基于深度学习的超临界裂解煤油流量特性研究

Flow Characteristics of Supercritical CrackedKerosene Based on Deep Learning

关键词:  超临界裂解态  煤油  流量特性  预测模型  深度学习

Key words:  Supercritical cracked state  Kerosene  Flow characteristics  Prediction model  Deep learning

摘要:

为了明晰超临界裂解煤油的流量特性，实现对超临界裂解煤油流量的准确预测，采用实验方法测量了较大压力和温度范围内超临界裂解煤油的流量，对超临界裂解煤油流量特性进行了分析，基于多元线性回归方法，多元多项式回归方法和深度学习方法分别建立了超临界裂解煤油流量预测模型并给出了模型评估指标。研究结果表明，超临界裂解煤油流量主要与压降和温度有关，并且流量与压降和温度之间存在着很强的非线性关系；基于深度学习方法的深度神经网络模型性能优于多元线性回归模型和多元多项式回归模型，能够更加准确地刻画超临界裂解煤油的流量特性，其平均相对预测误差在1.1%左右，最大相对预测误差在7%以下。

Abstract:

In order to figure out flow characteristics and implement accurate prediction of flow rate of supercritical cracked kerosene， flow rate of supercritical cracked kerosene in a large range of pressure and temperature was measured and flow characteristics were analyzed. Furthermore， prediction models respectively based on multivariate liner regression method， multivariate polynomial regression method and deep learning method were built and several evaluation criteria were given. The results indicate that flow rate of supercritical cracked kerosene varies with pressure drop and temperature， and there exists strong nonlinear relationship between flow rate and pressure drop as well as temperature. The results also show that deep neural network model based on deep learning method with mean relative prediction error at about 1.1% and max relative prediction error lower than 7%， is better than multivariate liner regression model as well as multivariate polynomial regression model and can more accurately reflect flow characteristics of supercritical kerosene.