Gray-Related Support Vector Machine Optimization Strategy and Its Implementation in Forecasting Photovoltaic Output Power

• Published in: International Journal of Photoenergy Vol. 2022; pp. 1 - 9
• Main Authors: Xiao, Bo, Zhu, Hai, Zhang, Sujun, OuYang, Zi, Wang, Tandong, Sarvazizi, Saeed
• Format: Journal Article
• Language: English
• Published: New York Hindawi 25.02.2022; John Wiley & Sons, Inc; Wiley
• Subjects: Accuracy; Air temperature; Analysis; Artificial intelligence; Deconstruction; Economic forecasting; Electric power grids; Evaluation; Genetic algorithms; Irradiance; Machine learning; Mathematical models; Optimization; Optimization techniques; Parameters; Photovoltaic cells; Sensitivity analysis; Solar energy; Solar energy industry; Support vector machines; Time series; China
• ISSN: 1110-662X; 1687-529X
• DOI: 10.1155/2022/3625541

Reliable and accurate photovoltaic (PV) output power projection is critical for power grid security, stability, and economic operation. However, because of the indirectness, unpredictability, and solar energy volatility, predicting precise and reliable photovoltaic output power is a complicated subject. The photovoltaic output power variable is evaluated in this study using a powerful machine learning approach called the support vector machine model based on gray-wolf optimization. A vast dataset of previously published papers was compiled for this purpose. Several studies were carried out to assess the suggested model. The statistical evaluation revealed that this model predicts absolute values with reasonable accuracy, including R2 and RMSE values of 0.908 and 74.6584, respectively. The practical input data were also subjected to sensitivity analysis. The results of this analysis showed that the air temperature parameter has a greater effect on the target parameter than the solar irradiance intensity parameter (relevancy factor equal to 0.75 compared to 0.49, respectively). The leverage approach was also used to test the accuracy of actual data, and the findings revealed that the vast majority of data is accurate. This basic but accurate model may be quite effective in predicting target values and could be a viable substitute for laboratory data.