An ai-based power system fault prediction and diagnosis system and method

Abstract

The application provides an AI-based power system fault prediction and diagnosis system and method. The system includes: real-time acquisition of current, voltage and grounding resistance data, and identification of arc fault and single-phase grounding fault by using a deep learning model; an arc fault detection module accurately detects low-voltage and medium-voltage arc faults through HAVOK analysis and a multi-level dynamic feature extraction method MDFE; a fault location module locates the fault position by collecting voltage, current and phase data through PMU and combining an optimal positioning algorithm; an AI comprehensive analysis and judgment module analyzes the fault type, position and severity based on reinforcement learning, and generates an intelligent control strategy; and a control execution module automatically performs fault isolation and power restoration according to the generated strategy, and optimizes system operation efficiency, recovery speed and equipment protection through a multi-objective optimization algorithm. The application significantly improves fault diagnosis accuracy, recovery speed, equipment life and system stability, and has a wide application prospect.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system fault prediction and diagnosis, and particularly relates to an AI-based power system fault prediction and diagnosis system and method. BACKGROUND

[0002] With the continuous development of power systems, traditional power fault prediction and diagnosis methods face many challenges. Traditional methods usually rely on manual judgment and empirical data, relying on manual inspection, device alarms, and traditional fault location technology to handle power system faults. Although this method can complete basic fault identification to some extent, its shortcomings are also obvious: on the one hand, fault identification has a certain lag, especially in complex systems, simple manual operation cannot monitor every fault signal in real time; on the other hand, the accuracy of fault diagnosis and positioning is limited by the capabilities of existing monitoring equipment, making it difficult to effectively identify complex fault types, especially in low-voltage and medium-voltage systems, subtle problems such as medium-voltage high-impedance arc faults often cannot be accurately detected and located.

[0003] In addition, existing systems lack intelligent fault prediction capabilities, resulting in a failure to take effective preventive measures before a fault occurs, and often only passive repair processing after the fault occurs. For recovery operations after a fault occurs, most existing systems still rely on manual decision-making and fixed emergency response strategies, lacking the ability of real-time dynamic optimization and automated control.

[0004] Therefore, the present application proposes an AI-based power system fault prediction and diagnosis system, aiming to provide an efficient and intelligent fault prediction, diagnosis, and recovery solution by integrating advanced technologies such as deep learning, reinforcement learning, and phasor measurement units (PMU). SUMMARY

[0005] The purpose of the present application is to provide an AI-based power system fault prediction and diagnosis system and method that can accurately identify fault types, quickly locate fault positions, and generate AI-based control strategies based on real-time monitoring and analysis of power system data, achieving automatic isolation and power restoration, thereby improving the intelligent and automated level of power system fault handling.

[0006] In order to achieve the above-mentioned purpose, the present application realizes the technical scheme as follows:

[0007] An AI-based power system fault prediction and diagnosis system, comprising:

[0008] a fault monitoring module for real-time acquisition of current, voltage, and grounding resistance data, and identification of arc faults and single-phase grounding faults through a deep learning model;

[0009] This module is responsible for real-time monitoring of the operation of the power system. By installing sensors, the system collects key data such as current, voltage and ground resistance. Next, these data will be analyzed by a deep learning model to determine whether there is an arc fault (such as an arc fault in the power system) or a single-phase ground fault (i.e. a single ground fault in the system). The deep learning model is like an "intelligent brain" that can learn different fault patterns and automatically identify faults.

[0010] The arc fault detection module uses an arc fault detection method based on HAVOK analysis to accurately detect low-voltage arc faults and medium-voltage high-impedance arc faults.

[0011] The arc fault detection module uses an arc fault detection method based on HAVOK analysis to accurately detect low-voltage arc faults and medium-voltage high-impedance arc faults. This module includes a low-voltage arc fault detection unit, a medium-voltage arc fault detection unit and a data fusion unit. The low-voltage arc fault detection unit and the medium-voltage arc fault detection unit are responsible for detecting low-voltage and medium-voltage arc faults, respectively, and the data fusion unit fuses the detection results of the two to improve the accuracy and reliability of the detection.

[0012] The fault location module collects real-time voltage, current and phase data of the power system through the phasor measurement unit (PMU) and combines the optimal positioning algorithm to locate the fault and output the fault location.

[0013] The fault location module collects real-time voltage, current and phase data of the power system through the phasor measurement unit (PMU) and combines the optimal positioning algorithm to locate the fault and output the fault location. This module includes a PMU acquisition unit, a data processing unit, a fault location calculation unit and a PMU arrangement optimization unit. The PMU acquisition unit is responsible for real-time acquisition of voltage, current and phase data of the power system, the data processing unit preprocesses the collected data, the fault location calculation unit locates the fault according to the preprocessed data combined with the optimal positioning algorithm, and the PMU arrangement optimization unit optimizes the arrangement of the PMU to improve the accuracy and reliability of the fault location.

[0014] The AI comprehensive analysis module uses reinforcement learning algorithms to analyze the type, location and severity of the fault, and generates control strategies based on real-time and historical data.

[0015] The fault type, location and severity are analyzed using a reinforcement learning algorithm to generate a control strategy based on real-time and historical data. This module includes a data fusion unit, a reinforcement learning unit and a fault classification unit. The data fusion unit is responsible for fusing data from the fault monitoring module, arc fault detection module and fault location module, the reinforcement learning unit analyzes and learns based on the fused data using a reinforcement learning algorithm to generate a control strategy, and the fault classification unit classifies the fault so that the subsequent control execution module can take appropriate control measures according to the fault type.

[0016] The control execution module automatically operates the fault indicator, disconnecting switch and backup power supply equipment according to the control strategy generated by the AI comprehensive judgment module to perform fault isolation and power supply recovery tasks.

[0017] According to the control strategy generated by the AI comprehensive judgment module, the fault indicator, disconnecting switch and backup power supply equipment are automatically operated to perform fault isolation and power supply recovery tasks. This module includes a fault isolation unit, a power supply recovery unit and a device optimization unit. The fault isolation unit is responsible for automatically operating the fault indicator and disconnecting switch according to the control strategy to isolate the fault area from the normal area to prevent the fault from expanding, the power supply recovery unit is responsible for automatically operating the backup power supply equipment to restore power supply to the fault area according to the control strategy, and the device optimization unit is responsible for optimizing the devices during control execution to improve the utilization rate and life of the devices.

[0018] The multi-objective optimization module generates a control strategy through a multi-objective optimization algorithm to optimize multiple objectives such as fault handling efficiency, power supply recovery speed and device wear, and generates a control scheme that adapts to the current power system state.

[0019] A control strategy is generated through a multi-objective optimization algorithm to optimize multiple objectives such as fault handling efficiency, power supply recovery speed and device wear, and a control scheme that adapts to the current power system state is generated. This module includes a target setting unit and a multi-objective optimization objective function. The target setting unit is responsible for setting multiple optimization targets such as fault handling efficiency, power supply recovery speed and device wear, and the multi-objective optimization objective function generates a control strategy using a multi-objective optimization algorithm according to the set optimization targets.

[0020] As a preferred scheme of the present application, the fault monitoring module comprises:

[0021] The data preprocessing unit is used for standardizing, denoising and normalizing the collected current, voltage, grounding resistance and environmental temperature data, and outputs the processed data.

[0022] The fault recognition unit analyzes the processed data based on the trained deep learning model to identify arc faults and single-phase ground faults, and outputs the fault recognition result.

[0023] The deep learning model employs a multi-level dynamic feature extraction model (MDFE), and the calculation formula is as follows:

[0024] y (l) =σ(W (l) *x (l-1) +b (l) Input feature x (l-1) For: x (l-1) =[I t V t ,R g ,T];

[0025] Among them: I t It is a current signal; V t It is a voltage signal; R g T is the grounding resistance signal; W is the temperature signal; (l) b is the weight matrix for the signal characteristics of the power system; (l) σ is the bias term; σ is the nonlinear activation function.

[0026] As a preferred embodiment of the present invention, the arc fault detection module includes:

[0027] The low-voltage arc fault detection unit uses a multi-level dynamic feature extraction model (MDFE) to extract and classify features from the collected current and voltage data, identify arc fault signals in the low-voltage power system, and output arc fault detection results.

[0028] The medium-voltage arc fault detection unit performs nonlinear dynamic analysis on arc faults in medium-voltage systems based on the HAVOK analysis method, and outputs the detection results of medium-voltage high-impedance arc faults.

[0029] The arc fault result data fusion unit merges the detection results of the low-voltage arc fault detection unit and the medium-voltage arc fault detection unit, and outputs the final detection result of the arc fault.

[0030] The HAVOK analysis method uses the following dynamic model formula to analyze arc faults:

[0031]

[0032] Where: V is the voltage signal of the arc fault, reflecting the voltage change caused by the arc fault in the power system; ξ is the damping coefficient of the arc fault signal, representing the attenuation rate of the voltage signal; ω n F(t) represents the natural frequency of the arc fault signal, indicating the frequency of signal oscillation and reflecting the dynamic characteristics of the arc fault; F(t) represents the external driving force, indicating the external disturbance or excitation caused by the fault in the power system.

[0033] As a preferred scheme of the present application, the fault location module comprises:

[0034] A PMU acquisition unit acquires current, voltage and phase information by deploying phasor measurement units (PMUs) at each node of the power distribution network, the current and voltage are acquired by current sensors and voltage sensors, and the phase information is acquired by PMU sensors and calculated and output by the PMU;

[0035] A data processing unit is configured to filter, denoise and standardize the acquired current, voltage and phase data, and output the processed data;

[0036] A fault location calculation unit is configured to calculate and output the accurate location of the power system fault occurrence in combination with the voltage, current and phase information using a positioning algorithm;

[0037] A PMU arrangement optimization unit is configured to calculate and output an optimal PMU arrangement scheme according to the topology and load distribution of the power distribution network to ensure accurate coverage of the monitoring data.

[0038] The PMU data acquisition and fault location calculation formula is as follows:

[0039] Fault Location = f(V abc ,I abc ,θ);

[0040] Wherein: V abc and I abc represent three-phase voltage and current data; and θ represents phase information.

[0041] As a preferred scheme of the present application, the AI comprehensive analysis module comprises:

[0042] A data fusion unit fuses the fault identification result output by the fault monitoring module and the fault location result output by the fault location module, analyzes and outputs the fault type, location and severity in combination with the current, voltage, grounding resistance and other data of the power system;

[0043] A reinforcement learning unit analyzes the type, location and severity of the power system fault through an enhanced algorithm, generates a dynamic control strategy based on real-time system state and historical data;

[0044] A fault classification unit classifies different types of faults including arc faults and single-phase ground faults based on reinforcement learning technology, and generates corresponding control strategies to optimize the fault handling process of the power system;

[0045] The update formula of the reinforcement learning is as follows:

[0046]

[0047] wherein: V(s t ,a t ) is the state-action value function of the power system state s t , after taking action a t , r t+1 is the reward after the current fault diagnosis, γ is the discount factor, α is the learning rate, β is the incremental adjustment factor, and △S t is the power system state change.

[0048] As a preferred scheme of the present application, the control execution module comprises:

[0049] A fault isolation unit automatically operates the fault indicator and the disconnecting switch device according to the fault handling decision generated by the AI comprehensive judgment module, performs the fault isolation operation, and outputs the fault isolation instruction.

[0050] A power supply recovery unit automatically operates the standby power supply to start power supply recovery according to the power system recovery strategy generated by the AI comprehensive judgment module, outputs the power supply recovery signal, and generates the recovery plan and controls the standby power supply to be enabled based on the fault type, location, and system load condition.

[0051] A device optimization unit adjusts the power device load according to the device load optimization strategy, outputs the device adjustment signal, and adjusts the device load based on the real-time system state to improve the device operation efficiency and ensure the stable operation of the power system.

[0052] As a preferred scheme of the present application, the multi-objective optimization module comprises:

[0053] A target setting unit sets multiple targets according to different fault handling requirements, and optimizes the control strategy through the following multi-objective optimization target function:

[0054] f(x)=w1f1(x)+w2f2(x)+…+w n f n (x);

[0055] wherein: f1(x), f2(x), …, f n (x) are different optimization target functions, specifically: f1(x) represents the accuracy of fault diagnosis; f2(x) represents the accuracy of fault positioning; f n (x) represents the minimization of device loss, w1, w2, …, w n are weight coefficients of the optimization target functions.

[0056] By adjusting the weight coefficients of each objective function, the overall control strategy is optimized by considering the power system operation efficiency, fault recovery time and equipment protection factors.

[0057] As a preferred embodiment of the present application, the system further comprises:

[0058] An environment adaptation unit adjusts the control strategy based on the collected environmental temperature data to adapt to different environmental conditions.

[0059] A data updating unit updates the training data set according to real-time data and outputs updated training data for optimizing the fault monitoring and fault detection module.

[0060] An AI-based power system fault prediction and diagnosis method, comprising the following steps:

[0061] Step 1: Real-time acquisition of current, voltage and grounding resistance data in the power system;

[0062] Step 2: Analyze the collected current, voltage and grounding resistance data through a deep learning model to identify arc faults and single-phase grounding faults;

[0063] Step 3: Based on the HAVOK analysis of arc fault detection method, combined with the current and voltage dynamic characteristics extracted by the MDFE model, analyze low-voltage arc faults and medium-voltage high-impedance arc faults;

[0064] Step 4: Use the phasor measurement unit PMU to real-time collect voltage, current and phase data in the power system;

[0065] Step 5: Use the optimal positioning algorithm to determine the fault location according to the collected voltage, current and phase data, and output the fault positioning result;

[0066] Step 6: Use the reinforcement learning algorithm to analyze the fault type, location and severity, and generate a control strategy;

[0067] Step 7: According to the generated control strategy, automatically operate the fault indicator, disconnecting switch, standby power supply equipment, and perform fault isolation and power supply recovery tasks;

[0068] Step 8: Use a multi-objective optimization algorithm to optimize fault handling efficiency, power supply recovery speed, and equipment loss to generate a control scheme that adapts to the current power system state.

[0069] The application can significantly improve the accuracy of fault diagnosis and reduce the delay caused by human error by real-time collection of current, voltage and grounding resistance data through the fault monitoring module and automatic identification of arc fault and single-phase grounding fault using a deep learning model. Secondly, the arc fault detection module uses a method based on HAVOK analysis, which can accurately detect low-voltage arc faults and medium-voltage high-impedance arc faults, making up for the shortcomings of traditional detection methods. The fault location module collects voltage, current and phase data in real time through the phasor measurement unit (PMU) and combines the optimal positioning algorithm to quickly and accurately locate the fault position, shortening the fault recovery response time. At the same time, the AI comprehensive judgment module uses reinforcement learning algorithm to analyze the fault type, location and severity, and generates intelligent control strategy to improve the system's automated decision-making ability. The control execution module automatically operates the fault indicator, disconnecting switch and standby power equipment according to the generated strategy to realize the rapid isolation of faults and power supply recovery, ensuring the continuity and reliability of power supply. Finally, through the multi-objective optimization module, the efficiency of fault handling, the speed of power supply recovery and the equipment loss are optimized, and the optimal control scheme is generated to adapt to the current power system state, improving the overall operation efficiency of the system and reducing equipment loss. The application significantly improves the accuracy, speed and intelligence level of power system fault diagnosis and recovery, reduces power outage time, optimizes equipment usage efficiency, and has wide application prospects and important social and economic benefits. BRIEF DESCRIPTION OF DRAWINGS

[0070] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0071] Among them:

[0072] Figure 1 The system modular structure diagram of the present application is shown in the figure.

[0073] Figure 2 The method flow chart of the present application is shown in the figure. DETAILED DESCRIPTION

[0074] In order to make the purpose, technical scheme and advantages of the embodiments of the present application more clear, the technical scheme of the embodiments of the present application will be described clearly and completely in the following with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all. Based on the described embodiments of the present application, all other embodiments obtained by those skilled in the art belong to the scope of protection of the present application.

[0075] As Figure 1As shown, this is an embodiment of the present invention, which provides an AI-based power system fault prediction and diagnosis system, including:

[0076] (1) Fault monitoring module

[0077] It is used to collect current, voltage and grounding resistance data in real time, and to identify arc faults and single-phase grounding faults through deep learning models;

[0078] The fault monitoring module includes:

[0079] The data preprocessing unit is used to standardize, denoise, and normalize the collected current, voltage, grounding resistance, and ambient temperature data, and output the processed data.

[0080] The fault identification unit analyzes and processes the data based on a trained deep learning model, identifies arc faults and single-phase grounding faults, and outputs the fault identification results.

[0081] The deep learning model employs a multi-level dynamic feature extraction model (MDFE), and the calculation formula is as follows:

[0082] y (l) =σ(W (l) *x (l-1) +b (l) Input feature x (l-1) For: x (l-1) =[I t V t ,R g ,T];

[0083] Among them: I t It is a current signal; V t It is a voltage signal; R g T is the grounding resistance signal; W is the temperature signal; (l) b is the weight matrix for the signal characteristics of the power system; (l) σ is the bias term; σ is the nonlinear activation function.

[0084] In this embodiment, this module is responsible for collecting real-time data on current, voltage, and grounding resistance in the power system. Using this data, the system employs a deep learning model to automatically identify arc faults and single-phase grounding faults. Simply put, the fault monitoring module acquires the operating status of the power system through real-time sensors and uses intelligent algorithms to determine whether an arc or grounding fault has occurred.

[0085] Some code optimizations are shown below:

[0086]

[0087]

[0088] (2) Arc fault detection module

[0089] The arc fault detection method based on HAVOK analysis is adopted to accurately detect low-voltage arc faults and medium-voltage high-impedance arc faults.

[0090] The arc fault detection module comprises:

[0091] A low-voltage arc fault detection unit adopts a multi-level dynamic feature extraction model MDFE to extract and classify the collected current and voltage data, identify arc fault signals in a low-voltage power system, and output arc fault detection results.

[0092] A medium-voltage arc fault detection unit performs nonlinear dynamic analysis on arc faults in a medium-voltage system based on the HAVOK analysis method, and outputs detection results of medium-voltage high-impedance arc faults.

[0093] An arc fault result data fusion unit fuses the detection results of the low-voltage arc fault detection unit and the medium-voltage arc fault detection unit, and outputs the final detection results of arc faults.

[0094] The HAVOK analysis method uses the following dynamic model formula to analyze arc faults:

[0095]

[0096] Where: V is the voltage signal of the arc fault, reflecting the voltage change caused by the arc fault in the power system; ξ is the damping coefficient of the arc fault signal, indicating the decay rate of the voltage signal; ω n is the natural frequency of the arc fault signal, indicating the frequency of signal oscillation, embodying the dynamic characteristics of the arc fault; F(t) is the external driving force, indicating the external disturbance or excitation caused by the fault in the power system.

[0097] In this embodiment, this module detects arc faults through two techniques. One is a multi-level dynamic feature extraction method for low-voltage power systems, and the other is a HAVOK analysis method for detecting medium-voltage high-impedance arc faults in medium-voltage systems. Through these methods, arc faults can be accurately identified and fault information can be output. In other words, this module can effectively distinguish different types of arc faults and ensure accurate response to various faults in the power system.

[0098] (3) Fault location module

[0099] The phasor measurement unit PMU collects voltage, current, and phase data of the power system in real time, and combines the optimal positioning algorithm to locate the fault, and outputs the fault location.

[0100] The fault location module comprises:

[0101] A PMU acquisition unit acquires current, voltage, and phase information by deploying phasor measurement units (PMUs) at each node of the power distribution network. The current and voltage are acquired by current and voltage sensors, and the phase information is acquired by PMU sensors and calculated and output by the PMU.

[0102] A data processing unit is configured to filter, denoise, and standardize the acquired current, voltage, and phase data, and output the processed data.

[0103] A fault location calculation unit combines voltage, current, and phase information to calculate and output the accurate location of a power system fault using a positioning algorithm.

[0104] A PMU arrangement optimization unit calculates and outputs an optimal PMU arrangement scheme based on the topology and load distribution of the power distribution network to ensure accurate coverage of the monitoring data.

[0105] The PMU data acquisition and fault location calculation formula is as follows:

[0106] Fault Location = f(V abc ,I abc ,θ);

[0107] Where V abc and I abc represent three-phase voltage and current data, and θ represents phase information.

[0108] In this embodiment, the fault location module acquires real-time current, voltage, and phase data in the power system through PMUs (phasor measurement units) installed at each node. Using an optimized positioning algorithm, the system can quickly and accurately determine the specific location of the fault. Specifically, the PMU collects data from different sensors, then analyzes the data through an algorithm to accurately identify the area where the fault occurred.

[0109] Some code optimizations are as follows:

[0110]

[0111] (4) AI comprehensive analysis module

[0112] Using reinforcement learning algorithms to analyze fault types, locations, and severity, and generate control strategies based on real-time and historical data.

[0113] The AI comprehensive analysis module comprises:

[0114] a data fusion unit that fuses the fault identification result output by the fault monitoring module and the fault location result output by the fault location module, combines current, voltage, grounding resistance and other data of the power system, analyzes and outputs the fault type, location and severity;

[0115] a reinforcement learning unit that analyzes the type, location and severity of the power system fault through a reinforcement algorithm, generates a dynamic control strategy based on real-time system state and historical data;

[0116] a fault classification unit that classifies different types of faults including arc fault and single-phase ground fault based on reinforcement learning technology, and generates corresponding control strategies to optimize the fault handling process of the power system;

[0117] The update formula of the reinforcement learning is as follows:

[0118]

[0119] wherein V(s t ,a t ) is the state-action value function of the power system state s t under action a t , r t+1 is the reward after current fault diagnosis, γ is the discount factor, α is the learning rate, β is the incremental adjustment factor, and △S t is the state change of the power system.

[0120] In this embodiment, the AI comprehensive judgment module uses a reinforcement learning algorithm to comprehensively analyze the type, location and severity of the fault. The role of this module is to dynamically generate the best solution to handle the fault based on real-time data and historical data. For example, if the system detects an arc fault or a ground fault, it will automatically calculate the most appropriate handling method to reduce power outage time and system loss.

[0121] (5) Control execution module

[0122] According to the control strategy generated by the AI comprehensive judgment module, the fault indicator, disconnecting switch and standby power supply equipment are automatically operated to perform fault isolation and power supply recovery tasks;

[0123] The control execution module comprises:

[0124] a fault isolation unit that automatically operates the fault indicator and disconnecting switch equipment according to the fault handling decision generated by the AI comprehensive judgment module, performs fault isolation operation, and outputs fault isolation instructions;

[0125] The power supply recovery unit automatically operates the standby power supply to start power supply recovery according to the power system recovery strategy generated by the AI comprehensive judgment module, and outputs a power supply recovery signal. The power system recovery strategy is based on the fault type, location, and system load condition to generate a recovery plan and control the standby power supply to start.

[0126] The device optimization unit adjusts the power device load according to the device load optimization strategy, and outputs a device adjustment signal. The device load optimization strategy is based on real-time system state to adjust the device load, improve the device operation efficiency, and ensure the stable operation of the power system.

[0127] In this embodiment, the control execution module automatically operates related devices such as fault indicators, disconnectors, and standby power supplies according to the control strategy generated by the AI comprehensive judgment module. The automatic start of these devices can ensure that the fault area is quickly cut off and power supply is restored after a fault occurs. The core function of this module is to quickly execute fault isolation and power recovery operations according to the AI strategy, reducing power interruption time.

[0128] (6) Multi-objective optimization module

[0129] The control strategy is generated by a multi-objective optimization algorithm to optimize the efficiency of fault handling, power supply recovery speed, and device loss, and to generate a control scheme that adapts to the current power system state.

[0130] The multi-objective optimization module includes:

[0131] The target setting unit sets multiple targets according to different fault handling requirements, and optimizes the control strategy through the following multi-objective optimization target function:

[0132] f(x) = w1f1(x) + w2f2(x) + … + w n f n (x);

[0133] Where: f1(x), f2(x), …, f n (x) are different optimization target functions, specifically: f1(x) represents the accuracy of fault diagnosis; f2(x) represents the accuracy of fault location; f n (x) represents the minimization of device loss, w1, w2, …, w n are the weight coefficients of each optimization target function.

[0134] By adjusting the weight coefficients of each target function, the power system operation efficiency, fault recovery time, and device protection factors are considered comprehensively to optimize the overall control strategy.

[0135] In this embodiment, the module optimizes the control strategy in the power system through a multi-objective optimization algorithm, with objectives including improving fault handling efficiency, accelerating power supply recovery speed, and reducing equipment loss, etc. It sets multiple optimization objectives according to different needs, and finds the best control scheme by adjusting the weights between these objectives. In short, the role of this module is to optimize the operating efficiency of the power system while ensuring the protection of equipment.

[0136] Part of the code optimization is as follows:

[0137]

[0138] Further, the system further comprises:

[0139] An environment adaptation unit adjusts the control strategy based on the collected environmental temperature data to adapt to different environmental conditions.

[0140] A data updating unit updates the training data set according to real-time data, outputs the updated training data, and uses it to optimize the fault monitoring and fault detection module.

[0141] As Figure 2 shown, another embodiment of the present application provides an AI-based power system fault prediction and diagnosis method, comprising the following steps:

[0142] Step 1: Real-time acquisition of current, voltage and grounding resistance data in the power system;

[0143] Through the data acquisition unit in the fault monitoring module, real-time acquisition of current, voltage and grounding resistance data in the power system is performed. The collected data will be used for subsequent fault identification and diagnosis.

[0144] Step 2: Analyze the collected current, voltage and grounding resistance data through a deep learning model to identify arc faults and single-phase ground faults;

[0145] The fault identification unit in the fault monitoring module inputs the collected current, voltage and grounding resistance data into the trained deep learning model for fault identification. The deep learning model can automatically learn the features in the data and accurately identify arc faults and single-phase ground faults.

[0146] Step 3: Based on the HAVOK analysis of the arc fault detection method, combined with the current and voltage dynamic characteristics extracted by the MDFE model, analyze low-voltage arc faults and medium-voltage high-impedance arc faults;

[0147] The low-voltage arc fault detection unit and the medium-voltage arc fault detection unit in the arc fault detection module use the arc fault detection method based on HAVOK analysis, combined with the current and voltage dynamic features extracted by the MDFE (Multi-scale Dynamic Feature Extraction) model, to accurately detect low-voltage arc faults and medium-voltage high-impedance arc faults. Then, the data fusion unit fuses the detection results of the two, improving the accuracy and reliability of the detection.

[0148] Step 4: Real-time collection of voltage, current, and phase data in the power system using the phasor measurement unit (PMU);

[0149] The PMU acquisition unit in the fault location module collects voltage, current, and phase data in the power system in real time. The PMU can accurately measure the voltage, current, and phase of the power system, providing accurate data support for fault location.

[0150] Step 5: Using the optimal positioning algorithm, determine the fault location based on the collected voltage, current, and phase data, and output the fault location result;

[0151] The fault location calculation unit in the fault location module determines the fault location based on the collected voltage, current, and phase data, combined with the optimal positioning algorithm. The optimal positioning algorithm can consider multiple factors such as fault type, fault distance, power system topology, etc., accurately determine the fault location, and output the fault location result.

[0152] Step 6: Using the reinforcement learning algorithm, analyze the fault type, location, and severity, and generate a control strategy;

[0153] The reinforcement learning unit in the AI comprehensive analysis module uses the reinforcement learning algorithm to analyze the fault type, location, and severity based on the data provided by the fault monitoring module, arc fault detection module, and fault location module. The reinforcement learning algorithm can automatically learn the fault handling strategy and generate the optimal control strategy based on real-time and historical data.

[0154] Step 7: According to the generated control strategy, automatically operate the fault indicator, disconnecting switch, and backup power equipment, and perform fault isolation and power supply recovery tasks;

[0155] The fault isolation unit and the power supply recovery unit in the control execution module automatically operate the fault indicator, the disconnecting switch and the standby power supply equipment according to the control strategy generated by the AI comprehensive judgment module. The fault indicator can intuitively display the fault location, facilitating the staff to quickly locate and handle the fault; the disconnecting switch can isolate the fault area from the normal area to prevent the fault from expanding; and the standby power supply equipment can restore power supply to the fault area, thereby shortening the power outage time.

[0156] Step 8: Using a multi-objective optimization algorithm, the fault handling efficiency, power supply recovery speed and equipment loss are optimized, and a control scheme suitable for the current power system state is generated.

[0157] The multi-objective optimization target function in the multi-objective optimization module uses a multi-objective optimization algorithm to generate a control strategy according to the set fault handling efficiency, power supply recovery speed and equipment loss. The multi-objective optimization algorithm can comprehensively consider the mutual influence between multiple optimization objectives, generate a control scheme suitable for the current power system state, and improve the overall performance and economic benefits of the power system.

[0158] In summary, the present application significantly improves the performance of the power system in fault detection, positioning, recovery and optimization through the integration of deep learning, reinforcement learning, phasor measurement unit (PMU) and multi-objective optimization algorithm. The system can accurately identify arc faults and single-phase ground faults, accurately locate the fault occurrence position, and generate intelligent fault handling strategies to realize automatic isolation and power supply recovery. In addition, the multi-objective optimization algorithm used makes the system perform well in improving fault handling efficiency, speeding up power supply recovery and reducing equipment loss. Compared with traditional technologies, the present application can effectively shorten the fault detection time, improve the fault positioning accuracy, reduce the equipment loss, and significantly improve the overall operation efficiency and equipment life of the power system. This technology has wide application prospects, not only can improve the intelligent level of the power system, but also can save costs, improve operation reliability and ensure the stability of power supply for the power industry.

[0159] In the description of the present application, the description of the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, the skilled in the art can combine and combine the different embodiments or examples described in the present application and the features of the different embodiments or examples without contradiction.

[0160] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process. Furthermore, the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functionality involved.

[0161] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in this application, and these should all be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An AI-based power system fault prediction and diagnosis system, characterized by, Comprise: Fault monitoring module for real-time acquisition of current, voltage, ground resistance data, and identification of arc fault and single-phase ground fault through deep learning model; Data preprocessing unit for standardization, denoising, and normalization processing of collected current, voltage, ground resistance, and environmental temperature data, and output of processed data; Fault identification unit based on trained deep learning model to analyze processed data, identify arc fault and single-phase ground fault, and output fault identification result; The deep learning model uses a multi-level dynamic feature extraction model MDFE; Arc fault detection module adopts arc fault detection method based on HAVOK analysis to accurately detect low-voltage arc fault and medium-voltage high-impedance arc fault; Low-voltage arc fault detection unit uses multi-level dynamic feature extraction model MDFE to extract and classify features of collected current and voltage data, identify arc fault signals in low-voltage power system, and output arc fault detection result; Medium-voltage arc fault detection unit performs nonlinear dynamic analysis of arc fault in medium-voltage system based on HAVOK analysis method, and outputs detection result of medium-voltage high-impedance arc fault; Arc fault result data fusion unit fuses detection results of low-voltage arc fault detection unit and medium-voltage arc fault detection unit, and outputs final detection result of arc fault; Fault location module acquires voltage, current, and phase data of power system in real time through a phase measurement unit PMU, and performs fault location combined with optimal positioning algorithm, and outputs fault location; AI comprehensive analysis module uses reinforcement learning algorithm to analyze fault type, location, and severity, and generates control strategy based on real-time and historical data; Data fusion unit fuses fault identification result output by fault monitoring module and fault location result output by fault location module, analyzes and outputs fault type, location, and severity based on current, voltage, and ground resistance data of power system; Reinforcement learning unit analyzes type, location, and severity of power system fault through reinforcement algorithm, and generates dynamic control strategy based on real-time system state and historical data; Fault classification unit classifies different types of faults including arc fault and single-phase ground fault based on reinforcement learning technology, and generates corresponding control strategy to optimize fault handling process of power system; Control execution module automatically operates fault indicator, disconnecting switch, and backup power supply equipment according to control strategy generated by AI comprehensive analysis module, and performs fault isolation and power supply recovery task; Multi-objective optimization module generates control strategy through multi-objective optimization algorithm to optimize fault handling efficiency, power supply recovery speed, and equipment loss, and generates control scheme adaptive to current power system state. 2.The AI-based power system fault prediction and diagnosis system of claim 1, wherein, The deep learning model uses a multi-level dynamic feature extraction model MDFE, and the calculation formula is as follows: ; input features are: ; wherein: is a current signal; is a voltage signal; is a ground resistance signal; is a temperature signal; is a weight matrix of power system signal features; is a bias term; is a nonlinear activation function. 3.The AI-based power system fault prediction and diagnosis system of claim 1, wherein, The HAVOK analysis method uses the following dynamic model formula to analyze arc fault: ； wherein: is the voltage signal of the arc fault, reflecting the voltage change caused by the arc fault in the power system; is the damping coefficient of the arc fault signal, indicating the decay rate of the voltage signal; is the natural frequency of the arc fault signal, indicating the frequency of the signal oscillation, embodying the dynamic characteristics of the arc fault; is the external driving force, indicating the external disturbance or excitation caused by the fault in the power system. 4.The AI-based power system fault prediction and diagnosis system of claim 1, wherein, The fault location module comprises: A PMU acquisition unit acquires current, voltage, and phase information by deploying phasor measurement units (PMUs) at each node of the power distribution network. The current and voltage are acquired by current and voltage sensors, and the phase information is acquired by a PMU sensor. The PMU calculates and outputs the data. A data processing unit filters, denoises, and standardizes the acquired current, voltage, and phase data, and outputs the processed data. A fault location calculation unit calculates and outputs the accurate location of a power system fault using a positioning algorithm based on voltage, current, and phase information. A PMU arrangement optimization unit calculates and outputs the optimal PMU arrangement scheme based on the topology and load distribution of the power distribution network to ensure accurate coverage of the monitoring data. The PMU data acquisition and fault location calculation formula is as follows: ； wherein: and represent three-phase voltage and current data, respectively; is the phase information. 5.The AI-based power system fault prediction and diagnosis system of claim 1, wherein, The reinforcement learning update formula is as follows: ； wherein: is the power system state action is taken post-action state-value function, is the reward after the current fault diagnosis, is the discount factor, is the learning rate, is the incremental adjustment factor, is the power system state change. 6.The AI-based power system fault prediction and diagnosis system of claim 1, wherein, The control execution module includes: A fault isolation unit automatically operates fault indicators and disconnecting switch devices based on the fault handling decisions generated by the AI comprehensive analysis module, executes fault isolation operations, and outputs fault isolation instructions. A power supply recovery unit automatically operates standby power sources to restore power supply based on the power system recovery strategy generated by the AI comprehensive analysis module, outputs power supply recovery signals, and generates a recovery plan based on the fault type, location, and system load to control the standby power source to be enabled. A device optimization unit adjusts the load of power equipment based on the device load optimization strategy, outputs device adjustment signals, and adjusts the load of the device based on the real-time system state to improve the operating efficiency of the device and ensure the stable operation of the power system. 7.The AI-based power system fault prediction and diagnosis system of claim 1, wherein, The multi-objective optimization module includes: A target setting unit sets multiple targets based on different fault handling requirements and optimizes the control strategy through the following multi-objective optimization target function: ； wherein: are different optimization objective functions, in particular: denotes the accuracy of the fault diagnosis; denotes the precision of the fault localization; denotes the minimization of the equipment wear, are weight coefficients for the respective optimization objective functions; By adjusting the weight coefficients of each target function, the power system operating efficiency, fault recovery time, and device protection factors are considered comprehensively to optimize the overall control strategy. 8.The AI-based power system fault prediction and diagnosis system of claim 1, wherein, The system also includes: An environmental adaptation unit adjusts the control strategy based on the collected environmental temperature data to adapt to different environmental conditions. A data update unit updates the training data set based on real-time data and outputs updated training data for optimizing the fault monitoring and detection module.

9. The fault prediction and diagnosis method of the AI-based power system fault prediction and diagnosis system according to any one of claims 1-8, characterized in that, The method includes the following steps: Step 1: Real-time acquisition of current, voltage, and grounding resistance data in the power system; Step 2: Analysis of the collected current, voltage, and grounding resistance data using a deep learning model to identify arc faults and single-phase ground faults; Step 3: Analysis of low-voltage arc faults and medium-voltage high-impedance arc faults based on the HAVOK analysis arc fault detection method combined with the current and voltage dynamic characteristics extracted by the MDFE model; Step 4: Real-time acquisition of voltage, current, and phase data in the power system using a phasor measurement unit (PMU); Step 5: Determination of the fault location based on the collected voltage, current, and phase data using an optimal positioning algorithm, and output of the fault location result. Step 6: Use reinforcement learning algorithm to analyze fault type, location and severity, and generate control strategy; Step 7: According to the generated control strategy, automatically operate fault indicator, disconnecting switch, standby power supply equipment, and perform fault isolation and power supply recovery tasks; Step 8: Use multi-objective optimization algorithm to optimize fault handling efficiency, power supply recovery speed, and equipment loss, and generate control scheme suitable for current power system state.

Images