Multi-fidelity Gaussian processes for wind power curves

ASME IDETC 2023, first author

Wind uncertainty
Can multi-fidelity Gaussian process regression, combining a theoretical power curve with limited measured data, improve probabilistic wind turbine power-curve estimation?

First author

  • Open SDWPF SCADA data
  • Nine test turbines
  • Kernel density sampling
  • Multi-fidelity Gaussian processes
  • Point and probabilistic metrics

Research question

Can multi-fidelity Gaussian process regression, combining a theoretical power curve with limited measured data, improve probabilistic wind turbine power-curve estimation?

Key insight

A theoretical power curve helped, but the gain depended on the metric

The method fuses about two hundred sampled measurements with the theoretical power curve of the turbine type, which covers all wind speeds but does not show how a turbine actually runs. In the reported nine-turbine test, nonlinear multi-fidelity regression had a lower RMSE than standard Gaussian process regression for most turbines, but the ranking depended on the evaluation metric.

How does a theoretical curve help a few measurements?

A Gaussian process fitted to a few measured points is uncertain wherever the points are sparse, so its band widens between them. A multi-fidelity model also learns from the turbine's theoretical power curve, which covers every wind speed, and uses it to fill in the shape between measurements, so the band can narrow there. In the paper's test, the result depended on the evaluation metric.

Illustrative model with stylized numbers: it shows the mechanism described in the paper, not the study's data or results.

My contribution

First author. I did most of the work: I implemented and evaluated the models and did the writing, with funding from the U.S. Department of Energy through the National Renewable Energy Laboratory.

How I approached it

  1. ChallengeMeasured SCADA records show real behaviour, including scatter, stops and curtailment, but are noisy and very numerous.

    What I didI drew a few hundred representative points from a kernel density estimate of the records, which keeps the shape of the scatter without manual cleaning.

    Why it matteredA Gaussian process solve grows with the cube of the training points, so a compact training set matters.

  2. ChallengeThe theoretical curve covers every wind speed but not how a turbine actually operates.

    What I didI compared linear and nonlinear multi-fidelity models, which link the curve to the measurements, with a standard Gaussian process on measurements alone, on point and probabilistic metrics.

    Why it matteredTesting both kinds of metric showed that the gain depended on which error matters.

Method

Linear (autoregressive) and nonlinear multi-fidelity Gaussian process regression that fuse the turbine type's theoretical power curve (low fidelity) with a few hundred kernel-density-sampled SCADA points from the open SDWPF dataset (high fidelity), compared with single-fidelity regression on point and probabilistic metrics.

Key findings

  1. In the reported nine-turbine test, nonlinear multi-fidelity regression achieved lower RMSE for most turbines; performance depended on the evaluation metric.

Boundary conditions

One dataset and one generic low-fidelity curve.

Publications

  1. IDETC 2023
    A Multi-Fidelity Gaussian Process Regression Method for Probabilistic Wind Farm Power Curve Estimation
    H. Li, C. Feng, J. Zhang
    Proceedings of the ASME 2023 IDETC-CIE, Paper No. IDETC2023-114762, article V03AT03A034, 2023.
    DOI Accepted manuscript (PDF) Code Cited by 1

Citation counts: Google Scholar, October 30, 2026.