A novel adaptive discrete grey model with time-varying parameters for long-term photovoltaic power generation forecasting. (1st January 2021)

• Record Type: Journal Article
• Authors: Ding, Song; Li, Ruojin; Tao, Zui
• Is Part Of: Energy conversion and management. Volume 227(2021)
• Publication Date: 2021-01-01
• Subjects: Adaptive discrete grey model -- Genetic algorithm -- Long-term forecasting -- Photovoltaic power generation; Direct energy conversion -- Periodicals; Energy storage -- Periodicals; Energy transfer -- Periodicals; Énergie -- Conversion directe -- Périodiques; Direct energy conversion; Periodicals; 621.3105
• DOI: 10.1016/j.enconman.2020.113644
• Languages: English
• ISSNs: 0196-8904

Abstract

Highlights: An adaptive discrete grey model is proposed for long-term photovoltaic power generation forecasting. The proposed model can grasp nonlinear, fluctuant, and periodic patterns in various datasets. The genetic algorithm is utilized to determine the adaptive parameters. The proposed method strikingly outperforms a range of prevalent benchmarks. Abstract: The rapidly growing photovoltaic power generation (PPG) instigates stochastic volatility of electricity supply that may compromise the power grid's stability and increase the grid imbalance cost. Therefore, accurate predictions of long-term PPG are of essential importance for the capacity deployment, plan improvement, consumption enhancement, and grid balance in systems with high penetration levels of PPG. Artificial neuron networks ( ANN s) have been widely utilized to forecast the short-term PPG due to their strong nonlinear fitting competence that corresponds to the prerequisite for handling PPG samples characterized by volatility and nonlinearity. However, under the circumstances of the large time span, the insufficient data samples, and the periodicity existing in the long-term PPG datasets, the ANN s are easily stuck in overfitting and generate large forecasting deviations. Given this situation, a novel discrete grey model with time-varying parameters is initially designed to deal with various PPG time series featured with nonlinearity, periodicity, and volatility, which widely exist in the long-term PPG sequences. To be specific, improvements in this proposed model lie in the following aspects: first, the time-power item and periodic item are designated to compose the time-varying parameters to capture the nonlinear, periodic, and fluctuant developing trends of various time series. Second, owing to the complex nonlinear relationships between the above parameters and forecasting errors, the genetic algorithm applies shortcuts to seek optimum solutions and thereby enhances the prediction precision. Third, several practical properties of the proposed model are elaborated to further interpret the feasibility and adaptability of the proposed model. In experiments, a range of machine learning methods, autoregression models, and grey models are involved for comparisons to validate the feasibility and efficacy of the novel model, through the observations of the PPG in America and China. Finally, a superlative performance of the proposed model with the highest forecasting precision, small volatility of empirical results, and generalizability are confirmed by the aforementioned cases.