Modeling the output power of heterogeneous photovoltaic panels based on artificial neural networks using low cost microcontrollers

Abstract

Many implementations of artificial neural networks have been reported in scientific papers. However, few of these implementations allow the direct use of off-line trained networks. Moreover, no implementation reported the use of relatively small network adequate to run on low cost microcontroller. Hence, this work, which presents a small artificial neural network, which models the output power of heterogeneous photovoltaic panel. In addition, the work discuss the hardware implementation that allows such network to run on low cost microcontroller. The hardware implementation has the ability to model heterogeneous photovoltaic panel's output power with very high accuracy and fast response time. Feedforward back propagation has been used because of its high resolution and accurate activation function. Real-time measured parameters can be used as inputs for the developed system. The resulting hardware data is tested with data from real photovoltaic panels; to confirm that it can efficiently implement the models prepared off-line with Matlab. The comparison revealed the robustness of the proposed heterogeneous photovoltaic model system at different conditions. The proposed heterogeneous photovoltaic model system offer a proper and efficient tool that can be used in monitoring photovoltaic panels, such as the ones used in smart-house applications.