基于集成多模型的煤制乙炔浓度软测量

Integrated Multi-model Soft Sensor of Acetylene Concentration Prediction for Coal Pyrolysis to Acetylene Process

目前,工业大量制取乙炔的方法主要是水解电石法,该方法成本高,污染严重。等离子体裂解煤制乙炔的技术具有高效清洁的特点。但该工艺中煤裂解过程机理未知且反应快速,乙炔浓度需要离线化验检测,时间滞后严重。针对上述问题,提出了一种集成多模型的乙炔浓度软测量算法。该算法采用两阶段级联策略,第一阶段采用自适应高斯混合模型(Gaussian mixture model,GMM)对样本聚类,第二阶段采用多个支持向量回归模型对每个簇建模。最终软测量结果基于后验概率权重得到。现场案例的仿真验证表明,所提方法在准确度和稳定性方面均能满足工业现场的需求。

At present,the main method for large-scale industrial production of acetylene is the hydrolysis calcium carbide method,which has high cost and serious pollution.The technology of plasma cracking coal pyrolysis to acetylene has the characteristics of high efficiency and cleanliness.However,the mechanism of the coal cracking process is unknown and the reaction is fast in this process.The concentration of acetylene needs to be tested by offline laboratory tests,and the time lag is serious.To solve the above problems,an integrated multi-model soft sensor algorithm for acetylene concentration is proposed in this paper.The algorithm adopts a two-stage cascade strategy.An adaptive Gaussian mixture model(GMM)is used to cluster samples in the first stage,and multiple support vector regression(SVR)models are used to model each cluster in the second stage.The final soft sensor result is obtained based on the posterior probability weight.The simulation verification of the field case shows that the proposed method can meet the needs of the industrial field in terms of accuracy and stability.