A LSTM-RNN based intelligent control approach for temperature and humidity environment of urban utility tunnels

Temperature and relative humidity are important indicators of utility tunnel indoor atmosphere hazards and operational risks, which can be effectively mitigated by accurate forecasting and corrective control measures. To this end, this paper proposed a multi-layer long short-term memory (LSTM) recurrent neural network (RNN) architecture to forecast the changing trend of temperature and relative humidity inside utility tunnels with distant past monitoring data. Based on the forecasting architecture, an intelligent control approach was designed, including early warning and ventilation control measures. Case study results showed that the proposed architecture fit the training dataset well and the prediction accuracy on testing datasets of temperature and relative humidity exceeded 98% and 99%, respectively. Meanwhile, the proposed LSTM-RNN architecture can also be used to simulate and evaluate the ventilation effects on the temperature and relative humidity environment of urban utility tunnels. Findings of this paper provide a reference for the safe, efficient and energy-saving indoor environment control of urban utility tunnels.