Research On Threshold Optimization In Chemical Process

With the development of modern science and technology, the chemical system is also moving towards fine, large-scale direction. In order to detect faults in time and ensure normal production, the alarm system is indispensable. However, there are a series of problems in the current alarm system, such as the alarm number, false or invalid alarms.These problems seriously interfere with the normal operation of the alarm system, increasing the burden on the staff, bringing additional industrial losses. How to use the current increasingly sophisticated data mining technology, and gradually optimize the industrial production problems,has become an important issue.Alarm threshold is directly related to the performance of the alarm system. When the threshold is set too low, the normal state of the data is also mistaken for the alarm data, resulting in a large number of false alarms. As a result, the important alarm data submerged in the mass of false positives, and the performs of the alarm system is reduced .If the alarm threshold is set too high, it will ignore the important alarm data and increase the false alarm. In order to make the alarm system better, it is necessary to optimize the alarm threshold reasonably.Based on the data mining method and selecting the chemical process as the research object, this paper presents a new alarm threshold setting method. Firstly, based on the historical data, the kernel density estimation technique is used to estimate the probability distributions of false alarms and invalid alarms. Then the loss function is used as the standard to explain its characteristics in terms of value and benefit. Finally, an efficient artificial fish swarm algorithm is used to get the best alarm threshold. TE simulation results show that the new method can quickly and effectively find the best threshold, balance the relationships of false and invalid alarms, and finally optimize the performance of the alarm system.