Electric power system fault diagnosis and early warning system based on AI

Abstract

The invention relates to the technical field of fault diagnosis, in particular to an AI-based power system fault diagnosis and early warning system, which comprises a data acquisition module, a data processing module, a fault mode identification module, a fault diagnosis module, an early warning module and a decision support module, multi-source data related to power system fault diagnosis are collected in real time, the data processing module carries out preprocessing and feature extraction on the collected data, and the fault mode recognition module constructs a fault mode recognition model based on a long short-term memory network, a random forest and a convolutional neural network and recognizes a fault mode. The fault diagnosis module carries out fault diagnosis based on real-time data, the early warning module carries out early warning on possible faults in advance according to a diagnosis result, and the decision support module generates an emergency scheduling scheme according to fault diagnosis and early warning information. And the stability and safety of the power system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to an AI-based power system fault diagnosis and early warning system. Background Art

[0002] As the scale of power systems continues to expand and the level of intelligence increases, the operation of power systems becomes more complex, and fault diagnosis and emergency response capabilities are crucial to the safety, stability and reliability of power systems. In recent years, with the development of big data and artificial intelligence technologies, more and more data-driven fault diagnosis methods have been applied.

[0003] Although power system fault diagnosis methods based on machine learning and deep learning have been gradually applied, existing technologies still face some challenges. First, existing fault diagnosis methods often have great limitations when dealing with complex fault modes, and the recognition ability of a single model cannot adapt to diverse fault types and complex operating environments. Even with the use of machine learning and deep learning methods, the recognition of fault modes still has the problem of insufficient accuracy, especially for nonlinear and time-varying fault modes. Secondly, the emergency response capability of existing technologies after fault diagnosis is weak, and it is impossible to quickly generate a reasonable emergency dispatch plan based on the diagnosis results, resulting in the inability of the power system to recover in a timely and effective manner when a fault occurs, increasing the risk of system outage. Summary of the invention

[0004] The present invention provides an AI-based power system fault diagnosis and early warning system.

[0005] The AI-based power system fault diagnosis and early warning system includes a data acquisition module, a data processing module, a fault pattern recognition module, a fault diagnosis module, an early warning module and a decision support module, among which;

[0006] The data acquisition module collects multi-source data related to power system fault diagnosis in real time by arranging sensors in the power system, and the multi-source data includes partial discharge, mechanical properties, sheath circulation, temperature and vibration;

[0007] The data processing module is used to preprocess the collected multi-source data, including denoising, missing value filling and data standardization, and perform feature extraction on the preprocessed multi-source data;

[0008] The fault mode recognition module constructs a fault mode recognition model based on the extracted features, and the fault mode recognition model includes a long short-term memory network sub-model, a random forest sub-model and a convolutional neural network sub-model;

[0009] The fault diagnosis module performs real-time fault diagnosis on the power system based on multi-source data monitored in real time by means of a constructed fault pattern recognition model;

[0010] The early warning module warns of possible faults in advance based on the fault diagnosis results and generates an early warning report;

[0011] The decision support module generates a fault emergency dispatch plan using an optimization algorithm based on the information output by the fault diagnosis module and the early warning module.

[0012] Optionally, the data acquisition module includes:

[0013] Arrange multiple sensors at key nodes and equipment of the power system, wherein the sensors include partial discharge sensors, mechanical property sensors, sheath circulation sensors, temperature sensors and vibration sensors;

[0014] The partial discharge sensor is installed on the main lines and equipment in the power system to monitor the partial discharge data in real time;

[0015] The mechanical characteristic sensor is arranged on the switch cabinet circuit breaker and is used to monitor the operating status of the circuit breaker in real time;

[0016] The sheath circulation sensor is used to monitor the grounding current leakage condition of the cable;

[0017] The temperature sensor is arranged on the power equipment (such as transformer, switchgear, etc.) to monitor the temperature change of the equipment;

[0018] The vibration sensor is installed on key equipment (such as generators, transformers, etc.) to collect vibration signals of the equipment in real time;

[0019] After being collected by sensors, the multi-source data are transmitted to the data processing module in real time through wired or wireless communication technology.

[0020] Optionally, the data processing module includes:

[0021] De-noising the received multi-source data;

[0022] Linear interpolation is used to fill missing values ​​in the denoised data;

[0023] Standardize the processed data to eliminate the dimensional differences between different sensors and data types;

[0024] The preprocessed multi-source data is reduced in dimension by principal component analysis method;

[0025] Feature extraction is performed on the data after principal component analysis, and key statistical features of the data, including mean, variance and kurtosis, are calculated through frequency domain analysis and time domain analysis.

[0026] Optionally, the long short-term memory network sub-model in the fault mode recognition module includes:

[0027] Construct the structure of the LSTM sub-model, including the input layer, LSTM network layer, and output layer.

[0028] Train the LSTM sub-model, use historical fault data for supervised learning, and use the back-propagation algorithm to optimize the model weights;

[0029] Use Adam optimization algorithm to adjust parameters;

[0030] After the training is completed, the trained long short-term memory network sub-model is used to perform fault diagnosis on the real-time data in the power system and output the corresponding fault mode and occurrence probability.

[0031] Optionally, the random forest sub-model in the fault mode recognition module includes:

[0032] Build a random forest sub-model. Random forest is an ensemble learning algorithm composed of multiple decision trees.

[0033] Train the random forest sub-model and use the training dataset for supervised learning;

[0034] Perform cross-validation on the trained random forest sub-model to evaluate its performance;

[0035] The prediction results of each decision tree are combined through a voting mechanism to obtain the final prediction result;

[0036] The trained random forest sub-model is used for real-time fault pattern recognition. The real-time characteristic data of the power system is input, and the model outputs the fault diagnosis results according to the learning results in the training process.

[0037] Optionally, the convolutional neural network sub-model in the fault mode recognition module includes:

[0038] Construct the structure of the convolutional neural network sub-model, including the input layer, convolution layer, pooling layer, fully connected layer and output layer;

[0039] Feature extraction is performed using convolution operations, which are used to identify spatially local features in the input data;

[0040] The pooling layer downsamples the feature map output by the convolutional layer to reduce the dimension of the data and the amount of calculation;

[0041] The fully connected layer combines the features extracted by the convolutional layer and the pooling layer to generate the final classification result;

[0042] The cross entropy loss function is used for training to calculate the difference between the model output and the true label;

[0043] The trained convolutional neural network sub-model is used for real-time fault pattern recognition, which inputs the feature data of real-time monitoring and outputs the fault diagnosis results.

[0044] Optionally, the fault diagnosis module includes:

[0045] Receiving preprocessed multi-source data output by a data processing module;

[0046] The preprocessed multi-source data is input into a fault pattern recognition model, which includes a long short-term memory network sub-model, a random forest sub-model and a convolutional neural network sub-model;

[0047] Through the long short-term memory network sub-model, the modeling ability of time series data is used to analyze the time series correlation in the input data and identify possible failure modes;

[0048] Through the random forest sub-model, multiple features of the input data are combined to classify and select features, further improving the accuracy of fault diagnosis;

[0049] Through the convolutional neural network sub-model, the local pattern of the input data is analyzed by using its automatic extraction ability of local features to capture potential fault signs in the system;

[0050] The output results of the long short-term memory network sub-model, the random forest sub-model and the convolutional neural network sub-model are integrated, and the final fault diagnosis result is obtained through weighted average or voting mechanism;

[0051] Based on the diagnosis results, the fault types are classified, such as normal, minor fault, medium fault, major fault, etc., and a fault diagnosis report is generated;

[0052] If a fault is detected, the fault diagnosis module transmits the fault information to the early warning module and the decision support module.

[0053] Optionally, the early warning module includes:

[0054] Receive the real-time fault diagnosis results output by the fault diagnosis module, the fault diagnosis results including the current power system status and fault type;

[0055] According to the fault diagnosis results, identify the possible fault trends in the current power system and analyze the probability and severity of the fault;

[0056] Use the long short-term memory network sub-model to predict the time series of power system operation data, and combine the fault diagnosis results to evaluate the occurrence time, type and risk level of potential faults;

[0057] Based on the prediction results of the long short-term memory network sub-model and the historical fault data of the power system, the possibility of fault occurrence is calculated and a fault warning signal is generated;

[0058] Based on the possibility of failure and the predicted failure type, determine whether an early warning is needed. If the probability of failure exceeds the preset threshold, a warning report is generated.

[0059] Send early warning reports to maintenance personnel and relevant management departments to provide a basis for timely preventive measures and emergency response.

[0060] Optionally, the decision support module includes:

[0061] Receive the real-time fault diagnosis results output by the fault diagnosis module and the fault warning signal output by the warning module;

[0062] Comprehensively analyze fault diagnosis results and warning information to assess the potential impact of faults on the power system;

[0063] According to the evaluation results, a fault emergency dispatch model is constructed, and the fault emergency dispatch model is solved by particle swarm optimization algorithm.

[0064] Based on the optimal solution obtained by the particle swarm optimization algorithm, a fault emergency dispatch plan is generated, which includes specific fault handling procedures, dispatch sequence, backup power configuration, maintenance personnel allocation, etc. The generated fault emergency dispatch plan is transmitted to the control center for dispatchers to execute.

[0065] Beneficial effects of the present invention:

[0066] In the present invention, the fault pattern recognition module combines the long short-term memory network sub-model, the random forest sub-model and the convolutional neural network sub-model, which can effectively identify and analyze different types of fault modes in the power system. The three sub-models respectively use their advantages to perform multi-dimensional analysis. The long short-term memory network sub-model can capture the dynamic changes of time series data, the random forest sub-model provides powerful classification capabilities, and the convolutional neural network sub-model can process complex spatial features, thereby enhancing the system's ability to recognize fault patterns and reducing the risks of misdiagnosis and missed diagnosis.

[0067] The present invention uses a particle swarm optimization algorithm to solve the fault emergency dispatching scheme. The decision support module can generate the optimal dispatching scheme based on the fault diagnosis and warning information. The scheme not only takes into account the time and cost of fault handling, but also optimizes the system recovery speed and the utilization efficiency of power resources to the greatest extent. Through real-time feedback and dynamic adjustment, the system can quickly respond to sudden faults in the power system and provide targeted emergency dispatching schemes to ensure the stability and sustainability of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0069] Figure 1 Schematic diagram of a system flow of an embodiment of the present invention. DETAILED DESCRIPTION

[0070] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0071] It should be noted that the references to "one embodiment", "an embodiment", "an exemplary embodiment", "some embodiments" and the like in the specification indicate that the embodiments described may include specific features, structures or characteristics, but not every embodiment may include the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the art to implement such feature, structure or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0072] In general, a term can be understood, at least in part, from its use in context. For example, depending, at least in part, on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending, at least in part, on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0073] like Figure 1As shown, the AI-based power system fault diagnosis and early warning system includes a data acquisition module, a data processing module, a fault pattern recognition module, a fault diagnosis module, an early warning module and a decision support module, wherein;

[0074] The data acquisition module collects multi-source data related to power system fault diagnosis in real time by arranging sensors in the power system. The multi-source data includes partial discharge, mechanical properties, sheath circulation, temperature and vibration.

[0075] The data processing module is used to preprocess the collected multi-source data, including denoising, missing value filling and data standardization, and to extract features from the preprocessed multi-source data;

[0076] The fault pattern recognition module builds a fault pattern recognition model based on the extracted features. The fault pattern recognition model includes a long short-term memory network sub-model, a random forest sub-model and a convolutional neural network sub-model;

[0077] The fault diagnosis module uses the constructed fault pattern recognition model to perform real-time fault diagnosis on the power system based on multi-source data monitored in real time;

[0078] The early warning module warns of possible faults in advance based on the fault diagnosis results and generates a warning report;

[0079] The decision support module generates a fault emergency dispatch plan using an optimization algorithm based on the information output by the fault diagnosis module and the early warning module.

[0080] The data acquisition module includes:

[0081] Multiple sensors are deployed at key nodes and equipment in the power system, including partial discharge sensors, mechanical property sensors, sheath circulation sensors, temperature sensors and vibration sensors, to ensure comprehensive monitoring of the operating status of the power system;

[0082] Partial discharge sensors are installed on the main lines and equipment in the power system to collect partial discharge data in real time, reflect the fault conditions of the equipment, and avoid serious accidents. Mechanical characteristic sensors can accurately monitor the operating status of the circuit breaker when the switch cabinet is in operation;

[0083] Sheath circulation sensors are arranged at important nodes in the cable system to monitor the ground current leakage in real time. Once a cable fault causes the ground leakage current to increase, the system will immediately issue an alarm.

[0084] Temperature sensors are placed on power equipment (such as transformers, switchgear, etc.) to monitor the temperature changes of the equipment and avoid equipment failures caused by overheating. Temperature sensors use thermocouples or thermistors, which have high accuracy and a wide temperature monitoring range.

[0085] Vibration sensors are installed on key equipment (such as generators, transformers, etc.) to collect vibration signals of the equipment in real time. Vibration sensors use piezoelectric accelerometers or optical fiber sensors to capture mechanical failures or abnormal conditions by monitoring the vibration frequency and amplitude changes of the equipment.

[0086] After multi-source data is collected by sensors, it is transmitted to the data processing module in real time through wired or wireless communication technology. Data transmission can adopt industrial standard protocols such as Modbus, CAN, MQTT, etc. to ensure the stability and reliability of data transmission.

[0087] The data processing module includes:

[0088] The received multi-source data is denoised to remove noise signals caused by electromagnetic interference, sensor errors or environmental noise. The specific denoising method is as follows:

[0089] Use Kalman filtering to denoise the signal. Kalman filtering is a recursive filtering algorithm that can effectively estimate the system state and remove noise. The calculation is:

[0090]

[0091] in, is the predicted value of the last state, is the estimated value of the current state, K k is the Kalman gain, y k is the measured value, H k is the measurement matrix;

[0092] The linear interpolation method is used to fill the missing values ​​of the denoised data. The linear interpolation formula is expressed as:

[0093]

[0094] Where x(t) is the interpolation result of the missing value, t0, t1 are the timestamps of the known data points, x(t0), x(t1) are the known data values ​​of the corresponding timestamps, and t is the timestamp of the missing value;

[0095] This interpolation method linearly estimates the missing data through known data points to ensure data continuity;

[0096] The processed data is standardized to eliminate the dimensional differences between different sensors and data types. The standardization method uses Z-score standardization, which is expressed as:

[0097]

[0098] Among them, z iis the standardized data value, x i is the original data value, μ is the mean of the data, σ is the standard deviation of the data. Through Z-score standardization, the data is converted into a standard normal distribution with zero mean and unit variance, thereby eliminating the dimensional differences of different data dimensions;

[0099] The preprocessed multi-source data is reduced in dimension by principal component analysis. The goal of principal component analysis is to transform the original features into new irrelevant features through linear transformation, which is calculated as:

[0100] Z = X·W;

[0101] Among them, Z is the data matrix after dimension reduction, X is the original data matrix, and W is the principal component matrix (composed of eigenvalues ​​and eigenvectors);

[0102] Feature extraction is performed on the data after principal component analysis. Through frequency domain analysis and time domain analysis, the key statistical features of the data are calculated, including mean, variance and kurtosis. The calculation formula is as follows:

[0103]

[0104] Among them, x i is the data point, N is the total number of data points, Mean is the mean, Variance is the variance, Std is the standard deviation, and Kurtosis is the kurtosis.

[0105] The long short-term memory network sub-model in the fault mode recognition module includes:

[0106] The structure of the long short-term memory network sub-model is constructed, including an input layer, a long short-term memory network layer and an output layer. The input layer receives the feature vector transmitted from the data processing module, the long short-term memory network layer is used to capture the long-term dependencies in the time series data, and the output layer is used to output the fault diagnosis results.

[0107] The long short-term memory network sub-model is trained, and supervised learning is performed using historical fault data. The model weights are optimized using the back propagation algorithm. The goal of the training process is to minimize the prediction error. The loss function used is the mean square error (MSE), and the formula is as follows:

[0108]

[0109] Where N is the number of samples, y i is the true label, is the predicted value;

[0110] The Adam optimization algorithm is used to adjust parameters. The Adam optimization algorithm improves the convergence speed of training by automatically adjusting the learning rate. The update rules of the Adam algorithm are as follows:

[0111]

[0112] Among them, θ t is the model parameter, m t , v t are the first-order and second-order moment estimates, η is the learning rate, ∈ is a small constant to prevent division by zero;

[0113] After the training is completed, the trained LSTM sub-model is used to perform fault diagnosis on the real-time data in the power system, and the corresponding fault mode and probability of occurrence are output. Combined with the output of other sub-models, the final fault diagnosis result is comprehensively decided.

[0114] The LSTM sub-model can effectively capture the long-term dependencies in the power system through time series learning. The optimization and loss function design during its training process ensures the accuracy of the model and can quickly respond to system state changes during real-time diagnosis, thereby effectively identifying fault modes.

[0115] The random forest sub-model in the fault pattern recognition module includes:

[0116] Construct a random forest sub-model. Random forest is an integrated learning algorithm composed of multiple decision trees. In this model, each decision tree is constructed by randomly selecting features from the training data, and finally outputs the diagnosis result through a voting mechanism.

[0117] The random forest sub-model is trained and supervised learning is performed using the training data set. During training, each tree randomly extracts samples and features from the original data set to generate multiple different decision trees. Each decision tree is constructed using the CART algorithm, which recursively divides the data set into different subsets to minimize the Gini index of each node. The goal of the CART algorithm is to minimize the Gini index of the node. The calculation formula is as follows:

[0118]

[0119] Among them, Gini(t) is the Gini index of node t, p i is the probability of category i;

[0120] Perform cross-validation on the trained random forest sub-model to evaluate its performance, select the optimal number of decision trees and maximum depth parameters to avoid overfitting or underfitting, and use k-fold cross-validation to determine the optimal model parameters;

[0121] The prediction results of each decision tree are combined through the voting mechanism to obtain the final prediction result. Specifically, for the classification task, the following formula is used to calculate the final classification result:

[0122]

[0123] Among them, y1, y2, ..., y N is the prediction output of each decision tree, is the final predicted category;

[0124] The trained random forest sub-model is used for real-time fault pattern recognition. The real-time characteristic data of the power system is input. The model outputs the fault diagnosis results, fault type or occurrence probability according to the learning results in the training process.

[0125] The random forest sub-model can improve the robustness and accuracy of the model by integrating multiple decision trees. The model's adaptability to different fault modes is ensured through random feature selection and random data sampling, and the reliability of the final diagnosis result is enhanced through the voting mechanism.

[0126] The convolutional neural network sub-model in the fault mode recognition module includes:

[0127] Construct the structure of the convolutional neural network sub-model, including the input layer, convolution layer, pooling layer, fully connected layer and output layer. The input layer receives the feature vector transmitted by the data processing module, the convolution layer extracts local features, the pooling layer reduces the size and calculation amount of the feature map, and the fully connected layer performs classification output;

[0128] Convolution operation is used for feature extraction. The convolution operation is used to identify the spatial local features in the input data. The calculation formula of the convolution layer is as follows:

[0129] y i =∑ j x j ·w ij +b i ;

[0130] Among them, y i is the convolution result, x j is the input data, w ij is the convolution kernel weight, b i is the bias term;

[0131] The pooling layer downsamples the feature map output by the convolutional layer to reduce the dimension of the data and the amount of calculation. The pooling operation uses the maximum pooling, and its calculation formula is:

[0132] y max =max(x1, x2, ..., x k );

[0133] Among them, y max is the result after pooling, x1, x2, ..., x k is the element in the pooling window;

[0134] The fully connected layer combines the features extracted by the convolutional layer and the pooling layer to generate the final classification result. The fully connected layer uses the softmax function for classification, and the calculation formula is:

[0135]

[0136] Among them, P(y i |x) is the predicted probability of category i, z i is the output of category i;

[0137] The cross entropy loss function is used for training to calculate the difference between the model output and the true label. The formula is as follows:

[0138]

[0139] Among them, y i is the true label, is the model prediction value;

[0140] The trained convolutional neural network sub-model is used for real-time fault pattern recognition, inputs the feature data of real-time monitoring, and outputs the fault diagnosis results;

[0141] The convolutional neural network sub-model extracts local features from the data through convolution and pooling operations, and can perform well in complex fault pattern recognition tasks. Through the fully connected layer and softmax classification, the convolutional neural network can efficiently perform fault diagnosis on real-time power system data with high accuracy and real-time response capabilities.

[0142] The fault diagnosis module includes:

[0143] Receiving preprocessed multi-source data output by a data processing module;

[0144] The preprocessed multi-source data is input into a fault pattern recognition model, which includes a long short-term memory network sub-model, a random forest sub-model and a convolutional neural network sub-model;

[0145] Through the long short-term memory network sub-model, the modeling ability of time series data is used to analyze the time series correlation in the input data and identify possible failure modes;

[0146] Through the random forest sub-model, multiple features of the input data are combined to classify and select features, further improving the accuracy of fault diagnosis;

[0147] Through the convolutional neural network sub-model, the local pattern of the input data is analyzed by using its automatic extraction ability of local features to capture potential fault signs in the system;

[0148] The output results of the long short-term memory network sub-model, the random forest sub-model and the convolutional neural network sub-model are integrated, and the final fault diagnosis result is obtained through weighted average or voting mechanism;

[0149] Based on the diagnosis results, the fault types are classified, such as normal, minor fault, medium fault, major fault, etc., and a fault diagnosis report is generated;

[0150] If a fault is detected, the fault diagnosis module will pass the fault information to the early warning module and decision support module for subsequent early warning issuance and emergency dispatch;

[0151] By integrating the three models of long short-term memory network sub-model, random forest sub-model and convolutional neural network sub-model, their respective advantages are fully utilized to diagnose the fault types in the power system in real time and accurately. The long short-term memory network sub-model is good at processing time series data and can capture the time correlation of faults in the power system. The random forest sub-model can efficiently handle the classification problems of multi-dimensional features and improve the accuracy of fault identification. The convolutional neural network sub-model can discover local rules in the data through automatic feature extraction, further enhancing the robustness of fault diagnosis.

[0152] The early warning module includes:

[0153] Receive the real-time fault diagnosis results output by the fault diagnosis module, the fault diagnosis results including the current power system status and fault type;

[0154] According to the fault diagnosis results, identify the possible fault trends in the current power system and analyze the probability and severity of the fault;

[0155] Use the long short-term memory network sub-model to predict the time series of power system operation data, and combine the fault diagnosis results to evaluate the occurrence time, type and risk level of potential faults;

[0156] Based on the prediction results of the long short-term memory network sub-model, combined with the historical fault data of the power system, the possibility of fault occurrence is calculated and a fault warning signal is generated. The specific calculation method is as follows:

[0157] P(fault occurrence) = f(fault diagnosis results, time series prediction results, historical fault data);

[0158] Where f(·) represents the function of calculating the failure probability based on the prediction results of the long short-term memory network sub-model and historical failure data, and P(failure occurrence) is the probability of failure occurrence;

[0159] Determine whether to issue an early warning based on the probability of a fault and the predicted fault type. If the probability of a fault exceeding a preset threshold, generate an early warning report.

[0160] The generated early warning report includes the following: fault type, predicted range of occurrence time, fault severity assessment, recommended emergency measures and other relevant information.

[0161] Send early warning reports to maintenance personnel and relevant management departments to provide a basis for timely preventive measures and emergency response;

[0162] The early warning module can accurately assess the probability and risk level of potential faults in the power system based on the real-time fault diagnosis results and the time series prediction capabilities of the long short-term memory network sub-model. By providing early warning of faults, an early intervention mechanism can be provided for the power system to avoid or mitigate possible losses. The module combines historical fault data with the output of the prediction model to provide accurate fault warnings and generate detailed warning reports, providing strong support for subsequent emergency response and dispatch decisions.

[0163] The decision support modules include:

[0164] Receive the real-time fault diagnosis results output by the fault diagnosis module and the fault warning signal output by the warning module, including the fault type, fault probability, expected fault occurrence time, current system status and warning report;

[0165] Comprehensively analyze fault diagnosis results and warning information, evaluate the potential impact of the fault on the power system, and determine the urgency of the fault and the risks that may be caused;

[0166] According to the evaluation results, a fault emergency dispatch model is constructed, which includes the following objective functions and constraints:

[0167] Objective function: minimize fault handling time, minimize system loss, minimize maintenance cost, etc.;

[0168] Constraints: system load constraints, grid operation constraints, maintenance personnel resource constraints, etc.

[0169] Specifically, the objective function of the scheduling model can be expressed as:

[0170] min f(x)=α·T 故障处理 +β·L 系统损失 +γ·C 维修成本 ;

[0171] Where x represents the decision variables (such as scheduling order, maintenance plan, backup power, etc.), T 故障处理 is the fault handling time, L 系统损失 is the system loss, C 维修成本 is the maintenance cost, α, β, γ are the corresponding weight coefficients;

[0172] The fault emergency dispatch model is solved by particle swarm optimization algorithm. The steps of particle swarm optimization algorithm are as follows:

[0173] Initialization: A certain number of particles are randomly generated, each particle represents a potential scheduling solution, and the initial position and speed of the particle are determined according to the range of decision variables of the scheduling model;

[0174] Evaluation: Calculate the objective function value corresponding to each particle, including fault handling time, system loss, maintenance cost, etc. The fitness value of the particle is the objective function value;

[0175] Update: Update the particle's position and velocity based on the particle's current position and velocity, combined with individual experience and global best experience. The update formula is as follows:

[0176]

[0177] in, is the velocity of particle i in the kth iteration, is the position of particle i in the kth iteration, is the historical optimal position of particle i, gk is the global optimal position of all particles, w is the inertia weight, c1 and c2 are acceleration constants, r1 and r2 are random numbers;

[0178] Iteration: Through multiple iterations, the particle swarm gradually searches for the optimal solution in the solution space until the preset termination condition is met (such as reaching the maximum number of iterations or convergence criteria);

[0179] Output the optimal solution: After the particle swarm optimization algorithm stops iterating, it outputs the optimal solution as the fault emergency scheduling plan.

[0180] Based on the optimal solution obtained by the particle swarm optimization algorithm, a fault emergency dispatch plan is generated, which includes specific fault handling procedures, dispatch sequence, backup power configuration, maintenance personnel allocation, etc.

[0181] The generated fault emergency dispatch plan is transmitted to the control center for dispatch personnel to execute.

[0182] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0183] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. The AI-based power system fault diagnosis and early warning system is characterized by: It includes data acquisition module, data processing module, fault mode recognition module, fault diagnosis module, early warning module and decision support module, among which; The data acquisition module collects multi-source data related to power system fault diagnosis in real time by arranging sensors in the power system, and the multi-source data includes partial discharge, mechanical properties, sheath circulation, temperature and vibration; The data processing module is used to preprocess the collected multi-source data, including denoising, missing value filling and data standardization, and perform feature extraction on the preprocessed multi-source data; The fault mode recognition module constructs a fault mode recognition model based on the extracted features, and the fault mode recognition model includes a long short-term memory network sub-model, a random forest sub-model and a convolutional neural network sub-model; The fault diagnosis module performs real-time fault diagnosis on the power system based on multi-source data monitored in real time by means of a constructed fault pattern recognition model; The early warning module warns of possible faults in advance based on the fault diagnosis results and generates an early warning report; The decision support module generates a fault emergency dispatch plan using an optimization algorithm based on the information output by the fault diagnosis module and the early warning module.

2. The AI-based power system fault diagnosis and early warning system according to claim 1 is characterized in that: The data acquisition module comprises: Arranging a plurality of sensors in the power system, the sensors including a partial discharge sensor, a mechanical property sensor, a sheath circulation sensor, a temperature sensor and a vibration sensor; The partial discharge sensor is used to collect equipment partial discharge data in real time; The mechanical property sensor is used to monitor the operating status of the circuit breaker in real time; The sheath circulation sensor is used to monitor the cable sheath circulation leakage in real time; The temperature sensor is used to monitor the temperature change of the equipment; The vibration sensor is used to collect vibration signals of the equipment in real time; After being collected by sensors, the multi-source data are transmitted to the data processing module in real time through wired or wireless communication technology.

3. The AI-based power system fault diagnosis and early warning system according to claim 2 is characterized in that: The data processing module comprises: De-noising the received multi-source data; Linear interpolation is used to fill missing values ​​in the denoised data; Standardize the processed data to eliminate the dimensional differences between different sensors and data types; The preprocessed multi-source data is reduced in dimension by principal component analysis method; Feature extraction is performed on the data after principal component analysis, and key statistical features of the data, including mean, variance and kurtosis, are calculated through frequency domain analysis and time domain analysis.

4. The AI-based power system fault diagnosis and early warning system according to claim 3 is characterized in that: The long short-term memory network sub-model in the fault mode recognition module includes: Construct the structure of the LSTM sub-model, including the input layer, LSTM network layer, and output layer. Train the LSTM sub-model, use historical fault data for supervised learning, and use the back-propagation algorithm to optimize the model weights; After the training is completed, the trained long short-term memory network sub-model is used to perform fault diagnosis on the real-time data in the power system and output the corresponding fault mode and occurrence probability.

5. The AI-based power system fault diagnosis and early warning system according to claim 4 is characterized in that: The random forest sub-model in the fault mode recognition module includes: Build a random forest sub-model. Random forest is an ensemble learning algorithm consisting of multiple decision trees. Train the random forest sub-model and use the training dataset for supervised learning; Perform cross-validation on the trained random forest sub-model to evaluate its performance; The prediction results of each decision tree are combined through a voting mechanism to obtain the final prediction result; The trained random forest sub-model is used for real-time fault pattern recognition. The real-time characteristic data of the power system is input, and the model outputs the fault diagnosis results according to the learning results in the training process.

6. The AI-based power system fault diagnosis and early warning system according to claim 5 is characterized in that: The convolutional neural network sub-model in the fault mode recognition module includes: Construct the structure of the convolutional neural network sub-model, including the input layer, convolution layer, pooling layer, fully connected layer and output layer; Feature extraction is performed using convolution operations, which are used to identify spatially local features in the input data; The pooling layer downsamples the feature map output by the convolutional layer to reduce the dimension of the data and the amount of calculation; The fully connected layer combines the features extracted by the convolutional layer and the pooling layer to generate the final classification result; The cross entropy loss function is used for training to calculate the difference between the model output and the true label; The trained convolutional neural network sub-model is used for real-time fault pattern recognition, which inputs the feature data of real-time monitoring and outputs the fault diagnosis results.

7. The AI-based power system fault diagnosis and early warning system according to claim 6 is characterized in that: The fault diagnosis module comprises: Receiving preprocessed multi-source data output by a data processing module; The preprocessed multi-source data is input into a fault pattern recognition model, which includes a long short-term memory network sub-model, a random forest sub-model and a convolutional neural network sub-model; Through the long short-term memory network sub-model, the modeling ability of time series data is used to analyze the time series correlation in the input data and identify possible failure modes; Through the random forest sub-model, multiple features of the input data are combined to classify and select features, further improving the accuracy of fault diagnosis; Through the convolutional neural network sub-model, the local pattern of the input data is analyzed by using its automatic extraction ability of local features to capture potential fault signs in the system; The output results of the long short-term memory network sub-model, the random forest sub-model and the convolutional neural network sub-model are integrated, and the final fault diagnosis result is obtained through weighted average or voting mechanism; Based on the diagnosis results, the fault types are classified and a fault diagnosis report is generated; If a fault is detected, the fault diagnosis module transmits the fault information to the early warning module and the decision support module.

8. The AI-based power system fault diagnosis and early warning system according to claim 7 is characterized in that: The early warning module comprises: Receive the real-time fault diagnosis results output by the fault diagnosis module, the fault diagnosis results including the current power system status and fault type; According to the fault diagnosis results, identify the possible fault trends in the current power system and analyze the probability and severity of the fault; Use the long short-term memory network sub-model to predict the time series of power system operation data, and combine the fault diagnosis results to evaluate the occurrence time, type and risk level of potential faults; Based on the prediction results of the long short-term memory network sub-model and the historical fault data of the power system, the possibility of fault occurrence is calculated and a fault warning signal is generated; Determine whether to issue an early warning based on the probability of a fault and the predicted fault type. If the probability of a fault exceeding a preset threshold, generate an early warning report. Send early warning reports to maintenance personnel and relevant management departments to provide a basis for timely preventive measures and emergency response.

9. The AI-based power system fault diagnosis and early warning system according to claim 8, characterized in that: The decision support module includes: Receive the real-time fault diagnosis results output by the fault diagnosis module and the fault warning signal output by the warning module; Comprehensively analyze fault diagnosis results and warning information to assess the potential impact of faults on the power system; According to the evaluation results, a fault emergency dispatch model is constructed and solved by particle swarm optimization algorithm; Based on the optimal solution obtained by the particle swarm optimization algorithm, a fault emergency dispatch plan is generated, and the generated fault emergency dispatch plan is transmitted to the control center for dispatch personnel to execute.

Images