Estimation of supermarket flexibility through control of cooling machine with and without pre-cooling

As renewable energy sources increasingly integrate into the Norwegian national grid, identifying new flexibility sources becomes crucial. Supermarkets can contribute flexibility by utilizing the thermal inertia of their refrigeration and freezer cabinets. The main objective of this study is to examine the flexibility potential of supermarkets to engage in power markets. This paper outlines real-world control experiments conducted with and without pre-cooling on 37 supermarket cabinets to assess the flexibility of the supermarket’s cooling machine. The flexibility provided is the difference between the adjusted power consumption of the cooling machines and the baseline consumption. The baseline is estimated using various machine learning (ML) techniques, and the flexibility calculated through these ML methods is compared to the flexibility calculated by a naive persistence model. The results of the experiments show that the cooling machines can provide an average flexibility of 8.4–8.8 kW for 1.25 hours during a flexibility event, depending on the ML technique employed. This represents 81.5 % to 85.3 % of the cooling machine’s average power consumption. Moreover, the flexibility achieved through the pre-cooling experiments demonstrated greater stability compared to the non-pre-cooling experiments; nevertheless, the rebound effect linked to pre-cooling was more pronounced. Tree-based ML methods produced consistent results both with and without pre-cooling. The flexibility event is influenced by the methods employed to estimate baseline consumption, and the outdoor temperature. Additional experiments are needed to explore the flexibility potential of the cooling machines under specific methods and experimental conditions. • Real-world experiments were conducted to evaluate the flexibility potential of the supermarket’s cooling machine, both with and without pre-cooling the refrigeration and freezer cabinets. • Five ML methods were employed and evaluated to predict the baseline power consumption of the cooling machine. • Imputation methods were used to handle missing historical data when calculating the baseline prediction after flexibility events. • The volume, duration, and amplitude of the flexibility events derived from each realworld experiment were estimated, both with and without pre-cooling. • End-to-end flexibility estimation of a supermarket is showcased, integrating experiments, ML predictions, and demand response.