51 results about "Probabilistic neural network" patented technology

The invention relates to the technical field of automobile automatic transmission, and discloses an AMT bearing operation state detection system and method under an active and passive switching working condition, and the method comprises the steps: collecting real-time data through a vibration acceleration sensor, a current sensor, a temperature sensor and a rotating speed encoder, carrying out the preprocessing of envelope demodulation, Kalman filtering and the like, and carrying out the detection of the operation state of an AMT bearing; and generating a time-frequency characteristic matrix by using variational mode decomposition and Hilbert transform. And inputting the matrix into a probabilistic neural network model adopting a sliding time window mechanism, and outputting a bearing health state probability value. A multi-parameter state evaluation model optimized by a quantum genetic algorithm is constructed, an optimal feature combination is obtained, and early warning levels and maintenance suggestions are output through a belief rule base inference device in combination with a hierarchical diagnosis control model (including an acquisition layer, an analysis layer and an execution layer). The system is further provided with a signal verification module to ensure data reliability. According to the invention, multi-source data fusion and dynamic adaptive diagnosis are realized, and the state detection precision and real-time performance of the AMT bearing under complex working conditions are improved.

The application discloses a battery thermal runaway early warning method, device and equipment of energy storage power station and medium, using neural network to predict the battery thermal runaway risk of energy storage power station, early warning before the battery thermal runaway of energy storage power station, improve the safety.In addition, first train the probabilistic neural network with a small amount of training samples, and after the convergence of the probabilistic neural network, train the back propagation neural network with the soft label output by the probabilistic neural network, solve the problem of slow learning speed and insufficient training samples in the initial training of the back propagation neural network, and improve the model training efficiency.At the same time, the soft label can solve the problem that the back propagation neural network is easy to fall into local minimum value in the training process.

According to the geomagnetic matching integrated navigation method based on the improved probabilistic neural network, the improved probabilistic neural network is utilized to indicate a trajectory for inertial navigation, geomagnetic actual measurement vectors are recognized at intervals, the method is short in training time, fixed in structure, simple in learning process and minimum in expected risk of misclassification, a Bayesian posterior probability output result can be obtained, and the method is suitable for popularization and application. The method has the advantages of strong non-linear recognition capability, good real-time performance, high matching rate and high precision, and can overcome the influence of geomagnetic measurement errors; smooth factors of the improved probabilistic neural network method change adaptively according to different pattern types, the relevance between feature vectors and pattern states is better represented, the actual effect of the input feature vectors on correct classification results is reflected, the improved probabilistic neural network can effectively improve the accuracy of geomagnetic matching positioning, and the accuracy of geomagnetic matching positioning is improved. Therefore, the matching result is more accurate and reliable, and the accuracy of the geomagnetic matching rate and the positioning precision is effectively improved especially in a geomagnetic feature weak region.

This invention discloses a photovoltaic system islanding detection method and system based on Santlet transform and Ridglet probabilistic neural network. The method includes: collecting historical voltage data of the photovoltaic inverter, preprocessing it, and then sequentially acquiring the data to be analyzed using a sliding window; performing time-frequency analysis using Santlet continuous wavelet transform to obtain the corresponding time-frequency spectrum; extracting multi-dimensional time-frequency feature vectors and inputting them into a trained Ridglet probabilistic neural network classification model to obtain the corresponding probability values; if the probability value is greater than a probability threshold, it is determined that the photovoltaic system has experienced islanding, and a trip signal is generated and sent to the grid-connected circuit breaker to disconnect it. This invention significantly improves the accuracy, speed, and reliability of islanding detection, effectively reduces the detection blind zone, and has good adaptability to complex power grid environments.

The invention relates to the technical field of fault diagnosis, in particular to a rapier loom fault diagnosis system. According to the system, fault tree analysis based on whale algorithm optimization and a probabilistic neural network are fused; the fault diagnosis method comprises the following steps: step 1, establishing a rapier loom fault tree diagnosis model; step 2, optimizing parameters of the fault tree model by using a whale algorithm; step 3, constructing a sample data set and carrying out normalization processing; and step 4, carrying out probabilistic neural network training and fault diagnosis. According to the diagnosis model based on fusion optimization of the probabilistic neural network and the fault tree algorithm, on the basis of a rapier loom fault tree diagnosis model, real-time newly-added state monitoring data of a monitoring system is introduced; and converting the data set into a sample feature vector, performing normalization processing, importing the sample feature vector into a probabilistic neural network algorithm model for training, and calculating and outputting a fault symptom probability, thereby realizing real-time, complete and rapid fault identification and positioning of the rapier loom fault diagnosis system.

The invention belongs to the field of oil and gas field exploration and development, and discloses a numerical reservoir model optimization and history fitting method and related device.The method comprises the steps that firstly, geological exploration data are utilized, a spatial clustering algorithm is adopted to automatically determine the local updating range of a structural model, and structural layers and stratums are optimized through spatial interpolation; outputting an optimized construction model and uncertainty measurement; based on the model and logging lithofacies attribute data, adaptively identifying a lithofacies updating key area and variation function parameters by applying a machine learning algorithm, and performing lithofacies simulation to output an optimized lithofacies model; combining the porosity, permeability and saturation attribute data, adaptively determining an attribute optimization range and variation function parameters by adopting a probabilistic neural network, and outputting an optimized attribute model through phase control attribute modeling; a three-dimensional oil-water two-phase black oil seepage model is constructed, a global optimization algorithm is adopted to automatically adjust parameters based on a target function, simulation data is calculated until convergence, optimal parameters are output to complete historical fitting, and efficient and reliable intelligent support is provided for complex oil reservoir development.

This invention discloses a method and system for transformer partial discharge pattern recognition based on an optimized probabilistic neural network. The method first acquires the transformer partial discharge signal; then, it establishes a two-dimensional spectrum of the partial discharge phase distribution pattern based on the partial discharge signal and extracts discharge statistical features from this spectrum; finally, it inputs the discharge statistical features into the optimized probabilistic neural network for pattern recognition to obtain the partial discharge pattern. The optimized probabilistic neural network uses a pollination algorithm to optimize the smoothing factor, and the switching probability in the pollination algorithm is a nonlinear function that decreases with the number of iterations. This invention can more accurately and efficiently classify and identify different partial discharge patterns of transformers, providing a data foundation for transformer fault diagnosis and resolution.

This invention discloses a bolt loosening location method and system based on environmental vibration and multi-scale PNN. The location method includes the following steps: S1: Establishing a refined finite element simulation model of the transmission tower and dividing it into multi-scale hierarchical substructures; S2: Simulating bolt loosening conditions in the finite element simulation model and extracting operational modal analysis results; S3: Calculating the frequency change ratio and modal compliance residual matrix norm, and constructing a multi-source damage-sensitive feature vector after weighted fusion; S4: Constructing and training a probabilistic neural network with an adaptive smoothing factor adjustment mechanism; S5: Acquiring on-site environmental vibration data and identifying modal parameters; S6: Extracting and calculating the measured data of the tower using the random subspace identification method, and inputting it into the trained probabilistic neural network to obtain the location results of the main bolt loosening area. This method achieves accurate bolt loosening location through scientific substructure division and multi-source fusion strategies, providing a reference for the safe operation and maintenance of tower bolts.

This invention provides an intelligent evaluation method for rockburst hazard at working faces based on principal component analysis-probabilistic neural networks, belonging to the field of rockburst hazard evaluation technology. An evaluation index system is established and data is collected by combining rockburst hazard influencing factors, drill cuttings monitoring, critical stress index monitoring, and actual field conditions. Principal component analysis (PCA) is used to simplify the original evaluation index data, resulting in comprehensive evaluation index data containing information from the original evaluation indicators. The comprehensive evaluation index data is divided into a training set and a test set. A probabilistic neural network (PNN) is used to train the evaluation model on the data in the training set, and then the performance of the evaluation model is tested using data from the test set, and its accuracy is calculated. If the accuracy is greater than or equal to 90%, the evaluation model is considered acceptable; if the accuracy is less than 90%, it is considered unacceptable, and the PNN's smoothing factor needs to be modified and retrained until the accuracy of the evaluation result meets the set value.

The application discloses an acceleration-based personalized virtual reality black insertion method for the disabled, which comprises the following steps: step 1, constructing an acceleration-based and personalized black insertion duration dataset based on the types of the disabled users; wherein the acceleration is divided into T types according to the size, and the disabled users have N types; step 2, obtaining the acceleration of the user, and calculating the time domain features and the frequency domain features of the acceleration; step 3, constructing a probability neural network based on the time domain and frequency domain features of the acceleration, inputting the time domain features and the frequency domain features of the acceleration into the probability neural network, and classifying the acceleration; and step 4, judging the type of the disabled user, combining the acceleration classification obtained in step 3, and obtaining the personalized black insertion duration of the user through the personalized black insertion duration dataset. The probability neural network based on the time domain and frequency domain features of the acceleration can well reflect the differences between different accelerations, reduce the influence of abnormal values on the network accuracy, and improve the differences between different features.

The application provides an insulator contamination grade evaluation method based on an infrared image, and comprises the following steps: S1, acquiring an infrared image of a sample insulator by using an infrared imaging device, wherein the image is acquired under different temperatures, different humidities and different contamination grades; S2, preprocessing the infrared image; S3, extracting a temperature characteristic parameter from the preprocessed image; S4, constructing a probabilistic neural network, inputting the temperature characteristic parameter, the environmental temperature and the environmental humidity into the probabilistic neural network as input features, and training the probabilistic neural network; S4, acquiring an infrared image of a to-be-tested insulator in real time, inputting the temperature characteristic parameter of the to-be-tested insulator and the environmental temperature and the environmental humidity of the to-be-tested insulator into the trained probabilistic neural network after processing through steps S2 and S3, and outputting a contamination grade of the to-be-tested insulator.

One method includes training, using data collected from sensors, a probabilistic neural network (NN) model comprising a set of model weights. The probabilistic NN model is trained to filter out data samples that induce a threshold of model uncertainty. The method includes training, at each cycle of training the probabilistic NN model and based on the set of model weights, a joint estimator to generate gradient updates to the set of model weights that are used to predict whether model updates from the sensors are anomalous. The method includes assigning, to each sensor, a confidence coefficient value that estimates a degree of confidence of the model updates.The method includes transferring the set of model weights to a subset of the sensors for which the confidence coefficient value meets a threshold.

The present invention relates to a method, device, electronic device and storage medium for classifying spam text messages. The method for classifying spam text messages includes the following steps: S1, pre-processing text messages, downloading a text message data set from a communication network as a basic data set for generating word vectors, and using a vector space model to form a word-text matrix from the word set; S2, generating text message feature vectors including a topic feature vector and a context feature vector, the topic feature vector is obtained by performing a singular value decomposition of the word-text matrix, and the context feature vector is used to predict the probability of the context based on the central word; S3, neural network fusion and training, a training task for a classifier for processing text messages, the classifier is used to distinguish whether the text message is a spam text message; S4, text message classification and discrimination. According to the spam text message classification method of the present invention, the topic feature and the context feature can make up for each other's shortcomings, realize multi-feature fusion, and thus realize accurate classification of spam text messages.

The present invention discloses a method and device for matching well-seismic data, which includes: obtaining a reservoir parameter time-shift curve of a target well target layer; determining a porosity time-shift curve of the target well target layer using a trained probabilistic neural network model based on the reservoir parameter time-shift curve of the target well target layer; determining a well velocity time-shift curve of the target well target layer using a constructed rock physics model based on the predicted porosity time-shift curve of the target well target layer; and correcting an original velocity curve of the target well target layer using the well velocity time-shift curve of the target well target layer to obtain a corrected well velocity curve of the target well target layer. The present invention predicts the porosity time-shift curve using a probabilistic neural network, determines the well velocity time-shift curve using a rock physics model, and eliminates the time variation of the original velocity data through the well velocity time-shift curve correction, thereby improving the matching accuracy of well-seismic data in the time dimension.

The application discloses a kind of monitoring method and equipment for reducing the pressure of high-pressure pipeline of drilling and production ship, and relates to high-pressure pipeline technical field, the method comprises: obtaining the real-time performance data of high-pressure pipeline of drilling and production ship, the real-time performance data includes high-pressure pipeline pressure data, motor rotating speed data of high-pressure mud pump, drill pipe motion state data and top drive motion state data;The real-time performance data is preprocessed;The high-pressure pipeline pressure data after preprocessing is input into the high-pressure mud pipeline probabilistic neural network agent model trained in advance to predict the safety condition level;According to the safety condition level predicted, control signal is generated to control the high-pressure pipeline of drilling and production ship.The application can realize closed-loop control of control process, with high accuracy and strong pertinence.

The invention discloses a rotating machine fault feature optimization extraction method based on variational mode extraction and comprehensive detection indexes, and the method comprises the following steps: S1, decomposing an input vibration signal through employing a variational mode extraction method, and calculating a corresponding sample entropy value; s2, constructing a fitness function of a particle swarm optimization algorithm by using the comprehensive detection index; s3, parameters of the variational modal extraction method are optimized in combination with the comprehensive detection index and a particle swarm optimization algorithm; s4, using the variational mode under the optimal parameter to extract and decompose the input vibration signal; s5, calculating the sample entropy of each modal component to obtain a fault feature set; and S6, sending the fault feature set into a probabilistic neural network for fault classification. According to the method, a solution is provided for parameter selection of variational mode extraction, so that the optimal fault feature of the rotating machine is obtained, the signal decomposition efficiency and the fault classification precision are improved, and intelligentization of fault diagnosis of the rotating machine is promoted.

A flexible power distribution network voltage drop identification method based on a reinforcement learning theory comprises the following steps: carrying out dimension reduction processing on original data through principal component analysis, and inputting an optimized probabilistic neural network-learning vector quantization model to carry out voltage drop disturbance source classification; and voltage sag identification caused by three common single disturbance sources and two composite disturbance sources can be realized. The invention provides a principal component analysis-probabilistic neural network-learning vector quantization voltage sag source classification identification model based on learning algorithm optimization, which is based on analog voltage sag data, performs dimension reduction on the data by using principal component analysis, extracts data features based on a deep learning method probabilistic neural network, and performs classification identification on the voltage sag source based on the deep learning method probabilistic neural network. And classifying voltage drop types by utilizing learning vector quantization. The model can effectively identify the voltage drop type. Compared with other models which only use the probabilistic neural network, the principal component analysis-probabilistic neural network and the probabilistic neural network-learning vector quantization, the method has the best recognition precision on the voltage drop source, and has very good performance on small and medium-sized sample data sets.

The application discloses a fuel cell multi-dimensional state prediction method based on a multi-task probability network and dynamic gate pruning, belongs to the technical field of fuel cell health management, and specifically relates to the following: obtaining fuel cell state data, including multi-cell voltage, partitioned cell current density and multi-dimensional operating condition parameters, performing double-target collaborative feature screening on the multi-dimensional operating condition parameters to obtain an optimal collaborative feature subset; constructing a parallel multi-task probability neural network architecture containing multiple base models, and independently training each base model based on a training set; and utilizing an online rolling monitoring and dynamic gate collaborative pruning integrated mechanism to output final multi-cell voltage global probability prediction curves and cell current density global probability prediction curves. The application can improve the accuracy, robustness and reliability of fuel cell multi-dimensional state prediction.

The invention discloses a single-foot jumping robot motion control method based on coupling flow model driving, and the method comprises the steps: outputting a decision action through an intelligent agent strategy model, and carrying out the online interaction with a single-foot jumping robot, and collecting a real environment sample; an environment dynamic model is constructed through a probabilistic neural network, and the model is trained in a supervised learning mode to simulate the motion state of the robot; building a coupling flow model by utilizing a multi-layer coupling structure, converting the output distribution difference into a reward signal, and realizing dynamic optimization of an environment model by reconstructing a Markov decision process and combining with a reinforcement learning algorithm; and finally, interacting the intelligent agent strategy with the calibrated environment model to generate a high-precision simulation sample, and combining with a real sample to complete strategy iteration updating. According to the method, accumulated rewards equivalent to model-free reinforcement learning can be achieved only through a small amount of environment interaction, and the learning efficiency and the model generalization ability of a single-foot jumping robot strategy are remarkably improved.

The application discloses a high-voltage bushing state diagnosis method and system, relates to the technical field of high-voltage bushing state diagnosis, and aims to solve the technical problem of low accuracy of the existing diagnosis method in high-voltage bushing state diagnosis.The high-voltage bushing state diagnosis method comprises the following steps: obtaining a visible light image and a thermal image of a high-voltage bushing; performing feature fusion on the visible light image and the thermal image to obtain a fused image; extracting feature data in the fused image by using a scale-invariant feature transform; inputting the feature data into a probability neural network optimized by using a salp swarm algorithm; and outputting a high-voltage bushing state diagnosis result represented by the fused image by the probability neural network.The diagnosis result comprises normal, mechanical failure and electrical failure.The technical scheme of the application is used for providing a high-voltage bushing state diagnosis method and system.

The invention discloses a forklift attachment structural member stress analysis method based on ASA combined with PNN. The forklift attachment structural member stress analysis method comprises the steps that original stress data in a forklift attachment period are collected; decomposing the original stress data into a plurality of independent modes and residual modes; performing signal reconstruction to obtain a reconstructed signal; fusing the time domain features and the frequency domain features in the signals to obtain time domain and frequency domain combined features of the stress of the typical structural member of the attachment of the forklift truck; carrying out feature screening on the obtained time-frequency domain joint features of the stress based on an adaptive myxobacteria optimization algorithm ASA to obtain stress screening features; and taking the screened features as the input of a probabilistic neural network PNN model, constructing a stress analysis model of the typical structural member of the attachment of the forklift truck to analyze the stress data acquired in real time, and outputting an analysis result. According to the method, the model is constructed by collecting data, the stress characteristics of the forklift attachment structural member are effectively screened, and the requirement for online accurate analysis of the stress of the structural member can be met.

The invention provides a probabilistic neural network mineral product prediction method based on spatial awareness, and relates to the technical field of mineral product prediction, and the method comprises the steps: extracting mineral control elements from multi-source geological data as feature data, extracting mineral production area data as label data, and randomly dividing the feature data and the label data into a training set and a test set; constructing a mineral product prediction model based on the neural network, introducing a space kernel function to perform output category weighting on a neural network output layer, outputting a neural network regression coefficient and bias, and introducing Dirichlet distribution to model output; training the model by using the training set, constructing a loss function through negative logarithm likelihood of Dirichlet distribution and KL divergence regularization, and testing the trained model by using the test set; and performing mineral product prediction by using the tested model to obtain a prediction probability, and calculating the uncertainty of each sample. According to the invention, non-stationary prediction and uncertainty measurement and simulation of a predicted target space are realized.

The invention discloses a short-term photovoltaic power prediction method and system based on data mechanism combined driving. The method comprises the following steps: acquiring the geographic latitude, the solar declination angle and the solar hour angle of a photovoltaic power station to be predicted; calculating a solar elevation angle according to the geographic latitude, the solar declination angle and the solar hour angle, and determining the projection direction of the sun in the horizontal plane to obtain a solar azimuth angle; calculating the ideal irradiance of a photovoltaic panel plane based on the solar elevation angle and the solar azimuth angle in combination with the photovoltaic panel arrangement parameters of the photovoltaic power station to be predicted; introducing ideal irradiance of a photovoltaic panel plane to establish a photovoltaic power prediction probability neural network PNN model; and constructing a photovoltaic power prediction neural network PVFNN model by combining a long short-term memory network LSTM on the PNN model, and realizing short-term photovoltaic power prediction by using the PVFNN model. According to the method, the irradiance model in ideal weather is organically embedded into the neural network, learning in a data driving process is constrained, and a predication result with interpretability can be obtained.

The invention discloses a transformer winding anti-short circuit capability evaluation method and device, and belongs to the technical field of transformer operation states, and the method comprises the steps: carrying out the accelerated aging test of a winding component of a transformer, and obtaining a parameter set reflecting the mechanical performance; establishing an electromagnetic-structure coupling model of the target transformer, and calculating electromagnetic force distribution data acting on each part of the winding under a special short-circuit working condition based on an analytical method and a finite element method; a probabilistic neural network algorithm is adopted, the parameter set and the electromagnetic force distribution data serve as input, and a cumulative effect model of the transformer winding under multiple short circuit impacts is trained and established; and inputting actual operation history or preset short-circuit working condition information into the cumulative effect model, and predicting and evaluating the anti-short-circuit capability of the transformer. The transformer winding cumulative effect model and the winding state under the special working condition are evaluated, and safe and reliable operation of the transformer is guaranteed.

The invention discloses a hyperspectral remote sensing image super-resolution reconstruction method based on a self-attention total probability neural network, and the method comprises the steps: obtaining a hyperspectral image under the same sensor, carrying out the resolution reduction preprocessing of the hyperspectral image, and obtaining a low-resolution hyperspectral image; the method comprises the following steps of: inputting a low-resolution hyperspectral image into a pre-constructed and trained super-resolution network reconstruction network to obtain a reconstructed high-resolution image, and training the network: carrying out coarse extraction on the low-resolution hyperspectral image, obtaining a coarse spatial spectrum feature map, carrying out feature enhancement by utilizing a global self-attention mechanism, and obtaining a reconstructed high-resolution image; the method comprises the following steps: acquiring a fine spatial-spectral feature map, setting a discrete intensity interval, calculating the discrete intensity probability of the fine spatial-spectral feature map, combining the discrete intensity probability with low-resolution position prior, acquiring a total probability feature under high resolution, and reconstructing a high-resolution image according to the total probability feature. According to the invention, high-efficiency super-resolution reconstruction of the hyperspectral image is realized.

The application relates to a robot reinforcement learning control method, system, device and storage medium. The method comprises the following steps: obtaining real interaction data of a robot control task, constructing an environment dynamics model based on a probability neural network set by using the real interaction data, and capturing the uncertainty of a real environment and generating virtual data by using the environment dynamics model; combining the virtual data with the real interaction data to generate mixed training samples, training a parameterized policy network and a value network by using the mixed training samples; performing model-based reinforcement learning on the policy network and the value network, introducing a dynamic relative entropy regularization mechanism in the policy optimization process of the reinforcement learning, and generating an optimal policy of the robot control task. By introducing the dynamic relative entropy regularization constraint mechanism in the model-based policy optimization framework, the cumulative influence of the environment model prediction error is inhibited, and the training instability problem caused by the model bias is effectively alleviated.

The application discloses a power distribution system harmonic detection method and device based on a neural network, and belongs to the technical field of power quality disturbance detection. The method comprises the following steps: in step S1, voltage time domain signals are respectively acquired at R monitoring points of the power distribution system; in step S2, a neural network model outputs power quality disturbance event information according to the voltage time domain signals; in step S3, if the power quality disturbance event type is the power harmonic event, a current time domain signal is acquired by a power signal detection instrument; if it is other events, the power quality disturbance event information is displayed; in step S4, a probabilistic neural network outputs a harmonic component recognition result according to the current time domain signal and displays the result. The application can effectively reduce the complexity of power quality disturbance event judgment, and improve the accuracy of power harmonic detection of the power distribution system.

A method includes training, using data collected from sensors, a probabilistic neural network (NN) model including a set of model weights. The probabilistic NN model is trained to filter out data samples causing a threshold level of model uncertainty. The method includes training, at each cycle of training the probabilistic NN model and based on the set of model weights, a common estimator to generate gradient updates to the set of model weights that are to predict whether model updates from the sensors are anomalous. The method includes assigning, to each sensor, a trust coefficient value that estimates a level of trustworthiness of the model updates. The method includes transmitting the set of model weights to a subset of the sensors for which the trust coefficient value satisfies a threshold value.

The present application belongs to the field of unconventional and new energy technology, and discloses a shale oil soluble hydrocarbon prediction method, which comprises the following steps: S1, analyzing and obtaining total organic carbon content sensitive well logging curve, and combining with measured geochemical data to establish a calculation model of soluble hydrocarbon and total organic carbon content sensitive well logging curve; S2, extracting well logging data in a production area, and importing the calculation model to calculate soluble hydrocarbon single well data; S3, establishing a data set of seismic attributes and soluble hydrocarbon single well data, taking soluble hydrocarbon as target data, optimizing sensitive seismic attributes through intersection of target data and seismic attributes, applying a probability neural network method to machine learning to establish a mapping relationship between target data and sensitive seismic attributes, and then calculating a soluble hydrocarbon target probability body on a three-dimensional seismic body to obtain a planar prediction result of soluble hydrocarbon. The present application realizes effective prediction of soluble hydrocarbon by using geochemical data, well logging data and seismic data, and provides accurate data support for shale oil sweet spot optimization.

The application discloses a heavy truck electric drive axle fault diagnosis method based on data driving, first collects the running state data of the heavy truck electric drive axle in various use scenarios, and labels the fault categories to construct an original data set; a generative adversarial network based on weight noise is used for sample generation, data expansion is realized, and the expanded data and the original data are combined to form a training data set; a self-encoder based on boundary smoothing is selected as a feature dimension reduction model, and the feature dimension reduction model is trained by using the training data set; a probabilistic neural network based on quantum probability guidance is used as a classifier model, and the classifier model is trained by using the features output by the feature dimension reduction model; finally, the original data of the heavy truck electric drive axle to be evaluated are collected, and are input into the trained feature dimension reduction model for feature processing; then, the processed features are input into the trained classifier model for classification, and a classification result is obtained. The application can realize high-precision and high-robustness of electric drive axle fault diagnosis.