Design of a Dynamic Neural Network with Kalman Filter for the Identification of Nonlinear Systems. Application: prediction of the maximum power generated by a Photovoltaic module

In this paper, the design of dynamic neural networks with Kalman filter is proposed and applied to identify nonlinear dynamic systems. The optimal parameters of a dynamic neural network, which contains several autoregressive moving average (ARMA) sub models, weighs and biases, are obtained using the well known delta rule. Using the obtained parameters of the ARMA sub models, a new dynamic network based on Kalman filter is designed. The obtained system is able to filter noise, identify the complex parameters and predict one more step ahead comparing to the original dynamic network. Due to the existing denominators inside each sub model, the stability of the dynamic neural networks should be mentioned. A maximum power of the photovoltaic system was chosen as a realistic nonlinear system to demonstrate the identification performance. Several simulation results were carried throughout this paper.