Energy consumption prediction of cold storage based on LSTM with parameter optimization

Abstract

In refined energy management, accurate energy consumption prediction is crucial for fault diagnosis, optimizing system operations based on peak electricity prices, and reducing costs. This study proposes a short-term energy consumption prediction model for cold storage refrigeration systems based on Long Short-Term Memory (LSTM) neural networks. Tailored to the load features of cold storage, the model incorporates compressor unit operating features and air cooler features specific to cold storage factors, rarely addressed in other refrigeration scenarios, as inputs and replaces personnel activity features with time features. This allows the model to predict energy consumption for the next hour while analyzing the impact of each feature on model performance. Results show that compressor unit operating features and air cooler features are essential for prediction accuracy; without these features, the model's R² is only 0.739 and 0.854. Furthermore, compared to models like CNN, BILSTM, and GRU, the LSTM model demonstrates a significant advantage in predictive accuracy, with R² improved by 0.306 to 0.475, confirming its efficiency and reliability in cold storage energy consumption prediction. By introducing an LSTM model that incorporates specific features of cold storage, this study achieves an innovative breakthrough in prediction accuracy for high-energy-consumption cold storage, laying a solid foundation for energy management applications in this field.