Abstract:
In Texaco water-coal-slurry gasification process, operational parameter has direct impact on effective gas yields. Big data opens a new opportunity to machine learning models in operational parameter optimization. However, gasification is a multivariate-coupled stable process with big inertia. The incoming quality (ICQ) inspection is sparse. Considering these characteristics, mean and variance based time series segmentation is proposed to extract stable subsequences. And multi-variate temporal patterns are obtained by clustering over these subsequences′ feature space. GMM (Gaussian mixture model) is applied to divide ICQ into few categories. Under a given ICQ category, static control policies are obtained by Apriori association rule mining. Considering time lags between control parameter change and output improvement, dynamic control policy problem can also be formulated as an association rule mining problem, by including future results as the RHS(right hand side), and considering the average ICQ during the time lag in the LHS(left hand side). The approach is verified over a 27 months dataset of 3 gasifiers. The expected improvement of CO and H2 gas yields is around 1.38%. Time series clustering and association rule mining approach is a feasible method in coal gasifier operational parameter optimization.
