Machine Learning based Short Term Electric Load Forecasting for Domestic Users

Abstract

Pakistan is facing a lot of problems in generating the required amount of energy and supplying it to the people. Most of the utility companies use load forecasting to anticipate how much power they'll expect to fulfil the demand of the electricity. Load forecasting helps the utility companies to generate the required amount of energy which will benefit the companies economically. Load forecasting can be classified in three ways: Short Term Load Forecasting, Medium-Term Load Forecasting and Long Term Load Forecasting. In this research, we have focused on Short Term Load Forecasting (STLF). STLF is forecasting approach with a period of a couple of hours to a day. In past years, multiple machine learning algorithms and models are implemented and tested to predict the electric load accurately. It's possible that the forecasted outcomes might have yielded better results but the disadvantage is that they have used the whole data set to create the model and when the new data value is added, they reprogrammed the whole system from the start. This has resulted in several issues. When a large set of data is used to create a machine learning algorithm, it takes a lot of time to create and test the model; it requires great storage to store the big data and lastly it consumes power to process the large data set again and again for reprogramming which results in delayed data processing. The purpose of this research is to use an online machine learning method to create a model using STLF technique in which data is presented in a progressive sequence and the model seeks to learn and upgrade for the accurate prediction of new data points at each phase that will minutely forecast the electric load which would result in better power management. With Short-Term Electric Load Forecasting we can foretell the electric load and according to that we can smartly manage the power consumption, generate the electricity as per the demand and improve the situation of load shedding in Pakistan. We have implemented Recursive Least Square with Forgetting Factor algorithm to forecast the electric load for Domestic users using PRECON dataset. The results we have obtained from this model are quite promising. The average MAPE % ranges from 0.6996% to 2.162% in the month of June and in December ranges from 0.1567% to 4.2864%, which shows that our proposed model has outperformed and we are getting the optimal results.