基于机器学习的煤层气组成预测及液化过程的实时优化

The skid-mounted liquefaction device is a promising way to solve the utilization problem of small, remote, and distributed coalbed methane (CBM). The CBM properties change with time, which brings a challenge to the optimal operation of liquefaction process. The research on the prediction of CBM component flowrate can provide CBM parameters required for optimization in time and make real-time optimization possible. Based on the idea of process simulation and soft measurement, the CBM liquefaction process was simulated and the CBM component flowrate prediction was carried out. A real-time optimization method for the mixed refrigerant liquefaction process was established, and three gas sources generated randomly were analyzed. The results showed that the data set obtained by the process simulation had good consistency and reliability. The parameter tuning of random forest showed that the model can obtain the optimal or near optimal accuracy when the number of decision trees was between 20 and 40. If this parameter continued to increase, the accuracy improvement was limited. The method based on prediction-optimization could obtain near-optimal operating parameters for the CBM liquefaction process, which was of great significance to the real-time optimization of industrial production.