Abstract: The difficulty in balancing data sharing and security makes it difficult to effectively promote the application of supply chain financing in the agricultural field. The combination of blockchain and federated learning can fully leverage the technological advantages of both, effectively compensating for the shortcomings of blockchain in data privacy and security, and federated learning in data storage, synchronization, and tamper prevention. However, it still faces problems such as low computational efficiency and insufficient incentive mechanisms. The paper takes accounts receivable financing business as the research object and constructs an agricultural supply chain financing system empowered by hierarchical federated learning and blockchain; On this basis, based on the deep learning optimal auction algorithm, the financial resource allocation strategy is constructed as an auction model with the goal of maximizing seller returns, in order to achieve individual rationality, incentive compatibility, and maximization of cooperative auction revenue for financial institutions. Research has shown that through auction strategy optimization, the optimal auction mechanism based on deep learning can maximize seller returns; The auction revenue of cooperatives is directly proportional to their data coverage and the data coverage requirements of financial institutions; The higher the approximate optimization rate of neural networks, the greater the difficulty of system optimization. Therefore, cooperatives need to invest more computing resources and training time, which in turn affects their profits.

Key words: Accounts receivable auction financing, Blockchain, Federated Learning, Deep Learning