Design And Realization Of BP Algorithm Applied To IGBT Junction Temperature Prediction

IGBT(Insulated Gate Bipolar Transistor)is a voltage-controlled composite device,which is widely used in the field of power electronics.In actual use,power electronic devices often fail due to excessively high or low temperature.Therefore,it is of great significance to study the junction temperature prediction method during the operation of IGBTs.In this paper,we study the junction temperature prediction method of IGBT based on BP neural network.The saturation voltage drop and collector current of IGBT measured by the temperature-sensitive parameter method are used as the input of the BP neural network,and the junction temperature of IGBT module is used as the output of the network.The BP network is trained by MATLAB,and the prediction results are simulated.Since the BP neural network has its own defects that it is easy to fall into local minimum values and cannot reach the global optimum,the paper adopts genetic algorithm and the LM improved BP algorithm optimized by mind evolutionary algorithm.And the simulation results show that compared with the network optimized by genetic algorithm,the LM improved BP algorithm optimized by mind evolutionary algorithm has a lower fluctuation range and a better accuracy in terms of regression coefficient(R),prediction error,prediction error percentage and so on.The maximum fluctuation of prediction error does not exceed 15? and the maximum percentage of prediction error does not exceed 0.3%.Based on the optimized network structure,the paper extracts the optimal threshold value and the weight value of the network into the memory,and completes the FPGA design of the network using the verilog hardware description language.Based on the SPARTAN-6 development board from Xilinx,the paper carried out the hardware design and verification of the system,which includes data acquisition module,data storage module and neural network calculation module.The paper completed the design and simulation of each module,and finally verifed on the experimental platform.The results show that the junction temperature prediction of IGBT modules can be achieved within the error range,and the maximum error between the measured junction temperature and the calculated junction temperature is 7?,which meets the expected requirements of the algorithm.The research work of the paper has some reference significance for IGBT online junction temperature monitoring method.