Forecasting of PV Output Power in Cloudy Conditions by LOLIBEE, MLP-ABC and MLP Algorithms

Abstract

Forecasting PV power generated by photovoltaic panels (PV) in cloudy conditions is of great importance. The aim of this paper is to forecast the produced power by PV using LOLIBEE (Local Linear Bee Model), MLP-ABC (Multi-layer perceptron - Artificial Bee Colony) and MLP algorithms. Experimental data (ambient temperature, solar radiation, speed of wind and relative humidity) are collected at a five-minutes interval from Tehran University's PV laboratory from September 22nd, 2012 to January 14th, 2013. Upon validation of data gathered from the lab, 10665 data which are equivalent to 35 days are used in the analysis. The output power of PV was forecasted by constructing three models for different parts of a day using LOLIBEE, MLP-ABC and MLP algorithms (three models for each algorithm), which resulted in better precision by LOLIBEE with about 95% and 1.9 in terms of R² (Co-relation Co-efficient) and MBE (Mean bias error) respectively. The accuracy gained by our proposed model for dividing the day into three durations is also increased by about 1.5 percentage in comparison with the model which is covering the whole day.