A centrally located switchgear intelligent one-button control device

Abstract

This invention relates to the field of power distribution automation technology and discloses an intelligent one-button control device for mid-mounted switchgear, comprising a data fusion module, a parameter optimization module, an operation execution module, an evaluation and maintenance module, a collaborative scheduling module, a visualization module, and a safety protection module. The data fusion module collects switchgear operation data; the parameter optimization module obtains equipment fingerprint vectors and drive parameters through transfer learning; the operation execution module generates operation sequences using a deep Q-network; the evaluation and maintenance module evaluates equipment health through a multimodal fusion network; the collaborative scheduling module achieves equipment synchronization and cluster scheduling; the visualization module provides a graphical interface and an operation interface; and the safety protection module has emergency stop, overcurrent protection, and five-proof verification functions. This invention can realize intelligent and automated operation of switchgear, improving safety and maintenance efficiency.

Description

Technical Field

[0001] This invention relates to the field of power distribution automation technology, and more specifically, to a centrally located intelligent one-button control device for switchgear. Background Technology

[0002] Medium-voltage switchgear is a core piece of equipment in power distribution systems, widely used in industrial parks, commercial buildings, and other locations. Traditional operation relies on manual intervention to perform a series of actions, including circuit breaker opening and closing, trolley movement, and grounding switch operation. This process is complex and time-consuming. A single switchgear typically takes 6 to 8 minutes to complete a power supply or de-energization cycle, and operating multiple units can take 1 to 2 hours. Operation requires multiple people working together, is physically demanding, and carries the risk of errors due to fatigue or lack of experience.

[0003] Existing electric operating devices for switchgear can achieve partial automation through motor drives, but they generally use fixed drive parameters and fail to consider the differences in mechanical characteristics between different brands and years of equipment, which can easily lead to unstable operation or failure. At the same time, the devices lack deep perception of operating status and cannot achieve intelligent planning and adaptive adjustment of operating procedures, still requiring manual intervention. Furthermore, equipment health management functions are weak, mainly relying on operation count statistics, lacking analysis of multi-dimensional data such as current and temperature, making it difficult to predict faults in advance, and remaining at the stage of post-fault maintenance.

[0004] To address the aforementioned issues, there is an urgent need for a switchgear control device that integrates intelligent sensing, adaptive control, and health prediction in the operation and maintenance of power distribution systems. This device would enable fully automated operation and predictive maintenance, thereby improving the efficiency and reliability of power distribution system operation and maintenance. Summary of the Invention

[0005] This invention provides a centrally located intelligent one-button control device for switchgear, which solves the technical problems in related technologies, such as fixed drive parameters that cannot adapt to different equipment characteristics, lack of intelligent decision-making ability that requires manual judgment of operation steps, and weak equipment health management function that cannot predict potential faults.

[0006] This invention provides an intelligent one-button control device for a centrally located switchgear, comprising:

[0007] The data fusion module is used to collect switchgear operation data and obtain sensor data streams and real-time equipment status vectors after preprocessing.

[0008] The parameter optimization module is used to collect motor current curves and displacement curves, and uses adaptive and transfer learning methods to obtain device fingerprint vectors, personalized drive parameters and fingerprint drift warning signals.

[0009] The operation execution module is used to acquire sensor data, device state vectors, and personalized drive parameters, and uses a state-action value network to obtain a one-click operation sequence and operation completion signal.

[0010] The assessment and maintenance module is used to obtain the cumulative number of operations, motor current timing, temperature timing and environmental parameter timing, and to obtain the health assessment results by using multimodal fusion analysis.

[0011] The collaborative scheduling module is used to obtain bus status synchronization signals and cluster operation scheduling instructions based on the device status vector and one-click operation sequence, using status synchronization and cloud scheduling methods.

[0012] The visualization module is used to acquire all data generated during the operation process. It adopts a graphical layout to obtain a dynamic primary system diagram, operation control commands and parameter configuration data.

[0013] The safety protection module is used to obtain protection signals by adopting safety protection strategies based on the motor current curve and displacement curve, operation completion signal and real-time equipment status vector.

[0014] In a preferred embodiment, the data fusion module includes:

[0015] High-voltage energizing sensors are installed on each of the three phases in the busbar compartment. The main control unit performs a logical AND operation on the three-phase energizing signals and determines that the busbar is energized only when all three phases are energized simultaneously.

[0016] Limit switches for working and testing positions are installed on the handcart guide rail. Auxiliary contacts are led out from the circuit breaker and grounding switch operating mechanisms. Magnetic switches are installed at the cabinet door position. The contact signals are connected to the main control unit. The main control unit scans each contact signal and filters out contact jitter to obtain a stable set of equipment status signals.

[0017] Temperature and humidity sensors are installed in each functional room, and the main control unit polls each sensor to obtain environmental parameter data; wireless temperature sensors are installed at the busbar contact arms and cable joints, and the temperature data is forwarded to the main control unit through the receiving module.

[0018] In a preferred embodiment, the parameter optimization module includes:

[0019] A current sensor is connected in series in the power supply circuit of the handcart drive motor, and a displacement sensor is installed on the handcart guide rail. The main control unit collects motor current and displacement data synchronously when performing handcart propulsion operation.

[0020] When performing the handcart propulsion operation for the first time, the main control unit uses the default drive parameters to collect data and extracts multi-dimensional features from the current curve to construct the equipment fingerprint vector, including features of the start-up stage, acceleration stage, constant speed stage, and deceleration stage.

[0021] The main control unit pre-stores a general driving parameter library. After extracting the fingerprint vector of a new device, it uses the K-nearest neighbor algorithm to search for the most similar sample and performs a weighted average of the driving parameters of the nearest samples. The weight is inversely proportional to the distance to obtain the initial driving parameters of the new device.

[0022] In a preferred embodiment, the parameter optimization module further includes:

[0023] When performing subsequent operations, the main control unit continues to collect data and extract fingerprint vectors, uses Bayesian optimization algorithm to fine-tune the driving parameters, selects the next set of parameters to be tried through the expected improvement function, and converges to the local optimum driving parameters after iteration;

[0024] The main control unit stores the device fingerprint vector and driving parameters in the local fingerprint database and uploads them to the cloud platform periodically. The cloud platform builds a global device fingerprint knowledge base, uses clustering algorithms to identify clusters of similar devices, calculates the cluster center fingerprint and optimal parameters, and sends them to update the parameter database.

[0025] After each operation, the main control unit extracts the current fingerprint vector and compares it with the historical fingerprint vector to calculate the Euclidean distance. When the distance is less than the first preset threshold, it is determined that the fingerprint has not drifted. When the distance is between the first preset threshold and the second preset threshold, the fingerprint vector and driving parameters are updated using the exponential weighted moving average method. When the distance is greater than the second preset threshold, it is determined that the fingerprint has changed and a warning signal is displayed.

[0026] In a preferred embodiment, the operation execution module includes:

[0027] The switchgear operation process is modeled as a Markov decision process. The state space is defined as including cabinet door state, circuit breaker state, truck state, grounding switch state, and bus energized state. The action space is defined as including circuit breaker opening, circuit breaker closing, truck forward, truck backward, grounding switch opening, grounding switch closing, and waiting. The reward function is defined as consisting of three parts: goal achievement reward, operation efficiency reward, and safety constraint penalty.

[0028] A deep Q-network model is established in the main control unit. The input layer receives the one-hot encoding of the state vector, the hidden layer uses the activation function, and the output layer outputs the Q value for each action. The Q value represents the expected long-term cumulative reward for performing the corresponding action in the current state.

[0029] Before the device leaves the factory, the deep Q network is trained offline in a simulation environment. An experience pool is established using an experience replay mechanism to store a quadruple of state, action, reward, and next state. During the training process, the agent interacts with the simulation environment, selects actions according to a greedy policy, randomly samples and updates the deep Q network parameters from the experience pool, and stores the trained network parameters in the main control unit.

[0030] When the operator triggers the one-click power-on command, the main control unit reads the device state vector, encodes the state vector and inputs it into a deep Q network. The network calculates and outputs the Q value of each action. The main control unit selects the action with the largest Q value that satisfies the five-prevention constraints to execute. After execution, the new device state vector is obtained and input into the network again to calculate the next action. The above process is repeated until the power-on target state is reached.

[0031] In a preferred embodiment, the assessment and maintenance module includes:

[0032] The main control unit establishes operation count counters for circuit breakers, handcarts, and grounding switches respectively. When an operation is completed, the corresponding counter is incremented by one. The main control unit presets the rated operating life threshold for each operating object and calculates the life consumption ratio to obtain the life assessment index.

[0033] The main control unit collects the motor current curve and extracts the peak starting current parameter during each handcart propulsion operation. It stores the peak starting current and timestamp in the historical database and periodically uploads the current time series data to the cloud platform. The cloud platform establishes a time series convolutional network model to predict the trend. When the slope of the predicted curve is greater than the preset slope threshold, an early warning message is generated.

[0034] The cloud platform establishes a temperature anomaly detection model based on the Transformer architecture, which uses a self-attention mechanism to capture abnormal patterns in the temperature sequence. The model outputs an anomaly score at each time step, and triggers an anomaly alarm when the anomaly score exceeds a preset anomaly threshold.

[0035] In a preferred embodiment, the assessment and maintenance module further includes:

[0036] The cloud platform establishes a multimodal fusion neural network. It adopts a multi-branch architecture to process the number of operations, motor current timing, temperature timing, and environmental parameter timing by independent feature extraction branches. The feature vectors of each modality are concatenated to form a fusion feature vector. The fusion layer uses a multi-head attention mechanism to calculate the attention weights between features of different modalities, and finally outputs a health score.

[0037] The cloud platform uses a survival analysis model to estimate the remaining useful life of equipment. The survival analysis model is based on the Weibull distribution to model the equipment failure time distribution. The Cox proportional hazards regression method is used to calculate the risk ratio of the current equipment. The remaining useful life probability distribution of the equipment is calculated by combining the Weibull distribution and the risk ratio. The median is extracted as the predicted value of the remaining useful life.

[0038] In a preferred embodiment, the cooperative scheduling module includes:

[0039] A bus is used to build a communication network between devices. The communication modules of each device are connected in parallel to the bus. Each device is assigned a unique bus address. The incoming line cabinet device acts as the master station, and the outgoing line cabinet device acts as the slave station.

[0040] Define the data frame format for communication between devices. The data frame structure includes frame header, address code, function code, data length, data content, check code, and frame tail. When the incoming line cabinet circuit breaker is closed, the main control unit detects the change in the energized state of the bus and sends a status broadcast frame through the bus. All outgoing line cabinet devices receive the broadcast frame, parse the data content, and update the local bus energized state variables.

[0041] The cloud platform sends operation instructions to each device in the order of the device list. After receiving the instructions, the device executes the operation process and sends an operation completion message to the platform after the process is completed. After receiving the completion message, the platform sends instructions to the next device.

[0042] In a preferred embodiment, the visualization module includes:

[0043] The main page adopts a layout that simulates a primary system diagram, drawing graphic symbols for incoming busbars, circuit breakers, handcart contacts, grounding switches, and outgoing circuits, and dynamically coloring them according to the equipment status vectors;

[0044] A one-button power-on button and a one-button power-off button are set up in the one-button operation control area on the side of the system diagram.

[0045] In a preferred embodiment, the security protection module includes:

[0046] An emergency stop button is installed on the front panel of the device. The contacts of the emergency stop button are connected in series in the power supply circuit of the drive output module. When pressed, the power supply to the drive output module is cut off. A current sensor is connected in series in the motor power supply circuit to monitor the motor operating current. When the sampled current exceeds the safety threshold, the motor is determined to be overloaded and the motor drive output is cut off.

[0047] The main control unit sets a maximum allowed execution time for each action. If the timer expires and no action completion signal is received, the operation is deemed to have timed out and the current action is stopped. Before executing an operation, the main control unit calls the five-prevention logic verification function to verify whether the current state meets the preconditions for the operation.

[0048] The beneficial effects of this invention are as follows:

[0049] By establishing a parameter adaptive optimization mechanism based on device fingerprint recognition, and employing transfer learning to quickly match initial driving parameters, combined with Bayesian optimization algorithms for online fine-tuning with a small number of samples, the device can automatically adapt to circuit breakers and operating mechanisms of different brands and mechanical characteristics, solving the problem of unstable operation or failure caused by fixed driving parameters in existing technologies. Simultaneously, a fingerprint drift detection mechanism is established to monitor changes in device mechanical characteristics in real time. When the device's characteristics drift due to wear and aging, the driving parameters are automatically updated to maintain long-term operational stability, improving the device's adaptability to individual device differences and time-varying characteristics.

[0050] By constructing an operation decision model based on deep reinforcement learning, and embedding five-prevention rules into the reward function as safety constraints, the device can autonomously plan the optimal operation sequence based on the current actual state, achieving true one-click automatic operation without requiring manual judgment of operation steps by operators. Simultaneously, a multi-modal data fusion health assessment system is established, integrating multi-dimensional data such as operation frequency, motor current timing, and temperature timing. Temporal convolutional networks and Transformer models are used for trend prediction and anomaly detection, combined with a survival analysis model to estimate the remaining service life of the equipment. This achieves a shift from passive maintenance to predictive maintenance, predicting potential equipment failures and generating maintenance decisions 3 to 6 months in advance, thus improving the intelligence level and operational reliability of the power distribution system. Attached Figure Description

[0051] Figure 1 This is a block diagram of a centrally located switchgear intelligent one-button control device according to the present invention;

[0052] Figure 2 This is a flowchart of a centrally located intelligent one-button control device for a switchgear according to the present invention. Detailed Implementation

[0053] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0054] At least one embodiment of the present invention discloses an intelligent one-button control device for a centrally located switchgear, such as... Figures 1 to 2 As shown, it includes:

[0055] The data fusion module is used to collect switchgear operation data and obtain sensor data streams and real-time equipment status vectors after preprocessing.

[0056] Based on the internal spatial structure and primary equipment layout characteristics of the mid-mounted switchgear, a multi-source heterogeneous sensor network is deployed in key functional areas such as the truck compartment, circuit breaker compartment, busbar compartment, and cable compartment. Data preprocessing and adaptive fusion are achieved through edge intelligent nodes to collect holographic information on the switchgear's operating status. Specifically, the following steps are included:

[0057] S11. High-voltage energized sensors are installed on phases A, B, and C of the busbar compartment, respectively. These sensors utilize inductive electric field sensing principles and output switching signals. The three-phase sensor signals are connected to the digital input port of the microcontroller in the main control unit. The main control unit performs a logical AND operation on the three-phase energized signals, determining that the busbar is energized only when all three phases are simultaneously energized, thus obtaining the busbar energized status flag.

[0058] S12: Install working and test position limit switches on both sides of the handcart guide rail. Lead out open / close auxiliary contacts from the circuit breaker and grounding switch operating mechanisms. Install a magnetic switch at the cabinet door hinge position. Connect all passive contact signals to the opto-isolated input module of the main control unit, using an optocoupler chip to achieve a 2500-volt isolation voltage. The main control unit scans each contact signal at a 50-millisecond cycle, confirming signals with consistent results after three consecutive scans, filtering out contact jitter, and obtaining a stable set of equipment status signals.

[0059] S13 includes digital temperature and humidity sensors installed in the truck compartment, circuit breaker compartment, busbar compartment, and cable compartment. These sensors use a bus interface, with a temperature measurement range of -40 to 125 degrees Celsius and an accuracy of ±0.3 degrees Celsius, and a humidity measurement range of 0 to 100% and an accuracy of ±2%. The four sensors are connected in series on the same bus and are distinguished by different device addresses. The main control unit polls each sensor every 10 seconds to obtain real-time environmental parameter data for each functional compartment.

[0060] S14. Wireless temperature sensors are installed at the three-phase positions of the busbar contact arms and the three-phase positions of the outgoing cable joints. The sensors draw energy from the busbar current, with a measurement range of -40 to 150 degrees Celsius and an accuracy of ±1 degree Celsius. Data is transmitted using the 433 MHz frequency band, with a transmission distance of 50 meters. A wireless temperature receiving module is installed externally on the switchgear, which forwards the temperature data to the main control unit via the bus to obtain the temperature distribution data at key points.

[0061] The S15 main control unit uses a real-time clock chip to establish a unified time reference, with an accuracy better than 2 minutes per year. Millisecond-level timestamps are appended when acquiring sensor data, and data from different types of sensors is uniformly encapsulated into a data structure containing fields such as sensor type, address, data value, timestamp, and validity flag, resulting in a time-synchronized multi-source sensor data stream.

[0062] S16, deploy an adaptive weighted fusion algorithm in the main control unit. For high-voltage live sensor signals, an adaptive filtering based on signal-to-noise ratio is adopted. The signal spectrum is analyzed by fast Fourier transform, the 50 Hz power frequency component is the effective signal, and the high-frequency component is noise. Dynamic weights are assigned to each sensor according to the signal-to-noise ratio.

[0063] The three-phase energized signal employs a weighted voting mechanism. A weighted voting result greater than a second threshold indicates the signal is energized, less than a first threshold indicates it is not energized, and values ​​between the first and second thresholds retain the previous result. For temperature and humidity sensor data, a Kalman filter algorithm is used to fuse historical data with current measurements, outputting the optimal estimate. Values ​​deviating from the predicted value by more than three times the standard deviation are considered outliers and discarded.

[0064] For wireless temperature measurement data, a multi-sensor cross-validation mechanism is established. The Pearson correlation coefficient between the temperature data sequences of phases A, B, and C is calculated using statistical analysis methods. When the correlation coefficient drops below a preset threshold, the temperature measurement data of that phase is determined to be abnormal. The weighted average of the temperatures of adjacent phases is used for compensation and a sensor fault alarm is triggered.

[0065] The fused sensor data stream is stored in the main control unit's circular buffer with a capacity of 1000 data sets. At the same time, the sensor data is converted into a device state vector, which includes five dimensions: cabinet door state, circuit breaker state, handcart state, grounding switch state, and bus energized state. This results in a high-quality sensor data stream and a real-time updated device state vector.

[0066] The parameter optimization module is used to collect motor current curves and displacement curves, and uses adaptive and transfer learning methods to obtain device fingerprint vectors, personalized drive parameters and fingerprint drift warning signals.

[0067] Due to differences in mechanical resistance, travel distance, and inertia between the handcart propulsion mechanisms and grounding switch operating mechanisms of different brands of circuit breakers, using fixed drive parameters can lead to unstable operation or failure of some equipment. Therefore, a few-sample adaptive mechanism based on transfer learning is established to quickly match the optimal drive parameters through equipment fingerprinting technology; the specific steps include:

[0068] In step S21, to achieve equipment fingerprint recognition and parameter optimization, a current sensor is connected in series in the power supply circuit of the trolley drive motor, with a measurement range of -20 to +20 amperes and a sampling frequency of 1000 Hz. A magnetostrictive displacement sensor is installed on the trolley guide rail, measuring a stroke of 400 mm and outputting a current signal of 4 to 20 mA. The trolley propulsion process is divided into five stages: start-up, acceleration, constant speed, deceleration, and stop. During the subsequent step S3, when the trolley propulsion operation is performed, the main control unit synchronously collects motor current and displacement data to obtain the current curve and displacement curve of the trolley propulsion process.

[0069] S22, during the initial handcart propulsion operation, the main control unit uses default drive parameters and synchronously collects motor current and displacement data at a sampling frequency of 1000 Hz. Multi-dimensional feature extraction is performed on the current curve to construct a 12-dimensional equipment fingerprint vector, including features for the startup phase (peak startup current, rise time, integral area), acceleration phase (mean acceleration, standard deviation, acceleration duration), constant speed phase (mean steady-state current, standard deviation, duration), and deceleration phase (mean deceleration, standard deviation, braking duration). The feature vector is normalized to the range of 0 to 1 to obtain the equipment fingerprint feature extraction capability.

[0070] S23: The main control unit pre-stores a general driving parameter library, containing device fingerprint vectors and optimal driving parameters from mainstream brands, totaling 20 samples. After extracting the fingerprint vector of a new device, the K-nearest neighbor algorithm is used to search for the three most similar samples, calculate the Euclidean distance, and perform a weighted average of the driving parameters of the nearest samples, with the weights inversely proportional to the distance, to obtain the initial driving parameters of the new device, achieving fast matching based on transfer learning.

[0071] S24, during the 2nd to 5th operations, the main control unit continues to collect data and extract fingerprint vectors, and uses a Bayesian optimization algorithm to fine-tune the driving parameters. The algorithm establishes a Gaussian process regression model of driving parameters and operational performance indicators, and selects the next set of parameters to be tried through the Expected Improvement Function (EI function). The EI function is used to balance the exploration and utilization of the parameter space, calculates the expected improvement value of each candidate parameter relative to the current optimal parameter, and selects the parameter with the highest EI value for the next attempt. After 3 to 5 iterations, it converges to the local optimal driving parameters, and fine-tunes are performed according to the personalized characteristics of the current device to obtain the few-sample online optimization capability.

[0072] In step S25, the main control unit stores the device fingerprint vector and driving parameters in a local fingerprint database using a hash table structure with the device serial number as the key. This data is periodically uploaded to a cloud platform, where all device data is aggregated to build a global device fingerprint knowledge base. Clustering algorithms are used to identify clusters of similar devices, calculate the cluster center fingerprint and optimal parameters, and update the parameter database via OTA (Over-The-Air) updates, achieving swarm intelligence learning. As the system operates, the knowledge base is continuously enriched, and the initial parameter matching accuracy for new devices improves from 60% to over 90%.

[0073] In step S26, the operation of the grounding switch and circuit breaker also employs an equipment fingerprinting mechanism. A current sensor is connected in series in the power supply circuit of the grounding switch motor. During the subsequent S3 step of executing the grounding switch opening and closing operation, the current curve of the opening and closing process is collected, and 12-dimensional fingerprint features are extracted. The K-nearest neighbor algorithm is used to match the initial parameters, and Bayesian optimization is used for fine-tuning to obtain personalized drive parameters. The circuit breaker operation records the coil current curve and action response time, extracting features such as current peak value, decay time constant, and action delay to optimize the coil excitation parameters.

[0074] S27. As the equipment operates for longer periods, mechanical parts experience wear and aging, causing fingerprint features to drift. Therefore, a fingerprint drift detection and parameter adaptive update mechanism is established. After each operation, the main control unit extracts the current fingerprint vector, compares it with historical fingerprint vectors, and calculates the Euclidean distance.

[0075] When the distance is less than 0.2, it is determined that the fingerprint has not drifted significantly, and the parameters do not need to be updated. When the distance is between 0.2 and 0.5, it is determined that the fingerprint has drifted slowly. The fingerprint vector and driving parameters are updated using the exponentially weighted moving average method. The new fingerprint vector is equal to the historical fingerprint vector multiplied by 0.9 plus the current fingerprint vector multiplied by 0.1. The new driving parameters are updated with the same weight.

[0076] When the distance is greater than 0.5, a fingerprint mutation is detected, which may be caused by equipment overhaul, component replacement, sensor failure, etc. The main control unit displays a warning signal on the interface to check the equipment status due to a sudden change in equipment characteristics. The mutation event is uploaded to the cloud platform. After the maintenance personnel confirm that the equipment status is normal, they can choose to restart the migration learning process to establish a new fingerprint and parameters. Combining steps S21 to S27, the device fingerprint vector (12-dimensional feature vector), personalized driving parameters (start-up voltage, constant speed voltage, deceleration voltage, PID parameters, etc.) and fingerprint drift warning signals (including drift degree indicators and mutation alarms) are obtained.

[0077] The operation execution module is used to acquire sensor data, device state vectors, and personalized drive parameters, and uses a state-action value network to obtain a one-click operation sequence and operation completion signal.

[0078] Based on the safety requirements of switchgear operation and the need for one-button power outage functionality, a deep reinforcement learning model is constructed in the main control unit. The five-prevention rules are embedded as safety constraints into the reward function. Through a combination of offline training and online optimization, an optimal operation sequence that balances safety and efficiency is generated. Specifically, the following steps are included:

[0079] S31. Establish an operational decision-making model. The switchgear operation process is modeled as a Markov decision process. The state space is defined as having five dimensions: cabinet door state, circuit breaker state, trolley state, grounding switch state, and busbar energized state. The state space has 2 x 2 x 3 x 2 x 2 = 48 possible state combinations. The action space is defined as having seven basic actions: circuit breaker opening, circuit breaker closing, trolley forward, trolley backward, grounding switch opening, grounding switch closing, and waiting. The reward function is defined as having three parts: the first part is the goal achievement reward, which provides a positive reward when the power supply or de-energization goal is successfully achieved; the second part is the operational efficiency reward, which deducts 1 point for each action performed, encouraging the completion of the goal with the fewest steps; and the third part is the safety constraint penalty, which provides a negative penalty when the executed action violates the five-prevention rules, forcing the model to learn safe operation paths.

[0080] A Deep Q-Network (DQN) model is established in the main control unit. The network structure includes an input layer, two hidden layers, and an output layer. The input layer receives a one-hot encoded 48-dimensional state vector. The first hidden layer contains 64 neurons using the ReLU activation function, and the second hidden layer contains 32 neurons using the ReLU activation function. The output layer contains 7 neurons corresponding to 7 actions, outputting the Q-value of each action, which is the state-action value function. The Q-value represents the expected long-term cumulative reward for performing the action in the current state. The Deep Q-Network adopts a lightweight design with approximately 30,000 network parameters. The computational cost of a single forward propagation is approximately 100,000 floating-point operations. The microcontroller operates at a frequency of 72 MHz and can complete a single network inference within 50 milliseconds, meeting the requirements of real-time decision-making, thus obtaining the Deep Q-Network decision model.

[0081] S32, an operation strategy knowledge base is established based on offline training. Before the device leaves the factory, the deep Q-network is trained offline in a simulation environment. The simulation environment simulates the transition relationships of 48 states and 7 actions of the switch cabinet, and sets one-click power-on and one-click power-off tasks as training objectives.

[0082] An experience replay mechanism is adopted, and an experience pool is established to store a quadruple of state, action, reward, and next state. The experience pool has a capacity of 10,000 experiences. During training, the agent interacts with the simulation environment. In the current state, it selects an action according to the ε-greedy policy. It randomly selects an action for exploration with probability ε and selects the action with the largest Q value with probability 1-ε. The initial value of ε is set to 1.0 and decreases linearly to 0.1 during training. After the action is executed, the environment returns the reward and the next state, and the experience is stored in the experience pool. 32 experiences are randomly sampled from the experience pool to form a mini-batch. The parameters of the deep Q network are updated using the mini-batch data. The loss function is the mean squared error. The target Q value is equal to the immediate reward plus a discount factor multiplied by the maximum Q value of the next state. The discount factor is preset to 0.95. The Adam optimizer is used for gradient descent, and the learning rate is set to 0.001.

[0083] After multiple iterations of training, the deep Q-network converges, and the agent learns to plan the optimal operation sequence that satisfies the five-defense constraints in any initial state. The trained network parameters are then stored in the non-volatile memory of the main control unit, resulting in an offline-trained operation strategy knowledge base.

[0084] S33 generates a one-click power-on operation sequence based on online inference. When the operator triggers the one-click power-on command, the main control unit reads the latest equipment state vector from the circular buffer of step S1. The equipment state vector contains five dimensions: cabinet door state, circuit breaker state, truck state, grounding switch state, and bus energized state. The state vector is encoded into a 48-dimensional one-hot vector and input into a deep Q-network. The encoding method is that each state dimension corresponds to a set of one-hot codes: cabinet door state (2-dimensional), circuit breaker state (2-dimensional), truck state (3-dimensional), grounding switch state (2-dimensional), and bus energized state (2-dimensional). After concatenation, an 11-dimensional state vector is formed, which is then expanded into a 4-dimensional one-hot vector. An 8-dimensional input vector is used to match the network input layer dimension. The network forward propagation calculates the Q values ​​of the seven actions. The main control unit selects the action with the largest Q value that satisfies the five-prevention constraint as the current action to be executed. If the action with the largest Q value violates the five-prevention constraint, the legal action with the second largest Q value is selected. After executing the action, the new device state vector is obtained in real time from step S1. The new state is input into the network again to calculate the next action. The above process is repeated until the power supply target state is reached, i.e., the circuit breaker is closed, the handcart is in the working position, and the grounding switch is opened, thus obtaining a complete one-button power supply operation sequence.

[0085] For example, the initial state is: circuit breaker open, trolley in test position, grounding switch closed, cabinet door closed, busbar de-energized. The first step of network inference is to select the grounding switch open action. After execution, the new state obtained from S1 is: circuit breaker open, trolley in test position, grounding switch open, cabinet door closed, busbar de-energized. The second step of network inference is to select the trolley forward action. After execution, the new state obtained from S1 is: circuit breaker open, trolley in working position, grounding switch open, cabinet door closed, busbar de-energized. The third step of network inference is to select the circuit breaker close action. After execution, the energized target state is achieved. The entire sequence contains three actions, reducing the cabinet door detection step compared to the traditional fixed process, thus improving operational efficiency. Deep Q-networks can flexibly plan paths based on the current actual state. When the grounding switch is already open in the initial state, the network automatically skips the grounding switch open step and directly executes the trolley forward and circuit breaker close actions, achieving adaptive operation sequence generation.

[0086] In step S34, the main control unit executes each action command sequentially based on the action sequence obtained from the deep Q-network inference. When executing the grounding switch tripping action, the main control unit queries the locally stored device driver parameter library. If optimized driver parameters for the device are found, the optimized parameters (from the personalized parameter library established in step S2) are used; otherwise, the default driver parameters are used.

[0087] The main control unit sends a tripping command to the drive output module. The drive module uses a relay output with a 24-volt DC coil and a contact capacity of 10 amps (250 volts). The contacts are connected in series in the grounding switch motor control circuit. The grounding switch motor is a three-phase asynchronous motor with a rated voltage of 380 volts and a rated power of 200 watts. It drives the operating mechanism through a reduction gear to complete the tripping, which takes approximately 15 to 20 seconds. During the tripping process, the main control unit collects motor current data in real time through a current sensor connected in series in the motor power supply circuit. The collected current curve data is used for equipment fingerprint recognition and drive parameter optimization in the subsequent S2 step.

[0088] After the main control unit issues a command, it starts a timeout timer with a timeout period of 30 seconds. It periodically checks the grounding switch status signal acquired in step S1. Upon detecting a closed contact, it immediately stops the motor, delays for 0.5 seconds for confirmation, and then retrieves the updated status from S1, inputting the data for network inference to proceed to the next action. If no closed contact signal is detected after the timeout, the tripping is deemed a failure, the process stops, a timeout fault message is displayed on the interface, and the log is recorded and uploaded to the cloud platform.

[0089] S35, when the network inference obtains the forward movement of the trolley, the main control unit reads the trolley status from S1. If it is already in the working position, the action is skipped. If it is in the test position, the unit queries the locally stored device drive parameter library to obtain the trolley drive parameters (starting voltage, constant speed voltage, deceleration voltage, PID parameters, etc.). If optimized parameters are found (from step S2), the optimized parameters are used; otherwise, the default parameters are used.

[0090] The main control unit sends a forward command to the drive module, which then controls the drive motor of the handcart to run. The motor is a brushless DC motor with a rated voltage of 48 volts and a rated power of 150 watts. The main control unit controls the motor voltage in stages: during the start-up stage, a starting voltage is applied to overcome static friction; during the constant speed stage, a constant speed voltage is applied and PID closed-loop control is used; and during the deceleration stage, the voltage is reduced to smoothly decelerate.

[0091] The motor drives the trolley forward along the guide rail via a rack and pinion mechanism. The distance from the test position to the working position is approximately 400 mm, and the advancement time is approximately 20 to 25 seconds. During the advancement process, the main control unit collects motor current data in real time through a current sensor connected in series in the motor power supply circuit, and collects displacement data through a displacement sensor on the trolley guide rail. The sampling frequency is 1000 Hz. The collected current curve and displacement curve data are used for equipment fingerprint recognition and drive parameter optimization in the subsequent S2 step.

[0092] After the main control unit issues a command, it starts a timeout timer with a timeout period of 40 seconds. It periodically checks the handcart status signal collected by S1. Upon detecting that the work position limit switch is closed, it immediately stops the motor, delays for 0.5 seconds for confirmation, and then retrieves the updated status from S1, inputting it into the network inference for the next action. If the work position signal is not detected after the timeout, the process is considered to have failed, the process is stopped, and an alarm is triggered.

[0093] S36, when the network inference determines that the circuit breaker is closing, the main control unit reads the circuit breaker status from S1. If it is already closed, it determines that the power supply target has been reached and ends the process. If it is in the open state, it queries the locally stored device drive parameter library to obtain the circuit breaker drive parameters (closing coil excitation time, excitation voltage, etc.). If optimized parameters are found (from step S2), the optimized parameters are used; otherwise, the default parameters are used.

[0094] The main control unit sends a circuit breaker closing command to the drive module. The drive module controls the closing coil to be energized. The closing coil is an electromagnetic coil with a rated voltage of 220 volts DC and a rated power of 300 watts. After the coil is energized, it generates electromagnetic force to drive the operating mechanism to complete the closing action. The closing action time is about 80 milliseconds.

[0095] The main control unit controls the coil energizing time according to the drive parameters, which is 200 milliseconds by default and may be adjusted to 150 to 250 milliseconds after optimization. It automatically disconnects after energizing to avoid overheating. After the main control unit issues a command, it starts a timeout timer with a timeout period of 5 seconds. It periodically checks the circuit breaker status signal collected by S1, and confirms the closing of the closed contact after a delay of 0.5 seconds.

[0096] After confirming the closing, the one-button power-on process ends, a power-on completion prompt is displayed on the interface, and the operation success information is recorded in the log, resulting in a one-button power-on operation completion signal. If no closing signal is detected after the timeout, the closing is deemed a failure, and a circuit breaker closing failure fault prompt and alarm are displayed on the interface.

[0097] S37: After each one-click power-off operation, the main control unit stores the complete experience trajectory of this operation in a local experience pool with a capacity of 1000 trajectories. When the experience pool accumulates 100 new trajectories, the main control unit starts online learning, sampling small batches of data from the experience pool to update the parameters of the deep Q network. The learning rate is 0.0001, lower than the offline training learning rate, to maintain the stability of the learned knowledge.

[0098] Through online learning, the network can adapt to the personalized characteristics of specific switchgear. For example, if the trolley of a certain switchgear has greater resistance and requires more time to advance, the network can automatically adjust the advancement parameters by learning from the operating experience of that switchgear, thereby achieving continuous optimization of the strategy.

[0099] The main control unit periodically uploads the local experience pool to the cloud platform. The cloud aggregates experience data from multiple devices for large-scale offline training. The general policy network parameters obtained from the training are distributed to each device via OTA, realizing the co-evolution of collective intelligence and individual intelligence, and obtaining an adaptive operation policy optimization mechanism based on deep reinforcement learning.

[0100] S38: When the operator triggers a one-button power outage command, the main control unit uses the same deep Q-network inference mechanism as the one-button power-on command. It reads the current state input to the network, the network outputs the Q-values ​​of each action, and selects the action with the largest Q-value that satisfies the five-prevention constraints for execution. A typical one-button power outage sequence is circuit breaker tripping, truck reversing, and grounding switch closing. The network can flexibly adjust the sequence according to the current actual state. If the circuit breaker has already tripped in the initial state, it automatically skips the circuit breaker tripping step, achieving adaptive power outage operation.

[0101] S39 allows for the setting of adjustable delay and emergency stop mechanisms during the one-click power supply / suppression process. A delay time can be set between each step of the process, with a default of 0.5 seconds and an adjustable range of 0.3 to 2 seconds. During the delay, the operator can observe the execution result of the current step and decide whether to continue. If an abnormality is detected, the operator can press the emergency stop button to terminate the process.

[0102] A red mushroom-shaped emergency stop button is set on the front panel of the device. When pressed, the button is locked. The emergency stop button contacts are directly connected in series in the power supply circuit of the drive output module. When pressed, all drive output power is immediately cut off, and the emergency stop contact signal is connected to the priority interrupt input terminal of the main control unit.

[0103] Upon detecting an emergency stop signal, the main control unit immediately terminates the current process, clears all pending instructions, displays an emergency stop trigger prompt on the interface, and records the emergency stop event to the log, thus establishing a safety protection mechanism for the operation process. If an emergency stop signal is detected during deep Q network inference, network inference is immediately stopped and the action queue is cleared to ensure the real-time response of the system to emergency stop commands.

[0104] The assessment and maintenance module is used to obtain the cumulative number of operations, motor current timing, temperature timing and environmental parameter timing, and to obtain the health assessment results by using multimodal fusion analysis.

[0105] Based on multimodal data accumulated during the long-term operation of the switchgear, including the number of operations, motor current timing, key point temperature timing, and environmental parameter timing, a health assessment and fault prediction model is established to achieve a shift from passive maintenance to predictive maintenance. This model can predict potential faults such as wear of the operating mechanism and aging of contacts 3 to 6 months in advance and generate maintenance decisions. Specifically, the model includes the following steps:

[0106] In step S41, the main control unit establishes an operation count counter for each circuit breaker, truck, and grounding switch in each switchgear. These counters are stored in non-volatile memory and are not lost upon power failure. After step S3 is completed, the corresponding counter is incremented by 1, yielding the cumulative operation count for each object. The main control unit presets the rated operating life threshold for each object: 10,000 operations for circuit breakers, 5,000 operations for trucks, and 2,000 operations for grounding switches. The main control unit calculates the lifespan consumption ratio, which is equal to the cumulative operation count divided by the rated lifespan threshold, resulting in a lifespan assessment index based on the number of operations.

[0107] In step S42, the main control unit collects the motor current curve and extracts the peak starting current parameter during each handcart propulsion operation in steps S2 and S3. It stores the peak starting current and timestamp for each operation in the historical database, storing data from the most recent 100 operations. The main control unit periodically uploads the current time-series data to the cloud platform, where a temporal convolutional network (TCN) model is built for feature extraction and trend prediction.

[0108] The TCN model employs a causal convolution and dilated convolution architecture, containing four residual blocks with dilation factors of 1, 2, 4, and 8, respectively. Its receptive field covers historical data from the last 30 operations. The TCN model takes as input the peak initiation current sequence of the last 30 operations and outputs a predicted peak initiation current sequence for the next 10 operations.

[0109] The cloud platform inputs the uploaded current time series data into the trained TCN model to obtain the current trend prediction curve for the next 10 operations. When the slope of the prediction curve is greater than 0.05 amperes per operation, it is determined that there is a risk of poor lubrication or mechanical wear in the handcart propulsion mechanism. An early warning message is generated and sent to the main control unit. The main control unit displays the early warning message on the interface, thus obtaining the current trend prediction capability based on the temporal convolutional network.

[0110] In step S43, the main control unit receives the bus contact arm temperature data from the wireless temperature measurement module in step S1, stores the temperature value and timestamp in the historical database, and stores the temperature data for the most recent 30 days in chronological order, with a sampling period of 10 minutes, totaling 4320 data points over 30 days. The main control unit periodically uploads the temperature time-series data to the cloud platform, where the cloud platform establishes a temperature anomaly detection model based on the Transformer architecture.

[0111] The Transformer model employs a self-attention mechanism to capture long-term dependencies and anomalous patterns in temperature sequences. The model input consists of 1008 data points from the most recent 7 days of temperature time-series data, comprising a 4-layer Transformer encoder, 8 attention heads, and 128 hidden layers. The model output is an anomaly score for each time step, calculated using the reconstruction error. An anomaly alarm is triggered when the anomaly score exceeds 0.8.

[0112] The cloud platform inputs the uploaded temperature time series data into the trained Transformer model to obtain the anomaly score at each time step. The model can not only detect absolute temperature exceeding the limit, but also identify complex patterns such as abnormal temperature change trends, periodic anomalies, and sudden changes. The cloud platform generates anomaly alarm information and sends it to the main control unit. The main control unit displays the alarm information on the interface, thus obtaining the temperature anomaly pattern recognition capability.

[0113] S44, the cloud platform establishes a multimodal fusion neural network, which deeply integrates different modal data such as the number of operations, motor current timing, temperature timing, and environmental parameter timing uploaded by the main control unit, and outputs a comprehensive health assessment result. The multimodal fusion network adopts a multi-branch architecture, with each modal data processed by an independent feature extraction branch.

[0114] The operation count mode uses a fully connected network to extract features. The input consists of three values: the cumulative number of operations of the circuit breaker, the trolley, and the grounding switch. The output is a 64-dimensional feature vector. The motor current time series mode uses a TCN network to extract features. The input consists of the current sequence of the most recent 30 operations. The output is a 64-dimensional feature vector. The temperature time series mode uses a Transformer network to extract features. The input consists of the temperature sequence of the most recent 7 days. The output is a 64-dimensional feature vector. The environmental parameter time series mode uses an LSTM network to extract features. The input consists of the temperature and humidity sequence of the most recent 24 hours. The output is a 64-dimensional feature vector.

[0115] The feature vectors of the four modalities are concatenated to form a 256-dimensional fused feature vector, which is input into the fusion layer for cross-modal feature interaction. The fusion layer adopts a multi-head attention mechanism to calculate the attention weights between features of different modalities and automatically learn the correlation and complementarity between modalities.

[0116] The fusion layer output passes through two fully connected layers to finally output a health score, ranging from 0 to 100, where 100 indicates the device is in good condition and 0 indicates extremely poor condition. The cloud platform inputs the multimodal data uploaded by the main control unit into the trained fusion network, calculates the health score, and sends it to the main control unit. The main control unit displays the health score on the interface, thus obtaining the health assessment capability based on multimodal data fusion.

[0117] S45, the main control unit receives the health assessment results from the cloud platform and sets the health warning thresholds: Level 1 warning 80 points, Level 2 warning 60 points, and Level 3 warning 40 points. The main control unit retrieves the current health status from the cloud platform at midnight every day and compares the health status with the warning thresholds.

[0118] When the health level drops below 80 points for the first time, a Level 1 warning is triggered. The main control unit generates a pre-maintenance work order. The work order includes the work order number, generation time, equipment identification, health level, warning level, recommended maintenance items, and recommended maintenance time. The recommended maintenance time for a Level 1 warning is within 6 months.

[0119] The recommended maintenance items automatically identify the main degradation modes based on the attention weights of the multimodal fusion network on the cloud platform. If the current mode has the highest attention weight, lubrication maintenance is recommended; if the temperature mode has the highest attention weight, checking the contact condition is recommended. The cloud platform sends the maintenance recommendations along with the health status to the main control unit. The main control unit displays the work order information on the interface, and the cloud platform simultaneously pushes the work order to the operation and maintenance management system to generate a maintenance task.

[0120] When the health score drops below 60, a Level 2 warning is triggered, with a recommended maintenance period of 3 months. When the health score drops below 40, a Level 3 warning is triggered, with a recommended maintenance period of 1 month and a recommended maintenance item of comprehensive overhaul. This results in a tiered pre-maintenance work order generation mechanism.

[0121] S46, the cloud platform uses a survival analysis model to estimate the remaining useful life (RUL) of the equipment. The survival analysis model is based on the Weibull distribution to model the equipment failure time distribution. The Weibull distribution is a commonly used life distribution model in reliability engineering, which can describe the full life cycle characteristics of the equipment from initial failure, random failure to wear failure. The shape and scale parameters of the Weibull distribution are obtained by fitting large-scale historical failure data from the cloud platform.

[0122] The cloud platform inputs the multimodal feature vector of the current device into the survival analysis model. The model uses the Cox proportional hazards regression method, which is used to assess the impact of different factors on the risk of device failure. It calculates the risk ratio of the current device relative to the baseline risk. A risk ratio greater than 1 indicates that the failure risk is higher than the average level, and a risk ratio less than 1 indicates that the failure risk is lower than the average level. Combining the Weibull distribution and the risk ratio, the remaining service life probability distribution of the device is calculated, and the median is extracted as the predicted value of the remaining service life in days.

[0123] For example, if the current health score is 70, the multimodal eigenvectors show a clear upward trend in current, the survival analysis model calculates a risk ratio of 1.5, and the Weibull distribution predicts a median remaining useful life of 150 days with a confidence interval of 120 to 180 days. The cloud platform sends the predicted remaining useful life and confidence interval to the main control unit, which then displays the remaining useful life information on the health dashboard, thus obtaining the remaining useful life prediction result.

[0124] When the predicted remaining service life is less than 90 days, the cloud platform automatically raises the warning level and generates a pre-maintenance work order even if the current health status has not yet dropped to the warning threshold, thus achieving early warning based on trends.

[0125] S47, the main control unit monitors short-term fluctuations in health status. The main control unit compares the current day's health status with the previous day's health status, calculates the daily change in health status, and determines an abnormal health status fluctuation when the absolute value of the daily change exceeds 10 points. This fluctuation may be caused by sudden malfunctions, operational abnormalities, sensor failures, etc. The main control unit immediately generates an emergency alarm message, displaying a pop-up window on the interface prompting immediate equipment inspection for the abnormal health status fluctuation. Simultaneously, it sends an emergency alarm to the cloud platform via the communication module. The cloud platform notifies maintenance personnel via SMS or telephone. Upon receiving the alarm, maintenance personnel should immediately inspect the equipment on-site to investigate the cause of the fluctuation, thus implementing the emergency response mechanism for abnormal health status fluctuations.

[0126] The collaborative scheduling module is used to obtain bus status synchronization signals and cluster operation scheduling instructions based on the device status vector and one-click operation sequence, using status synchronization and cloud scheduling methods.

[0127] Based on the application scenario of a large industrial park power distribution system comprising one incoming switchgear and twelve outgoing switchgears, a local area communication network between devices and a wide area communication network from devices to the cloud are established to achieve real-time synchronization of the status of all devices on the same busbar section, and to enable the cloud platform to perform clustered and programmed operation scheduling of multiple devices; specifically, the following steps are included:

[0128] The S51, based on the need for local communication between devices, uses a bus to construct an inter-device communication network. The bus uses differential signal transmission, which has strong anti-interference capabilities, supports one master and multiple slave or multiple master and multiple slave topologies, and the transmission distance can reach 1200 meters. The baud rate is set to 9600 bits per second to meet the requirements of real-time status synchronization.

[0129] A bus is laid along the switch cabinet row in the power distribution room, using shielded twisted-pair cable. The communication module of each device is connected to the bus in parallel through an interface, and 120-ohm terminating resistors are installed at both ends of the bus to eliminate signal reflection.

[0130] Each device is assigned a unique bus address, ranging from 1 to 255. The address of the incoming line cabinet is set to 1, and the addresses of the 12 outgoing line cabinets are set to 2 to 13 respectively. The addresses are configured through the parameter setting function of the human-machine interface and stored in non-volatile memory after configuration, thus obtaining the local area communication network between devices.

[0131] S52 is designed based on a state synchronization protocol, defining the data frame format for communication between devices. The data frames adopt a master-slave communication mode, with the incoming line cabinet device acting as the master station and the 12 outgoing line cabinet devices acting as slave stations. The master station periodically polls each slave station or actively broadcasts status information. After receiving instructions from the master station, the slave stations respond or update their status after receiving broadcast information.

[0132] The data frame structure includes seven fields: frame header, address code, function code, data length, data content, checksum, and frame trailer. The frame header is fixed at 0xAA55 and occupies 2 bytes. The address code occupies 1 byte to indicate the destination or source address. The function code occupies 1 byte to define the data frame function type. The data length occupies 1 byte to indicate the number of bytes in the data content, which can be up to 64 bytes. The checksum occupies 2 bytes and uses the CRC16 checksum algorithm. The frame trailer is fixed at 0x55AA and occupies 2 bytes, thus obtaining the data frame format definition.

[0133] Define the function code type: 0x01 indicates reading status, 0x02 indicates writing control command, 0x03 indicates status broadcast, 0x04 indicates heartbeat hold, and 0x05 indicates time synchronization. Function code 0x03, status broadcast, is used by the incoming line cabinet to broadcast the change in the energized status of the bus to all outgoing line cabinets.

[0134] When the incoming line circuit breaker is closed, the main control unit detects a change in the energized status of the bus and immediately sends a status broadcast frame via the bus. The address code is set to broadcast address 0xFF to indicate that all slave stations receive the frame. The data content includes the bus energized status flag. After receiving the broadcast frame, all outgoing line cabinets parse the data content, update their local bus energized status variables, and trigger the human-machine interface to simulate a change in the bus display color, thereby achieving global status synchronization and obtaining the status synchronization protocol.

[0135] Based on the communication requirements between the device and the cloud, the S53 incorporates a built-in wireless communication module and an Ethernet interface to connect with the cloud-based smart energy management platform. The wireless communication module is a 4G LTE module, supporting the LTE CAT1 network standard with a maximum downlink speed of 10 Mbps and a maximum uplink speed of 5 Mbps. The module has a built-in TCP / IP protocol stack and connects to the main control unit via a UART serial port.

[0136] The main control unit establishes a connection with the cloud platform via dial-up internet access through the AT command set control module. The Ethernet interface uses a 10 / 100 Mbps adaptive Ethernet controller. This chip has a built-in full hardware TCP / IP protocol stack and connects to the main control unit via the SPI interface, supporting up to 8 independent socket connections.

[0137] When the power distribution room has wired network access, Ethernet connection is preferred. When the power distribution room only has wireless network access, 4G module connection is used. The main control unit automatically detects the availability of the network interface and selects the communication method upon startup, thus obtaining flexible cloud communication capabilities.

[0138] The S54 uses the MQTT protocol to enable data interaction between the device and the cloud platform. MQTT is a lightweight publish-subscribe messaging protocol suitable for communication between IoT devices and cloud platforms, offering low overhead and good real-time performance. The main control unit acts as an MQTT client, while the cloud-based smart energy management platform deploys an MQTT server. The MQTT server address is a domain name assigned by the cloud platform, and the port is 8883. TLS encryption is used to ensure data security.

[0139] After the network connection is established, the master control unit sends a connection request to the MQTT server. The connection request includes authentication information such as client identifier, username, and password. The client identifier uses a unique serial number of the device. After the server verifies the information, the MQTT connection is established.

[0140] The master control unit subscribes to command topics issued by the server. The topic name format is device serial number / command. When the server needs to issue control commands to the device, it publishes a message to this topic. After receiving the message, the master control unit parses the command content and executes it. The master control unit periodically publishes status topic messages to the server. The topic name format is device serial number / status. The message content includes information such as the current status of the device, health, and operation logs. The publication cycle is 30 seconds, thus achieving a cloud-based two-way communication mechanism based on MQTT.

[0141] S55, based on a cloud platform, enables cluster operation scheduling. The cloud-based smart energy management platform has the capability to perform programmed sequential operations on a group of switchgear. Platform operators select the equipment group to be operated on the platform interface, such as all 12 outgoing switchgear in a distribution room, and set the operation type to one-click power outage or one-click power on. They also set the operation sequence to either by equipment number or by load priority. The platform generates a cluster operation task, which includes information such as the equipment list, operation type, operation sequence, and task identifier.

[0142] The platform issues operation instructions to each device in the order of the device list. It sends a one-click power outage instruction message to the first device via the MQTT protocol. After receiving the instruction, the first device executes the one-click power outage process. After the process is completed, it sends an operation completion message to the platform. After receiving the completion message, the platform waits for 5 seconds to confirm stability, and then issues an instruction to the second device. The above process is repeated until all devices have completed the operation, thus obtaining the cluster sequential operation capability.

[0143] During cluster operation, the platform monitors the execution status of each device in real time. When a device fails to execute, the platform suspends subsequent operations and alerts the operator that the device has failed and requires manual handling. After troubleshooting, the operator can continue execution from the failed device or terminate the task, thus obtaining the cluster operation exception handling mechanism.

[0144] S56, based on the global visualization of bus status through inter-device coordination, after the circuit breaker of the incoming line cabinet is closed, the main control unit of the incoming line cabinet device detects that the bus energization status changes from non-energized to energized through step S1, and immediately sends a status broadcast frame through the bus. The address code of the broadcast frame is 0xFF, the function code is 0x03, and the data content includes 3 bytes of bus A phase energization flag, B phase energization flag, and C phase energization flag.

[0145] After receiving the broadcast frame, the main control unit of the 12 outgoing line cabinets performs CRC16 verification. After the verification is successful, the data content is parsed, the locally stored bus energized status variable is updated, and the human-machine interface is refreshed in step S6. The color of the simulated bus display area changes from gray to red. The entire synchronization process takes less than 1 second.

[0146] The simulated busbar displays on the 13 switchgear cabinets in the power distribution room change color synchronously. Maintenance personnel can observe the energized status of the entire busbar section by looking at the display interface of any cabinet, achieving global visualization of the busbar status. When the incoming line cabinet circuit breaker trips and the busbar is de-energized, the incoming line cabinet device also sends a status broadcast notification to all outgoing line cabinets, and the simulated busbar display on each outgoing line cabinet changes from red to gray, achieving global synchronization of the de-energized status.

[0147] S57, based on the fault tolerance mechanism of communication failure, when the bus communication failure or the communication module failure of a certain device causes the inability to receive status broadcasts, the device cannot know the change of the bus energization status through the bus. In order to avoid display errors, the device uses local sensor judgment as a backup plan.

[0148] The local high-voltage energized sensor continuously monitors the energized status of the incoming busbar side of the unit. If no status broadcast is received for more than 60 seconds and the local sensor detects a change in energized status, the unit updates the simulated busbar display with the local detection results and prompts a communication fault on the human-machine interface. After discovering the communication fault prompt, the operation and maintenance personnel should check the bus connection or the status of the communication module.

[0149] When a cloud communication failure causes the device to lose connection with the platform, the device retains its complete local operational capabilities. Operators can still perform a one-click power-off operation via the touchscreen. The device caches the operation log in its local storage and automatically uploads it to the cloud platform to replenish the data once the network is restored. Combining steps S51 to S57, the bus status synchronization signal (including the energized status flags of bus phases A, B, and C) and the cluster operation scheduling instructions (including the device list, operation type, operation sequence, and execution status feedback) are obtained.

[0150] The visualization module is used to acquire all data generated during the operation process. It adopts a graphical layout to obtain a dynamic primary system diagram, operation control commands and parameter configuration data.

[0151] Based on the operator's need for intuitive and user-friendly human-machine interaction, the device is equipped with a 10-inch color capacitive touch screen and designed with a multi-page graphical interactive interface to realize functions such as dynamic display of primary equipment, one-button operation control, parameter setting, and log query.

[0152] Based on hardware selection, the S61 features a 10-inch TFT-LCD color touchscreen with a resolution of 1024 x 600 pixels, a brightness of 400 candela per square meter, a contrast ratio of 800:1, a viewing angle of 160 degrees horizontally and 140 degrees vertically, and a capacitive multi-touch touch method with a touch response time of less than 10 milliseconds.

[0153] The touchscreen is connected to the main control unit via an RGB parallel interface. The RGB interface includes 18 data lines, clock lines, and synchronization signal lines. The main control unit has a built-in LCD controller peripheral that directly drives the touchscreen display. The touchscreen chip is connected to the main control unit via an I2C interface. The main control unit reads the touch coordinate data to determine the operator's touch position and action, thus obtaining the touchscreen hardware platform.

[0154] S62 is based on the main page design. The main page adopts a layout that simulates a primary system diagram, drawing graphic symbols for incoming busbars, circuit breakers, truck contacts, grounding switches, and outgoing circuits. The graphic symbols adopt standard graphic symbols for power systems. The circuit breaker symbol is two contacts plus a circuit breaker frame. The truck contact symbol is a movable contact and a stationary contact. The grounding switch symbol is a combination of the grounding symbol and the knife switch symbol.

[0155] Based on the dynamic coloring display of the equipment status vector collected in step S1, the busbar is drawn as a thick red line when it is energized and as a thin gray line when it is de-energized. When the circuit breaker is closed, the contacts are drawn as closed in red and as open in gray.

[0156] When the handcart is in the working position, the handcart contacts and stationary contacts are shown as being in contact. When the handcart is in the test position, the handcart contacts and stationary contacts are shown as being separated. When the grounding switch is closed, it is shown as closed in green. When the grounding switch is open, it is shown as open in gray. The changes in color and graphic status intuitively reflect the real-time operating status of the switchgear, resulting in a dynamic primary system diagram display.

[0157] S63, based on the main page layout, sets up a one-click operation control area on the right side of the primary system diagram, with a one-click power-on button and a one-click power-off button. The button size is 120 pixels wide and 60 pixels high. The background color of the one-click power-on button is green, and the background color of the one-click power-off button is red. The buttons display text labels and adopt a raised effect design to provide visual feedback.

[0158] Below the system diagram, a status information display area is set up to display text information such as cabinet door status, circuit breaker status, handcart status, grounding switch status, busbar energization status, current operating mode, and logged-in user. The information is displayed using text and icons. A green checkmark icon is displayed when the cabinet door is closed, and a red cross icon is displayed when the cabinet door is open.

[0159] A function menu bar is set at the bottom of the page, which includes five menu items: Home, Parameter Settings, Log Query, Health Monitoring, and About. The menu items are displayed as icons and text. When the operator clicks a menu item, they can switch to the corresponding page and get the complete layout of the main page.

[0160] S64, based on the one-click operation interaction process, the operator clicks the one-click power button on the main page. The main control unit receives the touch event, first determines the current logged-in user's permissions. If the user's permissions are at or above the operator level, the operation is allowed. If the user is not logged in, a login dialog box pops up asking the user to log in first.

[0161] After the permission verification is successful, the main control unit reads the current device status collected in step S1 to determine whether the power supply prerequisites are met. If not, a prompt dialog box will pop up to show that the current status does not allow power supply and the specific reason. If the prerequisites are met, an operation confirmation dialog box will pop up, displaying a text prompt that a one-click power supply operation is about to be performed and providing a confirmation button and a cancel button.

[0162] After the operator clicks the confirmation button, the main control unit starts the one-click power supply process of step S3. A semi-transparent operation progress prompt layer is overlaid on the primary system diagram area of ​​the main page. The prompt layer displays a text description of the currently executed step, such as opening the grounding switch. At the same time, an animation effect is displayed next to the corresponding equipment graphic symbol to indicate that the equipment is in operation.

[0163] Each time a step in the process is completed, the prompt layer updates to display the next step. Once the entire process is finished, the prompt layer displays "Operation Complete," and automatically closes after 3 seconds to return to normal display, providing a complete interactive process for one-click operation. The interactive process for the operator clicking the one-click power off button is similar to that of the one-click power on button, the difference being the execution of the one-click power off procedure.

[0164] The S65 is based on a parameter setting page design. Operators can access the parameter setting page by clicking the parameter setting menu on the main page. The page adopts a list plus input box layout, listing configurable parameter items, including temperature warning threshold, humidity control upper limit, humidity control lower limit, operation delay time, drive mode selection, operation mode selection, equipment brand selection, communication address, network configuration and other parameter items.

[0165] The current parameter value and input controls are displayed on the right side of each parameter item. The temperature warning threshold parameter displays the current value, such as 75 degrees Celsius. Clicking it will bring up a numeric keypad dialog box. The operator can enter the new threshold value by touching the numeric keys. After entering the value, click the OK button to save the new parameter value.

[0166] The drive mode selection parameter provides a drop-down menu. Clicking it expands the option list, including adaptive mode, standard mode, and low-speed mode. After the operator selects an option, the drop-down menu collapses and displays the selection result. At the bottom of the parameter setting page, there are save and restore default buttons. After the operator modifies the parameters and clicks the save button, the main control unit writes the new parameter values ​​to non-volatile memory for permanent storage and displays a dialog box indicating that the parameters were saved successfully.

[0167] Clicking the "Restore Defaults" button will restore all parameters to their factory default values ​​in the main control unit. Operator confirmation is required to prevent accidental operation and to provide interactive parameter setting capabilities. The parameter setting page is protected by access controls; only administrators can modify parameters, while operators can only view them. If an operator attempts to modify parameters, a message will appear indicating insufficient permissions; please log in with an administrator account.

[0168] S66 is based on the log query page design. The operator can enter the log query page by clicking the log query menu on the main page. The filter area is set at the top of the page, which provides filter conditions such as date range selection, operation type selection, and operation result selection. The date range selection provides two date selection controls: start date and end date. Clicking the date control will bring up a calendar dialog box, and the operator can select the year, month, and day to set the query date range.

[0169] The operation type selection offers multiple checkboxes, including options such as one-click power supply, one-click power outage, circuit breaker operation, handcart operation, grounding switch operation, and parameter modification. The operator selects the operation type to query, and the operation result selection offers three single options: all, success, and failure.

[0170] The central part of the page displays the log list, showing the query results in a table format. The table columns include serial number, date and time, operation type, operation user, operation result, and detailed information. Each page displays 20 records, and the bottom of the page provides previous and next page buttons.

[0171] After the operator sets the filter criteria and clicks the query button, the main control unit reads the log records that meet the criteria from the log storage and displays them in a list in reverse chronological order. When the operator clicks on a log record, a detailed information dialog box pops up to display the complete content of the log, including details such as the status before the operation, the operation steps, the status after the operation, and fault information, thus providing log query and interactive capabilities.

[0172] The log storage uses an independent partition of the on-chip Flash of the main control unit, with a capacity of 64 kilobytes, which can store about 2,000 log records. When the number of log records exceeds the capacity, a circular overwrite strategy is adopted to automatically delete the oldest log record.

[0173] S67, based on the health monitoring page design, allows operators to enter the health monitoring page by clicking the health monitoring menu on the main page. The comprehensive health dashboard is displayed in the center of the page. The dashboard uses a semi-circular dial style with a scale range of 0 to 100. The pointer points to the current health value. The dial is divided into 3 color areas: 0 to 40 is the red area indicating poor condition, 40 to 70 is the yellow area indicating average condition, and 70 to 100 is the green area indicating good condition. The pointer color is consistent with the color of the area it is in.

[0174] The dashboard displays the predicted remaining healthy lifespan in days, in the format of "estimated remaining lifespan X days". The health score and remaining lifespan values ​​are derived from the health assessment results calculated and distributed by the S4 step cloud platform.

[0175] The bottom of the page displays five sub-health indicators, each presented as a progress bar. The length of the progress bar represents the health score, and the color of the progress bar changes according to the score. These five progress bars are for circuit breaker life health, handcart life health, grounding switch life health, handcart current health, and temperature health. The right side of each progress bar displays the specific value and unit. For example, the right side of the circuit breaker life health progress bar shows that 3526 operations have been performed and 6474 operations remain.

[0176] Pre-maintenance work order information is displayed at the bottom of the page. If there are currently unprocessed pre-maintenance work orders, information such as work order number, generation date, warning level, suggested maintenance time, and suggested maintenance items will be displayed. Operators can click on the work order to view the work order details. If there are currently no pre-maintenance work orders, the system will display that the equipment is in good condition and there are no maintenance work order information, thus providing a visual display of health monitoring.

[0177] S68, based on user login and permission management, enters the lock screen state after the device is started. The touch screen displays the login interface, requiring the user to enter a username and password. Two preset accounts are provided: administrator and operator. The administrator account username is admin and the initial password is admin123. The operator account username is operator and the initial password is oper123. After the first login, the system will force you to change the initial password.

[0178] After the operator enters their username and password and clicks the login button, the main control unit verifies whether the username and password match. If the verification is successful, the login user's identity is recorded, and the user is unlocked and enters the main page. If the verification fails, a message will be displayed indicating that the username or password is incorrect and requesting a retry.

[0179] Administrator users have full access permissions, including one-click power outage operation, single-step operation, parameter modification, and user management. Operator users only have one-click power outage operation permissions and cannot modify parameters or manage users. The currently logged-in user's name is displayed in the upper right corner of the main page, and a logout button is provided. Clicking the logout button exits the current user and returns to the login page.

[0180] The parameter settings page provides a user management subpage, where administrators can modify administrator passwords, modify operator passwords, and disable or enable operator accounts to ensure system access security. Combining steps S61 to S68, a dynamic primary system diagram (including real-time status graphics of busbars, circuit breakers, handcarts, and grounding switches), operation control commands (including one-button power-on commands, one-button power-off commands, and emergency stop commands), and parameter configuration data (including temperature thresholds, humidity thresholds, delay times, drive modes, communication addresses, etc.) are obtained.

[0181] The safety protection module is used to obtain protection signals by adopting safety protection strategies based on the motor current curve and displacement curve, operation completion signal and real-time equipment status vector.

[0182] Based on the high safety requirements of switchgear operation, a multi-layered safety protection system covering hardware interlocking, software protection, real-time monitoring, and emergency response is established to ensure that no accidents endangering equipment and personal safety occur under any circumstances.

[0183] The S71 features a hardware-based emergency stop protection system with a prominent red mushroom-shaped emergency stop button on the front panel. The button has a diameter of 40 mm and is mechanically locked in place when pressed. It can only be reset by rotating it. The emergency stop button uses a double-break normally closed contact with a contact capacity of 10 amps and 250 volts. The two sets of contacts are connected in series to improve reliability.

[0184] The first set of contacts of the emergency stop button is connected in series in the power supply circuit of the drive output module. The drive output module supplies power to all relays and motor drive circuits. When the emergency stop button is pressed, the first set of contacts immediately disconnects, cutting off the power supply to the drive output module. All relay coils are de-energized and released, and all motor drives stop outputting, achieving a forced stop at the hardware level. The disconnection time is less than 10 milliseconds.

[0185] This hardware interlock does not rely on the main control unit software and remains effective even if the main control unit crashes or the program malfunctions, providing hardware emergency stop protection. The second set of contacts of the emergency stop button is connected to the external interrupt input pin of the main control unit. The interrupt is triggered by a falling edge and the interrupt priority is set to the highest priority.

[0186] When the emergency stop button is pressed, the contact opens and a falling edge is generated. The main control unit immediately responds to the interrupt. The interrupt service routine performs operations such as clearing all pending operation instructions, stopping the current process, setting the system to emergency stop state, and displaying emergency stop alarm information on the touch screen. The emergency stop event is recorded to the log storage, thus obtaining the emergency stop response capability at the software level.

[0187] S72, based on motor current monitoring and protection, connects current sensors in series in the power supply circuits of the handcart drive motor and the grounding switch motor to monitor the motor operating current in real time. The main control unit samples the motor current value at a frequency of 100 Hz and compares the sampled current with the safety threshold. The rated current of the handcart drive motor is 3 amps, and the safety threshold is set to 1.5 times the rated current, i.e., 4.5 amps. The rated current of the grounding switch motor is 0.5 amps, and the safety threshold is set to 0.75 amps.

[0188] When the sampled current exceeds the safety threshold, the main control unit determines that the motor is overloaded, immediately cuts off the motor drive output, displays a motor overload protection alarm on the touch screen, records the overload event to the log, and analyzes the possible causes of the overload, such as mechanical jamming, excessive load, motor failure, etc. Maintenance personnel need to check and troubleshoot on-site before operation can be restored and the motor overcurrent protection is obtained.

[0189] The main control unit also monitors abnormal motor current waveforms. Under normal circumstances, the motor operating current should change smoothly. If abnormal waveforms such as violent current fluctuations, sudden drops to zero, or sudden jumps are detected, the main control unit determines that the motor is operating abnormally, immediately stops the drive and issues an alarm, thus achieving abnormal motor current detection and protection.

[0190] S73, based on operation timeout protection, the main control unit sets the maximum allowable execution time for each action. The timeout time for the trolley pushing operation is set to 40 seconds, the timeout time for the trolley reversing operation is set to 40 seconds, the timeout time for the grounding switch opening operation is set to 30 seconds, the timeout time for the grounding switch closing operation is set to 30 seconds, the timeout time for the circuit breaker opening operation is set to 5 seconds, and the timeout time for the circuit breaker closing operation is set to 5 seconds.

[0191] The main control unit starts a software timer when it issues an action command. The timer value is equal to the timeout period. If an action completion signal is received before the timer expires, the timer is stopped and subsequent steps continue. If no action completion signal is received before the timer expires, the main control unit determines that the operation has timed out and immediately stops the current action.

[0192] The touchscreen displays an operation timeout alarm, showing the specific name of the timed-out action and the elapsed time. The timeout event is logged, and the one-click operation process is terminated, thus providing operation timeout protection. Timeout protection prevents equipment damage caused by the system continuously attempting to complete actions due to mechanical failure, sensor failure, or drive failure.

[0193] S74, based on signal consistency verification protection, the main control unit performs consistency logic verification on the collected equipment status signals. The open and closed contacts of the circuit breaker should be mutually exclusive. Under normal circumstances, only one contact is closed. If the main control unit detects that both the open and closed contacts are closed or both are open at the same time, it determines that the signals are contradictory and refuses to perform any operation.

[0194] If the touchscreen displays an abnormal circuit breaker status signal, please check the auxiliary contacts and log the abnormal signal event. The working position limit switch and the test position limit switch of the handcart should also be mutually exclusive. If both limit switches are triggered or neither is triggered at the same time and the handcart is in the middle position, the main control unit determines that the handcart position signal is abnormal, refuses operation, and issues an alarm.

[0195] The grounding switch signal verification is similar to that of the circuit breaker, employing mutually exclusive logic verification to achieve status signal consistency verification and protection. Signal consistency verification prevents malfunctions caused by erroneous signals due to sensor failure, wiring errors, contact adhesion, etc.

[0196] S75 is based on the five-proof logic forced verification protection. Before performing any operation, the main control unit forcibly calls the five-proof logic verification function to verify whether the current state meets the preconditions for the operation. For example, before performing the circuit breaker closing operation, it verifies whether the grounding switch is in the open state. If the grounding switch is in the closed state, the circuit breaker closing operation is rejected.

[0197] If the touchscreen displays a message indicating that the five-proof interlocking grounding switch is not open and circuit breaker closing is prohibited, the interlocking event will be recorded in the log. The operator must first open the grounding switch before closing the circuit breaker. Before closing the grounding switch, verify that the circuit breaker is in the open state, the trolley is in the test position, and the busbar is de-energized. If any condition is not met, the operation will be refused and the specific reason will be displayed.

[0198] The five-proof logic verification adopts a whitelist mechanism, which allows only explicitly permitted state transitions to be executed, and rejects all unauthorized state transitions. This mechanism eliminates the possibility of erroneous operations and provides protection through mandatory five-proof logic verification.

[0199] S76, based on the immutable recording of operation logs, the main control unit records all operation records, alarm events, parameter modifications, user logins and other behaviors to the log storage. Each log record contains complete information such as a timestamp accurate to the second, event type, operating user, operation content, pre-operation status, post-operation status, and operation result.

[0200] Log entries are written sequentially, allowing only appending and preventing modification or deletion, thus ensuring log integrity and immutability. The main control unit calculates the CRC32 checksum for each log entry during writing and appends it to the end. When reading the log, the CRC32 checksum is verified; if verification fails, it indicates the log has been tampered with or corrupted, and an alarm is displayed on the touchscreen.

[0201] The log storage uses a separate partition of Flash memory, physically isolated from the program code partition, to prevent accidental erasure of log data during program upgrades, thus ensuring an immutable recording mechanism for operation logs. Operation logs provide reliable evidence for accident tracing, liability determination, and fault analysis.

[0202] S77, based on environmental parameter abnormality alarm, the main control unit continuously monitors the room temperature and humidity parameters of each function. When the temperature of a function room exceeds the upper temperature threshold or falls below the lower temperature threshold, a temperature abnormality alarm is displayed on the touch screen, showing the name of the abnormal function room and the current temperature value. At the same time, the buzzer is controlled to sound intermittently to remind the operator.

[0203] When the temperature exceeds the upper temperature threshold, the fan in the functional room will automatically start to dissipate heat. When the temperature falls below the lower temperature threshold, the heater in the functional room will automatically start to heat up. The temperature will be restored to the normal range through automatic regulation.

[0204] When the humidity in a certain functional room exceeds the upper limit threshold, it is determined that the humidity is too high and there is a risk of condensation. The heater is automatically started to dehumidify. The heater is stopped after the humidity returns to the normal range, thus obtaining an environmental parameter abnormality alarm and automatic control capability.

[0205] The main control unit records abnormal environmental parameter events to the log storage and uploads them to the cloud platform through the communication module. The cloud platform analyzes the long-term trend of environmental parameters and generates environmental improvement suggestions when it finds that the room temperature and humidity of a certain function frequently exceed the limits. It prompts the operation and maintenance personnel to check the ventilation system or install environmental control equipment, thus obtaining the ability to record and analyze abnormal environmental parameters.

[0206] The S78, based on critical point temperature over-temperature alarms, continuously monitors the temperature data of busbar contact arms and cable joints in the main control unit, and sets two levels of temperature thresholds for each temperature measurement point. The first level alarm threshold is set to 85 degrees Celsius, and the second level alarm threshold is set to 105 degrees Celsius.

[0207] When the temperature at a certain temperature measurement point exceeds the first-level alarm threshold, a temperature alarm will be displayed on the touch screen, showing the name of the over-temperature measurement point and the current temperature value. The alarm information will be uploaded to the cloud platform and pushed to the maintenance personnel, who should arrange for an inspection within 24 hours.

[0208] When the temperature at a certain temperature measuring point exceeds the level 2 alarm threshold, a severe over-temperature alarm is displayed on the touchscreen and highlighted in red flashing mode. The buzzer sounds continuously, and the emergency alarm information is uploaded to the cloud platform to notify maintenance personnel via SMS or telephone. Maintenance personnel should immediately go to the site for inspection and, if necessary, execute an emergency power outage to prevent equipment burnout or fire accidents. Combining steps S71 to S78, the following signals are obtained: emergency stop protection signal (including hardware emergency stop contact signal and software emergency stop interrupt signal), overcurrent protection signal (including motor overload indicator and current abnormality indicator), and five-prevention interlock signal (including five-prevention logic verification result and interlock reason prompt).

[0209] In one embodiment of the present invention, an application example of a centrally located switchgear intelligent one-button control device is provided:

[0210] The example focuses on the intelligent operation and maintenance application of a 10 kV ring network power distribution system in a large industrial park. The industrial park covers an area of ​​5 square kilometers and houses more than 60 companies in precision manufacturing, electronic information, and biomedicine. The power distribution system adopts a dual-power ring network power supply method. The power distribution station is located in the center of the park and is equipped with 13 medium-voltage switchgear, including 1 incoming switchgear, 2 bus tie switchgear, and 10 outgoing switchgear. The switchgear model is KYN28A-12, with a rated voltage of 10 kV and a rated current of 630 amps to 1250 amps. The circuit breakers are of the following brands: 4 ABB VD4 type, 3 VM1 type, and 3 3AH type. The power distribution system supplies power to 12 distribution areas in the park, with a total power supply capacity of 38 MVA and an annual power supply of 210 million kWh.

[0211] The power distribution station in the industrial park originally operated using a traditional manual mode, with two full-time electricians working in three shifts. Regular inspections and maintenance were required every month, and there were four planned power outages per year. Each power outage required the electrician to sequentially operate 13 switchgear cabinets, including opening circuit breakers, cranking out handcarts, and closing grounding switches. The average operation time for a single cabinet was 6 to 8 minutes, and the total power outage operation time for the entire station was about 90 minutes. The operation required two electricians to work together, which posed a high risk of personal safety and the possibility of misoperation. Between 2019 and 2022, there were two unplanned power outages caused by operational errors, resulting in economic losses for companies in the industrial park.

[0212] To improve the intelligence level and operational safety of the power distribution system, the park carried out intelligent transformation of 13 switchgear in December 2023. The intelligent one-button control device described in this invention was installed in each switchgear, and the sensor layout, drive mechanism installation, communication network construction and other projects were completed. The system was connected to the park's smart energy management cloud platform. The transformation project was completed and put into trial operation in January 2024.

[0213] Tables 1 and 2 show examples of operational data obtained after the upgrade and commissioning:

[0214] Table 1. Comparison of power outage operation times for the entire station

[0215]

[0216] Table 2, Predictive Maintenance Work Order Data

[0217]

[0218] By applying the device described in this invention, the operation and maintenance efficiency of the park's power distribution system has been improved. The total downtime has been reduced from 90 minutes to about 18 minutes, the number of operators has been reduced from 2 to 1, and operational safety has been greatly improved. No misoperation accidents have occurred in the year since the upgrade was put into operation. The predictive maintenance function predicted 5 potential equipment hazards 3 to 6 months in advance and dealt with them in a timely manner, avoiding possible unplanned power outages. The reliability of the power distribution system has increased from 99.92% to over 99.99%, and the average annual downtime has been reduced from 70 hours to 8 hours, saving the park's enterprises about 8 million yuan in economic losses. The device operates stably and reliably, providing an effective technical means for the intelligent operation and maintenance of the power distribution system.

[0219] The above description is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the protection scope of the present invention.

[0220] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. An intelligent one-key control device for a metal-clad switchgear, characterized in that, include: The data fusion module is used to collect switchgear operation data and obtain sensor data streams and real-time equipment status vectors after preprocessing. The parameter optimization module is used to collect motor current and displacement curves, and employs adaptive and transfer learning methods to obtain the device fingerprint vector, personalized drive parameters, and fingerprint drift warning signals; specifically, it includes: A current sensor is connected in series in the power supply circuit of the handcart drive motor, and a displacement sensor is installed on the handcart guide rail. The main control unit collects motor current and displacement data synchronously when performing handcart propulsion operation. When performing the handcart propulsion operation for the first time, the main control unit uses the default drive parameters to collect data and extracts multi-dimensional features from the current curve to construct the equipment fingerprint vector, including features of the start-up stage, acceleration stage, constant speed stage, and deceleration stage. The main control unit pre-stores a general driving parameter library. After extracting the fingerprint vector of a new device, it uses the K-nearest neighbor algorithm to search for the most similar sample and performs a weighted average of the driving parameters of the nearest samples. The weight is inversely proportional to the distance to obtain the initial driving parameters of the new device. When performing subsequent operations, the main control unit continues to collect data and extract fingerprint vectors, uses Bayesian optimization algorithm to fine-tune the driving parameters, selects the next set of parameters to be tried through the expected improvement function, and converges to the local optimum driving parameters after iteration; The main control unit stores the device fingerprint vector and driving parameters in the local fingerprint database and uploads them to the cloud platform periodically. The cloud platform builds a global device fingerprint knowledge base, uses clustering algorithms to identify clusters of similar devices, calculates the cluster center fingerprint and optimal parameters, and sends them to update the parameter database. After each operation, the main control unit extracts the current fingerprint vector and compares it with the historical fingerprint vector to calculate the Euclidean distance. When the distance is less than the first preset threshold, it is determined that the fingerprint has not drifted. When the distance is between the first preset threshold and the second preset threshold, the fingerprint vector and driving parameters are updated using the exponential weighted moving average method. When the distance is greater than the second preset threshold, it is determined that the fingerprint has changed and a warning signal is displayed. The operation execution module is used to acquire sensor data, device state vectors, and personalized drive parameters, and uses a state-action value network to obtain a one-click operation sequence and operation completion signal. The assessment and maintenance module is used to obtain the cumulative number of operations, motor current timing, temperature timing and environmental parameter timing, and to obtain the health assessment results by using multimodal fusion analysis. The collaborative scheduling module is used to obtain bus status synchronization signals and cluster operation scheduling instructions based on the device status vector and one-click operation sequence, using status synchronization and cloud scheduling methods. The visualization module is used to acquire all data generated during the operation process. It adopts a graphical layout to obtain a dynamic primary system diagram, operation control commands and parameter configuration data. The safety protection module is used to obtain protection signals by adopting safety protection strategies based on the motor current curve and displacement curve, operation completion signal and real-time equipment status vector.

2. The intelligent one-key control device for a metal-clad switchgear according to claim 1, characterized in that, The data fusion module includes: High-voltage energizing sensors are installed on each of the three phases in the busbar compartment. The main control unit performs a logical AND operation on the three-phase energizing signals and determines that the busbar is energized only when all three phases are energized simultaneously. Limit switches for working and testing positions are installed on the handcart guide rail. Auxiliary contacts are led out from the circuit breaker and grounding switch operating mechanisms. Magnetic switches are installed at the cabinet door position. The contact signals are connected to the main control unit. The main control unit scans each contact signal and filters out contact jitter to obtain a stable set of equipment status signals. Temperature and humidity sensors are installed in each functional room, and the main control unit polls each sensor to obtain environmental parameter data; wireless temperature sensors are installed at the busbar contact arms and cable joints, and the temperature data is forwarded to the main control unit through the receiving module.

3. The intelligent one-key control device for a metal-clad switchgear according to claim 1, characterized in that, The operation execution module includes: The switchgear operation process is modeled as a Markov decision process. The state space is defined as including cabinet door state, circuit breaker state, truck state, grounding switch state, and bus energized state. The action space is defined as including circuit breaker opening, circuit breaker closing, truck forward, truck backward, grounding switch opening, grounding switch closing, and waiting. The reward function is defined as consisting of three parts: goal achievement reward, operation efficiency reward, and safety constraint penalty. A deep Q-network model is established in the main control unit. The input layer receives the one-hot encoding of the state vector, the hidden layer uses the activation function, and the output layer outputs the Q value for each action. The Q value represents the expected long-term cumulative reward for performing the corresponding action in the current state. Before the device leaves the factory, the deep Q network is trained offline in a simulation environment. An experience pool is established using an experience replay mechanism to store a quadruple of state, action, reward, and next state. During the training process, the agent interacts with the simulation environment, selects actions according to a greedy policy, randomly samples and updates the deep Q network parameters from the experience pool, and stores the trained network parameters in the main control unit. When the operator triggers the one-click power-on command, the main control unit reads the device state vector, encodes the state vector and inputs it into a deep Q network. The network calculates and outputs the Q value of each action. The main control unit selects the action with the largest Q value that satisfies the five-prevention constraints to execute. After execution, the new device state vector is obtained and input into the network again to calculate the next action. The above process is repeated until the power-on target state is reached.

4. The intelligent one-button control device for a centrally located switchgear according to claim 1, characterized in that, The assessment and maintenance module includes: The main control unit establishes operation count counters for circuit breakers, handcarts, and grounding switches respectively. When an operation is completed, the corresponding counter is incremented by one. The main control unit presets the rated operating life threshold for each operating object and calculates the life consumption ratio to obtain the life assessment index. The main control unit collects the motor current curve and extracts the peak starting current parameter during each handcart propulsion operation. It stores the peak starting current and timestamp in the historical database and periodically uploads the current time series data to the cloud platform. The cloud platform establishes a time series convolutional network model to predict the trend. When the slope of the predicted curve is greater than the preset slope threshold, an early warning message is generated. The cloud platform establishes a temperature anomaly detection model based on the Transformer architecture, which uses a self-attention mechanism to capture abnormal patterns in the temperature sequence. The model outputs an anomaly score at each time step, and triggers an anomaly alarm when the anomaly score exceeds a preset anomaly threshold.

5. The intelligent one-button control device for a centrally located switchgear according to claim 4, characterized in that, The assessment and maintenance module also includes: The cloud platform establishes a multimodal fusion neural network. It adopts a multi-branch architecture to process the number of operations, motor current timing, temperature timing, and environmental parameter timing by independent feature extraction branches. The feature vectors of each modality are concatenated to form a fusion feature vector. The fusion layer uses a multi-head attention mechanism to calculate the attention weights between features of different modalities, and finally outputs a health score. The cloud platform uses a survival analysis model to estimate the remaining useful life of equipment. The survival analysis model is based on the Weibull distribution to model the equipment failure time distribution. The Cox proportional hazards regression method is used to calculate the risk ratio of the current equipment. The remaining useful life probability distribution of the equipment is calculated by combining the Weibull distribution and the risk ratio. The median is extracted as the predicted value of the remaining useful life.

6. The intelligent one-button control device for a centrally located switchgear according to claim 1, characterized in that, The collaborative scheduling module includes: A bus is used to build a communication network between devices. The communication modules of each device are connected in parallel to the bus. Each device is assigned a unique bus address. The incoming line cabinet device acts as the master station, and the outgoing line cabinet device acts as the slave station. Define the data frame format for communication between devices. The data frame structure includes frame header, address code, function code, data length, data content, check code, and frame tail. When the incoming line cabinet circuit breaker is closed, the main control unit detects the change in the energized state of the bus and sends a status broadcast frame through the bus. All outgoing line cabinet devices receive the broadcast frame, parse the data content, and update the local bus energized state variables. The cloud platform sends operation instructions to each device in the order of the device list. After receiving the instructions, the device executes the operation process and sends an operation completion message to the platform after the process is completed. After receiving the completion message, the platform sends instructions to the next device.

7. The intelligent one-button control device for a centrally located switchgear according to claim 1, characterized in that, The visualization module includes: The main page adopts a layout that simulates a primary system diagram, drawing graphic symbols for incoming busbars, circuit breakers, handcart contacts, grounding switches, and outgoing circuits, and dynamically coloring them according to the equipment status vectors; A one-button power-on button and a one-button power-off button are set up in the one-button operation control area on the side of the system diagram.

8. The intelligent one-button control device for a centrally located switchgear according to claim 1, characterized in that, The security protection module includes: An emergency stop button is installed on the front panel of the device. The contacts of the emergency stop button are connected in series in the power supply circuit of the drive output module. When pressed, the power supply to the drive output module is cut off. A current sensor is connected in series in the motor power supply circuit to monitor the motor operating current. When the sampled current exceeds the safety threshold, the motor is determined to be overloaded and the motor drive output is cut off. The main control unit sets a maximum allowed execution time for each action. If the timer expires and no action completion signal is received, the operation is deemed to have timed out and the current action is stopped. Before executing an operation, the main control unit calls the five-prevention logic verification function to verify whether the current state meets the preconditions for the operation.

Images