Improving WRF model turbine-height wind-speed forecasting using a surrogate- based automatic optimization method

Abstract

Improving turbine-height wind-speed forecasting using a mesoscale numerical weather prediction (NWP) model is important for wind-power prediction because of the cubic correlation between wind power and wind speed. This study investigates how a surrogate-based automatic optimization method can be used to improve wind-speed forecasting by an NWP model by optimizing its parameters. A key challenge in optimizing NWP model parameters is the tremendous computational requirements of such an exercise. A global sensitivity method known as the Multivariate Adaptive Regression Spline (MARS) method was first used to identify the most sensitive parameters among all tunable parameters chosen from seven physical parameterization schemes of the Weather Research and Forecast (WRF) model. Then, a highly effective and efficient optimization method known as adaptive surrogate modeling-based optimization (ASMO) was used to tune the sensitive parameters. In a case study carried out over Eastern China, the seven parameters that were most sensitive to wind-speed simulation were identified from among 27 tunable parameters. Those seven parameters were optimized using the ASMO method. The present study indicates that the hourly wind-speed simulation accuracy was improved by 8.7% during the calibration phase and by 7.58% during the validation phase. In addition, clear physical interpretations were provided to explain why the optimal parameters lead to improved wind speed forecasts. Overall, this study has demonstrated that automatic optimization method is a highly effective and efficient way to improve wind-speed forecasting by an NWP model.