Variable structure systems approach for on-line learning in multilayer artificial neural networks

A new sliding mode control approach is proposed for on-line learning in multilayer feedforward neural networks having scalar output. Such neural structures are commonly used for on-line modeling, identification and adaptive control purposes in case variations in process dynamics or in disturbance characteristics are present. The network weights are assumed to have capabilities for continuous time adaptation. The zero level set of the learning error variable is considered as a sliding surface in the learning parameters space. The proposed approach represents a simple, yet robust, mechanism for guaranteeing finite time reachability of zero learning error condition. Results from simulation experiments related to the application of the proposed learning algorithm for neural on-line identification of manipulator dynamics are presented. They show that the neural model inherits some of the advantages of the sliding mode control approach, such as high speed of learning and robustness.