关键词: 模态分解, 多变量相空间重构, 热负荷预测, 双向长短时记忆神经网络, 园区综合能源系统

Abstract: Aiming at the problem that the heat load of the park-level integrated energy system is affected by multi-energy flow and the feature extraction capability of the prediction model is insufficient, this paper proposes a dual feature processing heat load prediction model integrating the improved Complete Ensemble Empirical Mode Decomposition with Adaptive White Noise (ICEEMDAN) and Multivariate Phase Space Reconstruction (MPSR). First, the heat load time series are decomposed into modal components of different frequencies by ICEEMDAN method, and the component reconstruction is carried out by calculating the entropy value of the samples; second, the C-C method is used to determine the optimal delay time and embedding dimensions of the reconstructed heat load components and other input feature sequences, and to construct the multivariate phase-space reconstructed datasets under the different frequencies of the heat load components; finally, each dataset is inputted into a bidirectional long short-term memory neural network (BiLSTM) model with optimized parameters by an arithmetic optimization algorithm, and the prediction results were superposed to get the ultimate heat load prediction value. The case results show that compared with other models, the proposed method has good prediction results.

Key words: mode decomposition, multivariate phase space reconstruction, heat load prediction, bidirectional long short-term memory, park-level integrated energy system