Can cross-sector information improve multi-energy demand forecasting accuracy?

More and more distributed energy resources are integrated into regional integrated energy systems (RIES), which poses great challenges to energy balance. RIES can coordinate power, gas, heat, and cooling systems jointly to enhance energy efficiency and explore flexibility for distributed renewable energy accommodation. Multi-energy demand forecasts are the basis of the flexible operation of RIES. However, the multi-energy demands are deeply coupled in RIES. Related researches utilize cross-sector information from different sectors to tackle the coupling relationship. Nevertheless, the unknown influence of cross-sector information offered by other sectors varies with the operating pattern, which is difficult to be evaluated. This work proposes an operating pattern recognition-based method for adaptive cross-sector information identification. Firstly, the K-means cluster algorithm is adopted to identify different operating patterns. After that, cross-sector information is selected based on the Pearson coefficient. Furthermore, two models, i.e., the local model and fine-tuned model, are modified with the assistance of selected cross-sector information. The proposed method is evaluated by a RIES with three energy types (electricity, chill water, and steam). The proposed two methods acquire better accuracy than the three benchmark models. Moreover, the Shapley value is applied to verify the contribution of selected cross-sector information. The result shows that all selected cross-sector information plays a significant role in load prediction. (c) 2022 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Automated summary

More and more distributed energy resources are integrated into regional integrated energy systems to enhance energy efficiency and flexibility. However, the unknown influence of cross-sector information from different sectors and its evaluation pose challenges. This study proposes an adaptive method for cross-sector information identification based on operating pattern recognition, which shows improved accuracy in load prediction.