Model and Data Dual-Driven Based Temperature Estimation Method for Wind Turbine Mechanical Components

Accurate temperature estimation of mechanical components in wind turbines is essential for extending their lifespan and optimizing maintenance strategies, given the complex and dynamic operational environment. This paper presents a model and data dual-driven based temperature estimation method aimed at improving the accuracy and adaptability of temperature estimation for wind turbine mechanical components. The model-driven part of the approach involves the design of a simplified lumped parameter thermal network model, which simulates the heat generation and dissipation behaviors of critical mechanical components such as the generator stator winding, gearbox bearing, and spindle bearing. This model provides a unified framework without prior model parameters for temperature estimation. The data-driven part utilizes real-time sensor data, where the forgetting factor recursive least squares (FFRLS) algorithm is applied to continuously estimate and update the model parameters. This dynamic updating enhances the adaptive nature of the model, enabling it to adjust to changing operating conditions. Finally, an adaptive Kalman filter is employed to combine these data-based estimates with the model-based unified framework to make adaptive state predictions, further improving the accuracy of temperature estimation. Experiments using data of a 6.25 MW H6250-186 wind turbine provided by DEC demonstrate that the proposed method significantly outperforms traditional approaches in terms of both accuracy and adaptability across various operational conditions.