Drop-out fuse intelligent control system
By combining signal receiving, intelligent decision-making, and safety verification modules, along with reinforcement learning and digital twin simulation models, the problems of intelligence and coordination in traditional drop-out fuse control systems have been solved. This enables intelligent decision-making and rapid fault isolation in the power grid, improving the safety and self-healing capabilities of equipment and the power grid.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-26
- Publication Date
- 2026-04-07
AI Technical Summary
Traditional drop-out fuse control systems lack intelligence and coordination, are unable to make safety judgments and coordinate operations based on the real-time status of the power grid, pose a risk of misoperation, and have poor self-healing capabilities.
It employs a combination of signal receiving module, intelligent decision-making module, security verification module and execution module, combined with reinforcement learning model and digital twin simulation model, to achieve intelligent decision-making and security verification, and supports the collaborative operation of multiple fuses.
It enables intelligent decision-making based on the real-time status of the power grid, avoids misoperation, ensures the safety of equipment and the power grid, quickly locates and isolates faults, shortens power outage time, and improves self-healing capabilities.
Smart Images
Figure CN121813700A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of intelligent protection and control technology for power distribution networks, and relates to an intelligent control system for drop-out fuses. Background Technology
[0002] Drop-out fuses are widely used in power distribution networks for the protection and control of lines and transformers. Traditional operation relies on maintenance personnel using insulated hooks for local switching, which is not only inefficient but also poses significant safety risks during inclement weather or at night. While some electric operating mechanisms have emerged, enabling remote switching, these solutions still suffer from the following prominent problems: 1. Simple control logic and lack of intelligence: Most existing electric operators are merely remote controls replacing manual operation, simply receiving remote commands and executing mechanical actions. The control system cannot make safety judgments based on the real-time state of the power grid, such as the presence of load or voltage stability; blindly closing the circuit on a faulty line could lead to a major accident. 2. Lack of pre-operation safety verification: Before performing critical operations, especially power restoration closing operations, there is no effective mechanism to predict the consequences of the operation. The system cannot identify potential risks such as whether there are permanent faults in the line or whether the inrush current exceeds the standard. It relies entirely on the experience of maintenance personnel, resulting in poor reliability. 3. Isolated operation and lack of coordination: The control unit of each fuse is isolated. When a line fault occurs, it cannot communicate and coordinate with other intelligent devices on the same line. It cannot achieve rapid automatic isolation of the faulty section and restoration of power supply to non-faulty sections, resulting in poor self-healing ability.
[0003] Therefore, there is an urgent need in this field for a drop-out fuse control system that can achieve intelligent decision-making, safety verification, and support collaborative operation. Summary of the Invention
[0004] This invention proposes an intelligent control system for drop-out fuses. This system improves the accuracy and adaptability of control, enabling multiple fuses to work together when a fault occurs, quickly locate and isolate the fault, automatically restore power supply to non-faulty areas, and shorten power outage time.
[0005] The technical solution of this invention is implemented as follows:
[0006] A smart control system for drop-out fuses includes:
[0007] The signal receiving module is used to receive remote control commands and / or real-time power grid status data monitored by local sensors;
[0008] The intelligent decision-making module, connected to the signal receiving module, is used to generate a tripping or closing control strategy based on the control command and real-time power grid status data through a built-in intelligent decision-making model.
[0009] A safety verification module, connected to the intelligent decision-making module, is used to simulate the operational consequences based on the control strategy using a built-in digital twin simulation model and output a safety verification signal.
[0010] The execution module, connected to the safety verification module, is used to drive the electric operating mechanism to perform opening or closing actions after receiving a verification signal indicating safety.
[0011] Preferably, the intelligent decision-making model is a reinforcement learning model, and the establishment of this model includes a model initialization step:
[0012] Define a state space S, whose elements include: line voltage value, load current value, and control terminal battery power.
[0013] Define action space A, whose elements include: immediate closing, delayed closing, immediate opening, and holding still;
[0014] Define a reward function R, which is calculated based on a weighted sum of operation success indicators, grid stability indicators, and system energy consumption indicators;
[0015] Initialize a deep Q-network whose network structure includes an input layer, at least one fully connected hidden layer, and an output layer.
[0016] Preferably, the reward function R is calculated as follows: R = W1 × Operation success reward + W2 × Voltage stability reward - W3 × Energy consumption penalty, where W1, W2, and W3 are preset weight coefficients.
[0017] Preferably, the training of the reinforcement learning model includes the following steps:
[0018] Data preparation: Collect experience data (s,a,r,s') and store it in the experience replay pool;
[0019] Model update: Sample small batches of data from the experience replay pool, and iteratively update the weight parameters of the Q network using gradient descent by minimizing the mean square error between the predicted value and the target Q value of the Q network.
[0020] Model deployment: When the average cumulative reward of the model in the test environment reaches a predetermined threshold after N consecutive rounds, training is stopped and the final model parameters are deployed in the intelligent decision-making module.
[0021] Preferably, the establishment of the digital twin simulation model includes a model initialization step:
[0022] Construct an equivalent circuit model that includes a drop-out fuse, line impedance, and load;
[0023] The equivalent circuit model is discretized and encoded into a simulation program that can be executed by a processor.
[0024] Preferably, the operation of the digital twin simulation model includes the following steps:
[0025] Input settings: The current power grid status data and the closing command to be executed are used as inputs to the simulation program;
[0026] Simulation execution: Run the simulation program to calculate the line current waveform within a specific time window after the closing operation, and extract the peak value of the closing inrush current from it;
[0027] Safety judgment: The peak value of the closing inrush current is compared with a preset safety threshold. If the peak value exceeds the threshold, a verification failure signal is output; otherwise, a verification success signal is output to the execution module.
[0028] Preferably, the system further includes a line coordination module for communicating with other intelligent control systems on the same power distribution line and executing the following coordinated control process:
[0029] Fault information sharing: When any control system detects a fault, it broadcasts the fault information through the communication network;
[0030] Fault segment location: Based on the received fault current direction information, each control system jointly determines the segment where the fault occurred through a distributed voting consensus algorithm;
[0031] Cooperative operation sequence generation: The control systems located on both sides of the faulty section automatically generate tripping commands to isolate the fault, while the control system upstream of the non-faulty section generates closing commands to restore power supply.
[0032] Preferably, the system further includes a dynamic energy management module for executing the following energy-saving management procedures:
[0033] Status monitoring: Real-time monitoring of the power supply voltage and current task queue of the control system;
[0034] Power mode switching: When there are no high-priority tasks, the main processor is switched to a low-power sleep mode and periodically woken up to monitor the communication module;
[0035] Dynamic voltage and frequency adjustment: During task execution, the processor's operating voltage and frequency are dynamically adjusted according to the real-time computing load of the intelligent decision-making module to match computing power with power consumption.
[0036] The beneficial effects of the present invention using the above technical solution are as follows:
[0037] The intelligent control system for drop-out fuses provided by this invention can perform intelligent analysis and decision-making based on real-time power grid status and control commands, and automatically generate or adjust opening and closing strategies. Before executing operations, it can perform simulation pre-runs and safety verifications to avoid misoperation and ensure the safety of equipment and the power grid. It can dynamically adjust its own power consumption according to task priority and energy status to meet the requirements of long-term stable operation in the field. It supports multi-agent collaboration, enabling multiple fuses on the same line to communicate with each other and work together to complete fault location, isolation and power restoration tasks. Attached Figure Description
[0038] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.
[0039] Figure 1 This is a schematic diagram of the intelligent control system for the drop-out fuse of the present invention. Detailed Implementation
[0040] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.
[0041] A smart control system for drop-out fuses includes:
[0042] The signal receiving module is used to receive remote control commands and / or real-time power grid status data monitored by local sensors;
[0043] The intelligent decision-making module, connected to the signal receiving module, is used to generate a tripping or closing control strategy based on the control command and real-time power grid status data through a built-in intelligent decision-making model.
[0044] A safety verification module, connected to the intelligent decision-making module, is used to simulate the operational consequences based on the control strategy using a built-in digital twin simulation model and output a safety verification signal.
[0045] The execution module, connected to the safety verification module, is used to drive the electric operating mechanism to perform opening or closing actions after receiving a verification signal indicating safety.
[0046] The intelligent decision-making model is a reinforcement learning model, and its establishment includes a model initialization step:
[0047] Define a state space S, whose elements include: line voltage value, load current value, and control terminal battery power.
[0048] Define action space A, whose elements include: immediate closing, delayed closing, immediate opening, and holding still;
[0049] Define the reward function R, which is calculated as follows: R = W1 × Operation success reward + W2 × Voltage stability reward - W3 × Energy consumption penalty, where W1, W2, and W3 are preset weight coefficients;
[0050] In practice, the calculation methods and weighting coefficients for the above items are set as follows:
[0051] How to calculate the reward for successful operation:
[0052] Successful closing reward: When the closing command is issued, if the system detects that the line voltage recovers to more than 85% of the rated voltage and remains stable within a preset time (e.g., 200ms), a reward of +100 will be given.
[0053] Successful tripping reward: When the tripping command is issued and the system detects that the fault current has been effectively cut off, a reward of +80 is given.
[0054] Dangerous operation penalty: When an "immediate closing" or "delayed closing" command is issued under the following circumstances, a severe penalty of -200 will be imposed: there is a continuous short-circuit fault current in the line; the safety verification module predicts that the closing inrush current exceeds the standard.
[0055] Weighting coefficient W 1： Set to 1.0. This means that the above reward or penalty values will be included in the total reward R at a 1:1 ratio.
[0056] Voltage stability bonus calculation method: At the end of each decision step (e.g., 1 second), calculate the absolute deviation between the current line voltage per unit value (based on rated voltage) and the ideal value of 1.0.
[0057] Voltage stability index = 1 - |current voltage per unit value - 1.0|. The closer this index is to 1, the more stable the voltage.
[0058] Weighting coefficient W 2： The value is set to 50. This means that the total reward R will be increased by 50 times the voltage stability index.
[0059] Energy consumption penalty calculation method:
[0060] Energy consumption per operation: The equivalent energy consumption for performing one opening or closing operation is defined as 1.0 unit.
[0061] Energy consumption index = If a tripping or closing action is performed within the current decision cycle, the value is 1.0; if it remains unchanged, the value is 0.
[0062] Weighting coefficient W 3：The value is set to 10. This means that the total reward R will be reduced by 10 times the energy consumption index, thereby encouraging the system to remain stationary when not necessary, thus avoiding unnecessary mechanical wear and tear and power consumption.
[0063] The specific calculation process of the reward function R within a decision-making cycle is as follows:
[0064] Initialize the total reward R = 0.
[0065] Based on the actions and results performed, the cumulative reward for successful operation is: R = R + (successful closing reward / successful opening reward / dangerous operation penalty) × 1.0.
[0066] Calculate the current voltage stability index and sum it up: R = R + 50 × (1 - |current voltage per unit value - 1.0|).
[0067] Determine whether an operation was performed. If so, subtract the energy consumption penalty: R = R - 10 × 1.0.
[0068] Through the above design, the training process will be guided to an optimal strategy: first, ensure the safety and success rate of the operation; second, strive to maintain the stability of the grid voltage; and third, minimize unnecessary operations to save energy while satisfying the first two requirements.
[0069] Initialize a deep Q-network whose network structure includes an input layer, at least one fully connected hidden layer, and an output layer.
[0070] Specifically, the intelligent decision-making module employs a deep reinforcement learning model, whose network structure includes the following layers:
[0071] Input layer: 3 nodes, which respectively receive the per-unit value of line voltage, the per-unit value of load current, and the remaining battery power percentage;
[0072] First hidden layer: 128 nodes, using the ReLU activation function, fully connected to the input layer;
[0073] Second hidden layer: 64 nodes, using the ReLU activation function, fully connected to the first hidden layer;
[0074] Output layer: 4 nodes, corresponding to the Q values of the four actions: immediate closing, delayed closing, immediate opening, and holding still.
[0075] The state space includes: line voltage value, load current value, and control terminal battery level;
[0076] The action space includes: immediate closing, delayed closing, immediate tripping, and holding still;
[0077] The model is trained according to the following specific steps:
[0078] Training environment construction: Establish a power distribution network simulation environment that includes 10 typical operating conditions and 5 fault types.
[0079] Training parameter settings: learning rate: 0.001, discount factor: 0.95, experience replay pool capacity: 10,000 records, target network update frequency: once every 100 steps, exploration rate ε: initially 1.0, decays by 0.01 every 1,000 steps, minimum 0.1.
[0080] Training data preparation: Conduct 5000 training cycles in a simulation environment, with a maximum of 100 steps per cycle, and collect experience data to store in the experience replay pool.
[0081] Model iterative training: Randomly select 32 data points from the experience replay pool as a mini-batch; calculate the mean square error between the current Q-network's predicted value and the target Q-value; iteratively update the network parameters using gradient descent; copy the main network parameters to the target network every 100 steps;
[0082] Training termination condition: Training stops when the model's average cumulative reward reaches 800 points over 100 consecutive training cycles in the test environment, and the final model parameters are deployed to the intelligent decision-making module.
[0083] The establishment of the digital twin simulation model includes the model initialization step:
[0084] Construct an equivalent circuit model that includes a drop-out fuse, line impedance, and load;
[0085] The equivalent circuit model is discretized and encoded into a simulation program that can be executed by a processor.
[0086] Specifically, establish an equivalent circuit model that includes the following elements:
[0087] Power supply section: 10.5kV ideal voltage source, internal resistance 0.1Ω;
[0088] Circuit section: Circuit resistance 0.5Ω, reactance 1.2Ω;
[0089] Load section: parallel load, resistor 100Ω, inductor 0.1H, capacitor 1μF;
[0090] Fuse section: resistance 0.01Ω when closed, resistance 1MΩ when open;
[0091] The equivalent circuit model is discretized and encoded into a simulation program that can be executed by a processor.
[0092] The operation of the digital twin simulation model includes the following steps:
[0093] Input settings: The current power grid status data and the closing command to be executed are used as inputs to the simulation program;
[0094] Simulation execution: Run the simulation program to calculate the line current waveform within a specific time window after the closing operation, and extract the peak value of the closing inrush current from it;
[0095] Safety judgment: The peak value of the closing inrush current is compared with a preset safety threshold. If the peak value exceeds the threshold, a verification failure signal is output; otherwise, a verification success signal is output to the execution module.
[0096] Specifically, the implementation steps for security verification are as follows:
[0097] Input settings: The current power grid status data and the closing command to be executed are used as inputs to the simulation program;
[0098] Initialize the simulation environment: Set the current power grid operating state as the initial conditions for the simulation;
[0099] Set the simulation parameters: simulation duration 100ms, step size 10μs, sampling frequency 100kHz;
[0100] Injection operation command: Execute the closing operation 5ms during the simulation time;
[0101] Simulation execution: Run the simulation program, use the fourth-order Runge-Kutta method to solve the differential equation, and calculate the line current waveform within a specific time window after the closing operation;
[0102] Extracting key features: Extracting the peak inrush current at closing from the simulation results;
[0103] Safety Judgment: The peak inrush current is compared with a safety threshold of 6 times the rated current. If the peak exceeds the threshold, a verification failure signal is output; otherwise, a verification success signal is output to the execution module.
[0104] The hardware of this system includes an electric operating mechanism installed on the fuse body, three voltage sensors, three current sensors, an edge computing terminal, and a cloud management platform.
[0105] The system also includes a line coordination module for communicating with other intelligent control systems on the same power distribution line. The specific communication network architecture is as follows: a distributed communication system based on a 4G private network is established; each intelligent fuse is configured with an independent IP address; communication cycle: 100ms; data packet format: includes device ID, timestamp, fault type, and current direction; timeout retransmission mechanism: automatic retransmission after 200ms of no response, up to 3 times.
[0106] The specific implementation process of collaborative control includes the following steps:
[0107] Fault information sharing: When any control system detects a fault, it broadcasts the fault information through the communication network;
[0108] Fault segment location: Based on the received fault current direction information, each control system jointly determines the segment where the fault occurred through a distributed voting consensus algorithm;
[0109] Cooperative operation sequence generation: The control systems located on both sides of the faulty section automatically generate tripping commands to isolate the fault, while the control system upstream of the non-faulty section generates closing commands to restore power supply.
[0110] Operation execution confirmation: Each node executes sequentially with an interval of 100ms to ensure the correct implementation of the strategy.
[0111] The system also includes a dynamic energy management module for executing the following energy-saving management procedures:
[0112] Status monitoring: Real-time monitoring of the power supply voltage and current task queue of the control system;
[0113] Power mode switching: When there are no high-priority tasks, the main processor is switched to a low-power sleep mode and periodically woken up to monitor the communication module;
[0114] Dynamic voltage and frequency adjustment: During task execution, the processor's operating voltage and frequency are dynamically adjusted according to the real-time computing load of the intelligent decision-making module to match computing power with power consumption.
[0115] Among them, status monitoring: real-time monitoring of the following power parameters: lithium battery remaining power percentage: sampled every 10 minutes, with an accuracy of 1%; system real-time power consumption: range 0-500mA, with an accuracy of 1mA; current task queue: monitoring the number of tasks to be processed.
[0116] Among them, power consumption mode switching:
[0117] Establish a three-level power management mode: Full power mode: power consumption 500mA, processor frequency 800MHz; trigger condition: execute high-priority tasks such as circuit breaker operation and model inference; hold time: maximum 30 seconds;
[0118] Monitoring mode: power consumption 50mA, processor frequency 200MHz; trigger condition: normal standby state; wake-up cycle: check communication every 5 seconds;
[0119] Sleep mode: power consumption 5mA, processor frequency 50MHz; trigger conditions: no high-priority tasks, battery level below 20% or no tasks for more than 5 minutes; wake-up method: timed wake-up or external interrupt.
[0120] Among them, dynamic voltage and frequency regulation: dynamically adjusts system performance according to task load.
[0121] Lightly loaded tasks (queue length < 5): frequency 400MHz, voltage reduced accordingly;
[0122] Medium-duration tasks (queue length 5-10): 600MHz frequency, moderate voltage;
[0123] Heavy-load tasks (queue length > 10): frequency 800MHz, full voltage.
[0124] Frequency switching delay: less than 10ms.
[0125] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A smart control system for drop-out fuses, characterized in that, include: The signal receiving module is used to receive remote control commands and / or real-time power grid status data monitored by local sensors; The intelligent decision-making module, connected to the signal receiving module, is used to generate a tripping or closing control strategy based on the control command and real-time power grid status data through a built-in intelligent decision-making model. A safety verification module, connected to the intelligent decision-making module, is used to simulate the operational consequences based on the control strategy using a built-in digital twin simulation model and output a safety verification signal. The execution module, connected to the safety verification module, is used to drive the electric operating mechanism to perform opening or closing actions after receiving a verification signal indicating safety.
2. The intelligent control system for a drop-out fuse according to claim 1, characterized in that, The intelligent decision-making model is a reinforcement learning model, and its establishment includes model initialization steps: Define a state space S, whose elements include: line voltage value, load current value, and control terminal battery power. Define action space A, whose elements include: immediate closing, delayed closing, immediate opening, and holding still; Define a reward function R, which is calculated based on a weighted sum of operation success indicators, grid stability indicators, and system energy consumption indicators; Initialize a deep Q-network whose network structure includes an input layer, at least one fully connected hidden layer, and an output layer.
3. The intelligent control system for a drop-out fuse according to claim 1, characterized in that, The training of the reinforcement learning model includes the following steps: Data preparation: Collect experience data and store it in the experience replay pool; Model update: Sample small batches of data from the experience replay pool, and iteratively update the weight parameters of the Q network using gradient descent by minimizing the mean square error between the predicted value and the target Q value of the Q network. Model deployment: When the average cumulative reward of the model in the test environment reaches a predetermined threshold after N consecutive rounds, training is stopped and the final model parameters are deployed in the intelligent decision-making module.
4. The intelligent control system for a drop-out fuse according to claim 1, characterized in that, The establishment of the digital twin simulation model includes the model initialization step: Construct an equivalent circuit model that includes a drop-out fuse, line impedance, and load; The equivalent circuit model is discretized and encoded into a simulation program that can be executed by a processor.
5. The intelligent control system for a drop-out fuse according to claim 1, characterized in that, The operation of the digital twin simulation model includes the following steps: Input settings: The current power grid status data and the closing command to be executed are used as inputs to the simulation program; Simulation execution: Run the simulation program to calculate the line current waveform within a specific time window after the closing operation, and extract the peak value of the closing inrush current from it; Safety judgment: The peak value of the closing inrush current is compared with a preset safety threshold. If the peak value exceeds the threshold, a verification failure signal is output; otherwise, a verification success signal is output to the execution module.
6. The intelligent control system for a drop-out fuse according to claim 1, characterized in that, The system also includes a line coordination module for communicating with other intelligent control systems on the same power distribution line and executing the following coordinated control process: Fault information sharing: When any control system detects a fault, it broadcasts the fault information through the communication network; Fault segment location: Based on the received fault current direction information, each control system jointly determines the segment where the fault occurred through a distributed voting consensus algorithm; Cooperative operation sequence generation: The control systems located on both sides of the faulty section automatically generate tripping commands to isolate the fault, while the control system upstream of the non-faulty section generates closing commands to restore power supply.
7. The intelligent control system for a drop-out fuse according to claim 1, characterized in that, The system also includes a dynamic energy management module for executing the following energy-saving management procedures: Status monitoring: Real-time monitoring of the power supply voltage and current task queue of the control system; Power mode switching: When there are no high-priority tasks, the main processor is switched to a low-power sleep mode and periodically woken up to monitor the communication module; Dynamic voltage and frequency adjustment: During task execution, the processor's operating voltage and frequency are dynamically adjusted according to the real-time computing load of the intelligent decision module to match computing power with power consumption.