Modeling techniques of electricity consumption forecasting

Abstract

Nowadays the problem of the efficient use of available energy resources is urgent. It causes the development of energy-saving technologies, including using renewable energy sources and their implementation in existing energy systems. Decentralized electricity generation becomes more expanded. Smart Grids are designed to provide real-time data management for the balance between electricity demand and generation level. To operate and maintain the Smart Grid decision support tools aimed to achieve grid reliability by reducing peak demands and increasing energy efficiency are designed. Decision-making support on the energy management of the hybrid power grid requires in particular determination of the level of electricity consumption, which will determine the effective mode of the hybrid power grid operation. The aim of the study is to compare different techniques and develop forecasting models of electricity consumption with high accuracy. Electricity consumption depends on many variables such as consumer load, environmental conditions, etc. It was considered such forecasting techniques as regression analysis, fuzzy regression analysis and machine-learning based models. The dataset for forecasting was created from archive data of the gas station daily electricity consumption. Autoregressive forecasting models, exponential smoothing models, and neural network models were built for short-term electricity consumption forecasting. The accuracy of proposed models was compared by MAPE criteria. The developed forecast models are possible to obtain a daily schedule of electricity consumption for the next months, as well as an operational forecast for the next day. Forecasting results are used in the decision support system on the operation of the power grid with renewable energy sources.