Traditional pigpen temperature control relies on manual adjustments or basic equipment operation, which suffers from significant temperature fluctuations, high energy consumption, and poor uniformity, making it difficult to meet the dynamic needs of pigs at different growth stages. This study proposes an intelligent temperature control optimization method to achieve coordinated optimization of temperature control accuracy, uniformity, and energy efficiency. The research constructs a CFD agent model based on Elman's framework, combined with an improved natural field optimization algorithm. The method employs Tent chaotic mapping to enhance population diversity during initialization, introduces adaptive t-distribution mutation to adjust search step size, and incorporates Pareto strategy to optimize solution sets. Experimental results demonstrate that the improved CFD agent model exhibits excellent temperature prediction performance, achieving an average absolute error of 0.15ºC, a coefficient of determination of 0.988, and inference speed of 10.3ms, effectively replacing traditional CFD models for real-time temperature field calculations. In dynamic scenario tests, when pig density varies between 1.0-2.5 heads/m² and the average temperature reaches 27.3ºC, the temperature uniformity index remains within a reasonable range of 1.3, with energy consumption increases matching pig density variations. The study reveals that this intelligent temperature control method efficiently balances temperature control accuracy, spatial uniformity, and energy consumption, effectively meeting the dynamic environmental needs of pigs at different growth stages. It provides an efficient and practical solution for intelligent temperature control in large-scale livestock farming scenarios.