Toward Greener and Sustainable Airside Operations: A Deep Reinforcement Learning Approach to Pushback Rate Control for Mixed-Mode Runways

Airside taxi delays have adverse consequences for airports and airlines globally, leading to airside congestion, increased Air Traffic Controller/Pilot workloads, missed passenger connections, and adverse environmental impact due to excessive fuel consumption. Effectively addressing taxi delays necessitates the synchronization of stochastic and uncertain airside operations, encompassing aircraft pushbacks, taxiway movements, and runway take-offs. With the implementation of mixed-mode runway operations (arrivals-departures on the same runway) to accommodate projected traffic growth, complexity of airside operations is expected to increase significantly. To manage airside congestion under increased traffic demand, development of efficient pushback control, also known as Departure Metering (DM), policies is a challenging problem. DM is an airside congestion management procedure that controls departure pushback timings, aiming to reduce taxi delays by transferring taxiway waiting times to gates. Under mixed-mode runway operations, however, DM must additionally maintain sufficient runway pressure—departure queues near runway for take-offs—to utilize available departure slots within incoming arrival aircraft steams. While a high pushback rate may result in extended departure queues, leading to increased taxi-out delays, a low pushback rate can result in empty slots between incoming arrival streams, leading to reduced runway throughput. This study introduces a Deep Reinforcement Learning (DRL) based DM approach for mixed-mode runway operations. We cast the DM problem in a markov decision process framework and use Singapore Changi Airport surface movement data to simulate airside operations and evaluate different DM policies. Predictive airside hotspots are identified using a spatial-temporal event graph, serving as the observation to the DRL agent. Our DRL-based DM approach utilizes pushback rate as agent’s action and reward shaping to dynamically regulate pushback rates for improved runway utilization and taxi delay management under uncertainties. Benchmarking the learnt DRL-based DM policy against other baselines demonstrates the superior performance of our method, especially in high traffic density scenarios. During a typical day at Singapore Changi Airport, DRL-based DM reduces peak taxi times by 1-3 minutes on average, saves 26.6% in fuel consumption, and contributes to more environmentally friendly and sustainable airside operations.