Simultaneous Data Reconciliation and Gross Error Detection in Olefin Plants by the Expectation Maximization Algorithm Based on Gaussian Mixture Distributions

Ethylene and its derivatives play an important role in the development of the petrochemical industry, and olefin plants for obtaining ethylene by steam cracking technology are widely used around the world. However, there are many variables involved in the hot section of the plant, and the flow connection relationships are complex. Therefore, when the measurement data contains many gross errors, it is difficult to eliminate the gross errors and obtain accurate data reconciliation results at the same time. However, Gaussian mixture distribution models have independent parameters to characterize gross errors, so they are able to detect multiple gross errors and achieve good reconciliation results, and the expectation maximization(EM) algorithm can efficiently derive each parameter and update the estimates of the parameters iteratively in the model. In this paper, it is proved experimentally that this method can be well used to solve the data reconciliation and gross error detection problems in the hot section of olefin plant.