Research on the Optimization Method of Spark-based Reinforcement Learning in Dynamic Pricing of Airline Tickets

Abstract

This paper discusses the difficulties of dynamic pricing of civil aviation. With more competition, the ticket won’t last long. Traditional pricing methods and solutions have limitations compared with single aircraft reinforcement learning. They are not fast enough to deal with complex situations well and can’t respond in real time. We are looking for a better way to dynamically price airline tickets. To this end, we combine Spark distributed computing with reinforcement learning depth. We create a hierarchical and distributed pricing system. We use Spark to process data, compress functions and access data in real time. We are improving the MDP pricing model and using DKN algorithm to build a decision engine. The experimental results show that our solution is much better. It can process data faster and respond in real time faster. Seat Occupancy Rate increased by 5 percentage points. It is also sensitive and robust when computing resources, market competition or scenario complexity change. This study provides a feasible technical path for the engineering application of reinforcement learning in dynamic airline ticket pricing and has important practical value for enhancing the revenue management efficiency of airlines.