The Deep Learning Algorithm And Application Of Permanent Magnet Synchronous Motor Performance Analysis

The deep integration of new generation artificial intelligence technology and electrical equipment optimization design is becoming the main driving force of new industrial revolution.As a new type of energy-saving motor,permanent magnet synchronous motor(PMSM)is widely used in electric vehicles,wind power generation,rail transit and other fields because of its simple structure,high efficiency,reliable operation and other advantages.In this paper,the agent model of motor parameters and performance parameters is established by using deep learning,which can achieve highprecision prediction of motor performance and effectively reduce the time required for motor performance analysis and optimization.Firstly,this paper introduces the related theories of deep learning from the aspects of the construction of neurons,the structure of perceptron,the working principle and limitations of BP neural network,and analyzes the characteristics of shallow learning and deep learning,which provides the theoretical support for the performance prediction method of PMSM Based on deep learning;Then,according to the needs of deep feature extraction and dimension resolution,a multi-objective performance prediction method of permanent magnet synchronous motor based on stack auto encoder neural network(SAE-NN)is proposed.7776 groups of sample data are obtained by finite element software simulation,and this data is used as the input and output data of deep learning model.The statistical method is used to analyze the super parameter sensitivity of SAE-NN model,and Taguchi method is used to optimize the super parameter of sae-nn prediction model.An improved scheme is proposed to obtain the optimal structure of SAE-NN model.Based on the results of optimization,a prototype experiment is made to verify the reliability and precision of the performance prediction method of permanent magnet synchronous motor proposed in this paper accuracy and can effectively reduce the time of motor performance analysis and optimization.Based on the reliability of the above algorithm,in order to improve the prediction accuracy,a long short term neural network is proposed Memory(LSTM)multi-objective performance prediction method for permanent magnet synchronous motor also takes 7776 groups of sample data as input and output data of LSTM prediction model.By optimizing the super parameters(optimizer,epoch,number of neurons,activation function,number of hidden layers)of LSTM model,the optimal structure of LSTM model is determined.Compared with BP neural network,radial basis function and support vector machine,the results show that the multi-objective performance prediction method based on LSTM model has higher accuracy and effectively reduces the dependence on label data.Finally,the pruis case is used to verify that the LSTM prediction model can not only ensure high accuracy,but also be applicable to different performance indexes of PMSM,and has superior data feature expression ability and strong generalization ability when dealing with multi input,multi output and non-linear data.With the advantages of multi input,multi output and non-linear data processing,deep learning can accurately express the complex non-linear relationship between the structural parameters and performance parameters of PMSM,and has a high prediction accuracy.At the same time,it is verified with the finite element analysis method of motor design,which lays a foundation for the multi input and multi output problems in motor design,and provides theoretical and practical support for the application and promotion of deep learning algorithm in the field of electrical equipment.