Abstract: To address the challenges of fault localization and limited diagnostic accuracy in analog circuit fault diagnosis, this paper proposes a novel fault diagnosis method that integrates symplectic geometric mode decomposition (SGMD) and refined composite multiscale dispersion entropy (RCMDE) for feature extraction, combined with a random forest (RF) classifier optimized by an improved whale optimization algorithm (IWOA). In this method, SGMD is first applied to decompose and reconstruct the raw signals, achieving effective denoising and extraction of intrinsic modal components. On this basis, RCMDE is introduced to quantify the complexity and dynamic characteristics of the signals from a multiscale perspective, thereby fully capturing fault-sensitive features at different levels. During the classifier construction stage, a nonlinear convergence factor and an evaluation function are incorporated to collaboratively optimize the global exploration and local exploitation capabilities of the whale optimization algorithm. Furthermore, a multiplexing strategy for the update path of the optimal solution is implemented to enhance population diversity and convergence efficiency, thereby improving the classification performance of the random forest model. Simulation experimental results demonstrate that the proposed method achieves a recognition accuracy of 99.70% in analog circuit fault diagnosis, exhibiting a significant accuracy advantage over traditional methods. These results confirm its effectiveness and potential for engineering applications in complex circuit fault scenarios.