Enhancing cold storage efficiency: Continuous deep deterministic policy gradient approach to energy optimization utilizing strategic sensor input data

Abstract

In this study, we present a continuous Deep Deterministic Policy Gradient (DDPG)-based control algorithm applied to extended-scale cold storage environments to optimize energy efficiency. A key innovation of this study is the use of strategically positioned temperature sensors, particularly sensors placed near the unit cooler, which enabled the algorithm to respond rapidly to temperature fluctuations, including abnormal conditions such as defrost cycles. The foundational framework, reward function, and data communication methods, previously optimized for small-scale facilities, were adjusted to suit the requirements of the extended cold storage settings. Unlike small-scale facilities with a single temperature sensor, the extended setup incorporated 20 strategically placed sensors, enabling an in-depth investigation into how sensor location influences algorithm performance and effectiveness. The proposed algorithm represents a significant advancement in the field of energy management for cold storage, combining real-time data-driven learning with robust control strategies. Therefore, Experimental results demonstrate a remarkable 19.8 % reduction in energy consumption compared to conventional methods while maintaining a stable target temperature of −10 °C. This study not only highlights the practical feasibility of AI-based control algorithms in real-world facilities but also emphasizes the significance of sensor location in enhancing algorithm performance and energy savings. The ability to achieve substantial energy savings while maintaining the target temperature highlights the potential of this advanced control method in actual practical applications.