Adaptive Bayesian Very Short-Term Wind Power Forecasting Based on the Generalised Logit Transformation

TL;DR

This work tackles the challenge of probabilistic very-short-term wind power forecasting by introducing an adaptive Bayesian framework that leverages the generalised logit transformation to handle double-bounded wind-power data. By casting the transformed series in an AR$(p)$ form and applying online Bayesian updates—with options for a fixed or varying shape parameter $\nu$—the method delivers robust probabilistic forecasts and improved calibration. The authors evaluate seven adaptive and classical approaches on data from over 100 UK wind farms, using $CRPS$ and functional reliability diagrams, and show that the Bayesian approach with adaptive $\nu$ consistently achieves the best or near-best performance with strong robustness. The findings highlight the practical value of online Bayesian learning for wind power decision-making and grid integration, while underscoring the importance of boundary treatment and data quality.

Abstract

Wind power plays an increasingly significant role in achieving the 2050 Net Zero Strategy. Despite its rapid growth, its inherent variability presents challenges in forecasting. Accurately forecasting wind power generation is one key demand for the stable and controllable integration of renewable energy into existing grid operations. This paper proposes an adaptive method for very short-term forecasting that combines the generalised logit transformation with a Bayesian approach. The generalised logit transformation processes double-bounded wind power data to an unbounded domain, facilitating the application of Bayesian methods. A novel adaptive mechanism for updating the transformation shape parameter is introduced to leverage Bayesian updates by recovering a small sample of representative data. Four adaptive forecasting methods are investigated, evaluating their advantages and limitations through an extensive case study of over 100 wind farms ranging four years in the UK. The methods are evaluated using the Continuous Ranked Probability Score and we propose the use of functional reliability diagrams to assess calibration. Results indicate that the proposed Bayesian method with adaptive shape parameter updating outperforms benchmarks, yielding consistent improvements in CRPS and forecast reliability. The method effectively addresses uncertainty, ensuring robust and accurate probabilistic forecasting which is essential for grid integration and decision-making.

Figures (7)

• Figure 1: Histogram of percentage of missing data over the four years.
• Figure 2: Ternary chart pythonternary for lower and upper bound data over four years, excluding missing data.
• Figure 3: Histogram of estimated shape parameter across 4 years. The x-axis is the bin of the estimated shape parameter, while the y-axis shows the frequency of wind farms corresponding to each bin.
• Figure 4: Violin plot of Skill Score relative to Persistence for all three test datasets.
• Figure 5: Reliability Diagram and Functional Reliability Diagram Sun01012011Hyndman01012010. The top-left chart provides a comparison of all seven methods for one wind farm, with the x-axis representing the nominal proportion (forecasted CDF) and the y-axis representing the estimated observed proportion (estimated observed CDF). The remaining charts are functional reliability diagrams for each method across all wind farms. The diagonal line represents the ideal result for reference. The dark line represents the median curve across all reliability curves. The dark gray envelope indicates the 50% central region. The light gray region illustrates the maximum non-outlying envelope. The red to blue single curves are examples of outliers.