Posters

Abstract

Doppler lidars are widely used for measuring wind speed in the wind energy applications but suffer from degraded accuracy in situations where the wind speed field is not homogeneous, such as in complex terrain. In this work, we cast the inverse problem of estimating wind speed from lidar measurements as a Bayesian state estimation problem. In general, Bayesian state estimation is based on formulating a state-space model for the problem. The state-space model consists of two parts: 1) an observation model, which, in this application, describes how the lidar measurements are connected to the 3D wind field, 2) an evolution model, which describes how the 3D wind field evolves in time. The state estimation problem is then solved by e.g. Kalman filter. As the full solution is a posterior probability distribution, we also get an error estimate for the wind speed. To make state estimation of 3D wind field feasible, we adopt a partially data-based approach. We build training data of simulated wind fields and their evolution over time using turbulent computational fluid dynamics simulations, which take the surrounding terrain into account. This training data is then used to build a reduced order basis that is based on principal component analysis for the wind field and a vector autoregressive (surrogate) evolution model. The proposed approach is tested in a synthetic hill setting with simulated Vaisala WindCube measurements. While still preliminary, the results show that state estimation can potentially reduce bias in 10-minute average wind velocity and turbulence intensity compared to traditional trigonometric wind vector reconstruction methods. The proposed approach is also flexible and can accommodate e.g. multiple lidar instruments.