In circuit design work, parameter optimization is an inevitable issue, especially in analog circuit design, which requires a high level of experience from designers. In traditional parameter optimization processes, designers may also rely on optimization algorithms to find optimal solutions. This paper uses reinforcement learning algorithms to find optimal strategies, exploring two functions under model-free reinforcement learning algorithms: the value function and the policy function. These functions are estimated using recursive forms and policy gradients. Using the Y parameter to extract equivalent circuit parameters in RF circuits, a frequency AI model is established to optimize the parameters of RF circuits. The optimization effect is verified through metrics such as gain and frequency, and the final optimized results of the RF circuit are calculated. The distribution of the receptive field in the value function method model tends toward a Gaussian distribution, exhibiting sparsity, with weight values primarily distributed at both ends of 0, and the frequency approaching 120. This paper proposes three optimization schemes for parameter tuning, with the optimal solution coordinates for Schemes 1 to 3 being [3.671, 0.749], [3.726, 0.834], and [3.847, 0.578], respectively. After optimization, the static power consumption of the RF circuit was reduced by over 54% compared to before optimization, and the circuit cost was reduced by over 40%, indicating that the method proposed in this paper has good optimization effects.