Intelligent transformation strategy for ecological agriculture based on deep reinforcement learning and transformer

Abstract

Effective agricultural production supply chain management is the key to optimizing the intelligent transformation strategy of ecological agriculture. Traditional research faces huge challenges in resource optimization and dynamic decision-making. Therefore, this paper first abstracts the intelligent management problem of ecological agriculture as a multi-agent sequential decision-making problem. Then, the cosine function is adopted to improve the Transformer model (EOTransformer). For the entire agricultural production supply chain involved in the cooperation, maintain the encoder and decoder structure of EOTransformer. At each time step, agent observations are encoded by passing them through the encoder to obtain their latent representations. The decoder generates optimal actions of each agent through sequential autoregression to obtain the best strategy. Experimental outcome demonstrates that the suggested approach achieves an approximate 90% reduction in parameter scale over standard approach, greatly reducing the complexity of the algorithm.