A SUPERIOR TRAINING STRATEGY FOR THREE-LAYER FEEDFORWARD ARTIFICIAL NEURAL NETWORKS

摘要

A new algorithm is proposed for the identification of three-layer feedforward artificial neuraludnetworks. The algorithm, entitled LLSSIM, partitions the weight space into two majorudgroups: the input- hidden and hidden -output weights. The input- hidden weights are trainedudusing a multi -start SIMPLEX algorithm and the hidden -output weights are identified usinguda conditional linear- least- square estimation approach. Architectural design is accomplishedudby progressive addition of nodes to the hidden layer. The LLSSIM approach provides globallyudsuperior weight estimates with fewer function evaluations than the conventional backudpropagation (BPA) and adaptive back propagation (ABPA) strategies. Monte -carlo testingudon the XOR problem, two function approximation problems, and a rainfall- runoff modelingudproblem show LLSSIM to be more effective, efficient and stable than BPA and ABPA.