A method for preventing and cutting off electric leakage based on current detection algorithm

Abstract

The application discloses a leakage prevention and cutting-off method based on a current detection algorithm, which comprises the following steps: when it is detected that the residual current in a line exceeds a preset trigger threshold, real-time fault waveform data of the residual current is acquired, and in combination with pre-stored line historical operation data, a state vector containing multiple data components is generated; the state vector is input into a pre-trained decision model, the decision model selects a to-be-executed action from a preset action set according to a preset decision rule; and the to-be-executed action is executed to implement cutting-off, reclosing and isolation operation on a specific branch of the line. Through the construction of the state vector and the use of the pre-trained decision model, the application can accurately analyze the fault nature, thereby realizing differentiated and intelligent protection actions, ensuring safety, improving power supply reliability, and further having the functions of fault branch diagnosis and line state active early warning, thereby improving operation and maintenance efficiency and power utilization safety level.

Description

Technical Field

[0001] This invention relates to the field of electrical safety technology, and in particular to a leakage current prevention and disconnection method based on a current detection algorithm. Background Technology

[0002] As a crucial component in ensuring electrical safety, the development of residual current circuit breaker (RCD / RCCB) technology has primarily focused on improving detection accuracy and operational reliability. Early RCDs / RCCBs mainly employed electromagnetic trip units, relying on zero-sequence current transformers to detect residual current. Once the current exceeded a set value, the circuit breaker would trip instantaneously, providing basic protection against electric shock and grounding faults. With the advancement of electronic technology, electronic RCDs became more widespread. Their built-in microprocessors could perform preliminary analysis of current signals, distinguishing between power frequency leakage current and high-frequency interference to some extent, thus reducing the false trip rate. Building upon this foundation, to further improve power supply continuity, automatic reclosing technology was introduced into the RCD field. This technology aims to address transient grounding faults caused by non-continuous factors such as lightning strikes and transient overvoltages. By automatically attempting reclosing after tripping, it avoids unnecessary prolonged power outages.

[0003] However, these technological improvements remain confined to pre-defined, fixed logical frameworks. Specifically, their decision-making logic exhibits significant "inadaptability" and "blindness." Whether it's a traditional residual current device (RCD) or a product integrating automatic reclosing functionality, its tripping threshold and reclosing strategy (such as the number of reclosing attempts and time intervals) are usually fixed. They cannot comprehensively assess the real-time health status of the line, the transient characteristics of the fault waveform, and historical operating data, thus making it difficult to effectively distinguish between dangerous permanent insulation damage and self-healing transient interference. This may lead to incorrect reclosing in dangerous situations (such as electric shock) or unnecessary power outages in safe situations (such as the starting of a large motor). Furthermore, existing technologies generally lack "diagnostic and predictive capabilities." When the main circuit protector trips, it often causes a shutdown of the entire power supply area, and the system itself cannot provide any information about which downstream branch the fault source is located on, heavily relying on manual troubleshooting, resulting in low maintenance efficiency. More importantly, the existing protection mechanism is entirely passive and can only be dealt with after a leakage event occurs. It cannot detect or predict progressive safety hazards caused by factors such as slow aging of insulation materials and continuous moisture, thus missing the best opportunity to prevent accidents. Summary of the Invention

[0004] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0005] In view of the aforementioned existing problems, this invention is proposed. Therefore, this invention provides a leakage current prevention and interruption method based on a current detection algorithm to solve the problems mentioned in the background art.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a leakage current prevention and disconnection method based on a current detection algorithm, comprising: When the residual current in the line is detected to exceed the preset trigger threshold, the real-time fault waveform data of the residual current is acquired, and combined with the pre-stored historical line operation data, a state vector containing multiple data components is generated. The state vector is input into a pre-trained decision model, which selects an action to be executed from a preset action set according to preset decision rules. The actions to be performed include disconnecting, reclosing, and isolating specific branches of the line.

[0007] As a preferred embodiment of the leakage current prevention and interruption method based on current detection algorithm described in this invention, the step of generating the state vector includes: Calculate the energy distribution value of the real-time fault waveform data on multiple preset frequency bands, and calculate the morphological parameters of the waveform data. Use the energy distribution value and the morphological parameters as data components of the state vector.

[0008] As a preferred embodiment of the leakage current prevention and disconnection method based on current detection algorithm described in this invention, the step of generating the state vector further includes: Based on the background high-frequency residual current value recorded in the historical operation data of the line, a health index reflecting the current state of the line is calculated, and this health index is used as another data component of the state vector.

[0009] As a preferred embodiment of the leakage current prevention and disconnection method based on current detection algorithm described in this invention, the step further includes: Historical health index time series data are processed using a time series processing algorithm to calculate predicted health index values ​​for one or more future time points, and these predicted values ​​are used as additional data components of the state vector.

[0010] As a preferred embodiment of the leakage current prevention and disconnection method based on current detection algorithm described in this invention, the preset action set includes at least: Action to lock the main switch in the off state; The action of closing the main switch after a preset delay; And, the action of disconnecting the designated downstream branch and then closing the main switch.

[0011] As a preferred embodiment of the leakage current prevention and disconnection method based on the current detection algorithm described in this invention, the pre-trained decision model is a reinforcement learning model, and the preset decision rules of the model are: Calculate an expected long-term return value for each action in the preset action set, and select the action with the highest expected long-term return value as the action to be executed.

[0012] As a preferred embodiment of the leakage current prevention and disconnection method based on the current detection algorithm described in this invention, the reinforcement learning model is obtained through offline training and includes: During training, the rewards used to guide the model's learning are set as a weighted sum of values ​​associated with the security objective, the reliability objective, and the line state maintenance objective, respectively. Among them, the numerical weight associated with security objectives is set to the highest.

[0013] As a preferred embodiment of the leakage current prevention and disconnection method based on current detection algorithm described in this invention, when the action to be performed is to disconnect a specified downstream branch and then close the main switch, the action to be performed includes disconnection, reclosing, and isolation operations for a specific branch on the line, including: A target branch is determined based on the output of the pre-trained decision model, and a disconnect command is sent to the switch controlling the target branch. After confirming that the target branch is disconnected, a closing command is sent to the main switch.

[0014] As a preferred embodiment of the leakage current prevention and disconnection method based on the current detection algorithm described in this invention, the method further includes, after sending a closing command to the main switch: Start a verification timer of a preset duration; If, during the verification timing, the residual current exceeds the trigger threshold again, the main switch is immediately disconnected, and the process returns to the step of generating the state vector. If the remaining current does not exceed the trigger threshold again when the verification time ends, an alarm notification containing the target branch information is sent to the peripheral device.

[0015] As a preferred embodiment of the leakage current prevention and disconnection method based on current detection algorithm described in this invention, the step further includes: During normal operation, when no residual current exceeding the trigger threshold is detected, the health index is continuously calculated and updated; When the health index is lower than the preset maintenance threshold, a line status warning message is generated and sent.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By constructing a multi-dimensional state vector that includes real-time fault waveform features, historical line health index, and future trend prediction, and inputting it into a pre-trained reinforcement learning model for decision-making, the model can deeply understand the physical meaning behind different states. After optimization training with safety as the highest priority, it can accurately distinguish between high-risk permanent faults, self-healing transient faults, and non-fault interference. It selects the optimal solution from the action set of locking disconnection, delayed reclosing, and branch isolation, fundamentally solving the "inadaptability" and "blindness" of traditional technology. While eliminating dangerous reclosing and ensuring personal safety, it also minimizes unnecessary power outages and greatly improves power supply reliability. 2. By enabling the decision model to identify and output specific faulty branches, when it is determined that the fault originates from a specific downstream branch, the system can automatically execute actions. This not only reduces the power outage area to the smallest unit, ensuring the normal power supply of the remaining healthy lines, but also, after successful isolation, the system will immediately send an alarm notification containing accurate branch information to the operation and maintenance platform or personnel. This allows operation and maintenance personnel to directly reach the fault point for handling without blindly investigating, significantly shortening the fault repair time and greatly improving operation and maintenance efficiency. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a flowchart illustrating the overall process of a leakage current prevention and disconnection method based on a current detection algorithm according to an embodiment of the present invention. Detailed Implementation

[0018] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0019] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0020] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0021] This invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of this invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not adhering to the usual scale. Furthermore, the schematic diagrams are merely examples and should not be construed as limiting the scope of protection of this invention. In actual fabrication, the three-dimensional spatial dimensions of length, width, and depth should be included.

[0022] Furthermore, in the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used solely for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In addition, the terms "first," "second," or "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0023] Unless otherwise explicitly specified and limited, the terms "installation," "connection," and "joining" in this invention should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; similarly, they can refer to mechanical connections, electrical connections, or direct connections, or indirect connections through an intermediate medium, or internal connections between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0024] Example 1 Reference Figure 1This is the first embodiment of the present invention, which provides a leakage current prevention and disconnection method based on a current detection algorithm, including: S1. When the residual current in the line is detected to exceed the preset trigger threshold, the real-time fault waveform data of the residual current is acquired, and combined with the pre-stored historical line operation data, a state vector containing multiple data components is generated. Specifically, the method of the present invention is implemented through a leakage current protection system, which includes a main intelligent protection unit installed at the main incoming line and multiple slave intelligent control units installed on each downstream branch. The main intelligent protection unit is responsible for collecting the residual current of the main line, executing the on / off control of the main switch, running the core decision model, and acting as a communication gateway. The slave intelligent control units are responsible for monitoring and controlling the switches of their corresponding branches. The main intelligent protection unit and each slave intelligent control unit exchange data and transmit commands through communication buses such as power line carrier (PLC) and LoRa. All historical operating data and pre-trained decision models are stored in the non-volatile memory of the main intelligent protection unit. It should be noted that this step is the data foundation of the entire intelligent decision-making process. The core task is to comprehensively capture the line status information from two dimensions at the moment the fault occurs: first, real-time fault waveform data obtained through high-fidelity sampling to characterize the transient characteristics of the fault moment; second, extracting key indicators from the line's historical operating data pre-stored in the system to characterize the line's long-term health background. Finally, this information is quantified into a standardized data structure, namely a state vector, for subsequent analysis by the decision-making model. Furthermore, the leakage current protection system or power supply protection system continuously collects the live wire current and neutral wire current through a high-precision synchronous sampling circuit and calculates the residual current in real time. When the residual current in the line exceeds the preset trigger threshold, the fault response mechanism is activated immediately. At this time, the system does not immediately perform the traditional tripping action, but stores the real-time fault waveform data of the residual current at a high fidelity and high sampling rate (e.g., 256kHz) for a preset time before and after the fault (e.g., 50ms before the fault to 100ms after the fault). Furthermore, to balance high-sensitivity analysis with absolute safety, this invention preferably employs a dual-threshold triggering mechanism: First-level threshold (wake-up threshold): Set a low current value, such as 10mA. When the residual current exceeds the wake-up threshold, the system immediately enters the event capture mode, performs waveform acquisition and state vector generation in step S1, and starts decision model analysis in step S2. The second-level threshold (mandatory protection threshold): Set a current value that meets personal safety standards, such as 30mA. The system monitors this threshold in parallel. If the remaining current exceeds 30mA before the intelligent decision in step S2 is completed, or if the result of the intelligent decision is delayed for some reason, the system will bypass the intelligent decision and forcibly execute the action of immediately disconnecting the main switch to ensure that the most basic safety protection requirements are met under any circumstances. In this way, the intelligent decision is an optimization above the safety baseline, rather than a replacement operation. Furthermore, based on the acquired real-time fault waveform data and pre-stored historical line operation data, the system generates a multi-dimensional state vector through parallel processing. The specific generation process is as follows: S101. In order to extract valuable information from the original waveform, the system processes the acquired real-time fault waveform data and extracts two types of key features: frequency band energy distribution value and waveform morphology parameters. For the frequency band energy distribution value, since different types of leakage current (such as metallic grounding, high-resistance grounding, arcing grounding, capacitive leakage current) or interference (such as lightning induction, inverter starting) exhibit significantly different energy characteristics in the frequency domain, wavelet packet transform (WPT) or fast fourier transform (FFT) can be used to decompose the fault waveform into multiple preset frequency bands. Specifically, in one feasible implementation of the present invention, the frequency band can be divided into: Power frequency band (45~55Hz): used to characterize the traditional power frequency leakage current component; Mid-frequency band (150Hz~2kHz): This band is related to harmonic leakage current generated by switching power supplies and rectifiers; High-frequency band (>2kHz): This band is closely related to arcing faults, equipment start-up and shutdown, or transient interference; Specifically, calculate each preset frequency band. The energy distribution value on the surface can be calculated using the following formula: in, It is the first Energy distribution values ​​of each frequency band It is the first result obtained after transforming real-time fault waveform data. There are 10 coefficients (which are Fourier coefficients or wavelet packet coefficients). and They are the first The start and end positions of the coefficient index corresponding to each frequency band; It should be noted that the set of all energy distribution values ​​constitutes the first component of the state vector; Specifically, since waveform morphology parameters need to describe the shape and statistical characteristics of the waveform from a time-domain perspective, the calculated waveform morphology parameters, in order to distinguish actual fault types and provide supplementary information, include, but are not limited to: Peak value: the maximum instantaneous amplitude of the waveform; Crest factor: the ratio of peak value to RMS value, used to measure the impact of a waveform; Kurtosis: A statistic used to measure the sharpness of a waveform; it is sensitive to impulsive pulses. Skewness: A statistic used to measure the symmetry of a waveform; Similarly, the set of all calculated waveform morphological parameters constitutes the second component of the state vector; Furthermore, by combining the set of energy distribution values ​​and the set of waveform morphology parameters, the data components of the state vector can be obtained; Furthermore, in order to enable subsequent decision-making models to understand the "inherent condition" of the line, this invention introduces a health index that reflects the long-term insulation status of the line. It should be noted that during normal operation, the system continuously monitors and records the background high-frequency residual current value in the line. This background high-frequency residual current value is usually generated by the inherent distributed capacitance of the line and the EMI filter of the electrical equipment (especially frequency converters and switching power supplies). Its amplitude will slowly increase as the line insulation ages, gets damp or contaminated. Therefore, it is necessary to calculate a standardized health index based on the average value of the background high-frequency residual current value recorded over a long period (e.g., the past few weeks or months). Specifically, in one feasible implementation of the present invention, the standardized health index can be designed as an inverse proportional function, where the higher the background current value, the lower the health index. The calculation formula can be expressed as: in, It is the current health index of the line, and the value range is usually between [0,1], where 1 represents very healthy and 0 represents extremely poor insulation condition; It is a moving average of historical high-frequency residual current values; It is a reference value for the background high-frequency residual current value under new line or ideal conditions; It is the upper limit of the background high-frequency residual current value set according to safety specifications or experience, based on the assumption that the line insulation has been severely deteriorated; Specifically, to ensure the operability and standardization of the health index calculation, a reference value is used in one feasible process of the present invention. and upper limit value The determination method is as follows: For reference values During the initial installation and commissioning of the equipment, measurements are taken on the brand-new, well-insulated lines, and the background high-frequency residual current value under stable operating conditions is taken as the initial reference value for that specific line and stored in the equipment. For the upper limit value This value is not set arbitrarily, but is based on the guidance of relevant electrical safety standards (such as GB / T6829) on the maximum allowable leakage current under specific environments, and is set by professional technicians in combination with the importance and safety level of the line. For example, for general civil lines, a more conservative upper limit value can be set. It should be noted that this health index, as another data component of the state vector, can provide a key basis for the subsequent decision-making model to determine whether the current failure is an accidental event that occurs on a "healthy" path or an inevitable result that occurs on a "pathological" path. In addition, to give the decision-making model a forward-looking perspective, it is also necessary to make future predictions about health indices. Specifically, historical health indices are stored as time-series data. The time series is learned and modeled using a time series processing algorithm, such as ARIMA (Autoregressive Integral Moving Average) or a more complex LSTM (Long Short-Term Memory) network. Specifically, when a fault occurs, the time series processing algorithm calculates predicted health index values ​​for one or more future time points (such as the next day or week) based on the latest historical data. Finally, these predicted values ​​are incorporated into the state vector as additional data components. It should be noted that these predictions reflect the deteriorating trend of the line insulation condition. If the predictions show that the health condition will decline rapidly, the model may tend to take a more conservative and safer action (such as locking out the disconnection) even if the current fault characteristics are not obvious. Furthermore, ultimately, the generated state vector It is a high-dimensional vector containing all the above information, and its structure can be represented as: S2. Input the state vector into a pre-trained decision model. The decision model selects an action to be executed from a preset action set according to preset decision rules. It should be noted that this step aims to receive the state vector finally generated in step S1, use the reasoning ability of the artificial intelligence model to deeply analyze the nature, severity and possible root causes of the current fault, and select the optimal response action, thereby getting rid of the rigid and blind logical thinking of traditional protection devices. Furthermore, in one feasible implementation of the present invention, the pre-trained decision model is a reinforcement learning (RL) model, which can be, for example, a deep Q-network (DQN) or a variant thereof, and the model is obtained through offline training. Furthermore, during training, the model (or agent) continuously receives states, attempts to perform actions, and obtains a reward from the environment in a simulated electrical environment. Its learning objective is to maximize the long-term cumulative reward. To this end, this invention designs a multi-objective reward function, defined as a weighted sum of values ​​associated with multiple core objectives: in, Is Moment, the immediate reward an agent receives after performing an action; It is a reward component associated with safety objectives. For example, when the simulated fault is of a high-risk type (such as the waveform characteristics of simulating electric shock), if the model selects any form of reclosing action, it will receive a huge negative reward (penalty); conversely, if it selects lockout, it will receive a positive reward or zero reward. It is a reward component associated with reliability objectives. For example, when the simulated fault is transient and self-healing (such as lightning induction), if the model selects delayed reclosing and successfully restores power supply, a positive reward is given; if an unnecessary lockout disconnection is selected, resulting in a long power outage, a negative reward is given. It is a reward component associated with the goal of maintaining the line status. For example, if the model successfully locates and isolates a faulty branch that is causing the line health index to continue to decline, a positive reward is given to encourage it to make decisions that are conducive to long-term maintenance. These are the weights corresponding to each reward component; it should be noted that in this invention, security is of paramount importance, therefore its weight is set to the highest, i.e. and To ensure that the model prioritizes safety under all circumstances; Specifically, the offline training process of the model is as follows: Construct a high-fidelity power system simulation environment based on professional electromagnetic transient simulation software (such as PSCAD / EMTP) or using the Simscape Electrical toolbox in Simulink. The simulation model should include at least: a power source with adjustable parameters, a distributed parameter model of the line, multiple independently controllable branch switches, and modules capable of simulating various loads (resistive, inductive, and nonlinear). In this simulation environment, training samples are generated on a large scale and parametrically. Automated scripts are used to simulate tens of thousands of different fault scenarios in the simulation model. Each scenario corresponds to a state-action-reward training sample. Specific generation methods include: Simulates metallic grounding, high-impedance grounding (achieved by connecting resistors of different resistance values ​​in series), arc grounding (achieved by introducing Mayr or Cassie arc models), surges caused by capacitive devices (such as EMI filters), and non-fault interference (such as lightning induced waves and inrush currents from starting large motors). Meanwhile, for each type of fault, the key parameters are systematically changed, such as the fault occurrence time (phase angle), grounding resistance value, fault duration, and line aging degree (achieved by adjusting the insulation resistance and distributed capacitance parameters of the line model); then, simulation is run for each simulated fault scenario, the residual current waveform generated is collected, and the corresponding state vector is extracted from the simulation waveform using the same algorithm as in step S1. The massive (state vector, real fault type) data pairs generated above are used to train the reinforcement learning model in this invention. For example, an experience replay mechanism is used to store the (state, action, reward, next state) tuples in a replay pool. During training, small batches of data are randomly drawn from the pool for gradient descent to update the weights of the Q network. Through several iterations, the model can learn to accurately map from the complex state vector to the action that can obtain the maximum long-term reward. Furthermore, once training is complete, the weight parameters of the obtained reinforcement learning model will be solidified and lightweighted. For example, through model pruning, quantization and other techniques, an inference engine suitable for embedded processor operation will be generated. This lightweight inference engine will be downloaded and stored in the firmware of the main intelligent protection unit. Therefore, in actual operation, the device only performs efficient forward inference computation (i.e. Q-value calculation in step S2) without the need for resource-intensive online training. Furthermore, a preset action set is defined, which is a collection of all operations that the model can choose to execute. This set includes at least the following types of actions: Action to lock the main switch in the open state: This action is the most conservative safety action. When the model judges the fault to be high-risk or permanent (such as severe insulation damage or direct electric shock to the human body), this action will be selected. After execution, the main switch will remain in the open state and any automatic reclosing will be prohibited. Manual inspection and manual reset must be waited for. The action of closing the main switch after a preset delay: This action is used to improve the reliability of power supply. When the model determines that the fault is transient and non-persistent (such as lightning strike induction or transient interference from the start of a large motor), this action will be selected. The system will disconnect the main switch and wait for a short delay (e.g., 0.5 seconds) before automatically attempting to close the main switch again to quickly restore power supply. Specifically, in one implementable process of the present invention, the preset delay time can be used as part of the output of the decision model, or it can be adaptively adjusted according to the fault waveform characteristics (such as energy magnitude) in the state vector. For example, for transient interference with low energy, a shorter delay time can be used; for interference with high energy, the delay time can be appropriately extended to ensure that the arc is completely extinguished. The action of disconnecting the designated downstream branch and then closing the main switch: This action is an intelligent action with both diagnostic and recovery functions. When the model determines that the fault is a permanent fault originating from a specific downstream branch, it will not only select this action type, but also output and explicitly specify the number of the faulty branch. Each number represents a specific action. It should be noted that executing this action means first isolating the source of the fault, and then restoring power supply to the main line and other healthy branches, which greatly reduces the scope of the power outage and provides maintenance personnel with accurate fault location information. Furthermore, after generating the state vector, the trained reinforcement learning model selects the action to be executed according to its preset decision rule. The core idea of ​​this decision rule is to calculate an expected long-term reward value for each action in the preset action set, which is usually called the Q value in the field of reinforcement learning. Specifically, the model receives the state vector as input through its internal neural network and outputs the Q-value for each action in the action set. It represents the current state vector. Next, execute the action. And thereafter, always follow the expected value of the future cumulative rewards that can be obtained by the optimal strategy; Specifically, the decision-making process of the model can be expressed mathematically as follows: in, These are the final selected actions to be performed. It is a preset action set. It is a decision model (Q-network) for the state vector Next action Evaluation value; Furthermore, the system will use the state vector The input decision model calculates the Q-values ​​of all possible actions in parallel, and selects the action with the highest Q-value as the final decision result. ; S3. Execute the pending actions, including disconnecting, reclosing, and isolating specific branches of the line; It should be noted that this step aims to transform the abstract instructions output by the decision model in step S2 into precise control operations on physical switching equipment (such as main circuit breaker and branch circuit breaker) in the line, thereby achieving differentiated and intelligent handling of leakage faults. Furthermore, depending on the action selected in step S2, the execution flow of this step is also divided into the following cases: Scenario 1: When the action to be executed is to lock the main switch to the open state, it belongs to the highest safety level response; the system immediately sends a permanent disconnect command to the actuator of the main switch. This command not only trips the main switch, but also activates a locking mechanism (which can be a mechanical device or software algorithm) to prohibit any automatic or remote reclosing operation. The line will remain in the de-energized state until the maintenance personnel arrive on site, complete the fault diagnosis and repair, and perform a manual reset to ensure absolute safety in the most dangerous situation. Scenario 2: When the action to be performed is to close the main switch after a preset delay, the system first sends an immediate disconnect command to the main switch to cut off the power supply and eliminate the transient fault. Subsequently, a preset delay timer is started (where the delay time of the timer is set to the same initial value as the action definition). This delay is designed to wait for the ionization path of the transient fault to dissipate naturally. When the delay timer expires, the system automatically sends a closing command to the main switch to attempt to restore power supply. This ensures that the continuity of power supply is maximized while ensuring safety, and avoids unnecessary power outages caused by non-continuous interference. Scenario 3: When the action to be performed is to disconnect the specified downstream branch and then close the main switch, the system performs the following operations: S301. First, the action to be executed output by the decision model is parsed. In this case, the action not only indicates that the action type is isolation and overlap, but also includes the number of the target branch. Then, the system sends a disconnect command to the switch controlling the target branch through the communication bus (such as LoRa, PLC, etc.).

[0025] S302. After sending the branch disconnection command, the system will enter a waiting state and monitor the status feedback signal of the branch switch. After confirming that the target branch is disconnected (for example, receiving a feedback signal from branch switch No. 3 confirming that it has been successfully disconnected), the system will immediately send a closing command to the main switch to restore power supply to all healthy lines except the faulty branch. S303. Upon sending a closing command to the main switch and restoring power to the main line, the system immediately initiates a verification timer of a preset duration to ensure diagnostic accuracy. It is important to emphasize that this preset duration is typically set to several seconds (e.g., 3-5 seconds), and this duration must be sufficient to cover the time required for the line current to reach a steady state after the main switch is closed. This ensures that abnormal residual current recurring due to incomplete fault clearing can be effectively detected during this period. During the verification timer, the system continuously monitors the residual current of the main line with high sensitivity. If, during this period, the residual current exceeds the trigger threshold again, it indicates either a previous diagnostic error or multiple fault points on the line. In this case, the system will immediately perform an emergency safety operation: immediately disconnect the main switch and terminate the current process. At this time, the system will record an isolation failure event and, upon returning to step S1 for a new round of decision-making, add a data component representing the recent number of isolation failures to the state vector. When the decision model observes... When this component is non-zero, the model's decision will tend to favor conservative actions (such as locking the disconnect). In addition, a maximum number of attempts can be preset (e.g., 3 times). When the number of isolation failures reaches this limit, the system will no longer attempt isolation operations, but will directly execute the action of locking the main switch to the disconnect state and send a serious alarm containing the information of "multiple faults or diagnostic difficulties". If no residual current exceeding the trigger threshold is detected again when the verification time ends, it proves that the fault source has been successfully isolated and the line has returned to stable operation. At this time, the system will generate an alarm notification and send an alarm notification containing the target branch information to an external device via a wireless module (such as 4G / 5G) or wired network (it should be emphasized that the external device can be a cloud server, the mobile terminal of the maintenance personnel, or the host computer of the monitoring center). The notification clearly contains the information of the isolated target branch (e.g., "Warning: Branch switch No. 3 has a permanent leakage fault and has been automatically isolated. Please send personnel to repair it as soon as possible"). In addition to the aforementioned response and handling methods for existing faults, this invention also introduces a proactive leakage current risk prevention step, designed to be continuously executed during normal operation of the power line (i.e., when no residual current exceeding a preset trigger threshold is detected), specifically including: It should be noted that this step is a parallel monitoring task that runs continuously in the background. Its purpose is to achieve early detection and warning of the deterioration trend of line insulation. The specific implementation is as follows: The background high-frequency residual current value of the line is continuously collected at a preset low frequency (e.g., once per hour or once per day), and the real-time health index of the line is calculated and updated according to the same calculation formula as in step S1. This index is stored as a time series data, reflecting the long-term trend of the insulation performance of the line. A maintenance threshold is preset (for example, it can be set to 0.4 based on experience, with a range of (0,1]). After each update of the health index, the system will compare the updated health index with this maintenance threshold. If the updated health index is greater than the maintenance threshold, it indicates that the line status is acceptable, and the system will continue to maintain silent monitoring.

[0026] If the updated health index is less than the maintenance threshold, it indicates that the insulation condition of the line has deteriorated significantly and there is a high risk of leakage, even if a tripping fault has not yet occurred. At this time, the system will immediately generate and send a line status warning message. Similarly, the warning message is sent to the cloud server or the designated operation and maintenance management platform through a wireless module (such as 4G / 5G) or wired network. The content of the warning message may be: "Warning: The line insulation health index is [0.38], which is lower than the maintenance threshold of 0.4. There is a potential risk of leakage. It is recommended to arrange preventive maintenance." Through the above steps, this invention achieves a shift from "passive response" to "proactive prevention," providing maintenance personnel with valuable early warning time before catastrophic failures occur, thereby effectively preventing accidents and ensuring electrical safety.

[0027] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this application can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0028] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0029] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0030] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0031] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0032] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A leakage current prevention and disconnection method based on a current detection algorithm, characterized in that, include: When the residual current in the line is detected to exceed the preset trigger threshold, the real-time fault waveform data of the residual current is acquired, and combined with the pre-stored historical line operation data, a state vector containing multiple data components is generated. The state vector is input into a pre-trained decision model, which selects an action to be executed from a preset action set according to preset decision rules. The actions to be performed include disconnecting, reclosing, and isolating specific branches of the line. The step of generating the state vector includes: Calculate the energy distribution value of the real-time fault waveform data on multiple preset frequency bands, and calculate the morphological parameters of the waveform data. Use the energy distribution value and the morphological parameters as data components of the state vector. The step of generating the state vector further includes: Based on the background high-frequency residual current value recorded in the historical operation data of the line, a health index reflecting the current state of the line is calculated, and the health index is used as another data component of the state vector. The step of generating the state vector further includes: Historical health index time series data are processed by a time series processing algorithm to calculate the predicted health index values ​​for one or more future time points, and these predicted values ​​are used as additional data components of the state vector. The preset action set includes: Action to lock the main switch in the off state; The action of closing the main switch after a preset delay; And, the action of disconnecting the designated downstream branch and then closing the main switch; The pre-trained decision model is a reinforcement learning model, and the preset decision rules of the model are as follows: Calculate an expected long-term return value for each action in the preset action set, and select the action with the highest expected long-term return value as the action to be executed; The reinforcement learning model is obtained through offline training. During the training process, the reward used to guide the model's learning is set as a weighted sum of values ​​associated with the security objective, the reliability objective, and the line status maintenance objective, respectively; wherein, the weight of the value associated with the security objective is set to the highest. When the action to be performed is to disconnect a specified downstream branch and then close the main switch, the action to be performed includes disconnection, reclosing, and isolation operations for a specific branch on the line, including: A target branch is determined based on the output of the pre-trained decision model, and a disconnect command is sent to the switch controlling the target branch. After confirming that the target branch is disconnected, a closing command is sent to the main switch; After sending a closing command to the main switch, the method further includes: Start a verification timer of a preset duration; If, during the verification timing, the residual current exceeds the trigger threshold again, the main switch is immediately disconnected, and the process returns to the step of generating the state vector. If the remaining current does not exceed the trigger threshold again when the verification time ends, an alarm notification containing the target branch information is sent to the peripheral device.

2. The leakage current prevention and disconnection method based on current detection algorithm as described in claim 1, characterized in that, The step of generating the state vector further includes: During normal operation, when no residual current exceeds the trigger threshold, the health index of the current state is continuously calculated and updated. When the health index of the current state is lower than the preset maintenance threshold, a line status warning message is generated and sent.

Images