consumption areas can still maintain stable power supply, greatly enhancing the resilience of the power system to cope with extreme challenges and demonstrating the excellent protection capabilities of the solution under complex real-world conditions.

[0117] 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. An intelligent measurement and control system for a withdrawable switch cabinet; characterized in that: The system comprises: The edge processing module is used to collect the operating status information of the switch cabinet in real time, including voltage, current, temperature, switch status, and protection action information, and to perform real-time analysis on the voltage, current, and temperature of the switch cabinet. When abnormal data is detected, the edge processing module is also used to issue an early warning notification through an early warning mechanism; A dynamic monitoring and analysis module is used to perform state prediction and anomaly detection through a long short-term memory network based on the real-time data collected by the edge processing module. If a potential short circuit or overload risk is detected, a warning signal is generated, and the optimal adjustment strategy is calculated through a switch cabinet power failure mapping model; An adaptive control module, used to adjust the output power and protection settings of the switch cabinet according to the output results of the dynamic monitoring and analysis module, and optimize the control strategy through a reinforcement learning algorithm during the adjustment process; Emergency response module, which is used to monitor environmental variables through the SCADA system. If an extreme weather warning is detected, the switch cabinet operation mode is adjusted through the intelligent scheduling algorithm, and backup equipment, including UPS uninterruptible power supply and diesel generators, is switched when data anomalies are detected; The remote diagnosis module is used to provide a remote access interface, allowing operation and maintenance personnel to monitor the status of the switch cabinet in real time, receive early warning information, and perform remote fault diagnosis.

2. The intelligent measurement and control system for a withdrawable switch cabinet according to claim 1, characterized in that: In the edge processing module, the working method for detecting data anomalies is: when collecting each set of voltage, current, and temperature data, a consistency check is performed through a data cross-validation algorithm to compare similar data collected by different sensors at the same time; if the voltage data difference, the current data difference, or the temperature data difference does not exceed the preset error range, continue to monitor and update the data; if any of the voltage data difference, the current data difference, or the temperature data difference exceeds the preset error range, send real-time detection data and historical detection data within a preset time period to the dynamic monitoring and analysis module, and send early warning information in real time to the central monitoring platform or the mobile device of the operation and maintenance personnel, and the information content includes the abnormality type, occurrence time, and abnormal value.

3. The intelligent measurement and control system for a withdrawable switch cabinet according to claim 1 is characterized in that: In the dynamic monitoring and analysis module, the method for state prediction and anomaly detection by the long short-term memory network is as follows: based on data timestamp information and historical operation data, through the forget gate mechanism of the long short-term memory network, a dynamic weight distribution algorithm is used to perform a weight distribution operation on the hidden state information, screen key information and output the optimized hidden state information to the hidden layer; the hidden layer is used to perform a nonlinear transformation operation through a tanh activation function based on the input data of the input gate and the optimized hidden state information output by the forget gate, and output the time series feature data to the standard feature template, perform a deviation value calculation operation, and obtain the deviation value data between the current data and the standard feature; the standard feature template is generated through historical data training and an expert system; if the deviation value exceeds a preset threshold, a risk judgment operation is performed through an anomaly detection algorithm based on probability distribution, and if the output risk judgment result is yes, a warning signal generation mechanism is triggered, and the output includes the fault probability and the affected range information to the switch cabinet power failure prediction model; if the output risk judgment result is no, the state prediction and anomaly detection operation is terminated; the anomaly detection algorithm based on probability distribution fuses historical fault data and current abnormal features through a Bayesian network to perform fault judgment.

4. The intelligent measurement and control system for a withdrawable switch cabinet according to claim 1, characterized in that: The switch cabinet power failure mapping model includes a state assessment layer, an RL environment simulation layer, a multi-objective optimization layer, a decision selection layer, an instruction generation layer, and a self-correction and feedback layer; the working method of the switch cabinet power failure mapping model is: The state evaluation layer performs real-time evaluation on the current power grid state; the state evaluation layer constructs a dynamic state-action value function through a deep Q network algorithm, quantifies the long-term benefits of different control actions, and outputs the data of the state-action value function to the RL environment simulation layer; The RL environment simulation layer simulates the system response under different control strategies; the RL environment simulation layer generates operation scenarios through simulation modules, including fast action of circuit breakers, load redistribution, and generator output adjustment; and outputs simulation results and evaluation data of each scenario to the multi-objective optimization layer; Finding a balance point between multiple conflicting objectives through the multi-objective optimization layer; The multi-objective optimization layer maps multiple control strategies into a multi-dimensional objective space through a non-dominated sorting genetic algorithm, calculates the fitness value of each strategy, and selects the optimal solution on the Pareto front as a recommended solution; after the solution is selected, the optimized control strategy list is output to the decision selection layer; The decision selection layer selects the optimal control strategy according to the optimization result; the decision selection layer selects the control strategy and outputs it to the instruction generation layer through a rule-based decision tree combined with real-time monitoring data and historical experience; Generate adjustment instructions through the instruction generation layer and record them in the log system at the same time; The self-correction and feedback layer monitors the actual feedback and performs self-correction. When the self-correction and feedback layer detects that the actual feedback deviates from the expected result, it identifies the error factor through the causal relationship diagram and the Bayesian network, and uses the gradient descent method to adjust the model parameters and feed back to the state evaluation layer.

5. The intelligent measurement and control system for a withdrawable switch cabinet according to claim 4 is characterized in that: The instruction generation layer performs fuzzification processing on the input state variables through membership functions, converts the values ​​into language variables, and performs reasoning operations through fuzzy reasoning algorithms to calculate preliminary fuzzy control outputs; then, the control quantity value is calculated based on the preliminary fuzzy control outputs through the center of gravity method to determine at least the adjustment position of the transformer tap in the switch cabinet and the switching capacity parameters of the reactive compensation device.

6. The intelligent measurement and control system for a withdrawable switch cabinet according to claim 1, characterized in that: The method for the adaptive control module to optimize the control strategy through the reinforcement learning algorithm is as follows: first, at least voltage, current, and temperature variables are converted into discrete state representations through a state encoder, and then a state-action value function is constructed to quantify the long-term benefits brought about by taking different control actions under each state; when a new state change is detected, the reinforcement learning algorithm selects the optimal action according to the latest state-action value function and executes the corresponding control instructions; during the execution process, the adaptive control module continuously monitors the actual feedback of the system and adjusts the parameters of the state-action value function through a self-correction mechanism.

7. The intelligent measurement and control system for a withdrawable switch cabinet according to claim 1, characterized in that: The intelligent dispatching algorithm first integrates the weather forecast data of the meteorological department, the geographic information data of the area where the switch cabinet is located, and the real-time operation data of the current power system into a unified format through data format conversion and cleaning methods, and filters out erroneous or redundant information; the real-time operation data of the current power system includes the load conditions of different lines and the temperature status of each switch cabinet; then, the potential risk scenarios corresponding to the current extreme weather are located in the knowledge base through the pattern matching algorithm and the expert system; Based on the matching risks, the power outage scope, overload risk and backup power usage time constraints are comprehensively considered through an optimization strategy based on mixed integer linear programming, and a scheduling plan including line switching and load allocation is generated through a linear programming solver.

8. The intelligent measurement and control system for a withdrawable switch cabinet according to claim 1, characterized in that: The remote diagnosis module is used to build a remote access channel through the WebRTC protocol, receive the output data of the edge processing module, encrypt the transmitted data through an asymmetric encryption algorithm, and verify the identity of the operation and maintenance personnel through multi-factor authentication; if an abnormal situation is detected, the early warning mechanism is triggered, and the abnormal alarm information is transmitted to the operation and maintenance terminal through the RESTful API interface. The remote diagnosis module stores the timestamp, type and detailed description of each abnormal event through a distributed log system; the remote diagnosis module is also used to allow the operation and maintenance personnel to transmit control instructions to the physical layer equipment for execution through the industrial communication protocol EtherCAT and the interactive interface.

9. An intelligent measurement and control method for a withdrawable switch cabinet, characterized in that: An intelligent measurement and control system for a withdrawable switch cabinet as claimed in any one of claims 1 to 8 comprises the following steps: a. In each preset measurement and control cycle, the edge processing module uses the sensor network to perform ultra-high frequency sampling of the electrical parameters in the withdrawable switch cabinet, and performs real-time filtering and noise reduction processing on the sampled data through an adaptive filtering algorithm to obtain pre-processed electrical parameter data; b. Determine the operating status of the switch cabinet by using a data cross-validation algorithm; if it is determined to be in a normal operating state, execute step c; otherwise, execute step d; c. Continue data collection and monitoring operations, store current data in the normal operation data history library, and exit the measurement and control process of the current measurement and control cycle; d. Perform trend analysis on abnormal state data and historical data within a certain period of time through long short-term memory network; e. According to the fault prediction result, the fault level is judged according to the fault severity classification standard; if the fault prediction time is judged to be less than the preset emergency warning time threshold, step f is executed; otherwise, step g is executed; f. Immediately trigger the emergency warning signal sending operation, and send an emergency warning message containing the fault type, expected occurrence time, and switch cabinet location information to the remote monitoring center through the wireless communication module. At the same time, start the emergency protection plan in the local switch cabinet, including cutting off the switch equipment operation of the fault line, and then exit the measurement and control process of the current measurement and control cycle; g. Send general warning information to remind operation and maintenance personnel to check and handle at the appropriate time, and continue to monitor the development trend of the fault. If the trend worsens, re-enter step d; h. Generate a control action combination through the switch cabinet power failure prediction model according to the current operating state of the switch cabinet, including the electrical parameters collected in real time, the operating state information output in step b, and the future state information output in step d; i. For each generated control action combination, the stability index, reliability index and economic index of the power system are calculated through the reinforcement learning model to obtain the comprehensive evaluation score of each control action combination; j. Based on the evaluation scores, the particle swarm optimization algorithm is used to search for the optimal control action combination; k. Apply the optimized control strategy to the control operation of the withdrawable switchgear, and use the fuzzy control strategy to output control signals to the corresponding switchgear and compensation device to achieve dynamic adjustment of the switchgear operating parameters. During the adjustment process, the control strategy is optimized through the reinforcement learning algorithm; l. When an extreme situation is detected, the intelligent scheduling algorithm will be used to match and compare the monitored data with the various extreme situation feature data in the extreme situation knowledge base, adjust the switch cabinet operation mode, and switch to backup equipment, including UPS uninterruptible power supply and diesel generators, when abnormal data is detected; m. During the emergency handling process, continuously determine whether the extreme situation has been resolved; if it has been resolved, execute step n; otherwise, continue to execute step l; n. Gradually restore the normal operation of the switch cabinet, including reclosing the previously disconnected switch and adjusting the equipment operating parameters to normal status. At the same time, store the data and operation records of the emergency handling process in the emergency handling history library.

Images