To address complex fault risk for wind turbine pitch systems working under the long-term operation and harsh environment, a fault prediction method based on Swarm Optimization (SO) algorithm and optimized Support Vector Machine (SVM) is proposed. Firstly, principal component analysis (PCA) is adopted to identify the core feature values from a datasets of wind turbine variable pitch systems containing operating data and feature extraction. Secondly, an Improved Swarm Optimization (ISO) algorithm is introduced to resolve sample imbalances and parameter selections in traditional SVM algorithms, which can improve the prediction accuracy and iteration speed. By utilizing Latin hypercube sampling, the algorithm searches for the optimal solution across all parameter spaces. It combines the frosting algorithm (RIME) with mirror imaging reverse learning to enhance adaptability to complex problems and improves the algorithm's convergence time by accelerating the iteration speed of the population. Lastly, the ISO-SVM algorithm is utilized to develop a fault prediction model for wind turbine pitch and performs excellently in fault prediction. The experimental results indicate the proposed method has achieved an increased prediction accuracy of 98.67% compared with other algorithms. Moreover, the proposed approach significantly improves the accuracy of fault prediction for the pitch system of wind turbines, addressing the primary issues in fault prediction for the pitch system of wind turbines.