Short term load forecasting based on ARIMA and ANN approaches

Abstract

Forecasting electricity demand requires accurate and sustainable data acquisition systems which rely on smart grid systems. To predict the demand expected by the grid, many smart meters are required to collect sufficient data. However, the problem is multi-dimensional and simple power aggregation techniques may fail to capture the relational similarities between the various types of users. Therefore, accurate forecasting of energy demand plays a key role in planning, setting up, and implementing networks for the renewable energy systems, and continuously providing energy to consumers. This is also a key element for planning the requirement for storage devices and their storage capacity. Additionally, errors in hour-to-hour forecasting may cause considerable economic and consumer losses. This paper aims to address the knowledge gap in techniques based on machine learning (ML) for predicting load by using two forecasting methods: Auto Regressive Integrated Moving Average (ARIMA) and Artificial Neural Network (ANN); and compares the performance of both methods using Mean Absolute Percentage Error (MAPE). The study is based on daily real load electricity data for 709 individual households were randomly chosen over an 18-month period in Ireland. The results reveal that the (ANN) offers better results than ARIMA for the non-linear load data.

Keywords

• Artificial Neural Network (ANN)
• Auto Regressive Integrated Moving Average (ARIMA)
• Smart grid
• Short time load forecasting (STLF)
• Storage device