Energy Efficiency Forecasting for Central Air-conditioning Refrigeration Systems Based on Deep Neural Network

Central air-conditioning is a complex system project, and the assessment and forecasting of its performance often involves a very large number of factors. Accurate assessment of a system's energy efficiency is crucial for system energy demand management and performance improvement. Therefore, researchers have done a lot of related work focusing on energy efficiency forecasting by constructing thermodynamic and mech-anistic models of central air conditioning, and recently there have been attempts to combine these models with methods based on data mining and machine learning as well. The problem of predicting energy efficiency in refrigeration systems can be, generally, viewed as a multivariate time-series(MTS) problem following a hidden Markov process. With the development of artificial intelligence techniques, deep neural network-based tem-poral prediction models have made important advances in many application areas. This study aims to achieve higher accuracy for energy efficiency forecasting tasks, using an improved model based on LSTNet. We conducted performance prediction on a dataset collected from a real-world central air-conditioning refrigeration circulation system. As a result, we found significant correlation between the prediction result and ground truth. Our method is compared with several common baseline methods for assessing and predicting the performance of refrigeration systems. Experimental results show that our method generally achieves better performance than those baseline methods.