Load Prediction in Smart Grid Networks

Abstract

Efficient forecasting and load prediction for maintaining the accurate DR (Demand Response) ratio is a key factor in implementing and deploying the Smart-Grid networks [1]. There are a plethora of techniques and models suggested by forecasters over the decades, the most accurate and feasible being – artificial neural networks, linear regression technique and the curve fitting algorithm. Researchers have demonstrated extreme zeal and effort in devloping algorithms which could derive the best effeciency, thus saving excess production than demand. For example, the work descrbied in the paper [2] puts forward the prediction values to be at an accuracy of around 95%. A hybrid algorithm has been presented in this paper, which has been practically proved to have a forecasting efficiency much higher than the conventional methods. Using the artificial neural networks for training the model with historical data and fluctuations in demand, the linear regression method has been used for implementing the temperature sensitivity, namely – dew point, humidity, wind speed, seasonal variations and location of the smart-meter. Together along with the curve fitting algorithm, the proposed hybrid algorithm has been practically implemented by taking data from smart-meters across the United States to determine their efficiency of implementation. The proposed algorithm described in this paper encountered a marvelous prediction accuracy of 99.2% - 99.45%, which promises vast reduction in the power wasted by power utility companies owing to the mismatch within the DR rates from the consumer end and is far accurate than the predictions made by [2].