The Research On The Electrical Locomotive's Permanent Magnet's Inverted Question

In this thesis, the principal problem that needs solution is to discovery the best method of the electric locomotive's excessive permanent magnet, that is to discovery the sample of the known magnetic field's inverted design's structure of the magnet. Based on this problem, there are three parts to solve it. Firstly, the method of the finite element analysis and the software of the two orientation finite element analysis have been introduced. Through creating the finite element model and assaying its data, the magnetic field intensity data of every specimen's are obtained. Based on this, using MATLAB/SIMULINK, the calculator emulation model about the import of sample's dimension and the output of the square of the magnetic field intensity's error is created. The specimen's warehouse about the magnetism field intensity is obtained by the calculator emulation model. Then, trying to set up a target function of inputting the size of the magnet's sample and outputting the summation of the square of error of the sample's magnetic intensity and the real magnetic intensity. Because of the complicated nonlinear relationship between the input and the output and the non-settled model's progress, the thesis adopts the method that Artificial Neural Networks (ANN) with the capability of high nonlinear recognition. That is to say, the model that reflects the relationship between the measurements and the summation of the square of error is obtained by Back Propagation Multi-layer feed forward net-work (BP neural network). Specified with the momentum gradient method and the L-M optimization calculation to train BP neural network, it adopts the training result of L-M optimization calculation according to the accuracy of the error to create the reflecting relationship between the inputting sample size and the outputting summation of the square of the error— the target function. Finally, it tries to create a optimization math model on the basis of taking the magnet's size as variable, taking the smallest summation of the square of error as the target function, and taking the size limit as the restriction. And then it tries to obtain the best method approaching to the real data with the optimization method of genetic algorithm(GA). In order to show the GA's validity, the thesis also achieves the optimization calculation with Quasi-Newton method. Two method's results are accordance. The method that ANN controls GA can solve such problems that knowing the sample's data and unknowing corresponding function or constitution. By using MATLAB/SIMULINK, MATLAB Neural Network Toolbox (Nnbox), MATLAB Optimization Toolbox and MATLAB Genetic Algorithm Toolbox(GA Toolbox), it can prevent repeated calculation and surplus way, find the corresponding relationshipmodel by the best way, and solve the specific optimization problem. The results shows the speculate manner and the optimization's feasibility and effective.