| In areas with concentrated production,well-developed logistics,and mature industrial chain operations,auction-based transaction models have been widely adopted for bulk transactions of fresh agricultural products such as fresh flowers and vegetables.In addition to improving the efficiency of transactions,the auction transaction mode can also promote the standardization of products and shorten the circulation chain of fresh agricultural products.At present,Kunming International Flower Auction Center is one of the best auction markets for flower auctions in China.As the demand for fresh flowers continues to expand,the auction model is developing from offline auctions to online remote auctions.The online auction mode breaks the time and space constraints of traditional auctions,and bidders can enter and exit the online auction platform at any time,which also causes the number of bidders to fluctuate greatly between different auction sessions,and the auction demand is difficult to predict.In addition,the large transaction volume and strong timeliness of flower products lead to a large gap in transaction prices in different auction sessions.Due to the above practical operational challenges,the auction center should guide the competition among bidders by setting reasonable auction lots for different auction sessions,so as to increase the auction turnover and the revenue of the auction center.Existing research on auction lots usually assumes that the demand of bidders and the distribution of bids are fixed,but in reality,the demand of bidders will change over time,making it difficult to apply the optimal strategies in existing research to actual situations.From the perspective of practical application,this thesis takes Kunming International Flower Auction Center as an object to study the multi-stage optimal auction lot decision problem of online flower auction platform under the condition of random arrival of demand.First of all,this thesis uses game theory to analyze the impact of the lot of flowers decided by the auction center on the bidding price of the bidder,builds a bidding game model under the given auction lot and bidder’s demand,and deduces the bidder’s optimal bidding strategy.Secondly,based on the evaluation distribution and demand distribution of bidders,the income of auctioning homogeneous flowers on the auction platform is analyzed,and the income model of the auction center on the auction lot is established.Then,based on the bidder’s optimal bidding strategy and revenue model,the sequential auction process is transformed into a finite-stage Markov decision process model,and the dynamic lot decision-making strategy of the auction center is solved by using the Qlearning algorithm in reinforcement learning.In the design of the algorithm,the Upper Confidence Bound(UCB)exploration strategy is used to find the optimal auction lot,which improves the efficiency of the algorithm’s exploration and learning.Finally,through numerical experiments,the difference between the-greedy exploration strategy commonly used in the literature and the UCB exploration strategy adopted in this paper is compared on the income of the auction center under different supply and demand conditions;and through the sensitivity analysis of other problem parameters,some management implications are extracted to provide reference for the multi-stage online auction operation of the auction center. |
