基于免疫遗传算法的化工过程微小差别异常诊断

化工过程中同一类型的异常往往也存在微小差别,将不同程度的同一类型异常进行有效识别和归属,掌握同一类型异常的多样化情况,对于化工过程安全监控有重要意义。设计了一种基于人工免疫和遗传思想的微小差别异常诊断方法,并以一个典型的精馏过程Chem CAD仿真模型进行同类型不同程度的阀门异常模式设置,获取正常情况和各异常情况下的样本数据,然后利用所提出的诊断方法计算不同程度异常对其异常类型的隶属概率。结果表明,同一阀门不同开度异常尽管会导致过程监控参数的变化,但该微小差别异常诊断方法却能对其进行有效归属,且归属准确度经启发式算法多次运行后可达98%以上。最后进一步运用TE过程中具有微小差异输出结果的4种异常模式进行了验证试验,结果表明,该方法也可以在较少的异常数据基础上完成对异常模式的准确归属。

Diagnosis of the anomalies of miniature differences in the chemical technological process based on the genetic immunology algorithm

In chemical process,small differences often exist between abnormal situations which belong to the same type of anomaly pattern.It is important to identify and affiliate the same type of anomalies with different varying degrees effectively and grasp the diversification of abnormal situations in process safety monitoring.However,few studies have focused on this issue currently.In this work,a diagnosis method for anomalies with small differences based on artificial immune genetic algorithm was proposed.In this algorithm,two specific judgment conditions were designed to generate appropriate antibodies,i.e.,the judgment for the individuals' fitness and the judgment for the distance between abnormal codes and antibodies.With this algorithm,the affiliation probability of anomalies with different degree belonging to their corresponding types can be obtained by heuristic calculation,as well as the evolutionary individuals' fitness in every generation can be obtained to prove the convergence of the algorithm.With sample data obtained through a dynamic ChemCAD simulation model of a typical distillation process in which some normal and abnormal situations with small differences in valves' opening de gree were considered,the proposed diagnosis method was used to calculate the affiliation probabilities.Results show that the proposed method can affiliate the same type of anomaly with different varying degrees effectively,achieving a correct rate of up to 98％.Furthermore,a validation test using the data of four abnormal patterns of TE process was conducted,in which the four abnormal patterns were of small differences in terms of the reactor feed flow (kg/h).It is found that the proposed method can also achieve accurate affiliation of abnormal patterns with quite few amounts of process data.Therefore,in the field of process safety.monitoring,it is hoped that the proposed diagnosis method of anomalies with small differences can provide some references for the safety control of relevant equipment and process.