Approximating Algorithm of Wavelet Neural Networks with Self-adaptive Learning Rate

This paper proposes a Wavelet Neural Networks (WNN) with self-adaptive learning rate. The algorithm can automatically change the learning rate with operational parameter, but without any artificial adjustments. Thus it once for ado overcomes the drawbacks of WNN, i.e. slow convergence, inability to determine the value of learning rate and easiness to fall into local minimum point. The results of simulation indicate that the algorithm is better than that of WNN with changeless learning rate when it is used in approaching non-linear functions, and is worth of promotion and popularization.