Medium-Term Hourly Electricity Tariff Forecasting Using Ensemble Models

Abstract

Forecasting electricity tariff rates is necessary for large suppliers, consumers, and power brokers working in the wholesale markets. Meanwhile, tariff rates of the retail market are also hourly changed for certain groups of electricity consumers. It creates more efficient electrical load regulation opportunities than the traditional load leveling approach. Power facilities that include controlled load consumers or local generation can use their capabilities by adjusting the load curve according to tariff rates. This work aims to study the potential for medium-term forecasting of retail electricity tariff rates and develop a predictive machine learning model. Hourly data on the retail market tariffs of the Novosibirsk region (Siberia) for four years were collected, several machine learning models were applied, and an analysis of the input parameters for forecasting was carried out. The most significant results are the proof of the possibility of obtaining the month ahead electricity tariff rate forecast with the mean absolute percentage error 4 %. It could be used for electricity costs reduction by regulating the load curve. It was shown that the discrete models based on ensembles of logical rules give higher accuracy than models based on continuous and piecewise continuous functions, such as neural networks. The significance of the obtained results is the proposed approach for month ahead electricity tariff rates forecasting, which was verified on a four-year dataset with an error of 4 %. The approach is based on open data and open-source machine learning models, which allow specialists with even a basic level of data science skills to put it into practice.