Solar Energy Prediction Using Particle Swarm Optimiser Trained Recurrent Neural Network

Ibrahim Makama Dattijo and Salihu Aish Abdulkarim

Department of Computer Science, Federal University Dutse, Nigeria

Corresponding Author:


In recent years, renewable energies have been covering an increasing part of the worldwide electrical power demand. The amount of solar radiation received by the earth’s surface is very important to numerous renewable energy applications. However, direct measurement of solar data is not widely available, especially in developing countries such as Nigeria. Numerous statistical methods have been developed and used in the areas of solarenergy prediction. Statistical approaches such as; Auto Regressive Moving Average (ARIMA), and linear regression (LR) are used. The challenges of using these statistical approaches are Linear methods (works best when the data is linear), high mathematical complexity and dependence on prior knowledge of how the time series was generated. In the meantime, data and computational poweravailability have enabled machine learning algorithms to perform improved predictions. The need to predict solar energy output is essential; machine learning models can be employed. In this work, Particle Swarm Optimization (PSO) and Back propagation (BP) algorithms were used to train a Recurrent Neural Network (RNN) to model monthly mean daily solar radiation values. The inputs to the network are month, sunshine duration, latitude, longitude, and altitude of the location. The BP algorithm is compared with the PSO algorithm trained by the RNN. The PSO outperformed the BP in most cases with average accuracy for the BP at about 0.71, 0.725, 0.76, 0.735 and 0.78, 0.8, 0.84, 0.815 for the PSO algorithm for the four (4) datasets used.

Keywords and Phrases:

Artificial Neural Networks,
Particle swarm optimisation,
Recurrent Neural Network,
Time series