A new technique for highly efficient sensor-less control of electric vehicles by using neural networks

A new controller based on artificial neural networks (ANNs) for induction machines is introduced and implemented. It is designed for efficiency maximization. It consists of two ANN for estimation of speed and torque, and efficiency optimization. The first neural network (ANN1) is trained to estimate speed and torque of the machine to use in the feedback loop of the control system. The second neural network (ANN2) is used for commanding optimum voltage and frequency that maximizes the drive efficiency. ANN1 has been trained with the data obtained using the machine model operating under different speed, load, temperature and flux density. To provide the required data to train and test ANN2, a simulation program was written to calculate commanded voltage and frequency that would drive the induction machine with maximum efficiency at different load, speed, and temperature conditions. The controller is able to control the induction machine over a wide speed range from standstill to high speeds in the flux weakening region. The trained neural networks are employed for the control of a 7.5 kW induction machine under different loads. The proposed controller is an appropriate technique for speed sensor-less control of an induction machine to drive an electric vehicle (EV). The performance of this control system has been found to be as good as those controllers which use the induction machine model. The description of the control system, training procedure of the neural network are also given. The test results obtained for a torque control scheme suitable for the control of an EV is also presented.