Torque Estimation Of Vehicle Engine Based On Artificial Neural Networks

How to realize smooth power transfer under state-switch is one of the difficult control issues of Parallel hybrid system. Electromotor torque compensation for the engine torque can be used to reduce the fluctuations of total torque and vehicle acceleration under state-switch, so a smooth power transfer under state-switch is ensured. Therefore, real-time engine torque estimation is required.This paper presents a torque estimation method by the direct description of the torque value. Use the neural network forecast function, let some certain steady operating condition data as the sample, to estimate all the steady operating condition torque value; choose 100%/s as the throttle opening change rate at state-switch and shifting process in hybrid system, and estimate the engine torque output characteristic at this dynamic operating condition.First, complete a series of steady and dynamic tests on TU5JP/K gasoline engine based on the PUMA Open automation test system. As the system has high dynamic response, it can complete the acquisition of various parameters at different frequencies, particularly to the parameters which change at high frequency in dynamic operating condition. So the system can acquire the data in the engine test accurately in real-time. The neural network can get sample from the data.Then, choose the BP neural network which has highly non-linear mapping function and can distinguish the noise or the distortion in sample carries out the torque estimate. Use MATLAB to establish the BP neural network forecast model, and determine its structure and the related parameters. Select the engine speed and throttle opening as inputs, because they have the greatest impact on the torque. Using interval extraction samples data as sample selection methods. Carrying out the normalized pretreatment to the inputs and the output data to improve the data quality, it can enhance the modeling training and forecast precision.Finally, complete the torque estimation of vehicle engine based on BP neural network forecast model. And then calculate and analyze the neural network's convergence precision and generalization effect, find the influence factors of torque estimation precision, and propose the optimization solution to perfect this method.