Development of daily three-layer neural networks to estimate PV power from global horizontal irradiation

Abstract

In this study, a daily three-layer artificial neural network (ANN) which converts from global horizontal irradiation (GHI) to photovoltaic (PV) power is developed. Here, each ANN estimates PV power of a specific day after the ANN learns the data from days around the specific day. Though the three-layer ANN has small numbers of the weight in the ANN, daily ANNs are expected to include the seasonal dependence which appears in the conversion from GHI to PV power. In addition to GHI, temperature, humidity, and solar azimuth are given as input variables. As a result of applying the developed daily ANN to a sunny day, a cloudy day, and a rainy day, it was confirmed that the PV power generation in these object days could be accurately estimated. The estimation errors were also evaluated for one year and effectiveness of daily three-layer ANNs with proposed input parameters was shown.