Accurate indoor air temperature (IAT) predictions for heating, ventilation, and air conditioning (HVAC) systems are challenging, especially for multi-zone building and for different HVAC system types. Moreover, the nonlinearity of the buildings thermal dynamics makes the IAT prediction more difficult since it is affected by complex factors such as controlled and uncontrolled points, outside weather conditions and occupancy schedule. This paper presents a long short-term memory (LSTM) model to predict IAT for multi-zone building based on direct multi-step prediction with sequence-to-sequence approach. Two strategies, LSTM-MISO and LSTM-MIMO, are built for multi-input single-output and multi-input multi-output, respectively. The performance of these two strategies has been evaluated based on two case studies on real smart buildings using variable air volume (VAV) and constant air volume (CAV) systems. For both buildings, experimental results showed that the LSTM models outperform multilayer perceptron models by reducing the prediction error by 50%.

对于供暖，通风和空调（HVAC）系统的准确室内空气温度（IAT）预测具有挑战性，尤其是对于多区域建筑和不同HVAC系统类型而言。此外，建筑物热力学的非线性使IAT预测更加困难，因为它受复杂因素的影响，例如受控点和非受控点，外界天气条件和入住时间表。本文提出了一个长期的短期记忆（LSTM）模型，以基于序列到序列方法的直接多步预测，预测用于多区域构建的IAT。分别针对多输入单输出和多输入多输出构建了两种策略LSTM-MISO和LSTM-MIMO。基于对使用可变风量（VAV）和恒定风量（CAV）系统的真实智能建筑的两个案例研究，评估了这两种策略的性能。对于这两座建筑物，实验结果表明，LSTM模型的预测误差降低了50％，胜过多层感知器模型。