This study addresses the issue of microclimate prediction in greenhouse environmental control in the southeastern Yunnan region by proposing a deep learning-enhanced CFD modeling method, the DeepCFD-OptNet model. Traditional CFD models have certain limitations when handling complex environmental changes, making it difficult to effectively capture the multidimensional variations in dynamic greenhouse environments. To address this, the study employs Convolutional Neural Networks (CNN) to extract spatial features from greenhouse environmental data and uses Temporal Convolutional Networks (TCN) to model time-series changes. Additionally, Particle Swarm Optimization (PSO) is integrated to optimize greenhouse control strategies. Experimental results show that the DeepCFD-OptNet model demonstrates high accuracy in predicting temperature and humidity, significantly reducing the Root Mean Square Error (RMSE) compared to traditional CFD models, and better simulates and predicts microclimate changes within the greenhouse. The study further confirms that deep learning techniques and optimization algorithms significantly enhance the performance of CFD simulations. This research provides a new technological approach for the development of smart agriculture in the region, contributing to improved crop yields, optimized resource efficiency, reduced energy consumption, and the promotion of sustainable agricultural production through smarter greenhouse management.