Energy Demand Prediction Considering Weather Change Effects using Artificial Neural Network

This study proposes the development of a comprehensive model of prediction for energy demand. The very first step is to do prediction using the energy need as the input data. During the second stage, weather variables such as temperature, precipitation, relative humidity, specific humidity, surface pressure, wind, and earth skin temperature are considered to forecast the energy requirement. The variability in these variables has a substantial impact on the needs of energy. The proposed approach is implemented for two southern States in India, namely Kerala and Tamil Nadu. We have considered monthly data for 11 years on energy requirement and weather variables of these two States, and the Artificial Neural Network technique is used for energy prediction. The training data are divided into two sets comprising low values of energy requirements and higher values of energy requirements to improve the prediction. Four more dual Artificial Neural Network models are constructed by considering the various possible combinations of these models. The results indicate that the dual model consisting of lower energy requirements values without weather and higher energy values with weather yields the best results for Kerala State and the dual model consisting of lower energy requirements values without weather and higher energy values without weather yields the best results for Tamil Nadu State. The findings show that the proposed Artificial Neural Network models can predict energy requirements for the State of Kerala with a Correlation Coefficient, Root Mean Square Error and Mean Absolute Error of about 0.99, 31.55 Million Unit and 16.61 Million Unit, while for the State of Tamil Nadu with a Correlation Coefficient, Root Mean Square Error and Mean Absolute Error of about 0.99, 46.93 Million Unit, and 31.49 Million Unit.