Abstract: Regarding drawbacks of the standard BP algorithm, a high-order derivative based multiple memory BP algorithm is proposed. It combines the n-th order of energy function with the direction of the fastest decline to construct a new direction of the fastest decline, and improve the learning speed of the neural network. The new algorithm is compared with the traditional gradient algorithm to show its high computation speed. Implementation of the new algorithm is given. Finally a neural network controller is designed to optimize the performance of the automobile suspension. The result shows that the new algorithm is convenient and effective.

Key words: neural network, BP algorithm, high order derivative, suspension