Validity index in dynamic cell structures

The appeal of including adaptive components in complex computational systems, such as flight control, is in their ability to cope with a changing environment. Neural networks are adopted as a popular soft-computing paradigm to carry out the adaptive learning. The dynamic cell structures (DCS) network is derived as a dynamically growing structure to achieve better adaptability and employed for online learning of the intelligent flight control system (IFCS). As a crucial component of a safety critical system, the DCS networks need to be validated. Within the scope of validating adaptive systems, the validation of neural networks is particularly challenging due to their complexity and nonlinearity. The predictions of DCS networks are difficult to warrant because of the locally poor fitting during a relatively short time of adaptive learning. In this paper, we present the validity index, an estimated confidence interval associated with each output, as a reliability-like measure of the network´s prediction performance. Experimental results of validity index on the flight condition data collected from an IFCS simulator demonstrate an effective validation scheme for DCS networks.