Identification of stationary/nonstationary systems using artificial neural networks

Multilayered perceptron type artificial neural networks were investigated as nonlinear signal processors used for the purpose of identification of stationary and nonstationary systems.; The ANN can generally generalize a set of given training data for the stationary system. In this regard, the ANN can be considered as a type of time series or regressional paradigm. In this study, an ANN used algorithm which combines both time series and regressional techniques is proposed.; Three real problems were considered: (1) the power system security assessment problem, (2) the electric load forecasting problem, and (3) the optical sensing of particle size distribution problem. Each was investigated for the application of the ANN to the identification of stationary systems.; The results show that the ANN is able to identify the functional relationship and properly generalize among the training data sets. The importance of the results is that, once trained, the ANN represents the complex mathematical relationships of the systems which otherwise must be explicitly simulated.; A training procedure is proposed that adapts the weights of a trained layered perceptron artificial neural network to training data originating from a slowly varying nonstationary process. The resulting adaptively trained neural network (ATNN), based on nonlinear programming techniques, is shown to adapt to new training data that is in conflict with earlier training data without affecting the neural networks' response to data elsewhere. The ATNN also allows for new data to be weighted in terms of its significance. The ATNN is applied to the problem of electric load forecasting and is shown to significantly outperform the conventionally trained layered perceptron.