In this paper, a novel wind turbine (WT) fault detection method, based on the Partial Least Squares (PLS), Firefly Algorithm (FA), Chaos Map (CP) and Extreme Learning Machine (ELM), which is proposed and explained in detail. The proposed method includes two procedures: a WT mathematical model with PLS and a prediction model with FA-CP-ELM. Since the WT system is modeled as a system using PLS, the ELM has been optimized by the FA and CP to improve the predictive performance. Then, it’s calculated the residual between the mathematical model and the predicted model. If a fault occurs, the residual will increase accordingly and exceed the tolerance range. Hence, a fault can be detected quickly. To demonstrate the feasibility and effectiveness of the proposed approach, the wind turbine system is tested with a fault point set in this system. According to the results of the example, this proposed method is found to achieve better performance.