基于BP神经网络的煤直接液化装置VOCs包袋法检测与预测建模

挥发性有机物（VOCs）是现代煤化工生产过程中产生的主要污染物之一。煤化工是我国石化领域的重要组成部分，通过煤制油装置可以实现煤炭的清洁利用，煤制油装置由于设备、物料、操作条件等与石化企业存在差异，目前我国针对石化的美国联邦环保署（EPA）相关方程不适用于煤直接液化装置，且目前国内并没有用于核算煤制油装置的相关方程。本文开展了煤制油装置VOCs检测方法研究，并修正了相关方程系数，得到了密封点VOCs核算方程。首先对煤制油装置进行泄漏与维修（LDAR）检测，得到了煤制油装置密封点的VOCs泄漏数据。在数据分析的基础上，改进了包袋法采样方法，对煤直接液化装置进行了密封点的包袋采样分析。在此基础上，得到了相关方程的修正系数，首次提出了适用于国内煤制油装置VOCs泄漏核算的相关方程。最后，考虑了移动距离、温度、压力及多处泄漏等参数对相关方程的影响，校准了煤直接液化装置的相关方程，建立了反向传播算法（BP）神经网络模型，得到了逼近结果较好的相关函数。

Method of checking and accounting for the bag method of VOCs in coal direct liquefaction unit based on BP neural network

Volatile organic compounds (VOCs) are one of the main pollutants in modern coal chemical production. Coal chemical industry is an important part of China's petrochemical fields, and through coal direct liquefaction device the clean utilization of coal can be realized. As the equipment, materials, operating conditions of the coal to oil plant are different from petrochemical enterprise, the related equation for petrochemical industry in China does not apply to coal direct liquefaction plant, and there is no relevant equation for the accounting of the coal to oil plant in China at present. Therefore, this paper studied VOCs detection method of coal-to-oil unit, modified the coefficient of relevant equation, and obtained the VOCs accounting equation of sealing point. Firstly, the leakage and maintenance (LDAR) detection was carried out on the coal-to-oil unit, and the VOCs leakage data of the sealing point of coal-to-oil unit was obtained. On the basis of data analysis, the sampling method of bagging method was improved, and the sealing point sampling and analysis of coal direct liquefaction unit was carried out. On this basis, the correction coefficient of the relevant equation of the U.S. federal environmental protection agency (EPA) was obtained, and the relevant equation applicable to the VOCs leakage accounting of domestic coal-to-oil plant was put forward. Finally, considering the influence of parameters such as moving distance, temperature, pressure and multiple leaks on the correlation equation, the EPA correlation equation of the coal direct liquefaction unit was calibrated, the BP neural network model was established, and correlation function with good approximation result was obtained.