Multilayer Perceptron Neural Network Supervised Learning Based Solar Radiation Prediction

• Published in: Distributed Computing and Optimization Techniques Vol. 903; pp. 625 - 634
• Main Authors: Shyamala Devi, M., Peter Soosai Anandaraj, A., Venkata Thanooj, K., Sandeep Guptha, P. V., Jayanth Reddy, A.
• Format: Book Chapter
• Language: English
• Published: Singapore Springer 2022; Springer Nature Singapore
• Series: Lecture Notes in Electrical Engineering
• Subjects: Activation layer; Correlation; Multilayer perceptron; Regressor
• ISBN: 9811922802; 9789811922800
• ISSN: 1876-1100; 1876-1119
• DOI: 10.1007/978-981-19-2281-7_58

Worldwide solar radiation is a basic parameter for the plan and operation of solar radiation frameworks. Long-standing records of worldwide solar radiation data are not accessible in many places because of the cost and maintenance of the measuring instruments. Sun based radiation expectation contains an incredible significance in power generation from sun based energy and makes a difference to measure photovoltaic control frameworks. With this overview, this paper enabled to predict the solar radiation using machine learning algorithms. With this context, we have utilized solar radiation dataset extracted from UCI Machine Learning. The proposed model Multilayer perceptron (MLP) based Neural Network for forecasting solar radiation are attained in four ways. Firstly, the data set is analyzed and preprocessed with Feature Scaling and missing values. Secondly, the correlation of the features is done and the relation of each features are visualized. Thirdly, the raw solar radiation data set is fitted to all the regressors and the implementation is furnished before and after scaling. Fourth, the raw data set is subjected to multilayer perceptron with various activation layers like identity, logistic, tanh and relu layers. The performance is analyzed with EVS, MAE, MSE, RScore and run time of the neural network layer. The execution is done using python language under Spyder platform with Anaconda Navigator. Experimental results show that the Gradient boost regressor have the RScore of 0.98 before and after feature scaling. The MLP regressor with TANH activation layer is tends to retain 0.99 Rscore before and after scaling.