An adaptive switching control model for air conditioning systems based on information completeness

As building energy management systems are widely applied, a large amount of operational data can be acquired and utilized for building load forecasting and energy system control. However, there's a lack of methods for assessing the completeness of operational data sets and the requisite data quality for various control models. Without these assessments, addressing the loss of model accuracy due to fluctuations in data quality becomes challenging, resulting in deviations from actual operating conditions and decreased control performance. To bridge the gaps, a clustering method is employed to categorize load forecasting training dataset into high and low completeness categories. By matching a reinforcement learning control model combining transfer learning and imitation learning to low completeness datasets and a model-based optimization control model to high completeness datasets, an adaptive switching control model is proposed for air conditioning systems in this study. Case study demonstrates that employing transfer imitation learning results in an 11.5% higher operation benefit compared to model-based optimization under low completeness data conditions. The adaptive switching control model can further reduce operational costs by 5.5% and energy consumption by 5.0% compared to the single control models using only model-based optimization or transfer imitation learning.