A street lighting based on solar energy: optimization of a neural network for the productivity evaluation in different climate conditions

This paper describes a stand-alone public solar street lighting system powered by photovoltaic (PV) cells with energy storage battery and an LED consumer installed along a street located in the center of Italy. The electric system parameters were collected during a year and simultaneously the weather conditions were monitored. The PV cells are applied to the cylindrical surface of the pole and the forecasting of the PV productivity is a critical aspect. Therefore, the annual monitored data were implemented in order to train an Artificial Neural Network (ANN) able to find the productivity of this application also in other locations. Correctly sizing ANNs is fundamental to have a trade-off among model accuracy and computational cost. After determining the best configuration of the ANN, different scenarios were compared to assess the validity of the developed tool.

I. Introduction

In recent decades, a significant increase in energy demand has been observed worldwide. This phenomenon, in conjunction with the increase in the greenhouse effect generated by the use of fossil fuels, has prompted the energy market to move towards the search for and effective integration of new, clean and cheap energy sources and to adopt energy-saving policies. The technological revolution that is taking place has mainly focused on optimising the use of traditional energy sources [1] and developing renewable energy sources (photovoltaic, solar thermal, wind, biomass, etc.) [2]-[5].