Comparative Study of Shortterm Electricity Price Forecasting Models to Optimise Battery Consumption

Electricity price prediction has been crucial for network operators in the UK energy market. There has been a need for effective and accurate electricity price prediction model which will help in optimisation of energy storage devices, and better bidding plans. The high accuracy of prediction assists the electricity generators and suppliers to maximise their benefits and offer better services for their clients. In this paper, we present the first ever comparative study of different time series techniques to predict short-term electricity prices for a use case of optimising on-site battery consumption. As a use case, the UK Market Index Data (MID) from Elexon API has been used. This dataset is a representation of half-hourly electricity prices where the price value varies with time. The pattern of time-series for electricity prices is quite erratic and complex to predict. For the short-term electricity price prediction, different forecasting models have been analysed, starting from simple models for univariate time-series to more advanced models with exogenous variables or multivariate time-series. Autoregressive Integrated Moving Average (ARIMA), Prophet and XGBoost algorithms are implemented, tested and compared to highlight advantages and weaknesses of these models for short-term electricity price prediction. The comparative analysis showed that the XGBoost model outperforms the ARIMA and Prophet models. Also, the XGBoost model has a faster execution time than the ARIMA and Prophet models. For validation and analytical experiments, in this paper, a use case has been applied with the dataset extracted from January 2017 to June 2019.