摘要：This paper describes the use of the LoopRank algorithm to determine the importance of control loops in large operating plants. Defining control loop importance is an important step towards the prioritization of control loop maintenance efforts, building on extensive existing research into controller performance monitoring, which can identify poorly-performing control loops. LoopRank uses a connectivity matrix, which can be extracted from operating data using techniques such as partial correlation or Granger causality. The algorithm was applied to synthetic data from a simulated tank network, and to industrial data from a minerals processing plant. For the simulated data, LoopRank with Granger causality provided no significant improvement over LoopRank with the relatively simple partial correlation results. However, for the industrial data, LoopRank with partial correlation was not able to correctly assign importance to variables which are critical to advanced process control, such as mill load, power and speed. LoopRank with Granger causality was able to identify these variables, suggesting that incorporating lagged information in the connectivity matrix provides better importance rankings than using simpler methods.
