Using sustainable performance prediction in data-scarce scenarios: A study of park-level integrated microgrid projects in Tianjin, China. (1st July 2021)

• Record Type: Journal Article
• Authors: Wang, Yuan; Li, Shuquan; Wu, Xiuyu; Zhang, Yan; Li, Baoluo; Gao, Lei
• Is Part Of: Journal of cleaner production. Volume 304(2021)
• Publication Date: 2021-07-01
• Subjects: Integrated energy system -- Sustainable performance prediction -- Conditional generative adversarial networks -- Long short-term memory -- Park level integrated energy system; Factory and trade waste -- Management -- Periodicals; Manufactures -- Environmental aspects -- Periodicals; Déchets industriels -- Gestion -- Périodiques; Usines -- Aspect de l'environnement -- Périodiques; 628.5
• DOI: 10.1016/j.jclepro.2021.127042
• Languages: English
• ISSNs: 0959-6526

Abstract

Abstract: Time series data of project performance in park-level integrated energy system projects are non-linear, difficult to collect and store, and scarce. Thus, it is difficult to carry out real-time prediction of project performance. In the context of energy and environmental crises, a real-time prediction method for the sustainable development performance of IES projects based on conditional generative adversarial networks-long short-term memory neural networks was proposed after a full study of the park-level IES projects in Tianjin, China. In this study, an evaluation system for the sustainable development performance of IES projects, such as "integrated energy efficiency, " was established to collect the monthly performance index values of 638 IES projects in Tianjin in 2017. The monthly performance evaluation index was calculated using the entropy weight method. After sorting, a binary method was applied to form the monthly performance evaluation label values, "1″ corresponding to the top 50% of the project, "0″ corresponding to the bottom 50% projects, establishing a database of historical project performance. The generator in CGAN game training was initially used to learn the mapping relationship between the noise distribution under the predicted conditions and the historical IES project performance data set, resulting in 6220 project data with similar distribution, with improved generalization ability of online data mining and accuracy of the stabilization algorithm. LSTM was then used to capture the time dependence in IES project performance data characteristics to predict monthly sustainable performance after 12 months of project operation. Compared with other machine learning models, this method is time-adaptive and the model structure is simple. The average response time of performance prediction for the same park was shortened to 2.76 months, and the prediction accuracy increased to 98.75%. Three schemes were designed to verify the effectiveness of the proposed method by comparing the real data of the park-level IES project in Tianjin with the predicted results. These results have practical significance for strengthening the real-time control of integrated energy projects and for effectively promoting the sustainable development of the integrated energy industry.