Short Term Load Forecasting: A Hybrid Approach Using Data Mining Methods

Load forecasting and analysis play a crucial role in electric power management and market planning of the grid system. In this paper, a hybrid approach using data mining methods is proposed for short term load forecasting. The approach uses the AFTER (Aggregated Forecast through Exponential Reweighting) algorithm to combine k-means clustering-Naive Bayes classification model and ARIMA (Autoregressive Integrated Moving Average) model forecasts to form a hybrid model for load forecasting from the supply side. The error in the forecast from the hybrid model is also corrected and reduced using the neural network. To validate the effectiveness of the proposed method a case study has been put through where load forecast is performed using actual load data from AEGCL (Assam Electricity Grid Corporation Limited), Assam. Load forecast is done for the first week of December 2012 and 2018 using training data set from January to November 2012 and 2018 respectively. The error of the forecast load is measured using MAPE (Mean Absolute Percentage Error) to test the accuracy of the proposed method. The hybrid model forecast with the neural network error model extensively improves prediction accuracy of the load forecast from the supply side when compared to individual forecast models considered.