The adaptive kernel-based extreme learning machine for state of charge estimation

The state of charge (SOC) is a key factor in the battery management, and the accuracy of its estimation plays an important role in battery-life prediction. This paper develops a kernel-based extreme learning machine for SOC estimation. The extreme learning machine is a single hidden layer feedforward neural network with a randomly initialized weight matrix and a bias vector. Unlike the traditional neural network, it does not need to update the network parameters. Then, the kernel-based method is combined with the extreme learning machine to avoid overfitting in parameter estimation. In the example, three kernel-based extreme learning machine methods and the regularized extreme learning machine method are used to train the model. The simulation results show the effectiveness of the proposed methods.