Research on Food Dynamic Pricing Algorithm Based on Deep Reinforcement Learning

In order to mitigate food waste resulting from irrational pricing and concurrently augment overall business revenue, this paper introduces a finite-shelf-life food dynamic pricing framework denoted as W-DQN, founded upon the principles of deep reinforcement learning theory. Initially, the dynamic pricing problem for perishable goods is formulated as a Markov decision process. Subsequently, a dynamic pricing algorithm model and corresponding reward function are devised to bolster business revenue while curtailing food waste. Experimental findings unequivocally demonstrate that, relative to tabular-based dynamic pricing algorithm models, W-DQN attains commendable returns. Furthermore, the proposed reward function effectively reduces waste. In comparison to conventional pricing approaches, W-DQN significantly diminishes food waste, thereby enhancing overall business revenue.