A combination model based on multi-angle feature extraction and sentiment analysis: Application to EVs sales forecasting. (15th August 2023)

• Record Type: Journal Article
• Authors: Liu, Jinpei; Chen, Lijuan; Luo, Rui; Zhu, Jiaming
• Is Part Of: Expert systems with applications. Volume 224(2023)
• Publication Date: 2023-08-15
• Subjects: Electric vehicles sales -- Multi-angle feature extraction -- Sentiment analysis -- Combination forecasting -- MEMD; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2023.119986
• Languages: English
• ISSNs: 0957-4174

Abstract

Abstract: The accurate forecasting of electric vehicles (EVs) sales is crucial for reasonably deploying charging infrastructure, improving industrial policies, and securing energy supply for transportation systems. However, most of the forecasting models are based on historical sales data, which results in a certain lag. Although some researchers further consider exploiting the Internet Search Index with high dimensions, they often ignore the extraction of nonlinear features in the process of dimension reduction. Therefore, a novel combination forecasting model based on multi-angle feature extraction and sentiment analysis is proposed. First, we collect Internet search index, online reviews, and sales volume data, thereby considering real-time information. Second, sentiment analysis is adopted to quantify online reviews. Meanwhile, a multi-angle feature extraction method is developed, which can extract the multi-frequency features from the Internet search index. Third, the decomposition methods are performed on the sequences. Finally, EVs sales prediction results are obtained through combination forecasting. To verify the effectiveness and stability of the proposed model, four groups of comparison experiments are conducted. As shown in the empirical results, the proposed model has higher prediction accuracy, which indicates that multi-source information fusion and multi-angle feature extraction enhance the forecasting performance and adaptability of the model.