基于FA-ISSA-PPR模型的旋风分离器分离效率预测

Prediction of the separation efficiency of a cyclone separator based on the FA-ISSA-PPR model

旋风分离器是气田开发中常用的气固分离设备,准确预测旋风分离器的分离效率对于指导其结构设计和方法优化具有重要意义。在对数据集进行相关性分析的基础上,采用因子分析(factor analysis, FA)简化变量,降低预测模型的复杂程度,利用改进的樽海鞘群算法(improved salp swarm algorithm, ISSA)对投影寻踪(projection pursuit regression, PPR)的模型参数进行优化,形成FA-ISSA-PPR组合模型。结果表明,利用FA模型,原数据集的10个变量可以简化合并为4个公因子,分别代表尺寸参数、颗粒沉降特性、粒子运行轨迹和等效分割粒径对分离效率的影响;与半经验模型和其余机器学习模型相比,组合模型在预测精度和训练时间上具有一定的优越性,在测试样本上的平均绝对误差(MAE)为0.005 91,R^(2)可达0.995,证明了其在小样本、非线性数据分析上的准确性、鲁棒性和泛化性。

Cyclone separators are commonly used for gas-solid separation in a gas field.It is of great significance to accurately predict the separation efficiency of a cyclone separator in order to guide its structure design and optimization.On the basis of correlation analysis of data sets,factor analysis(FA)was used to simplify the variables to reduce the complexity of the prediction model,and the improved salp swarm algorithm(ISSA)was used to optimize the model parameters of projection pursuit regression(PPR)to form a combinatorial optimization model.The results show that the ten variables in the original dataset can be simplified and merged into four common factors by the FA model,representing the effects of size parameters,particle settling characteristics,particle trajectories and equivalent particle size on separation efficiency.Compared with semi-empirical models and other machine learning models,our new combined model has advantages in prediction accuracy and training time.The MAE on test samples was 0.00591,and the R^(2) reached 0.995,demonstrating the accuracy,robustness and generalization of the model for small samples and nonlinear data analysis.