Uncertainty Quantification Analysis of Wind Power: A Data-Driven Monitoring-Forecasting Framework

Advances in wind power system modeling have produced widespread socioeconomic benefits for alleviating global environmental problems. However, previous studies mainly payed attention to point forecasts of wind power system, with the absence of its uncertainty quantification analysis and outlier detection, which cannot facilitate further development in this field. In this paper, a novel monitoring-forecasting system, including the analysis module, the outlier detection module, the probabilistic forecasting module, and the evaluation module, is proposed for uncertainty modeling of wind power. In the analysis module, recurrence analysis techniques are developed, aiming at characterizing complicated patterns of wind power. Furthermore, the interval partitioning-based isolation forest algorithm, which can effectively address the effects of swamping and masking, is first developed in the outlier detection module for wind power. Superior to the traditional point forecasting method that cannot perform quantitative characterization of the intrinsic uncertainties in wind power forecasting, an advanced probabilistic forecasting method based on Gaussian process regression (GPR) with an optimal kernel function scenario, cooperating with a feature selection method, is first presented in the probabilistic forecasting module, indicating that the forecast skill of GPR is significantly enhanced. Finally, the proposed system is validated using real wind power data with high resolution from Spain in the evaluation module, solidly demonstrating its high reliability and flexibility compared to benchmarks considered in this study.

Automated summary

Advancements in wind power system modeling have led to significant socioeconomic benefits in addressing global environmental issues. A novel monitoring-forecasting system is proposed in this study for uncertainty modeling of wind power, which includes modules for analysis, outlier detection, probabilistic forecasting, and evaluation. Through the use of advanced techniques such as Gaussian process regression and isolation forest algorithm, the system demonstrates improved reliability and flexibility in wind power forecasting.