A novel deep interval prediction model with adaptive interval construction strategy and automatic hyperparameter tuning for wind speed forecasting. (1st February 2021)

• Record Type: Journal Article
• Authors: Xie, Yuying; Li, Chaoshun; Tang, Geng; Liu, Fangjie
• Is Part Of: Energy. Volume 216(2021)
• Publication Date: 2021-02-01
• Subjects: Variational mode decomposition (VMD) -- Gated recurrent unit (GRU) -- Particle swarm optimization (PSO) -- Construct prediction interval -- Decomposition prediction aggregation (DPA); Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2020.119179
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: Wind energy is a renewable energy source with great development potential. However, its inherent instability and randomness have brought great challenges to the maximum utilization of wind energy. Wind speed forecasting is one of the most effective ways to mitigate these challenges, which plays an important role in the operational management and decision-making of wind power system operators. In this study, a novel wind speed interval prediction model based on gated recurrent unit, Variational Mode Decomposition, and Particle Swarm Optimization was proposed. The original wind speed sequence was decomposed into several smoother sub-sequences through the Variational Mode Decomposition algorithm, and corresponding sub-models were established based on the gated recurrent unit. To better supervise the training process, artificial prediction intervals with adaptive adjustment strategies were devised. Moreover, the Particle Swarm Optimization algorithm was adopted to search for the optimal superposition weights of PIs to achieve the integral optimization of the model. The qualitative and quantitative performance of the proposed method has been fully tested and verified in a series of real cases. Highlights: A novel hybrid wind speed interval prediction model was proposed. Adaptive interval construction and automatic hyperparameter tuning methods were devised. Intelligent superposition optimization scheme was designed. The qualitative and quantitative performance of the proposed method has been fully verified.