31 results about "Wavelet neural network" patented technology

The invention discloses an urban intelligent water affair risk dynamic identification and early warning method based on deep learning, and the method comprises the following steps: S1, collecting the water pressure, flow, residual chlorine concentration, elevation, rainfall, valve state, pump station state and accident label of each node in a water supply network, and constructing a time alignment data sequence; s2, constructing a dynamic adjacency matrix according to the pipe network connection relation and the event state information; s3, inputting the data sequence and the dynamic adjacency matrix into an improved space-time diagram wavelet neural network to generate space-time feature representation; s4, multi-scale features are extracted and fused through the high-frequency branches and the low-frequency branches; s5, constructing a hyperedge set, executing graph structure propagation, and generating a risk representation tensor; s6, inputting the risk representation tensor into the risk prediction network, and outputting a node risk probability and a confidence interval; and S7, determining a risk level according to the risk probability and the confidence interval, and generating a corresponding early warning signal. According to the invention, fine modeling and dynamic early warning of urban water supply risks are realized.

The invention provides a dynamic sequence recommendation mechanism for multi-scale wavelet transform of an intelligent maritime reconnaissance instrument, which belongs to the technical field of communication information services and comprises an embedded layer, a multi-scale wavelet decomposition layer, a wavelet neural network layer, a multi-view contrast learning module and a final recommendation layer. According to the method, the discrete wavelet transform is introduced to replace the traditional Fourier transform and discrete cosine transform, so that the dynamic interest change in the user behavior sequence is more accurately captured. According to the method, time and frequency localization analysis can be carried out on the signals at the same time, and the method is especially good at processing non-stable user behavior data containing mutation, so that long-term stable interests and short-term sudden interests of users are extracted and distinguished on different time scales. In addition, a wavelet neural network and an enhanced multi-view contrast learning mechanism are introduced, and the feature processing ability, generalization ability and recommendation precision of the model are further improved.

The invention discloses a centralized heat supply network pipe pressure monitoring method and system based on artificial intelligence, and relates to the technical field of heat supply system monitoring, and the method comprises the steps: firstly laying optical fibers and point pressure sensors on a network pipe, and collecting acoustic strain and pressure data; constructing a digital twin grid through same-frequency alignment and Delaunay triangulation, and reconstructing a continuous pressure field; calculating the residual error between the reconstruction pressure and the measured value at the monitoring node; extracting multi-scale residual features to form a feature matrix based on a wavelet operator of a pipe network topological structure map; inputting the features into a pre-trained spectrogram wavelet neural network, and outputting a normalized abnormal score; and finally positioning a fault pipe section according to the score and performing graded warning. Physical modeling and deep learning are fused in the whole process, closed-loop monitoring from data collection to fault diagnosis is achieved, and the precision and timeliness of pipe network anomaly detection are remarkably improved. Manual detection is not needed, and the monitoring process is more flexible.

The invention provides a gait evaluation method and system based on a human body nonlinear system analysis technology, and the method comprises the steps: collecting a gait cycle six-channel high-precision time sequence signal outputted by a wearable inertial measurement unit, eliminating noise and gait difference through wavelet threshold denoising and Z-score standardization, extracting a chaotic feature vector composed of a Lyapunov index spectrum and a Kolmogorov entropy value, and carrying out the recognition of a gait signal, a wavelet neural network is adopted to realize nonlinear mapping of chaotic features and phase-space reconstruction parameters, and a self-adaptive feedback mechanism is introduced to dynamically optimize modeling parameters, so that the accuracy and personalized matching capability of gait pattern recognition and stability evaluation are effectively improved; quantitative characterization of the gait chaos level and real-time online model optimization can be achieved, and high-robustness support is provided for rehabilitation training and exercise aided decision making.

The invention discloses a main bearing fault diagnosis and degradation prediction method for multi-source coupling signal sparse decoupling, and belongs to the technical field of fault diagnosis. The method comprises the following steps: acquiring a multi-source vibration signal of a rotating machinery main bearing; performing variable working condition anomaly detection based on a causal decoupling theory to obtain a health state index with independent working conditions; performing signal sparse separation and decoupling by using a multi-band wavelet sparse decomposition network; performing frequency domain feature enhancement and impact response sparse diagnosis based on a wavelet neural network, and identifying a fault type; and predicting a degradation trend by combining a deep transfer learning model, and estimating the remaining service life. The problems of poor adaptability to complex working conditions, weak noise interference suppression, insufficient signal sparse processing, difficulty in weak feature extraction and the like in the prior art are solved, the fault diagnosis precision and the prediction reliability are remarkably improved, and the method is suitable for health management of the rotating machine main bearing.

The application provides a power line carrier channel noise prediction method and system considering type matching deviation, and the method comprises the following steps: collecting original noise data of power line carrier channel noise; calculating the matching similarity between the original noise data and each noise type respectively, and determining the noise type; calculating the matching deviation of the corresponding noise type; inputting the original noise data into a pre-trained noise prediction model, and realizing the prediction of the power line carrier channel noise based on the output of the model; wherein the model is trained by a preset wavelet neural network based on the wavelet neural network prediction error of the corresponding noise type and the matching deviation. Compared with the prior art, by identifying the type of the power line carrier channel noise, training the preset wavelet network model based on the matching deviation, realizing the prediction of the power line carrier channel noise, the type of the noise and the prediction model can be adapted, and the prediction accuracy and precision are effectively improved.

The present application relates to a kind of discrete multi-motor servo system fuzzy wavelet neural network control method with event triggering mechanism, belong to multi-motor servo system control field, comprising the following steps: S1: the discrete time system model of multi-motor servo system is established;S2: design two type fuzzy wavelet neural network to estimate unknown nonlinear function caused by external interference and internal parameter perturbation;S3: based on the backstepping control framework, introduce event triggering mechanism with dead zone operator, design discrete time fuzzy wavelet neural network controller;S4: using the discrete time fuzzy wavelet neural network controller, control multi-motor servo system is carried out.

The present application relates to a cross-stitch network-based tool multi-task high-synergy monitoring method, comprising the following steps: S1, collecting tool vibration signals, constructing a training data set and a test data set; S2, using a wavelet neural network for feature extraction for the anomaly detection task, and using a gated recurrent network for feature extraction for the RUL prediction task; S3, constructing a cross-stitch network, performing deep feature extraction and shunt-type feature fusion; S4, constructing a regularization loss function, inputting the training data set, and using an automatic weight adjustment strategy to adaptively optimize the loss weights of different tasks, so that the regularization loss function reaches a minimum, and the network training is completed; S5, inputting the test data set into the trained network model to obtain the results of the anomaly detection task and the RUL prediction task. The present application innovatively introduces and optimizes and improves the cross-stitch network and applies it to tool multi-task monitoring, significantly improves the tool multi-task synergy, can achieve higher classification and prediction accuracy, and has good stability.

The invention provides an adaptive equalization method for a PON, and the method comprises the steps: decomposing an input unbalanced signal into a preset number of layers through employing a Mallat decomposition algorithm, and obtaining a high-frequency detail coefficient and a low-frequency approximation coefficient of each layer; performing denoising processing on the high-frequency detail coefficient of each layer: calculating a GCV score of the high-frequency detail coefficient of the current layer by using a generalized cross validation threshold method, determining an optimal denoising threshold value for denoising processing based on the GCV score, and performing denoising processing on the high-frequency detail coefficient based on the optimal denoising threshold value by using a nonlinear contraction function; based on the denoised high-frequency detail coefficient of each layer and the denoised low-frequency approximation coefficient of the corresponding layer, performing layer-by-layer iterative reconstruction by using a Mallat reconstruction algorithm to obtain a denoised data symbol sequence; and taking the denoised data symbol sequence as the input of a pre-trained embedded wavelet neural network, and outputting an equalized signal.

The invention relates to a converter end point carbon content prediction method based on a wavelet neural network, which comprises the following steps: firstly, carrying out data preprocessing on original data, sequentially finishing blank value elimination, abnormal value screening and normalization processing, then screening out weak correlation process parameters through Pearson's correlation coefficient analysis in combination with a metallurgy mechanism, and finally, carrying out prediction on the end point carbon content of a converter. Subsequently, dimensionality reduction is carried out through principal component analysis, and finally an optimized input parameter set is formed to replace original full-amount parameters. After the optimization parameter set is substituted into the wavelet neural network model, the model complexity is greatly reduced, and the training efficiency is improved. And finally, inputting production data into the wavelet neural network prediction model to obtain an end point carbon content prediction result. According to the method, modeling is carried out based on the wavelet neural network, deep learning driving replaces traditional mechanism driving, complex reactions and nonlinear relations in automatic learning data can be achieved, and the prediction precision of the converter end point carbon content is improved.

The invention relates to a wavelet neural network dynamic construction method, which comprises the following steps of: setting a nonlinear system, and processing an unknown nonlinear function in the nonlinear system by adopting a self-adaptive scaling wavelet network approximation framework; processing the non-linear function, namely determining an initial wavelet space for approaching the non-linear function by utilizing an initial wavelet frequency estimator of the adaptive scaling wavelet network approaching framework; on the basis of a scaling parameter self-adaptive adjustment mechanism, scaling parameters are dynamically updated on line so as to adjust the frequency bandwidth of the wavelet basis function and enable the frequency bandwidth to be matched with the spectrum distribution of the nonlinear function; and based on a wavelet basis function increasing and cutting mechanism of the adaptive scaling wavelet network approximation framework, increasing and cutting a wavelet basis function on the basis of the initial wavelet space so as to further approach the nonlinear function.

The application relates to a UPFC device fault prediction method based on a wavelet neural network, and comprises the following steps: defining a fault variable of a unified power flow controller; classifying the fault variable, and defining a data acquisition structure for device state monitoring and fault prediction; acquiring signal data of each node of the unified power flow controller, classifying and coding the signal data after filtering and A / D conversion; constructing a device fault prediction model by adopting a wavelet neural network; inputting real-time signal data into the trained device fault prediction model to obtain a device state prediction result output by the device fault prediction model; judging whether a device fault exists according to the device state prediction result; and issuing a fault early warning signal for the predicted device fault. The application realizes multi-dimensional and all-around UPFC online state monitoring, fault diagnosis and fault early warning, predicts specific fault information of the UPFC device in advance, effectively protects the operation safety of the device, and avoids damage to other power electronic elements caused by the UPFC device fault.

The invention relates to the field of Internet of Things production, in particular to a production equipment monitoring method, system and equipment based on the industrial Internet of Things and a medium, and the method comprises the following steps: obtaining coupling change information of a monitoring part obtained through combined measurement based on a displacement sensor and an eddy current sensor; the coupling change information comprises a change rate signal and an eddy current energy signal, and the monitoring part is a part needing to be monitored on the target production equipment; acquiring surface feature information of the monitored part obtained based on a phase-controlled eddy current sensor; fusing the coupling change information with the surface feature information to obtain a comprehensive feature signal; inputting the comprehensive characteristic signal into a preset wavelet neural network prediction model to output a risk level of the target production equipment; through dynamic fusion of feature enhancement driven by a physical mechanism and data driving, the dependence of traditional time domain analysis on a single signal amplitude is broken through, and the cross-physics coupling effect of weak damage is amplified from multiple dimensions.

The invention relates to the technical field of electric power data processing, in particular to an electric energy metering data anomaly detection method, system and device and a medium, and the method comprises the steps: carrying out the moving average filtering and standardization processing of original electric energy metering data, and generating preprocessing data; dynamically selecting a target wavelet basis function according to the spectrum kurtosis value of the preprocessed data, performing multilayer wavelet packet decomposition on the preprocessed data, and extracting a high-frequency detail component and a low-frequency approximate component; energy class features of the high-frequency detail components and statistical features of the low-frequency approximate components are extracted, fusion and normalization are carried out, and multi-dimensional feature vectors are constructed; inputting the multi-dimensional feature vector into a wavelet neural network model, and outputting an abnormal probability value of the current time point; and comparing the abnormal probability value of the current time point with a dynamic threshold value to judge whether the current time point is an abnormal point or not. The method has the beneficial effects that the accuracy, the stability and the self-adaptive capability of the abnormal detection of the electric energy metering data are remarkably improved.

The invention discloses a dissolved oxygen concentration control method based on a pipeline recursive wavelet neural network, and relates to the field of artificial intelligence. Firstly, a control factor of the DO concentration is analyzed based on a reaction mechanism; and then, constructing a pipeline recursive wavelet neural network to control the DO concentration. Meanwhile, an improved adaptive learning rate algorithm is adopted to train network parameters. Therefore, the DO control precision in the sewage treatment process is improved.

This invention discloses a method and device for early warning of energy consumption in air conditioning refrigeration systems based on genetic algorithms. It aims to solve the problems of traditional energy consumption prediction models, such as single feature set, weak nonlinear modeling ability, reliance on experience for parameter tuning, and susceptibility to local optima. The method first integrates operating parameters such as indoor and outdoor temperature difference, humidity, compressor performance, and refrigerant charge to construct input features that comprehensively reflect the system state. Second, it designs a wavelet neural network with Morlet wavelet function as the activation function and introduces a genetic algorithm to globally optimize network weights, scaling factors, and translation factors, automatically searching for the optimal parameter combination to improve model convergence speed and prediction accuracy. Third, it dynamically sets energy consumption thresholds based on historical data, compares the prediction results in real time, and triggers an early warning when the threshold is exceeded. This method ensures high accuracy while possessing good generalization ability and engineering applicability, providing effective support for intelligent operation and maintenance and energy-saving management.

This invention discloses a method for classifying coral reef substrate based on the fusion of lidar and side-scan sonar data, mainly including the following steps: (1) By analyzing the structure of the photon-counting lidar scanning system, a precise calculation model for the coordinates of underwater depth sounding points of UAV-borne lidar is constructed; (2) Through fine processing steps such as seabed line tracking and joint correction of radiation distortion, a geocoded side-scan sonar image is constructed; (3) The backscattering intensity of lidar and side-scan sonar is fused to construct a seabed topographic image; (4) The standard wavelet BP neural network is improved; (5) Based on the fused image, coral reef species samples are extracted, and feature parameters with high self-cohesion and resolution are selected for sample training. The sample training results are used to classify the coral reef substrate in the fused image. Through the above steps, coral reef substrate classification based on the fusion of lidar and side-scan sonar data can be realized, providing technical support for identifying coral reef types.

The application relates to the technical field of intelligent equipment manufacturing, and specifically discloses an AI-based equipment manufacturing process energy consumption optimization and carbon footprint tracking system, which comprises a central collaborative scheduler, and the central collaborative scheduler is communicatively connected with the following modules: a data self-checking reconstruction module, a wavelet neural network optimized by using an improved particle swarm algorithm is used to construct a data self-checking model, and time sequence analysis and correlation comparison are carried out on multi-source heterogeneous sensing data of an equipment manufacturing process; the application can dynamically identify and correct inherent errors of sensing data by deploying a multi-source heterogeneous sensor network, combining a precise time synchronization protocol, constructing a time sequence alignment sequence, using a wavelet neural network optimized by using an improved particle swarm algorithm to construct a data self-checking model, outputting a high-fidelity data stream, and solving the problem of insufficient reliability of carbon footprint accounting data in a traditional manufacturing process from the source, thereby providing a data basis for energy consumption optimization and carbon emission tracking, and ensuring the authenticity and effectiveness of accounting results.

The embodiment of the invention provides an engine fault diagnosis method and device based on a wavelet neural network, computer equipment and a medium, and the method comprises the following steps: constructing the wavelet neural network comprising an input layer, a hidden layer and an output layer, and determining the number of nodes of the input layer and the number of nodes of the output layer; performing normalization processing on a data set comprising a training set sample and a test set sample, inputting the normalized training set sample into a wavelet neural network, training the wavelet neural network by using an L-M algorithm, determining an optimal hidden layer node number p according to a node number of an input layer and a node number of an output layer, and generating a trained wavelet neural network; and inputting the normalized test set sample into the trained wavelet neural network to generate a diagnosis result of the engine fault. According to the scheme, the fault is diagnosed through the wavelet neural network, and the accuracy and efficiency of diagnosis of the rotor-rotor / rotor-static rub-impact fault of the aero-engine are improved.

The invention discloses a single-source satellite-borne GNSS-R soil humidity inversion method based on dynamic space-time sequence error propagation, and the method comprises the steps: carrying out the inversion of global soil humidity through employing a soil humidity inversion model of a conventional single CYGNSS reflectivity; calculating the space-time sequence deviation of the CYGNSS inversion soil humidity value in each sliding window by taking an SMAP soil humidity product as a reference; dividing the global grid into a plurality of analysis windows by adopting a sliding window algorithm, and performing denoising processing on the space-time sequence deviation by applying a double-branch wavelet neural network in each analysis window to obtain a denoised deviation change trend; and constructing a final soil humidity inversion model by using the denoised space-time sequence deviation variation trend and the CYGNSS effective reflectivity. According to the method, the single-source satellite-borne GNSS-R soil humidity inversion capability is further improved.

The invention discloses an unmanned aerial vehicle high-precision positioning and navigation method and device, a medium and a product, and relates to the technical field of unmanned aerial vehicle positioning and navigation.The method comprises the steps that data of an unmanned aerial vehicle multi-sensor system is obtained, and signal preprocessing is conducted; performing wavelet transformation on the preprocessed data; constructing a wavelet neural network model based on the data after wavelet transformation; fusing the data from the BDS, the IMU and the VIO, and carrying out positioning estimation on the fused data by adopting PF and EKF to obtain positioning optimization data; further, when the GNSS signal fails, a compensation signal is determined according to the INS, the BDS and the wavelet neural network model; based on the compensation signal, accurate positioning information is determined; and based on the accurate positioning information, calculating the optimal flight path of the unmanned aerial vehicle by adopting a path planning algorithm. According to the invention, the positioning precision and robustness of the unmanned aerial vehicle in a complex environment can be improved, so that the unmanned aerial vehicle can continuously and stably work in any environment.

The application discloses a milk product detection method and device based on near-infrared spectroscopy, equipment and medium, and belongs to the technical field of food detection. The method comprises the following steps: optimizing the next generation population evolved by a genetic algorithm GA based on a snake optimization algorithm SO to obtain an optimized next generation population; wherein the position of each individual in the next generation population is a set of network parameters of a wavelet neural network WNN; obtaining an optimal individual based on the optimized next generation population; obtaining a detection model based on the optimal individual, wherein the detection model is constructed based on the WNN; in response to the input of feature data of a milk product to be detected obtained by near-infrared spectroscopy, calling the detection model to obtain a detection result. The application greatly improves the convergence speed and also improves the recognition accuracy of the detection model.

The application discloses a kind of based on artificial intelligence's central heating network pressure monitoring method and system, it is related to heating system monitoring technical field, including, first in pipe network layout optical fiber and point pressure sensor, acoustic strain and pressure data are collected;By the same frequency alignment and Delaunay triangulation, construct digital twin grid, reconstruct continuous pressure field;Residual error of reconstructed pressure and measured value at monitoring node is calculated;Based on pipe network topology, construct graph wavelet operator, extract multi-scale residual feature to form feature matrix;Characteristic is input into pre-trained spectral graph wavelet neural network, and output normalized anomaly score;Finally, according to the score, locate fault pipe section and hierarchical alarm.The whole process combines physical modeling and deep learning, realizes closed-loop monitoring from data acquisition to fault diagnosis, significantly improves the precision and timeliness of pipe network anomaly detection.The application does not need artificial detection and monitoring process is more flexible.

The invention discloses a blast furnace air permeability index prediction method based on multi-scale time-delay characteristic mining and application, and belongs to the field of soft measurement modeling in the blast furnace ironmaking process. The invention provides a blast furnace air permeability index prediction framework combining wavelet decomposition, wavelet coherence analysis and a time-frequency fusion model aiming at the characteristics of nonlinearity, unsteady state and large time delay of blast furnace operation data. The method comprises the following steps: firstly, carrying out peak separation and multi-scale decomposition on data, and extracting multi-scale time delay information among variables by utilizing wavelet coherence analysis; and then modeling time domain global features and frequency domain local features by adopting orthogonal subspace analysis and a wavelet neural network, and performing model fusion through Gauss-Markov estimation to realize advanced multi-step prediction of the air permeability index. The dynamic time lag characteristic of the blast furnace process can be effectively captured, the prediction precision and robustness are improved, and technical support is provided for stable operation of the blast furnace.

The application discloses a multi-tooth cutter damage monitoring method based on ResNet-GAN and a wavelet neural network, belongs to the cutter monitoring technical field, and is used for improving the quality of generated signals in sample expansion, reducing the unbalanced ratio of sample training, and realizing visualization of deep features. The application performs sample expansion on damage samples, realizes data enhancement of few-class samples, makes an algorithm learn features of damage signals more fully, reduces the unbalanced ratio of sample training, and improves training accuracy. In addition, the application combines ResNet residual blocks and gradient penalty terms, so that gradient normal propagation can be ensured in a deep neural network, training is kept smooth and stable, the phenomenon of gradient explosion is reduced, and the quality of generated signals is improved. Deep features of signals are extracted from different dimensions through a multi-wavelet function kernel convolution module, feature visualization is realized, and different damage states can be better distinguished.

The invention is suitable for the technical field of track safety monitoring, and provides a turnout switch rail flaw detection method and device based on sweep frequency phased array ultrasonic.The method comprises the steps that at the preset detection position of a turnout switch rail, a phased array ultrasonic probe is controlled to scan the interior of the turnout switch rail at a plurality of different excitation frequencies, ultrasonic echo signals corresponding to different scanning angles under each excitation frequency are obtained; determining a first shape parameter of the injury through a preset injury shape function according to the ultrasonic echo signal; generating a two-dimensional scanning graph corresponding to each excitation frequency according to the ultrasonic echo signal; inputting the two-dimensional scanning graphs corresponding to all the excitation frequencies into a pre-trained wavelet neural network model to obtain a second shape parameter of the injury; and determining a final shape parameter of the injury based on the first shape parameter and the second shape parameter. The problems that an existing turnout switch rail flaw detection method is low in detection efficiency, high in omission ratio and inaccurate in flaw evaluation can be solved.

The application discloses a genetic algorithm optimized wavelet neural network greenhouse temperature prediction modeling method, which comprises the following steps: obtaining multiple greenhouse temperature related variables at the current time, generating the initial genetic population of the genetic algorithm according to multiple initial weight parameters of a wavelet neural network (WNN), constructing the fitness function of the genetic algorithm according to the prediction result and the actual output result of the wavelet neural network (WNN) and using the fitness function to evaluate the individuals of the genetic population, screening the optimal initial weight parameter through the genetic algorithm, constructing the greenhouse temperature prediction model based on the wavelet neural network (WNN) by using the optimal initial weight parameter, inputting the multiple greenhouse temperature related variables at the current time into the greenhouse temperature prediction model based on the wavelet neural network (WNN) and outputting the greenhouse temperature prediction value at a future time. The method combines the wavelet neural network and the genetic algorithm, and a more accurate model can be obtained.

The invention discloses a research area seasonal rainfall prediction method and system, and relates to the technical field of rainfall prediction. Comprising the following steps: taking monthly average reanalysis data of historical ERA5 data of a research area as a historical forecasting factor set; dividing the research area according to longitude lines and latitude lines, determining a seasonal rainfall index according to the rainfall sum of each radial grid point and each latitudinal grid point at the historical moment, and decomposing the seasonal rainfall index into a plurality of physical components; determining a causal relationship between the plurality of physical components and a historical forecast factor set, and determining a key influence area of each historical forecast factor influencing each physical component; constructing a historical rainfall forecast sample set according to the key influence region, and training the wavelet neural network to obtain a regional seasonal rainfall forecast model; and inputting the real-time forecasting factor data into the regional seasonal rainfall forecasting model to obtain the seasonal rainfall forecasting intensity of the research region. The method can effectively improve the prediction accuracy of seasonal rainfall.

A non-contact method for monitoring the icing state of wind turbine blades includes: acquiring wind turbine images; conducting noise tests on the wind turbine rotor composed of iced blades to obtain the wind turbine noise signal under icing conditions; processing the wind turbine noise signal using an eigenvalue-optimized beamforming algorithm to obtain sound pressure level spectral characteristics; acquiring noise sources based on the wind turbine images and identifying the coordinates of the maximum noise source, establishing the relationship between the noise source and different icing states in non-icing locations; constructing a positive relationship between icing states at different locations, with different masses, and with different incoming flow angles of attack, and the increase in the total sound pressure level of the wind turbine; inputting the positive relationship into a BP neural network model and a local linear wavelet neural network model respectively to obtain monitoring results. This invention utilizes acoustic array technology to achieve non-contact measurement of horizontal axis iced wind turbine noise and thus monitor its icing state, realizing the monitoring of wind turbine blade noise under different icing states.

The application provides a power supply system operation and maintenance monitoring method and platform based on a neural network, relates to the technical field of power supply system operation and maintenance, collects historical operation data of a power supply system, and constructs an operation and maintenance management database according to the historical operation data, so as to mark the historical operation state of each power supply station in the power supply system; based on historical sensing detection images and historical operation and maintenance management parameters, the current sensing detection image and operation and maintenance management parameter are integrated into a multi-source splicing feature map through a wavelet neural network model; the multi-source splicing feature map is classified by a softmax classifier, and the operation state of each power supply station is updated in real time according to the classification result of the skip-convolution network model, so as to obtain the real-time operation state of each power supply station; the historical operation state and the real-time operation state of each power supply station are compared, and an operation and maintenance report and a fault trend prediction of the corresponding power supply station are generated. The application is helpful to the supervision efficiency and reliability of power supply system operation and maintenance.