Inter-turbine Modelling of Wind-Farm Power using Multi-task Learning

TL;DR

This work tackles the challenge of predicting wind-farm power under wake-effects with limited failure data by introducing a probabilistic BSBR model and a metamodelling framework. The BSBR model uses a Beta distribution with a mean $\mu$ and precision $\phi$, modeled with B-spline basis functions to capture heteroscedastic, asymmetric power curves. A hierarchical Bayesian metamodelling approach then leverages spatial correlations to share information across turbines, enabling probabilistic predictions at unobserved locations. The metamodel consistently outperforms no-pooling and fully-pooled baselines, especially for unobserved turbines, while maintaining reasonable computational efficiency. The approach offers a practical pathway to reduce telemetry needs across wind-farm populations and can extend to other population-based SHM problems where inter-structure correlations are informative.

Abstract

Because of the global need to increase power production from renewable energy resources, developments in the online monitoring of the associated infrastructure is of interest to reduce operation and maintenance costs. However, challenges exist for data-driven approaches to this problem, such as incomplete or limited histories of labelled damage-state data, operational and environmental variability, or the desire for the quantification of uncertainty to support risk management. This work first introduces a probabilistic regression model for predicting wind-turbine power, which adjusts for wake effects learnt from data. Spatial correlations in the learned model parameters for different tasks (turbines) are then leveraged in a hierarchical Bayesian model (an approach to multi-task learning) to develop a "metamodel", which can be used to make power-predictions which adjust for turbine location - including on previously unobserved turbines not included in the training data. The results show that the metamodel is able to outperform a series of benchmark models, and demonstrates a novel strategy for making efficient use of data for inference in populations of structures, in particular where correlations exist in the variable(s) of interest (such as those from wind-turbine wake-effects).

Figures (21)

• Figure 1: Data from all turbines in wind-farm Ciabatta showing the power-curve relationship. Blue markers show the raw data prior to filtering, and orange markers show the data after filtering.
• Figure 2: Density plots for the raw data (blue), filtered data (orange) and the subsampled data (green) for the generated power, yaw angle and wind-speed variables.
• Figure 3: B-spline basis-functions over the interval of zero and one, with four interior knots.
• Figure 4: Comparison of NMSE score vs. the number of (evenly-spaced) interior knots per feature, for the no-pooling BSBR model described in Section \ref{['section:bsbr_demo']}.
• Figure 5: A graphical representation of the B-spline beta regression model.