Accurate sensor measurements are important to normal operation and energy-saving control of air-conditioning systems in buildings. However, there is currently a lack of faulty sensor data recovery approach for building air-conditioning systems. This study aims to provide a robust faulty sensor data recovery approach for air-conditioning systems in different real buildings. A faulty sensor data recovery method is presented in this paper for air-conditioning systems based on transfer entropy (TE), unfold principal component analysis (UPCA), and grey wolf algorithm optimizing long short-term memory neural networks (GT-LSTM). The faulty sensor data recovery method consists of a TE-based fault feature variable extraction module, a UPCA-based sensor fault diagnosis module, and a GW-LSTM-based faulty sensor data recovery module. The proposed faulty sensor data recovery method was validated by using faulty sensor data from real air-conditioning systems. The results showed that the ranking of accuracy in recovering fault data was TE-GW-LSTM > PCA-GW-LSTM > PCA-GA-LSTM > TE-ANN. The TE-GW-LSTM method had the highest accuracy in recovering fault data from faulty sensors. There were slight differences in the accuracy rates of recovering fault data from different types of sensors. Compared with the PCA method, the TE-based fault feature variable extraction method improved the accuracy in recovering fault data by at least 3.1 %. Grey wolf optimization algorithm improved the accuracy of LSTM neural network models by 6 % through optimizing their weights and thresholds. The presented faulty sensor data recovery method provides an effective tool to accurately reconstruct faulty sensor data of air-conditioning systems in real buildings.

中文翻译：

---

基于转移熵和 LSTM 深度学习的空调系统故障传感器数据恢复方法


精确的传感器测量对于建筑物中空调系统的正常运行和节能控制非常重要。然而，目前缺乏用于构建空调系统的故障传感器数据恢复方法。本研究旨在为不同真实建筑物中的空调系统提供一种稳健的故障传感器数据恢复方法。该文提出了一种基于传递熵（TE）、展开主成分分析（UPCA）和优化长短期记忆神经网络（GT-LSTM）的空调系统故障传感器数据恢复方法。故障传感器数据恢复方法由基于 TE 的故障特征变量提取模块、基于 UPC 的传感器故障诊断模块和基于 GW-LSTM 的故障传感器数据恢复模块组成。通过使用来自真实空调系统的故障传感器数据验证了所提出的故障传感器数据恢复方法。结果表明，故障数据恢复精度排序为 TE-GW-LSTM > PCA-GW-LSTM > PCA-GA-LSTM > TE-ANN。TE-GW-LSTM 方法在从故障传感器恢复故障数据方面具有最高的准确性。从不同类型的传感器恢复故障数据的准确率略有不同。与 PCA 方法相比，基于 TE 的故障特征变量提取方法将恢复故障数据的准确性提高了至少 3.1 %。Grey wolf 优化算法通过优化 LSTM 神经网络模型的权重和阈值，将模型的准确性提高了 6%。该文提出的故障传感器数据恢复方法为准确重建真实建筑物中空调系统的故障传感器数据提供了一种有效的工具。