Short Term Forecasting of Solar Power Using Harmony Search based Extreme Learning Machine

The grid incorporation of solar power production creates nonlinearity, instability and entails improper energy management, which is still a challenging situation while installing a large solar power plant. Thus prediction of solar power generation is highly essential in different time horizons for maintaining proper power adjustment techniques and management. In this work a smart and competent prediction model has been implemented on a real time photovoltaic power plant. The Extreme learning machine (ELM) which is a new prediction technique is applied in this work. The weights of ELM are selected randomly in general. The performance of the forecasting model also depends on proper weight selection. Thus a new optimization technique such as harmony search optimization is applied to select the optimized weights. The forecasting model is implemented on an historical data set of real time solar power plant whose geographical location is given in the last part of section-II. The ELM model is activated to mobilize the feed forward neural network, iteratively, to achieve better forecasting error in each step. ELM model and the Harmony Search optimized ELM model are simulated for error calculation and their results are compared in terms of different measuring indices and their forecasting errors are compared.