Quantification of implicit price flexibility of household customers’ load demand with machine learning and Shapley analysis

The demand for understanding household customers’ flexibility on the electricity consumption side is growing as the proportion of weather-dependent electricity production is increasing in energy systems. Therefore, to enhance this understanding, a study based on hourly electricity consumption data on over 6000 customers with information of their electricity contract types was conducted applying machine learning (ML) algorithms and Shapley values to define spot-price-based flexibility from the data. 26 ML models were studied to identify the most suitable model for further study of the load dynamics of household consumption during the year 2022. The predictor variables in the models were time of day, weekday, outdoor temperature, electricity spot price, and the difference of the spot price from the daily average. The Exponential Gaussian Process Regression model showed the best performance with respect to error parameters. Shapley values of the exponential GPR models of three customer groups were analyzed. The results show that a Gaussian process regression model can recognize the price flexibility of electricity demand. In the Nordic conditions, the outdoor temperature had the most significant effect on electricity consumption, but for the spot contract customers the spot market price had almost as high effect as the hour of the day. • Spot-price-based flexibility was identified from load data with machine learning. • Exponential Gaussian process regression performed the best of the 26 trained models. • SHAP values help user to interpret the model and understand the load dynamics. • The most important external predictor variable was outdoor temperature. • Spot-contract customers’ load demand had lower dependence on time of day.