A multi-input and three-output wind speed point-interval prediction system based on constrained many-objective optimization problem

•The first three-output neural network for wind speed point-interval forecast.•A many-objective optimization problem with coverage constraint is developed.•The proposed CKNSGA-III perfectly solves many-objective optimization problem.•The theoretical proof of the Pareto front existence in CKNSGA-III...

Celý popis

Uloženo v:

Podrobná bibliografie
Vydáno v:	Information sciences Ročník 720; s. 122531
Hlavní autoři:	Lv, Mengzheng, Wang, Jianzhou, Wang, Shuai, Wang, Kang, Zhao, Yang, Gao, Jialu
Médium:	Journal Article
Jazyk:	angličtina
Vydáno:	Elsevier Inc 01.12.2025
Témata:	Advanced optimization algorithm Constrained many-objective optimization problem Interval prediction Three-output neural network Wind speed point prediction Three-output neural network Constrained many-objective optimization problem Advanced optimization algorithm Interval prediction Wind speed point prediction
ISSN:	0020-0255
On-line přístup:	Získat plný text
Tagy:	Přidat tag Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!

Popis
Shrnutí:	•The first three-output neural network for wind speed point-interval forecast.•A many-objective optimization problem with coverage constraint is developed.•The proposed CKNSGA-III perfectly solves many-objective optimization problem.•The theoretical proof of the Pareto front existence in CKNSGA-III is given. Neural networks play a key role in wind speed deterministic and uncertainty analysis, significantly improving wind energy utilization efficiency and reducing power system costs. However, existing studies often rely on complex neural architectures, leading to excessive computational time, and fail to integrate point and interval predictions, disconnecting deterministic and uncertainty analysis, thereby affecting prediction efficiency. To address these issues, this paper presents a novel three-output wind speed prediction system via constrained multi-objective optimization. The framework minimizes mean squared error (MSE), mean absolute error (MAE), and prediction interval normalized average width (PINAW) while maximizing prediction interval coverage (PICP), with adjustable coverage constraints for diverse interval demands. Leveraging the outlier-robust extreme learning machine (ORELM) as the predictor, the system outputs point values and interval bounds simultaneously, addressing the volatility of wind speed time series and multi-output complexity. To solve the optimization problem, an improved non-dominated sorting genetic algorithm (CKNSGA-III) is proposed, integrating Henon chaotic mapping and a knee-oriented mechanism to boost optimization efficiency and accuracy. Experimental results show that, compared with existing methods, the proposed prediction system has significant advantages in interval prediction performance, point prediction accuracy, and runtime, and has passed significance and robustness tests.
ISSN:	0020-0255
DOI:	10.1016/j.ins.2025.122531