基于PMHA的CNN-BiLSTM-XGBoost催化裂化装置产率预测

PMHA-Based CNN-BiLSTM-XGBoost Catalytic Cracking Unit Yield Prediction

摘要: 催化裂化装置是石油炼制过程中至关重要的设备之一，其石油产品产率的准确预测对于优化生产和资源利用至关重要。针对传统的机理模型方式存在的建模困难、参数估计困难、对复杂系统的局限性以及难以处理非线性关系和高维数据等劣势，数据驱动的建模方式更灵活适应复杂系统和大量数据，能够有效处理非线性关系和高维数据，具有更强的预测准确性和实时性。因此，采用并行结构多头注意力机制，构建了一种基于CNN-BILSTM-XGBoost的催化裂化装置产率预测模型，以应对这些挑战。实验结果表明，该模型的预测准确率能达到98%，显著优于SVR、RF、AdaBoost等传统回归模型，为催化裂化过程的智能化升级提供新的思路。

Abstract: The catalytic cracking unit is one of the most crucial equipment in the petroleum refining process, and accurate prediction of its petroleum product yield is crucial for optimizing production and resource utilization. In view of the disadvantages of traditional mechanism model methods such as modeling difficulties, parameter estimation difficulties, limitations in complex systems, and difficulty in processing nonlinear relationships and high-dimensional data, data-driven modeling methods are more flexible and adaptable to complex systems and large amounts of data. It can effectively handle nonlinear relationships and high-dimensional data, and has stronger prediction accuracy and real-time performance. Therefore, a parallel structure multi-head attention mechanism was used to construct a catalytic cracking unit yield prediction model based on CNN- BILSTM-XGBoost to address these challenges. Experimental results show that the prediction accuracy of this model can reach 98%, which is significantly better than traditional regression models such as SVR, RF, and AdaBoost, and provides new ideas for the intelligent upgrade of the catalytic cracking process.