文献类型：期刊文献

英文题名：Hybrid Modeling Based on Multi-Strategy Enhanced Dung Beetle Optimization Integrating Seventeen-Lump Kinetic and Light Gradient Boosting Machine

作者：Chen, Xiangming[1]; Luo, Kai[1]; Huang, Cheng[1]; Gong, Jiahao[1]; Long, Jian[1,2]; Liang, Chen[1]

机构：[1] Key Laboratory of Smart Manufacturing in Energy Chemical Process, Ministry of Education, East China University of Science and Technology, Shanghai, 200237, China; [2] State Key Laboratory of Chemical Engineering, China

年份：2024

外文期刊名：SSRN

收录：EI(收录号：20240428292)

语种：英文

外文关键词：Adaptive boosting - Bayesian networks - Kinetic parameters - Multiobjective optimization - Prediction models

摘要：Fluid catalytic cracking is a crucial procedure in the petrochemical industry, but understanding and predicting its process remains challenging as its high degree of complexity and nonlinearities. This paper proposes a hybrid model incorporating a seventeen-lump kinetic model (SLKM) and light gradient boosting machine (LightGBM). Firstly, as the difficulty of parameter determination of SLKM, this paper improves the dung beetle optimization algorithm for solving the parameters in SLKM. Consequently, the inaccuracy and time-consuming parameter solving of the mechanistic model are solved. Secondly, the challenge of hyperparameter optimization in LightGBM, which greatly influences the precision of model predictions, is addressed in this paper by applying Bayesian optimization with Hyperband. The prediction performance of the LightGBM model improves significantly following this hyperparameter optimization. SLKM is backed by knowledge of existing theories, so the outputs of the models are consistent with physicochemical laws and require less data, but its prediction performance is not good. Finally, three hybrid models are constructed by combining six-lump kinetic model and LightGBM: the series model, the parallel model, and the hybrid model with adaptive weights to realize the complementary of the two models mentioned above. The industrial data validation results demonstrate that the adaptive weighted hybrid model achieves the best prediction performance, with the Mean Relative Error, Mean Squared Error, and Mean Absolute Error reaching 1.36%, 0.0169, and 0.077 on the industrial real-world dataset, respectively. ? 2024, The Authors. All rights reserved.