Research On The Efficiency Of Motor Under Power Quality Disturbance Based On Deep Learning

The large-scale use of high-power nonlinear loads,power electronic devices and distributed new energy sources has brought many power quality problems and additional losses to the distribution network.As the most demanded motor,the operating efficiency of the asynchronous motor will directly affect the efficiency of the distribution network.Therefore,the accurate evaluation of the efficiency of asynchronous motors under the influence of power quality factors is of great guiding significance for the follow-up work on energy conservation and loss reduction in distribution networks.At present,the research on the efficiency of asynchronous motors under the influence of power quality factors mainly focuses on three-phase unbalance and harmonics.However,there are few researches on voltage fluctuation and flicker and comprehensive power quality factors.What’s worse,there is also a lack of enough field experimental data to support theoretical research.In addition,the T-type equivalent circuit is mainly used for the efficiency calculation of the asynchronous motor at present,which has the problems of complicated algebraic calculation,too many internal parameters of the motor and low accuracy.In response to the above problems,starting from the loss composition of asynchronous motors and the definitions of three-phase unbalance,harmonics,voltage fluctuations and flicker,this thesis introduces the efficiency calculation model of asynchronous motors under the influence of the above three power quality factors alone.Then,decoupling the comprehensive power quality factors,this thesis derives the efficiency calculation model under the influence of the comprehensive power quality factors,and uses MATLAB programming software to simulate and analyze each efficiency calculation model.On the basis of using the traditional T-type equivalent circuit to calculate the efficiency of the asynchronous motor,this thesis proposes a method for evaluating the efficiency of the motor based on the crisscross algorithm to optimize the feature attention mechanism and the convolutional neural network.Firstly,the probability weights between different input features and efficiency are calculated through the feature attention mechanism,and the important features are given higher weights.Secondly,used to learn and process the data characteristics of the measured motor data,the convolutional neural network fully extracts the nonlinear relationship between the input parameters and efficiency.Finally,the crisscross algorithm is used to perform secondary training on the weights and thresholds of the fully connected layer and the output layer,so that the model can further converge in the training process.In order to verify the model proposed,through the large-capacity power quality experimental platform,this thesis conducts a large number of field experiences on the power quality disturbance of asynchronous motors.By adjusting the disturbance source power cabinet,control cabinet and other equipment to achieve three-phase unbalance,harmonic and voltage fluctuation and flicker custom output functions,this thesis collects and summarizes the data of power quality disturbance of asynchronous motors.Finally,in order to establish the deep learning model proposed in this thesis,a large number of collected experimental data are used as training samples,when the average voltage,load rate and various power quality indicators of the asynchronous motor are used as input characteristic parameters,and the loss of asynchronous motor is used as the output target parameter.Comparing the error with T-type equivalent circuit and traditional machine learning method,the result shows that the efficiency evaluation method of the asynchronous motor proposed in this thesis has better performance in terms of efficiency evaluation accuracy,generalization and stability.