Single-phase earth fault AI comprehensive research and judgment system for power distribution network

Abstract

The invention relates to the technical field of power maintenance, and discloses a power distribution network-oriented single-phase earth fault AI comprehensive research and judgment system, which integrates a power distribution network monitoring module, a fault diagnosis equipment module and an AI comprehensive research and judgment module. The system can accurately judge the type, position and severity of the single-phase earth fault by monitoring and collecting all-dimensional data of the power distribution network in real time and combining intelligent analysis of the AI comprehensive research and judgment module, and the accuracy of fault positioning and isolation is remarkably improved. The AI comprehensive research and judgment module adopts a Markov decision process for modeling, and a deep reinforcement learning algorithm is used for training a research and judgment model, so that rapid fault positioning, isolation and power supply recovery are realized, the fault processing efficiency and the power supply recovery speed are effectively improved, and the power failure time and the influence on users are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of power maintenance, and in particular to an AI comprehensive analysis and judgment system for single-phase grounding faults in a distribution network. Background Art

[0002] As an important part of the power system, the stable operation of the distribution network is crucial to ensure the continuity and reliability of power supply. However, in actual operation, the distribution network is often affected by various factors, leading to faults, among which single-phase grounding fault is a common one. Single-phase grounding fault not only affects the quality of power supply, but also may damage the equipment in the distribution network and even cause more serious power accidents.

[0003] At present, the processing of single-phase grounding faults mainly relies on traditional fault diagnosis algorithms. Although traditional fault diagnosis algorithms, such as overcurrent protection and distance protection, can locate and isolate faults to a certain extent, their accuracy and efficiency are often affected by the complexity and variability of the distribution network. These algorithms are usually based on preset thresholds and rules for judgment, and it is difficult to adapt to the specific conditions of different distribution networks and the changes in fault types. Moreover, traditional algorithms often lack global considerations when handling faults, and it is difficult to comprehensively evaluate multiple factors such as the efficiency of fault handling, the speed of power supply recovery, equipment loss, and power supply reliability. For example, in a certain city's distribution network, single-phase grounding faults occur frequently due to aging lines and obsolete equipment. When handling these faults, traditional fault diagnosis algorithms are often unable to accurately determine the fault location and type, resulting in long fault handling time and slow power supply recovery. Summary of the invention

[0004] The purpose of the present invention is to provide a single-phase grounding fault AI comprehensive analysis system for distribution network to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solution: an AI comprehensive analysis and judgment system for single-phase grounding faults in distribution networks, the system comprising:

[0006] The distribution network monitoring module integrates current sensors, voltage sensors, ground resistance monitors and environmental parameter sensors to monitor and collect the current, voltage, ground resistance, and environmental temperature and humidity parameters of the distribution network in real time;

[0007] Fault diagnosis equipment module, including fault indicator, disconnect switch and backup power supply device, used for fault location, isolation and power restoration when single-phase grounding fault occurs;

[0008] The AI ​​comprehensive analysis module combines the data collected by the distribution network monitoring module, the real-time topology of the distribution network and the current status of the fault diagnosis equipment to automatically analyze and determine the type, location and severity of the single-phase grounding fault, providing accurate control strategies for the fault diagnosis equipment.

[0009] Preferably, the implementation method of the AI ​​comprehensive analysis module is:

[0010] A single-phase grounding fault analysis model is constructed, and the single-phase grounding fault processing process is modeled as a Markov decision process. The specific definitions include:

[0011] State space: including the current, voltage, ground resistance value, ambient temperature and humidity parameters of the distribution network, topological structure information of the distribution network, and the status of the fault diagnosis equipment;

[0012] Action space: contains the possible actions that the fault diagnosis equipment may take when handling a single-phase grounding fault. The possible actions include starting or stopping the fault indicator, operating the isolation switch to isolate the fault, and enabling or switching the backup power supply device;

[0013] Reward: A reward function designed based on the efficiency of fault handling, speed of power supply restoration, equipment loss, and power supply reliability indicators to evaluate the results of actions;

[0014] Policy: The rule for choosing an action in a given state, optimized through learning to maximize the cumulative reward;

[0015] Design a reward function. Based on the goal of single-phase grounding fault handling, design a reward mechanism that includes rapid fault location, isolation and power restoration, reduced equipment loss, and improved power supply reliability. At the same time, set penalties for failed fault handling or equipment misoperation.

[0016] To train the single-phase grounding fault analysis model, we use a deep reinforcement learning algorithm, perform iterative training on actual distribution network operation data, and optimize the learning rate, discount factor, and exploration rate.

[0017] Evaluate the single-phase grounding fault diagnosis model, and compare the model recommendation scheme with the traditional fault diagnosis algorithm in terms of fault handling efficiency, power supply restoration speed, equipment loss and power supply reliability indicators through a combination of simulation tests and field tests;

[0018] Based on the trained single-phase grounding fault analysis model, the real-time data, topology structure and status information of the distribution network and the fault diagnosis equipment are analyzed to provide a precise control strategy for the fault diagnosis equipment to achieve rapid location, isolation and power restoration of single-phase grounding faults.

[0019] Preferably, a reward function based on multi-objective optimization is designed to comprehensively evaluate the effect of single-phase grounding fault handling. The function comprehensively considers four main aspects: fault handling efficiency, power supply recovery speed, equipment loss and power supply reliability. The specific design includes:

[0020] Fault handling efficiency reward: Assume the fault handling time is T f , the target processing time is T target , then the fault handling efficiency reward R efficiency It can be expressed as:

[0021]

[0022] Among them, α and β are positive coefficients, representing the weights of rewards and penalties for fault handling efficiency, respectively;

[0023] Power supply recovery speed reward: Suppose the power supply recovery time is T r , the target recovery time is T restore , then the power supply recovery speed bonus is T restore It can be expressed as:

[0024]

[0025] Among them, γ and δ are positive coefficients, representing the weights of rewards and penalties for power supply restoration speed, respectively;

[0026] Equipment loss penalty: Let the equipment loss be L, then the equipment loss penalty is P loss It can be expressed as:

[0027] P loss =-∈·L

[0028] Among them, ∈ is a positive coefficient, indicating the penalty weight of equipment loss;

[0029] Power supply reliability reward: Let the power supply reliability index be R and the target reliability be R target , then the power supply reliability reward R reliability It can be expressed as:

[0030]

[0031] Among them, θ is a positive coefficient, which indicates the reward weight of power supply reliability;

[0032] Combining the rewards and penalties of the above four aspects, the total reward function R total It can be expressed as:

[0033] R total =R efficiency +R restore -P loss +R reliability

[0034] This function provides an evaluation basis for action selection in the process of single-phase grounding fault handling by weighing four factors: fault handling efficiency, power supply restoration speed, equipment loss and power supply reliability.

[0035] Preferably, the AI ​​comprehensive analysis module constructs a single-phase grounding fault analysis model in the following way: a Q network is constructed using a deep neural network, which takes a state vector s as input and outputs a Q value corresponding to each action a, i.e., Q(s,a), where the Q value represents the expected benefit of taking action a under state s.

[0036] Preferably, the training step of the single-phase grounding fault analysis model includes:

[0037] S1: Collect distribution network operation data, including current, voltage, ground resistance, ambient temperature, humidity parameters, as well as status information of fault diagnosis equipment and fault handling records, and perform data denoising and standardization on the collected data to form a training data set.

[0038] S2: Initialize the Q network parameters, including network weights and biases; at the same time, initialize the experience replay pool to store state transition samples during training;

[0039] S3: Use the ε-greedy strategy to select actions, randomly select an action with probability ε, and select the action with the largest current Q value with probability 1-ε; after executing the selected action, observe the reward and the next state, and transfer the state sample (s, a, r, s ′ ) is stored in the experience replay pool;

[0040] S4: Randomly extract a batch of state transition samples from the experience replay pool, and use the target Q network to calculate the target Q value. The calculation formula is: r+γmax a′ Q ′ (s ′ ,a ′ ), where Q ′ represents the target Q network, γ represents the discount factor;

[0041] Then, the loss function is used to calculate the difference between the Q network and the target Q network, and the Q network parameters are updated through the gradient descent algorithm;

[0042] S5: At regular intervals, the parameters of the Q network are copied to the target Q network to maintain the stability of the target Q network;

[0043] S6: Repeat the steps of policy iteration, network update, and target network update until the preset number of training rounds is reached.

[0044] Preferably, in step S4, the formula for updating the parameters of the gradient descent algorithm is:

[0045]

[0046] Among them, θ represents the model parameters, including all weights and biases that the model needs to learn; α represents the learning rate, which is used to control the step size of parameter update; It represents the gradient of the loss function J(θ) with respect to the parameter θ, which is a vector pointing to the direction in which the loss function grows fastest; := represents the assignment operation, that is, updating the value of the parameter θ.

[0047] Preferably, in step S1, for the noise data, a sliding average filtering algorithm is used to perform data denoising, and the algorithm formula is:

[0048]

[0049] Among them, y[n] is the denoised data, x[n] is the original data, and N is the size of the sliding window.

[0050] Preferably, in step S1, the Z-score method is used to standardize the data, and the mean of each numerical data is subtracted and divided by its standard deviation, so that the processed data conforms to the standard normal distribution. The formula for standardization is: Z = (X-μ) / σ, wherein X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data.

[0051] Preferably, when evaluating the single-phase grounding fault analysis model, the simulation test implementation steps include:

[0052] Use MATLAB / Simulink power system simulation software to build a distribution network simulation model, including power supply, line, load, fault diagnosis equipment elements, and set up a single-phase grounding fault scenario;

[0053] In the simulation environment, the single-phase ground fault analysis model recommendation scheme and the traditional fault diagnosis algorithm are implemented, including overcurrent protection and distance protection;

[0054] Based on the four indicators of fault handling efficiency, power supply restoration speed, equipment loss and power supply reliability, the specific test method and calculation formula are defined: fault handling efficiency is measured by calculating the time from the occurrence of the fault to the fault being located and isolated; power supply restoration speed is measured by calculating the time from the fault being isolated to the power supply being restored; equipment loss is measured by counting the energy consumption and wear of the equipment during the fault handling process; power supply reliability is measured by calculating the number and duration of power supply interruptions;

[0055] A single-phase grounding fault is triggered in the simulation environment, and the single-phase grounding fault judgment model recommendation scheme and the traditional fault diagnosis algorithm are run respectively. The various indicator data of the two algorithms in the process of handling the fault are recorded.

[0056] Preferably, when evaluating the single-phase grounding fault analysis model, the steps of performing field testing include:

[0057] Select representative test scenarios in actual distribution networks, including urban distribution networks or rural distribution networks;

[0058] Deploy fault diagnosis equipment in the test scenario, including fault indicators and smart switches;

[0059] According to the results and experience of simulation tests, formulate a specific plan for field testing, including fault triggering mode, data collection method and indicator calculation method;

[0060] A single-phase grounding fault is triggered in the actual distribution network, and the single-phase grounding fault analysis model recommendation scheme and the traditional fault diagnosis algorithm are run respectively. The indicator data of the two algorithms in the process of handling the fault are recorded and compared and analyzed.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] This system can monitor and collect all-round data of the distribution network in real time. Combined with the intelligent analysis of the AI ​​comprehensive analysis module, it can accurately determine the type, location and severity of single-phase grounding faults, thereby effectively improving the accuracy of fault location and isolation. The AI ​​comprehensive analysis module adopts Markov decision process modeling. The analysis model obtained through deep reinforcement learning algorithm training can quickly select the optimal action under a given state to achieve rapid fault location, isolation and power restoration, significantly improving the efficiency of fault handling and the speed of power supply restoration, and reducing the power outage time caused by faults and the impact on users.

[0063] When designing the reward function, this system fully considers the factors of reducing equipment loss and improving power supply reliability. By optimizing the control strategy, the system can minimize equipment loss and extend the service life of the equipment while handling faults. At the same time, rapid fault handling and power supply recovery also effectively improve the reliability of power supply and reduce power accidents and losses caused by faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 This is the working principle diagram of the AI ​​comprehensive analysis module;

[0065] Figure 2 Schematic diagram of the design of the reward function;

[0066] Figure 3This is the training flow chart of the single-phase grounding fault analysis model;

[0067] Figure 4 Flowchart for performance evaluation of single-phase grounding fault analysis model. DETAILED DESCRIPTION

[0068] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0069] See also Figure 1-4 The present invention provides a technical solution: an AI comprehensive analysis and judgment system for single-phase grounding faults in distribution networks, the system comprising:

[0070] The distribution network monitoring module integrates current sensors, voltage sensors, ground resistance monitors and environmental parameter sensors to monitor and collect key parameters in the distribution network in real time. Specifically, the current sensor and voltage sensor are used to measure the current and voltage values ​​in the distribution network, the ground resistance monitor is used to monitor the changes in ground resistance, and the environmental parameter sensor is responsible for collecting parameters such as ambient temperature and humidity.

[0071] Fault diagnosis equipment module, which includes hardware devices such as fault indicator, isolation switch and backup power supply device. The fault indicator is used to provide intuitive fault location indication when a fault occurs, and the isolation switch is used to isolate the fault area when necessary to prevent the fault from expanding. The backup power supply device provides emergency power support when the main power fails to ensure the continuous operation of key equipment.

[0072] AI comprehensive analysis module. This module is responsible for comprehensive analysis of single-phase grounding faults based on the data collected by the distribution network monitoring module, the real-time topology of the distribution network, and the current state of the fault diagnosis equipment. In specific implementation, a single-phase grounding fault analysis model is first constructed, and the fault handling process is modeled as a Markov decision process. The state space includes the current, voltage, grounding resistance value, ambient temperature and humidity parameters of the distribution network, the topology information of the distribution network, and the state of the fault diagnosis equipment. The action space contains the actions that the fault diagnosis equipment may take, such as starting or stopping the fault indicator, operating the disconnector for fault isolation, enabling or switching the backup power supply device, etc. The reward function is designed based on indicators such as the efficiency of fault handling, the speed of power supply recovery, equipment loss, and power supply reliability, and is used to evaluate the results of the action. The strategy is the rule for selecting actions under a given state, which is optimized through learning to maximize the cumulative reward.

[0073] In the model training phase, a deep reinforcement learning algorithm is used to perform iterative training on actual distribution network operation data. The performance of the model is optimized by continuously adjusting parameters such as learning rate, discount factor, and exploration rate. After training, the model is deployed to the AI ​​comprehensive analysis module to analyze the operation data of the distribution network in real time and provide accurate control strategies.

[0074] Finally, the system is evaluated and optimized. Through a combination of simulation tests and field tests, the model recommendation scheme is compared with the traditional fault diagnosis algorithm in terms of fault handling efficiency, power supply recovery speed, equipment loss, and power supply reliability indicators. According to the test results, the system is adjusted and optimized as necessary to ensure its stability and reliability in actual operation.

[0075] The present invention will be further described below in conjunction with Examples 1 to 3:

[0076] Embodiment 1:

[0077] The design of the reward function is used to comprehensively evaluate the effect of fault handling to guide the system to make the best decision. In this embodiment, the design method of the reward function for the four key indicators of fault handling efficiency, power supply recovery speed, equipment loss and power supply reliability is described in detail.

[0078] Fault handling efficiency reward: Assume the fault handling time is T f , the target processing time is T target , then the fault handling efficiency reward R efficiency It can be expressed as:

[0079]

[0080] Among them, α and β are positive coefficients, representing the weights of rewards and penalties for fault handling efficiency, respectively.

[0081] Assume that in a single-phase ground fault processing, the system detects the fault processing time T f The target processing time T is 30 minutes. target = 20 minutes. According to the design of the reward function, if the fault handling time is less than or equal to the target time, a positive reward is given; if it is greater than the target time, a negative penalty is given. Assuming α = 10, β = 5, the fault handling efficiency reward R efficiency The calculation is as follows:

[0082]

[0083] Power supply recovery speed reward: Suppose the power supply recovery time is T r , the target recovery time is T restore , then the power supply recovery speed bonus is T restore It can be expressed as:

[0084]

[0085] Among them, γ and δ are positive coefficients, representing the weights of rewards and penalties for power supply restoration speed, respectively.

[0086] Assume that the power supply recovery time is T r The target recovery time is 15 minutes. restore Assume γ = 8, δ = 4, then the power supply recovery speed reward R restore The calculation is as follows:

[0087] Due to T r >T restore ,so:

[0088]

[0089] Equipment loss penalty: Let the equipment loss be L, then the equipment loss penalty is P loss It can be expressed as:

[0090] P loss =-∈·L

[0091] Among them, ∈ is a positive coefficient, which indicates the penalty weight of equipment loss.

[0092] Assume that during the fault handling process, the equipment loss L is 0.05 (expressed in some standardized unit). Let ∈ = 20, then the equipment loss penalty P loss The calculation is as follows:

[0093] P loss =-∈·L=-20·0.05=-1

[0094] Power supply reliability reward: Let the power supply reliability index be R and the target reliability be R target , then the power supply reliability reward R reliability It can be expressed as:

[0095]

[0096] Among them, θ is a positive coefficient, which represents the reward weight of power supply reliability.

[0097] Assuming that the power supply reliability index R is 0.95, the target reliability R target is 0.98. Assuming θ = 15, the power supply reliability reward R reliability The calculation is as follows:

[0098]

[0099] Combining the rewards and penalties of the above four aspects, the total reward function R totalIt can be expressed as:

[0100] R total =R efficiency +R restore -P loss +R reliability =-2.5-2-1+14.54=9.04

[0101] In the AI ​​comprehensive judgment module, the above reward function will be used to evaluate the impact of different actions (such as starting the fault indicator, operating the disconnector, etc.) on the fault handling effect. The system will select those actions that can maximize the cumulative reward through learning optimization, thereby achieving fast and efficient handling of single-phase grounding faults.

[0102] Embodiment 2:

[0103] This embodiment is used to explain in detail the construction and training process of the single-phase grounding fault analysis model in combination with Python code examples.

[0104] The steps of constructing a single-phase grounding fault analysis model include:

[0105] State space definition: Taking a distribution network as an example, the state space is defined to include current, voltage, ground resistance, ambient temperature, humidity, and the state of fault diagnosis equipment. Specifically, the current and voltage are obtained through real-time monitoring by sensors, the ground resistance is measured by a ground resistance monitor, the ambient temperature and humidity are collected by environmental parameter sensors, and the state of the fault diagnosis equipment includes the switch state of the fault indicator, the position of the isolation switch, and the working state of the backup power supply device.

[0106] Action space definition: For the above state space, the action space is defined to start or stop the fault indicator, operate the disconnector for fault isolation (such as closing or opening), enable or switch the backup power supply, etc. Each action corresponds to a specific operation instruction, which is selected and executed by the AI ​​comprehensive judgment module according to the current state.

[0107] Reward function design: Reward function is designed based on four main aspects: fault handling efficiency, power supply restoration speed, equipment loss, and power supply reliability. For example, a positive reward is given when the fault handling time is less than or equal to the target handling time, and a negative reward is given when it is greater than the target handling time; a positive reward is given when the power supply restoration time is less than or equal to the target restoration time, and a negative reward is given when it is greater than the target restoration time; a negative reward is given for equipment loss based on the actual loss value; and a positive reward is given for power supply reliability based on the power supply reliability index.

[0108] Q network construction: A deep neural network is used to construct a Q network, where the input layer is the components of the state vector, and the output layer is the Q value corresponding to each action. The network structure can be adjusted according to specific needs, such as using a multi-layer fully connected network, a convolutional neural network, etc.

[0109] The steps of training the single-phase grounding fault analysis model include:

[0110] S1: Collect distribution network operation data, including current, voltage, ground resistance, ambient temperature, humidity parameters, as well as status information of fault diagnosis equipment and fault handling records, and perform data denoising and standardization on the collected data to form a training data set.

[0111] For noisy data, the sliding average filtering algorithm is used for data denoising, and the algorithm formula is:

[0112]

[0113] Among them, y[n] is the denoised data, x[n] is the original data, and N is the size of the sliding window.

[0114] The Z-score method is used to standardize the data. The mean of each numerical data is subtracted and divided by its standard deviation so that the processed data conforms to the standard normal distribution. The standardization formula is: Z = (X-μ) / σ, where X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data.

[0115] S2: Initialize the Q network parameters, including network weights and biases; at the same time, initialize the experience replay pool to store state transition samples during training.

[0116]

[0117] S3: Use the ε-greedy strategy for action selection. In the early stages of training, a larger ε value is set to encourage exploration; as training progresses, the ε value is gradually reduced to exploit known information. After executing the selected action, observe the reward and the next state, and transfer the state sample (s, a, r, s ′ ) is stored in the experience replay pool.

[0118]

[0119] S4: Randomly extract a batch of state transition samples from the experience replay pool, and use the target Q network to calculate the target Q value. The calculation formula is: r+γmax a′ Q ′ (s ′ ,a ′ ), where Q ′Represents the target Q network, and γ represents the discount factor. Then, the mean square error loss function is used to calculate the difference between the Q network and the target Q network, and the Q network parameters are updated by the gradient descent algorithm. The formula for updating the parameters of the gradient descent algorithm is:

[0120]

[0121] Among them, θ represents the model parameters, including all weights and biases that the model needs to learn; α represents the learning rate, which is used to control the step size of parameter update; It represents the gradient of the loss function J(θ) with respect to the parameter θ, which is a vector pointing to the direction in which the loss function grows fastest; := represents the assignment operation, that is, updating the value of the parameter θ.

[0122]

[0123] S5: Every certain number of steps (such as every 100 steps), copy the parameters of the Q network to the target Q network to maintain the stability of the target Q network.

[0124] S6: Repeat the steps of policy iteration, network update, and target network update until the preset number of training rounds (such as 10,000 rounds) is reached or the convergence condition is met (such as the loss function value is less than a certain threshold). During the training process, hyperparameters such as learning rate, discount factor, and exploration rate are continuously adjusted to optimize the performance of the model. For example, you can try different learning rates (such as 0.01, 0.001, etc.) and discount factors (such as 0.9, 0.99, etc.), and observe the performance of the model on the validation set to determine the best hyperparameter combination.

[0125]

[0126] Embodiment 3:

[0127] In order to evaluate the performance of the single-phase grounding fault analysis model (hereinafter referred to as the "analysis model"), this embodiment adopts a combination of simulation testing and field testing to verify the performance of the analysis model by comparing the recommended solution of the analysis model with the traditional fault diagnosis algorithm in terms of fault handling efficiency, power supply recovery speed, equipment loss and power supply reliability indicators.

[0128] The implementation steps of the simulation test include:

[0129] ①Simulation environment construction:

[0130] Use MATLAB / Simulink software to build a distribution network simulation model that includes power supply, lines, loads, fault diagnosis equipment, etc. Set up single-phase grounding fault scenarios to simulate single-phase grounding faults in different locations and types.

[0131] In the simulation environment, the recommended solution of the judgment model and the traditional fault diagnosis algorithm (such as overcurrent protection algorithm) are implemented respectively. Ensure that the two algorithms can correctly receive the input data of the simulation model (such as current, voltage, ground resistance, etc.) and output the corresponding fault processing actions (such as starting the fault indicator, operating the isolation switch, etc.).

[0132] ②Test indicator definition and calculation:

[0133] Fault handling efficiency: This is measured by calculating the time from when a fault occurs to when it is located and isolated. Specifically, the time when the fault occurs and when it is isolated are recorded, and the time difference is calculated as the fault handling efficiency indicator.

[0134] Power supply restoration speed: measured by calculating the time from fault isolation to power supply restoration. Specifically, the time when the fault is isolated and the time when power supply is restored are recorded, and the time difference is calculated as the power supply restoration speed indicator.

[0135] Equipment loss: This is measured by counting the energy consumption and wear of the equipment during the fault handling process. Specifically, the energy consumption and wear data of each device during the fault handling process are recorded and accumulated.

[0136] Power supply reliability: measured by calculating the number and duration of power supply interruptions. Specifically, the number of power supply interruptions and the duration of each interruption during the simulation test are recorded and statistically calculated.

[0137] ③Simulation test execution:

[0138] A single-phase grounding fault is triggered in the simulation environment, and the recommended solution of the analysis model and the traditional fault diagnosis algorithm are run respectively. The various indicator data of the two algorithms in the process of handling the fault are recorded and compared and analyzed.

[0139] The implementation steps of the field test include:

[0140] ①Test scenario selection:

[0141] A city distribution network is selected as the test scenario. The distribution network contains multiple feeders, substations and load points, which can represent the operation of the actual distribution network.

[0142] ② Equipment deployment and debugging:

[0143] Deploy fault diagnosis equipment (such as fault indicators, smart switches, etc.) in the test scenario and ensure that the equipment can work normally. At the same time, debug the research and judgment model to ensure that it can adapt to the operating environment of the actual distribution network.

[0144] ③Test plan formulation:

[0145] According to the results and experience of simulation tests, a specific plan for field testing is formulated, including fault triggering methods (such as manually setting fault points), data collection methods (such as collecting current, voltage and other data through fault diagnosis equipment), and indicator calculation methods (such as calculating according to the definition in the simulation test).

[0146] ④ On-site test execution and data recording:

[0147] A single-phase grounding fault is triggered in the actual distribution network, and the recommended solution of the judgment model and the traditional fault diagnosis algorithm are run respectively. The various indicator data of the two algorithms in the process of handling the fault are recorded, including fault handling time, power supply restoration time, equipment energy consumption and wear data, and the number and duration of power supply interruptions.

[0148] ⑤Data analysis and comparison:

[0149] The data collected from the field test were collated and analyzed, and the statistical quantities such as the mean and standard deviation of each indicator were calculated. The performance of the recommended solution of the judgment model and the traditional fault diagnosis algorithm in various indicators was compared and analyzed to verify the superiority and practicality of the judgment model.

[0150] By comparing the results of simulation tests and field tests, it can be concluded that the research and judgment model is superior to traditional fault diagnosis algorithms in terms of fault handling efficiency, power supply recovery speed, equipment loss and power supply reliability. Therefore, the research and judgment model has higher practicality and promotion value.

[0151] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0152] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. The AI ​​comprehensive analysis system for single-phase grounding faults in distribution networks is characterized by: The system comprises: The distribution network monitoring module integrates current sensors, voltage sensors, ground resistance monitors and environmental parameter sensors to monitor and collect the current, voltage, ground resistance, and environmental temperature and humidity parameters of the distribution network in real time; Fault diagnosis equipment module, including fault indicator, disconnect switch and backup power supply device, used for fault location, isolation and power restoration when single-phase grounding fault occurs; The AI ​​comprehensive analysis module combines the data collected by the distribution network monitoring module, the real-time topology of the distribution network and the current status of the fault diagnosis equipment to automatically analyze and determine the type, location and severity of the single-phase grounding fault, providing accurate control strategies for the fault diagnosis equipment.

2. According to the single-phase grounding fault AI comprehensive analysis system for distribution network according to claim 1, it is characterized in that: The implementation method of the AI ​​comprehensive analysis module is: A single-phase grounding fault analysis model is constructed, and the single-phase grounding fault processing process is modeled as a Markov decision process. The specific definitions include: State space: including the current, voltage, ground resistance value, ambient temperature and humidity parameters of the distribution network, topological structure information of the distribution network, and the status of the fault diagnosis equipment; Action space: contains the possible actions that the fault diagnosis equipment may take when handling a single-phase grounding fault. The possible actions include starting or stopping the fault indicator, operating the isolation switch to isolate the fault, and enabling or switching the backup power supply device; Reward: A reward function designed based on the efficiency of fault handling, speed of power supply restoration, equipment loss, and power supply reliability indicators to evaluate the results of actions; Policy: The rule for choosing an action in a given state, optimized through learning to maximize the cumulative reward; Design a reward function. Based on the goal of single-phase grounding fault handling, design a reward mechanism that includes rapid fault location, isolation and power restoration, reduced equipment loss, and improved power supply reliability. At the same time, set penalties for failed fault handling or equipment misoperation. To train the single-phase grounding fault analysis model, we use a deep reinforcement learning algorithm, perform iterative training on actual distribution network operation data, and optimize the learning rate, discount factor, and exploration rate. Evaluate the single-phase grounding fault diagnosis model, and compare the model recommendation scheme with the traditional fault diagnosis algorithm in terms of fault handling efficiency, power supply restoration speed, equipment loss and power supply reliability indicators through a combination of simulation tests and field tests; Based on the trained single-phase grounding fault analysis model, the real-time data, topology structure and status information of the distribution network and the fault diagnosis equipment are analyzed to provide a precise control strategy for the fault diagnosis equipment to achieve rapid location, isolation and power restoration of single-phase grounding faults.

3. The single-phase grounding fault AI comprehensive analysis and judgment system for distribution network according to claim 2 is characterized in that: A reward function based on multi-objective optimization is designed to comprehensively evaluate the effect of single-phase grounding fault handling. The function comprehensively considers four main aspects: fault handling efficiency, power supply restoration speed, equipment loss, and power supply reliability. The specific design includes: Fault handling efficiency reward: Assume the fault handling time is T f , the target processing time is T target , then the fault handling efficiency reward R efficiency It can be expressed as: Among them, α and β are positive coefficients, representing the weights of rewards and penalties for fault handling efficiency, respectively; Power supply recovery speed reward: Suppose the power supply recovery time is T r , the target recovery time is T restore , then the power supply recovery speed bonus is T restore It can be expressed as: Among them, γ and δ are positive coefficients, representing the weights of rewards and penalties for power supply restoration speed, respectively; Equipment loss penalty: Let the equipment loss be L, then the equipment loss penalty is P loss It can be expressed as: P loss =-∈·L Among them, ∈ is a positive coefficient, indicating the penalty weight of equipment loss; Power supply reliability reward: Let the power supply reliability index be R and the target reliability be R target , then the power supply reliability reward R reliability It can be expressed as: Among them, θ is a positive coefficient, which indicates the reward weight of power supply reliability; Combining the rewards and penalties of the above four aspects, the total reward function R total It can be expressed as: R total =R efficiency +R restore -P loss +R reliability This function provides an evaluation basis for action selection in the process of single-phase grounding fault handling by weighing four factors: fault handling efficiency, power supply restoration speed, equipment loss and power supply reliability.

4. The single-phase grounding fault AI comprehensive analysis and judgment system for distribution network according to claim 3 is characterized in that: The AI ​​comprehensive analysis module constructs a single-phase grounding fault analysis model in the following way: a Q network is constructed using a deep neural network. The network takes a state vector s as input and outputs a Q value corresponding to each action a, i.e., Q(s, a). The Q value represents the expected benefit of taking action a under state s.

5. The single-phase grounding fault AI comprehensive analysis and judgment system for distribution network according to claim 4 is characterized in that: The training steps of the single-phase grounding fault analysis model include: S1: Collect distribution network operation data, including current, voltage, ground resistance, ambient temperature, humidity parameters, as well as status information of fault diagnosis equipment and fault handling records, and perform data denoising and standardization on the collected data to form a training data set. S2: Initialize the Q network parameters, including network weights and biases; at the same time, initialize the experience replay pool to store state transition samples during training; S3: Use the ε-greedy strategy to select actions, randomly select an action with probability ￡, and select the action with the largest current Q value with probability 1-ε; after executing the selected action, observe the reward and the next state, and store the state transition sample (s, a, r, s′) in the experience replay pool; S4: Randomly extract a batch of state transition samples from the experience replay pool, and use the target Q network to calculate the target Q value. The calculation formula is: r+γmax a′ Q′(s′, a′), where Q′ represents the target Q network and γ represents the discount factor; Then, the loss function is used to calculate the difference between the Q network and the target Q network, and the Q network parameters are updated through the gradient descent algorithm; S5: At regular intervals, the parameters of the Q network are copied to the target Q network to maintain the stability of the target Q network; S6: Repeat the steps of policy iteration, network update, and target network update until the preset number of training rounds is reached.

6. The single-phase grounding fault AI comprehensive analysis system for distribution network according to claim 5 is characterized in that: In step S4, the formula for updating parameters of the gradient descent algorithm is: Among them, θ represents the model parameters, including all weights and biases that the model needs to learn; α represents the learning rate, which is used to control the step size of parameter update; represents the gradient of the loss function J(θ) with respect to the parameter θ, which is a vector pointing to the direction in which the loss function grows fastest; := represents the assignment operation, that is, updating the value of the parameter θ.

7. The single-phase grounding fault AI comprehensive analysis and judgment system for distribution network according to claim 5 is characterized in that: In step S1, for the noise data, a sliding average filtering algorithm is used to denoise the data, and the algorithm formula is: Among them, y[n] is the denoised data, x[n] is the original data, and N is the size of the sliding window.

8. The single-phase grounding fault AI comprehensive analysis and judgment system for distribution network according to claim 5 is characterized by: In step S1, the Z-score method is used to standardize the data. The mean of each numerical data is subtracted and divided by its standard deviation so that the processed data conforms to the standard normal distribution. The formula for standardization is: Z = (X-μ) / σ, where X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data.

9. The single-phase grounding fault AI comprehensive analysis and judgment system for distribution network according to claim 1 is characterized in that: When evaluating the single-phase grounding fault analysis model, the simulation test implementation steps include: Use MATLAB / Simulink power system simulation software to build a distribution network simulation model, including power supply, line, load, fault diagnosis equipment elements, and set up a single-phase grounding fault scenario; In the simulation environment, the single-phase ground fault analysis model recommendation scheme and the traditional fault diagnosis algorithm are implemented, including overcurrent protection and distance protection; Based on the four indicators of fault handling efficiency, power supply restoration speed, equipment loss and power supply reliability, the specific test method and calculation formula are defined: fault handling efficiency is measured by calculating the time from the occurrence of the fault to the fault being located and isolated; power supply restoration speed is measured by calculating the time from the fault being isolated to the power supply being restored; equipment loss is measured by counting the energy consumption and wear of the equipment during the fault handling process; power supply reliability is measured by calculating the number and duration of power supply interruptions; A single-phase grounding fault is triggered in the simulation environment, and the single-phase grounding fault judgment model recommendation scheme and the traditional fault diagnosis algorithm are run respectively. The various indicator data of the two algorithms in the process of handling the fault are recorded.

10. The single-phase grounding fault AI comprehensive analysis and judgment system for distribution network according to claim 1 is characterized in that: When evaluating the single-phase ground fault analysis model, the steps for field testing include: Select representative test scenarios in actual distribution networks, including urban distribution networks or rural distribution networks; Deploy fault diagnosis equipment in the test scenario, including fault indicators and smart switches; According to the results and experience of simulation tests, formulate a specific plan for field testing, including fault triggering mode, data collection method and indicator calculation method; A single-phase grounding fault is triggered in the actual distribution network, and the single-phase grounding fault analysis model recommendation scheme and the traditional fault diagnosis algorithm are run respectively. The indicator data of the two algorithms in the process of handling the fault are recorded and compared and analyzed.

Images