Intelligent switch circuit AI safety management system and method based on machine learning

Abstract

The invention relates to the technical field of intelligent safety electrical engineering management, and discloses an intelligent switch circuit AI safety management system and method based on machine learning. The AI safety management method for the intelligent switch circuit is applied to circuit management equipment and specifically comprises the following steps: acquiring original data of a multi-source sensor from the intelligent switch circuit in real time through a high-precision sensor and high-frequency sampling; the original data of the multi-source sensor comprises current data, voltage data, temperature data, humidity data, vibration data and switch state data. According to the invention, by introducing multi-modal sensor fusion and high-frequency data acquisition and combining a streaming processing architecture and a machine learning model, accurate and real-time sensing and evaluation of the safety state of the intelligent switch circuit are realized, and circuit behaviors are comprehensively described by using time domain, frequency domain and time-frequency domain multi-dimensional feature extraction, so that the safety of the intelligent switch circuit is improved. And the limitation of traditional single threshold detection is overcome.

Description

Technical Field

[0001] This invention relates to the technical field of intelligent safety electrical engineering management, and more specifically, to an AI-based safety management system and method for intelligent switching circuits based on machine learning. Background Technology

[0002] With the rapid development of IoT technology and smart grids, smart switching circuits, as key nodes in power distribution and control, are of paramount importance in terms of safety. Traditional circuit protection mechanisms mainly rely on physical circuit breakers, fuses, and electronic protection relays based on fixed thresholds. These devices achieve protection by cutting off the circuit when faults such as overcurrent and short circuits are detected. Their core technology is analog circuits or simple digital logic judgments, which are fast-responding but have limited functionality.

[0003] In recent years, with the advancement of digitalization, microprocessor-based monitoring systems have emerged. These systems can collect basic voltage and current data and compare them with preset safety thresholds, triggering alarms or actions if these limits are exceeded. These existing technologies constitute the basic defense line for current circuit safety; however, they are essentially a reactive, "after-the-fact" or "during-the-fact" response. More advanced solutions are beginning to incorporate simple data analysis, such as recording historical fault data and calculating basic power quality parameters. However, these analyses often remain at the statistical level, lacking true predictive and cognitive capabilities. In summary, existing technologies mainly suffer from the following key technical problems: First, its protection logic relies on preset, fixed thresholds, lacking adaptive and learning capabilities. The operating state of the circuit system is affected by multiple dynamic factors such as load changes, environmental conditions, and equipment aging. Fixed thresholds cannot accurately reflect these changes, easily leading to two extremes: one is that the threshold setting is too sensitive, generating false alarms and frequently cutting off the circuit, affecting normal power supply; the other is that the threshold setting is too lenient, failing to detect slowly developing hidden faults in time, until an accident occurs. Second, its data utilization capability is low, usually focusing only on individual parameters while ignoring the correlation between multiple sources of information. Circuit faults are often the result of the combined effects of multiple physical fields such as electrical, mechanical, and thermal. It is difficult to conduct accurate root cause analysis and large-scale fault early warning based solely on current or voltage data. Third, the system's response strategy is simplistic and passive, unable to achieve predictive maintenance and optimized operation.

[0004] These problems have led to blind spots, delayed responses, and insufficient intelligence in the current safety management of intelligent switching circuits, which cannot meet the urgent needs of future high-reliability smart grids for proactive safety and predictive operation and maintenance. Therefore, developing a closed-loop intelligent safety management method that can deeply integrate multi-source data, has self-learning and self-adaptive capabilities, and can realize closed-loop intelligent safety management from perception to decision-making has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] The purpose of this invention is to provide an AI safety management system and method for intelligent switching circuits based on machine learning. By introducing multimodal sensor fusion and high-frequency data acquisition, and combining streaming processing architecture and machine learning models, it achieves accurate and real-time perception and evaluation of the safety status of intelligent switching circuits. By utilizing time domain, frequency domain, and multi-dimensional feature extraction in the time and frequency domains, it comprehensively characterizes the circuit behavior, aiming to solve the problems in the prior art.

[0006] This invention is implemented as follows: a machine learning-based AI-powered safety management method for intelligent switching circuits, applied to circuit management devices, specifically includes the following steps: S101: Real-time acquisition of multi-source sensor raw data from the intelligent switch circuit through high-precision sensors and high-frequency sampling. The multi-source sensor raw data includes current data, voltage data, temperature data, humidity data, vibration data, and switch status data. S102: Clean the collected raw data from multiple sources of sensors to remove outliers, fill in missing values ​​and smooth noise. Use the sliding window averaging method to handle transient interference. Then, use minimum-maximum scaling to convert data of different dimensions into a unified range to achieve data normalization. Finally, extract features from the processed data to generate a structured dataset and label it as normal or abnormal. S103: Use historically obtained structured data to train a supervised machine learning model. Training supervision includes dataset partitioning, model selection, and hyperparameter tuning. The training process uses cross-validation and utilizes cross-entropy or mean squared error as the loss function to output a safety score or anomaly probability. After training, the machine learning model is deployed to edge devices or cloud servers and is continuously updated and trained regularly. The machine learning model achieves adaptive learning to adapt to circuit aging or environmental changes. S104: Input the pre-processed structured dataset into the trained machine learning model for continuous safety assessment and real-time monitoring. Use a streaming data processing architecture to achieve continuous, low-latency safety assessment of the state of the smart switch circuit and output real-time safety indicators. If the indicators output abnormal index or risk level, visualize them through the dashboard and trigger low-level warnings or high-level alarms. S105: Executes automated response safety operations based on low-level warnings or high-level alarms. The response safety operations are graded according to the severity of the anomaly to achieve grid-level coordination, ensuring stable circuit operation. During the response safety operation process, maintenance personnel are allowed to manually intervene or confirm the operation to perform response processing that minimizes safety risks.

[0007] Furthermore, in S101, the acquisition methods for the raw data from the multi-source sensors, including current data, voltage data, temperature data, humidity data, vibration data, and switch status data, include: Current data is acquired through a Hall effect sensor, voltage data is acquired through a voltage divider circuit and an analog-to-digital converter, temperature data is acquired through a thermocouple or a digital temperature sensor, humidity data is acquired through a capacitive humidity sensor, vibration data is acquired through a MEMS accelerometer, and switch status data is acquired through an optocoupler or a relay status detection circuit. The data acquisition frequency is set to 1000-1500 times per second to ensure the capture of fast transient events in the circuit. The acquired data is transmitted to the microcontroller via SPI or I2C interface and temporarily stored in a circular buffer. At the same time, it is uploaded to the cloud server or local server via Wi-Fi for further processing.

[0008] Furthermore, the collected data is transmitted to the microcontroller via SPI or I2C interface and temporarily stored in a circular buffer. Simultaneously, it is uploaded to a cloud server or local server via Wi-Fi for further processing. Prior to this, the circuit management device performs a self-test, including: Perform sensor self-tests regularly to detect faults, ensure data accuracy, and record timestamps for time series analysis. The collected data includes sensor ID metadata, collection time, and geographic location information, which are used for multi-device collaborative analysis. A backup sensor is also provided to prevent the main sensor from failing and to improve system reliability. Data transmission employs encryption protocols to prevent data tampering or leakage. Through comprehensive data acquisition strategies, the integrity, accuracy, and security of the data are ensured, providing a reliable data source for subsequent machine learning analysis.

[0009] Furthermore, in S102, feature extraction is performed on the processed data to generate a structured dataset and label it as normal or abnormal, including: After data cleaning and normalization, multi-dimensional feature extraction is performed to extract core features from the preprocessed sensor data: Time-domain features, including mean, variance, peak value, and waveform factor, are used to directly characterize the amplitude variation of a signal. Frequency domain characteristics are obtained by converting the signal to the frequency domain through Fast Fourier Transform and calculating its dominant frequency, spectral centroid, and bandwidth energy to reveal potential periodic faults or resonance phenomena. Furthermore, for non-stationary signals, wavelet transform time-frequency analysis tools are applied to extract the joint time-frequency domain features of wavelet coefficient energy entropy and scale graph moments, accurately capturing the time-frequency domain features of local time of instantaneous anomalies; By fusing time-domain features, frequency-domain features, and time-frequency-domain features with circuit topology information and environmental context data, a structured feature dataset is constructed. Each data sample is labeled as normal or abnormal based on historical fault records, providing high-quality and comprehensive input for subsequent model training.

[0010] Furthermore, in S103, training supervision includes dataset partitioning, model selection, and hyperparameter tuning. The training process employs cross-validation, utilizing cross-entropy or mean squared error as the loss function, and outputting a safety score or anomaly probability, including: First, the labeled structured dataset is divided into a preset ratio: 70% training set, 15% validation set, and 15% test set. Based on the characteristics of circuit safety problems, a preliminary model architecture is selected from the candidate model library, and a search space and optimization target are set for hyperparameter tuning. Using training set data, the generalization performance of the model under different hyperparameter combinations is evaluated through cross-validation. The model parameters are continuously iterated and updated to minimize the prediction error for the cross-entropy of the classification task or the mean squared error of the regression task based on the selected loss function. Once trained, the machine learning model can output a quantified safety score or anomaly probability for the input real time series data, providing accurate judgment criteria for real-time monitoring.

[0011] Furthermore, in S104, the pre-processed structured dataset is input into the trained machine learning model for continuous security assessment and real-time monitoring. A streaming data processing architecture is used to achieve continuous, low-latency security assessment of the intelligent switching circuit state, outputting real-time security indicators, including: The structured data stream after real-time preprocessing is continuously fed into the trained machine learning model deployed in the streaming engine. The streaming engine uses a sliding window mechanism to dynamically extract and maintain features of the continuous data stream, enabling the machine learning model to perform inference based on the latest continuous time-series data, thus achieving true continuous security assessment. The machine learning model performs instantaneous reasoning on each data window to calculate a security score or real-time security index reflecting the current circuit state or the probability of anomalies. The security indicators, along with the original context information, are immediately output to the downstream complex event processing end for advanced logic judgment, and simultaneously distributed to the visualization dashboard and the automated response end, forming an end-to-end low-latency intelligent monitoring system from data input to result distribution.

[0012] Furthermore, the machine learning model performs instantaneous inference on each data window to calculate a security score or real-time security indicator reflecting the current circuit state, including: The machine learning model receives standardized feature vectors within a specific time window from the stream processing engine and immediately performs a forward propagation computation. The forward propagation computation utilizes the complex nonlinear mapping relationships learned by the machine learning model to comprehensively analyze the multi-dimensional features within the window and outputs an unfiltered logistic regression result reflecting the preliminary probability of anomalies. The initial results of the unfiltered logistic regression, which reflect the probability of anomalies, are then standardized and mapped to a preset quantitative range. At this point, the safety indicators are appended with timestamps and circuit node identifiers, forming a complete monitoring event, which is immediately pushed to the downstream alarm judgment and visualization end, providing the most direct and quantitative decision basis for whether to trigger a response action.

[0013] Furthermore, in S105, automated response safety operations are executed based on low-level warnings or high-level alarms. These response safety operations are graded according to the severity of the anomaly to achieve grid-level coordination and ensure stable circuit operation, including: Based on the severity of the security indicators output by real-time monitoring, a preset tiered response strategy is triggered. For low-level warnings, logs are automatically recorded and notifications are sent to the mobile devices of operations and maintenance personnel; for high-level alerts, local automated operations are immediately initiated. Based on this, the response end further uploads local alarms and operation instructions to the higher-level power grid management system through a standard interface. The power grid management system performs wide-area analysis and performs grid-level optimization and adjustment based on the collaborative information from multiple smart switches, completes rapid switching of power supply lines, adjusts distributed energy output, and realizes multi-level linkage from a single node to the entire network.

[0014] Compared with existing technologies, the AI-based safety management system and method for intelligent switching circuits based on machine learning provided by this invention have the following beneficial effects: 1. By introducing multimodal sensor fusion and high-frequency data acquisition, and combining streaming processing architecture and machine learning models, accurate and real-time perception and evaluation of the safety status of intelligent switching circuits are achieved. By using time domain, frequency domain and multi-dimensional feature extraction, the circuit behavior is comprehensively characterized, overcoming the limitations of traditional single threshold detection. Through sliding window mechanism and instantaneous model inference, millisecond-level low-latency anomaly detection and alarm are achieved, significantly improving the timeliness of fault detection, transforming from passive response to proactive early warning, and effectively avoiding the expansion of accidents caused by response lag. 2. Through the adaptive learning and continuous updating mechanism of the machine learning model, the system can dynamically adapt to circuit aging and environmental changes, maintain long-term monitoring accuracy, combine local automated hierarchical response with grid-level collaborative control, and perform differentiated operations from log recording to rapid load shedding according to risk level. Furthermore, through functions such as digital twins and predictive maintenance, it achieves comprehensive protection from nodes to the network. This not only greatly improves the operational reliability of individual circuits, but also optimizes the resilience and stability of the overall power grid through system-level collaborative optimization, realizing an intelligent upgrade of the safety management model.

[0015] A machine learning-based intelligent switching circuit AI safety management system is used to execute the aforementioned intelligent switching circuit AI safety management method. The system includes: The data acquisition module is used to collect raw data from the intelligent switching circuit in real time through multi-source sensors; The data preprocessing module is used to clean, normalize, and extract features from the raw data to generate labeled structured datasets. The model training and management module is used to train, deploy, and continuously update machine learning models, enabling them to have adaptive learning capabilities. The real-time monitoring and alarm module is used to continuously assess the circuit status and trigger alarms through a streaming processing architecture. The intelligent response and collaborative control module is used to perform hierarchical automated response operations based on alarm levels and to achieve grid-level collaborative control.

[0016] Specifically, the real-time monitoring and alarm module includes: The streaming data access and processing unit is used to receive the structured data stream after real-time preprocessing, and to dynamically extract and maintain the features of the continuous data stream using a sliding window mechanism. The real-time inference unit of the model is used to perform instantaneous forward propagation calculations on the standardized feature vectors within each data window, and to perform comprehensive analysis using the complex nonlinear mapping relationships learned by the model, outputting preliminary results that reflect the possibility of anomalies. The safety indicator generation and post-processing unit is used to standardize and post-process the preliminary results, map them to a preset quantitative safety score or anomaly probability range, and attach timestamps and circuit node identifiers to generate complete monitoring events. The alarm and distribution unit is used to immediately push the security indicators to the downstream complex event processing logic for advanced judgment, and simultaneously distribute them to the visualization interface and the intelligent response and collaborative control module to form an end-to-end low-latency intelligent monitoring closed loop. Attached Figure Description

[0017] Figure 1This is a flowchart illustrating the AI-based safety management method for intelligent switching circuits based on machine learning proposed in this invention. Figure 2 This is a schematic flowchart illustrating the self-testing process of the circuit management device in the AI ​​safety management method for intelligent switching circuits based on machine learning proposed in this invention. Figure 3 This is a schematic diagram of the structure of the AI ​​safety management system for intelligent switching circuits based on machine learning proposed in this invention. Figure 4 This is a schematic diagram of the real-time monitoring and alarm module in the AI ​​safety management system for intelligent switching circuits based on machine learning proposed in this invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] The implementation of the present invention will be described in detail below with reference to specific embodiments.

[0020] In the accompanying drawings of this embodiment, the same or similar reference numerals correspond to the same or similar components. In the description of this invention, it should be understood that if terms such as "upper," "lower," "left," and "right" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting this invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0021] Reference Figure 1-2 As shown, the AI-based safety management method for intelligent switching circuits, based on machine learning, is applied to circuit management devices and specifically includes the following steps: S101: Through high-precision sensors and high-frequency sampling, it collects raw data from multiple sensor sources in real time from the intelligent switch circuit. The raw data from multiple sensor sources includes current data, voltage data, temperature data, humidity data, vibration data, and switch status data. The raw data from the multi-source sensors includes current data, voltage data, temperature data, humidity data, vibration data, and switch status data, acquired through various methods, including: Current data is acquired through a Hall effect sensor, voltage data is acquired through a voltage divider circuit and an analog-to-digital converter, temperature data is acquired through a thermocouple or a digital temperature sensor, humidity data is acquired through a capacitive humidity sensor, vibration data is acquired through a MEMS accelerometer, and switch status data is acquired through an optocoupler or a relay status detection circuit. The data acquisition frequency is set to 1000-1500 per second to ensure that fast transient events in the capture circuit are captured. The acquired data is transmitted to the microcontroller via SPI or I2C interface and temporarily stored in a ring buffer. At the same time, it is uploaded to the cloud server or local server via Wi-Fi for further processing. Specifically, Kalman filters or particle filters are used to fuse multi-source sensor data to improve data accuracy and reliability. The fusion process includes data alignment, correlation, and estimation. The sensor types are expanded to include infrared thermal imagers for detecting hotspots, acoustic sensors for capturing arc noise, and radio frequency sensors for electromagnetic interference monitoring. Intelligent diagnostic functions are integrated, and the health status of sensors is detected in real time through a rule engine. The transmission protocol adopts MQTT or CoAP to adapt to low-power IoT environments. S102: Clean the collected raw data from multiple sources of sensors to remove outliers, fill in missing values ​​and smooth noise. Use the sliding window averaging method to handle transient interference. Then, use minimum-maximum scaling to convert data of different dimensions into a unified range to achieve data normalization. Finally, extract features from the processed data to generate a structured dataset and label it as normal or abnormal. Finally, feature extraction is performed on the processed data to generate a structured dataset, which is then labeled as normal or abnormal, including: After data cleaning and normalization, multi-dimensional feature extraction is performed to extract core features from the preprocessed sensor data: Time-domain features, including mean, variance, peak value, and waveform factor, are used to directly characterize the amplitude variation of a signal. Frequency domain characteristics are obtained by converting the signal to the frequency domain through Fast Fourier Transform and calculating its dominant frequency, spectral centroid, and bandwidth energy to reveal potential periodic faults or resonance phenomena. Furthermore, for non-stationary signals, wavelet transform time-frequency analysis tools are applied to extract the joint time-frequency domain features of wavelet coefficient energy entropy and scale graph moments, accurately capturing the time-frequency domain features of local time of instantaneous anomalies; By fusing time-domain features, frequency-domain features, and time-frequency-domain features with circuit topology information and environmental context data, a structured feature dataset is constructed. Each data sample is labeled as normal or abnormal based on historical fault records, providing high-quality and comprehensive input for subsequent model training. The preprocessing workflow integrates real-time quality monitoring, uses statistical process control (SPC) charts to detect data drift, and dynamically adjusts preprocessing parameters. In addition, the preprocessing supports multi-resolution analysis, adopts adaptive filtering for data with different sampling rates, optimizes data representation, improves model performance in complex scenarios, and ensures that the safety management method is more generalizable and robust. S103: Use historically obtained structured data to train a supervised machine learning model. Training supervision includes dataset partitioning, model selection, and hyperparameter tuning. The training process uses cross-validation and utilizes cross-entropy or mean squared error as the loss function to output a safety score or anomaly probability. After training, the machine learning model is deployed to edge devices or cloud servers and is continuously updated and trained regularly. The machine learning model achieves adaptive learning to adapt to circuit aging or environmental changes. The training process employs cross-validation, utilizing cross-entropy or mean squared error as the loss function. Specifically, it uses a safe state as the environment and switching actions as agents, learning the optimal safe policy through Q-learning or policy gradient methods. The training process takes place in a simulated environment, using digital twin technology to create a virtual circuit model, integrating transfer learning to transfer knowledge from pre-trained models in similar circuit domains, reducing training time and data requirements. Hyperparameter tuning employs an automated machine learning (AutoML) framework for efficient optimization. S104: Input the pre-processed structured dataset into the trained machine learning model for continuous safety assessment and real-time monitoring. Use a streaming data processing architecture to achieve continuous, low-latency safety assessment of the state of the smart switch circuit and output real-time safety indicators. If the indicators output abnormal index or risk level, visualize them through the dashboard and trigger low-level warnings or high-level alarms. Specifically, the pre-processed structured dataset is input into a trained machine learning model for continuous safety assessment and real-time monitoring. A streaming data processing architecture is used to achieve continuous, low-latency safety assessment of the intelligent switching circuit state, outputting real-time safety indicators, including: The structured data stream, which is preprocessed in real time, is continuously fed into the trained machine learning model deployed in the streaming engine. The streaming engine uses a sliding window mechanism to dynamically extract and maintain features of the continuous data stream, enabling the machine learning model to perform inference based on the latest continuous time-series data, thus achieving true continuous security assessment. The machine learning model performs instantaneous reasoning on each data window to calculate a safety score or real-time safety indicator reflecting the current circuit state or the probability of anomalies. Security indicators, along with the original context information, are immediately output to the downstream complex event processing end for advanced logic judgment, and simultaneously distributed to the visualization dashboard and the automated response end, forming an end-to-end low-latency intelligent monitoring system from data input to result distribution. The real-time monitoring process includes a federated learning mechanism, specifically: the monitoring process performs local model inference on distributed edge nodes and periodically aggregates model updates to the central server to achieve privacy protection and reduce communication overhead. The monitoring integrates a real-time feedback loop to feed back false alarms or missed alarms to the model for online learning and continuous performance optimization. In addition, the monitoring supports a multi-tenant architecture, allowing different users to customize security policies and thresholds. S105: Automated response safety operations are executed based on low-level warnings or high-level alarms. These response safety operations are graded according to the severity of the anomaly to achieve grid-level coordination, ensuring stable circuit operation. During the response safety operation process, maintenance personnel are allowed to manually intervene or confirm the operation, performing minimal safety risk response processing. This technical solution introduces multi-modal sensor fusion and high-frequency data acquisition, combined with a streaming processing architecture and machine learning models, to achieve accurate and real-time perception and assessment of the safety status of intelligent switching circuits. Utilizing time-domain, frequency-domain, and multi-dimensional feature extraction, it comprehensively characterizes circuit behavior, overcoming the limitations of traditional single-threshold detection. Through a sliding window mechanism and instantaneous model inference, it achieves millisecond-level low-latency anomaly detection and alarm, significantly improving the timeliness of fault discovery, transforming from passive response to proactive early warning, and effectively preventing the escalation of accidents due to response lag. The response steps include predictive maintenance capabilities, specifically: estimating the remaining lifespan of circuit components based on historical data and model predictions, and planning maintenance activities in advance; maintenance strategies include replacement recommendations, lubrication schedules, or cleaning reminders, integrated through the CMMS system; and response actions are combined with blockchain technology to record security events to an immutable ledger to ensure audit trails.

[0022] In S101 of this embodiment, the collected data is transmitted to the microcontroller via SPI or I2C interface and temporarily stored in a circular buffer. Simultaneously, it is uploaded to a cloud server or local server via Wi-Fi for further processing. Before this, the circuit management device performs a self-test, including: Perform sensor self-tests regularly to detect faults, ensure data accuracy, and record timestamps for time series analysis. The collected data includes sensor ID metadata, collection time, and geographic location information, which are used for multi-device collaborative analysis. A backup sensor is also provided to prevent the main sensor from failing and to improve system reliability. Data transmission employs encryption protocols to prevent data tampering or leakage. Through comprehensive data acquisition strategies, the integrity, accuracy, and security of the data are ensured, providing a reliable data source for subsequent machine learning analysis.

[0023] In S103 of this embodiment, training supervision includes dataset partitioning, model selection, and hyperparameter tuning. The training process employs cross-validation, utilizing cross-entropy or mean squared error as the loss function, and outputting a safety score or anomaly probability, including: First, the labeled structured dataset is divided into a preset ratio: 70% training set, 15% validation set, and 15% test set. Based on the characteristics of circuit safety problems, a preliminary model architecture is selected from the candidate model library, and a search space and optimization target are set for hyperparameter tuning. Using training set data, the generalization performance of the model under different hyperparameter combinations is evaluated through cross-validation. The model parameters are continuously iterated and updated to minimize the prediction error for the cross-entropy of the classification task or the mean squared error of the regression task based on the selected loss function. Once trained, the machine learning model can output a quantified safety score or anomaly probability for the input real time series data, providing accurate judgment criteria for real-time monitoring.

[0024] In S104 of this embodiment, the machine learning model performs instantaneous inference on each data window to calculate a security score or real-time security indicator reflecting the current circuit state, including: The machine learning model receives standardized feature vectors within a specific time window from the stream processing engine and immediately performs a forward propagation computation. The forward propagation computation utilizes the complex nonlinear mapping relationships learned by the machine learning model to comprehensively analyze the multi-dimensional features within the window, outputting an unfiltered logistic regression result reflecting the preliminary probability of anomalies. The initial results of the unfiltered logistic regression, which reflect the probability of anomalies, are then standardized and mapped to a preset quantitative range. At this point, the safety indicators are appended with timestamps and circuit node identifiers, forming a complete monitoring event. This event is immediately pushed to the downstream alarm judgment and visualization end, providing the most direct and quantitative decision-making basis for whether to trigger a response action.

[0025] In S105 of this embodiment, automated response safety operations are performed based on low-level warnings or high-level alarms. These response safety operations are graded according to the severity of the anomaly to achieve grid-level coordination and ensure stable circuit operation, including: Based on the severity of the security indicators output by real-time monitoring, a preset tiered response strategy is triggered. For low-level warnings, logs are automatically recorded and notifications are sent to the mobile devices of operations and maintenance personnel; for high-level alerts, local automated operations are immediately initiated. Based on this, the response end further uploads local alarms and operation instructions to the higher-level power grid management system through a standard interface. The power grid management system performs wide-area analysis and performs grid-level optimization and adjustment based on the collaborative information from multiple smart switches, completing the rapid switching of power supply lines and adjusting the output of distributed energy, realizing multi-level linkage from a single node to the entire network.

[0026] This technical solution utilizes the adaptive learning and continuous updating mechanism of machine learning models to enable the system to dynamically adapt to circuit aging and environmental changes, maintaining long-term monitoring accuracy. It combines local automated hierarchical response with grid-level collaborative control, performing differentiated operations from log recording to rapid load shedding based on risk levels. Furthermore, it achieves comprehensive protection from nodes to the network through functions such as digital twins and predictive maintenance. This not only greatly improves the operational reliability of individual circuits but also optimizes the resilience and stability of the overall power grid through system-level collaborative optimization, realizing an intelligent upgrade of the safety management model.

[0027] Reference Figure 3-4 As shown, the AI ​​safety management system for intelligent switching circuits based on machine learning is used to execute the aforementioned AI safety management method for intelligent switching circuits. The system includes: a data acquisition module for real-time acquisition of raw data from intelligent switching circuits through multi-source sensors; a data preprocessing module for cleaning, normalizing, and extracting features from the raw data to generate labeled structured datasets; a model training and management module for training, deploying, and continuously updating machine learning models to enable adaptive learning; a real-time monitoring and alarm module for continuous safety assessment of circuit status and triggering alarms through a streaming processing architecture; and an intelligent response and collaborative control module for performing hierarchical automated response operations based on alarm levels and achieving grid-level collaborative control.

[0028] In this embodiment, the real-time monitoring and alarm module includes: a streaming data access and processing unit, used to receive structured data streams after real-time preprocessing, and to dynamically extract and maintain features of the continuous data stream using a sliding window mechanism; a real-time model inference unit, used to perform instantaneous forward propagation calculations on the standardized feature vectors within each data window, and to perform comprehensive analysis using the complex nonlinear mapping relationships learned by the model, outputting preliminary results reflecting the probability of anomalies; and a safety indicator generation and post-processing unit, used to perform standardized post-processing on the preliminary results, mapping them to a preset quantitative safety score or anomaly probability range, and adding timestamps and circuit node markers. The system identifies and generates complete monitoring events; the alarm and distribution unit immediately pushes security indicators to the downstream complex event processing logic for advanced judgment, and simultaneously distributes them to the visualization interface and intelligent response and collaborative control module, forming an end-to-end low-latency intelligent monitoring closed loop. Through the adaptive learning and continuous update mechanism of the machine learning model, the system can dynamically adapt to circuit aging and environmental changes, maintain long-term monitoring accuracy, and combine local automated hierarchical response with grid-level collaborative control. It performs differentiated operations from log recording to rapid load shedding according to the risk level, and achieves all-round protection from nodes to the network through functions such as digital twins and predictive maintenance.

[0029] By introducing multimodal sensor fusion and high-frequency data acquisition, and combining streaming processing architecture and machine learning models, accurate and real-time perception and assessment of the safety status of intelligent switching circuits are achieved. Utilizing multi-dimensional feature extraction in the time domain, frequency domain, and time-frequency domain, circuit behavior is comprehensively characterized, overcoming the limitations of traditional single-threshold detection. Through a sliding window mechanism and instantaneous model inference, millisecond-level low-latency anomaly detection and alarm are achieved, significantly improving the timeliness of fault discovery and transforming passive response into proactive early warning, effectively preventing the escalation of accidents due to response lag. Through the adaptive learning and continuous update mechanism of the machine learning model, the system can dynamically adapt to circuit aging and environmental changes, maintaining long-term monitoring accuracy. Combining local automated hierarchical response with grid-level collaborative control, differentiated operations from log recording to rapid load shedding are performed according to the risk level.

[0030] In this embodiment, the entire operation process can be controlled by a computer to provide signal feedback and implement the steps sequentially. These are all conventional knowledge in current automation control, and will not be elaborated on in this embodiment.

[0031] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A machine learning-based AI-powered safety management method for intelligent switching circuits, characterized in that: Applied to circuit management devices, the specific steps include: S101: Real-time acquisition of multi-source sensor raw data from the intelligent switch circuit through high-precision sensors and high-frequency sampling. The multi-source sensor raw data includes current data, voltage data, temperature data, humidity data, vibration data, and switch status data. S102: Clean the collected raw data from multiple sources of sensors to remove outliers, fill in missing values ​​and smooth noise. Use the sliding window averaging method to handle transient interference. Then, use minimum-maximum scaling to convert data of different dimensions into a unified range to achieve data normalization. Finally, extract features from the processed data to generate a structured dataset and label it as normal or abnormal. S103: Use historically obtained structured data to train a supervised machine learning model. Training supervision includes dataset partitioning, model selection, and hyperparameter tuning. The training process uses cross-validation and utilizes cross-entropy or mean squared error as the loss function to output a safety score or anomaly probability. After training, the machine learning model is deployed to edge devices or cloud servers and is continuously updated and trained regularly. The machine learning model achieves adaptive learning to adapt to circuit aging or environmental changes. S104: Input the pre-processed structured dataset into the trained machine learning model for continuous safety assessment and real-time monitoring. Use a streaming data processing architecture to achieve continuous, low-latency safety assessment of the state of the smart switch circuit and output real-time safety indicators. If the indicators output abnormal index or risk level, visualize them through the dashboard and trigger low-level warnings or high-level alarms. S105: Executes automated response safety operations based on low-level warnings or high-level alarms. The response safety operations are graded according to the severity of the anomaly to achieve grid-level coordination, ensuring stable circuit operation. During the response safety operation process, maintenance personnel are allowed to manually intervene or confirm the operation to perform response processing that minimizes safety risks.

2. The AI-based safety management method for intelligent switching circuits based on machine learning as described in claim 1, characterized in that, In S101, the acquisition methods for the raw data from the multi-source sensors, including current data, voltage data, temperature data, humidity data, vibration data, and switch status data, include: Current data is acquired through a Hall effect sensor, voltage data is acquired through a voltage divider circuit and an analog-to-digital converter, temperature data is acquired through a thermocouple or a digital temperature sensor, humidity data is acquired through a capacitive humidity sensor, vibration data is acquired through a MEMS accelerometer, and switch status data is acquired through an optocoupler or a relay status detection circuit. The data acquisition frequency is set to 1000-1500 times per second to ensure the capture of fast transient events in the circuit. The acquired data is transmitted to the microcontroller via SPI or I2C interface and temporarily stored in a circular buffer. At the same time, it is uploaded to the cloud server or local server via Wi-Fi for further processing.

3. The AI-based safety management method for intelligent switching circuits based on machine learning as described in claim 2, characterized in that, The collected data is transmitted to the microcontroller via SPI or I2C interface and temporarily stored in a circular buffer. Simultaneously, it is uploaded to a cloud server or local server via Wi-Fi for further processing. Prior to this, the circuit management device performs a self-test, including: Perform sensor self-tests regularly to detect faults, ensure data accuracy, and record timestamps for time series analysis. The collected data includes sensor ID metadata, collection time, and geographic location information, which are used for multi-device collaborative analysis. A backup sensor is also provided to prevent the main sensor from failing and to improve system reliability. Data transmission employs encryption protocols to prevent data tampering or leakage. Through comprehensive data acquisition strategies, the integrity, accuracy, and security of the data are ensured, providing a reliable data source for subsequent machine learning analysis.

4. The AI-based safety management method for intelligent switching circuits based on machine learning as described in claim 3, characterized in that, In S102, the processed data is finally subjected to feature extraction to generate a structured dataset and labeled as normal or abnormal, including: After data cleaning and normalization, multi-dimensional feature extraction is performed to extract core features from the preprocessed sensor data: Time-domain features, including mean, variance, peak value, and waveform factor, are used to directly characterize the amplitude variation of a signal. Frequency domain characteristics are obtained by converting the signal to the frequency domain through Fast Fourier Transform and calculating its dominant frequency, spectral centroid, and bandwidth energy to reveal potential periodic faults or resonance phenomena. Furthermore, for non-stationary signals, wavelet transform time-frequency analysis tools are applied to extract the joint time-frequency domain features of wavelet coefficient energy entropy and scale graph moments, accurately capturing the time-frequency domain features of local time of instantaneous anomalies; By fusing time-domain features, frequency-domain features, and time-frequency-domain features with circuit topology information and environmental context data, a structured feature dataset is constructed. Each data sample is labeled as normal or abnormal based on historical fault records, providing high-quality and comprehensive input for subsequent model training.

5. The AI-based safety management method for intelligent switching circuits based on machine learning as described in claim 4, characterized in that, In S103, training supervision includes dataset partitioning, model selection, and hyperparameter tuning. The training process employs cross-validation, utilizing cross-entropy or mean squared error as the loss function, and outputting a safety score or anomaly probability, including: First, the labeled structured dataset is divided into a preset ratio: 70% training set, 15% validation set, and 15% test set. Based on the characteristics of circuit safety problems, a preliminary model architecture is selected from the candidate model library, and a search space and optimization target are set for hyperparameter tuning. Using training set data, the generalization performance of the model under different hyperparameter combinations is evaluated through cross-validation. The model parameters are continuously iterated and updated to minimize the prediction error for the cross-entropy of the classification task or the mean squared error of the regression task based on the selected loss function. Once trained, the machine learning model can output a quantified safety score or anomaly probability for the input real time series data, providing accurate judgment criteria for real-time monitoring.

6. The AI-based safety management method for intelligent switching circuits based on machine learning as described in claim 5, characterized in that, In S104, the pre-processed structured dataset is input into the trained machine learning model for continuous security assessment and real-time monitoring. A streaming data processing architecture is used to achieve continuous, low-latency security assessment of the smart switch circuit state, outputting real-time security metrics, including: The structured data stream after real-time preprocessing is continuously fed into the trained machine learning model deployed in the streaming engine. The streaming engine uses a sliding window mechanism to dynamically extract and maintain features of the continuous data stream, enabling the machine learning model to perform inference based on the latest continuous time-series data, thus achieving true continuous security assessment. The machine learning model performs instantaneous reasoning on each data window to calculate a security score or real-time security index reflecting the current circuit state or the probability of anomalies. The security indicators, along with the original context information, are immediately output to the downstream complex event processing end for advanced logic judgment, and simultaneously distributed to the visualization dashboard and the automated response end, forming an end-to-end low-latency intelligent monitoring system from data input to result distribution.

7. The AI-based safety management method for intelligent switching circuits based on machine learning as described in claim 6, characterized in that, The machine learning model performs instantaneous inference on each data window to calculate a security score or real-time security indicator reflecting the current circuit state, including: The machine learning model receives standardized feature vectors within a specific time window from the stream processing engine and immediately performs a forward propagation computation. The forward propagation computation utilizes the complex nonlinear mapping relationships learned by the machine learning model to comprehensively analyze the multi-dimensional features within the window and outputs an unfiltered logistic regression result reflecting the preliminary probability of anomalies. The initial results of the unfiltered logistic regression, which reflect the probability of anomalies, are then standardized and mapped to a preset quantitative range. At this point, the safety indicators are appended with timestamps and circuit node identifiers, forming a complete monitoring event, which is immediately pushed to the downstream alarm judgment and visualization end, providing the most direct and quantitative decision basis for whether to trigger a response action.

8. The AI-based safety management method for intelligent switching circuits based on machine learning as described in claim 7, characterized in that, In S105, automated response safety operations are executed based on low-level warnings or high-level alarms. These response safety operations are graded according to the severity of the anomaly to achieve grid-level coordination, ensuring stable circuit operation, including: Based on the severity of the security indicators output by real-time monitoring, a preset tiered response strategy is triggered. For low-level warnings, logs are automatically recorded and notifications are sent to the mobile devices of operations and maintenance personnel; for high-level alerts, local automated operations are immediately initiated. Based on this, the response end further uploads local alarms and operation instructions to the higher-level power grid management system through a standard interface. The power grid management system performs wide-area analysis and performs grid-level optimization and adjustment based on the collaborative information from multiple smart switches, completes rapid switching of power supply lines, adjusts distributed energy output, and realizes multi-level linkage from a single node to the entire network.

9. A machine learning-based intelligent switching circuit AI safety management system, characterized in that, The system is used to perform the AI ​​safety management method for intelligent switching circuits according to any one of claims 1-8, the system comprising: The data acquisition module is used to collect raw data from the intelligent switching circuit in real time through multi-source sensors; The data preprocessing module is used to clean, normalize, and extract features from the raw data to generate labeled structured datasets. The model training and management module is used to train, deploy, and continuously update machine learning models, enabling them to have adaptive learning capabilities. The real-time monitoring and alarm module is used to continuously assess the circuit status and trigger alarms through a streaming processing architecture. The intelligent response and collaborative control module is used to perform hierarchical automated response operations based on alarm levels and to achieve grid-level collaborative control.

10. The AI ​​safety management system for intelligent switching circuits based on machine learning as described in claim 9, characterized in that, The real-time monitoring and alarm module includes: The streaming data access and processing unit is used to receive the structured data stream after real-time preprocessing, and to dynamically extract and maintain the features of the continuous data stream using a sliding window mechanism. The real-time inference unit of the model is used to perform instantaneous forward propagation calculations on the standardized feature vectors within each data window, and to perform comprehensive analysis using the complex nonlinear mapping relationships learned by the model, outputting preliminary results that reflect the possibility of anomalies. The safety indicator generation and post-processing unit is used to standardize and post-process the preliminary results, map them to a preset quantitative safety score or anomaly probability range, and attach timestamps and circuit node identifiers to generate complete monitoring events. The alarm and distribution unit is used to immediately push the security indicators to the downstream complex event processing logic for advanced judgment, and simultaneously distribute them to the visualization interface and the intelligent response and collaborative control module to form an end-to-end low-latency intelligent monitoring closed loop.

Images