Soft sensor validation for monitoring and resilient control of sequential subway indoor air quality through memory-gated recurrent neural networks-based autoencoders

Indoor air quality (IAQ) measurements play an important role in the subway ventilation system control, influencing over crucial factors as ventilation energy consumption and commuters' health. Therefore, faulty sensors may result in misinterpreting the IAQ conditions and misoperating the air delivery rate level in subway stations. However, due to the IAQ data properties of dynamism and non-Gaussian distribution. Linear and fixed structures are not sufficient to extract essential features from the IAQ data. This paper presents a machine learning-based soft sensor validation technique to detect, diagnose, identify, and reconstruct faulty measurements of the multivariate IAQ data in subway stations. The proposed method is memory-gated recurrent neural networks-based autoencoders (MG-RNN-AE), which are capable of processing sequential and dynamic IAQ information. The performance of the sensor validation was evaluated through several metrics to consequently be compared among different methods, being the batch normalization-based gated recurrent unit (BN-GRU) method, the most effective to detect (DRSPE = 100%) and reconstruct faulty IAQ sensors (R-2 = 0.45-0.79). Additionally, the effects of the faulty and repaired measurements in the ventilation system were evaluated to determine that the proposed method is capable of finding a sustainable balance between energy demand and commuters' health level.