Description
An Artificial Intelligence (AI) model has been proposed to control the tunnel ventilation system of a road tunnel during normal operations, as an alternative to widely used PID control algorithms. The performance of the proposed AI model has been compared against that of a PID control developed for this study. The AI model has been developed based on synthetic data, including the vehicle traffic, the ventilation capacity and configuration, and corresponding in-tunnel air quality. This data has been generated in IDA Tunnel software through the simulations of a road tunnel with multiple inlets and exits. A random traffic and ventilation input have been considered to obtain the corresponding in-tunnel air quality so that the data generated and used for AI training was not biased with the operation of a control system. Among many potential AI models, a random forest predictor and a random forest classifier have been used together to develop a novel control approach. The predictor model is used to predict the conditions inside the tunnel sections in the next time stage and to propose the ventilation strategy to be applied next, and the classifier model has been used to map the control system output within the pre-defined operational boundaries. The classifier model has been incorporated to address the potential concerns regarding use of AI in safety critical applications. Overall results of the study have shown that the AI model could control the ventilation system in an efficient way requiring less intervention on the operational capacity and configuration, making it a strong candidate for existing tunnels, a detailed computational model of which is hard to construct. Furthermore, the classifier model of the developed AI has demonstrated that the AI response could be safely restricted within manually set operational limits.
