Univrsity of Tehran Press Industrial Management Journal 3115-7386 16 4 2024 11 21 Dynamic Pricing of Customer Classes in Rail Transportation Systems Using Deep Q Network Algorithm قیمت‌گذاری پویای کلاس‌های مشتریان در سیستم‌های حمل‌ونقل ریلی با استفاده از الگوریتم شبکۀ عمیق Q 597 630 99727 10.22059/imj.2024.377050.1008164 EN Omid Niknami MSc. Student, Department of Industrial Engineering, Faculty of Industrial and Systems Engineering, Tarbiat Modares University, Tehran, Iran. Elham Akhondzadeh Noughabi Assistant Prof., Department of Information Technology, Faculty of Industrial Engineering and Systems, Tarbiat Modares University, Tehran, Iran. Journal Article 2024 05 27 <strong>Objective</strong> This research investigates the problem of dynamic pricing in rail transportation systems using advanced deep reinforcement learning techniques. The main goal is to optimize the revenue of railway transport companies by developing a ticket sales policy that dynamically adjusts ticket prices based on service classes. This approach allows rail transport companies to enhance revenue and profitability by accurately aligning prices with passenger demand.   <strong>Methods</strong> To solve the problem of dynamic pricing, this research utilizes the Q deep network algorithm, which combines deep neural networks with Q-learning. Deep neural networks approximate Q values instead of using a costly Q table. The Q deep network algorithm is widely used due to its ability to learn optimal policies in complex environments. As reinforcement learning models are often too complex to analyze, numerical experiments and simulations are used to analyze different pricing strategies.   <strong>Results</strong> The simulations demonstrate that the Q deep network algorithm successfully converges to a stable pricing policy. Various performance indicators were investigated, including such as total revenue, remaining capacity, average prices offered to customers, and the number of tickets sold in each service class. The algorithm showed improvement in the early stages and gradually achieved stability. The average total revenue converges to 225,000 after 5,000 iterations, indicating that the company earns an average of 225,000 monetary units from each train. The average residual capacity approaches zero after approximately 3,000 iterations, indicating that the reinforcement learning agent learns to sell all available tickets to maximize total revenue. The average price index offered to customers stabilizes after approximately 7,500 iterations, indicating that the algorithm has converged to an optimal pricing policy. In this state, the average prices remain within the range of 680 to 700 monetary units, with no significant fluctuations observed. In other words, the reinforcement learning model has successfully converged based on the average proposed price index.<strong><em> </em></strong>Finally, after about 5,000 iterations, the average number of tickets sold for all service classes reaches a stable level. The average number of tickets sold for economy class is around 175 to 180 tickets, for business class is around 130 to 135 tickets, for special<strong><em> </em></strong>class is around 60 to 65 tickets, and for hotel class is around 23 to 25 tickets.   <strong>Conclusion</strong> The findings of this study suggest that employing the Deep Q-Network algorithm in dynamic pricing can lead to substantial optimization in revenue management for railway transportation systems.<strong> </strong>The results of this research indicate that after approximately 7,500 iterations, the Q deep network algorithm reaches an optimal and stable policy with no significant changes in performance. It can be concluded that the use of the Q deep network algorithm in dynamic pricing can significantly improve the revenue management of rail transportation systems. This algorithm can learn and adapt to changing conditions, allowing for effective pricing policies to maximize revenue and determine the optimal number of tickets sold in each service class. The obtained findings can help rail transport companies improve pricing strategies and increase economic productivity. https://imj.ut.ac.ir/article_99727_90ae11416a117502b7e35f3ded8510f0.pdf