Decision tree based parameter identification and state estimation: Application to Reactor Operation Digital Twin
Rong Zhao, Lizhan Hong, Hongjun Ji, Qinyi Zhang, Shiquan Zhang, Qing Li, Helin Gong
Abstract
This study presents a comprehensive investigation into enhancing the performance of decision tree algorithms within the Reactor Operational Digital Twin (RODT) framework. Our previous work established the RODT and optimized the K-Nearest Neighbors (KNN) algorithm for its operation. Building on this foundation, we systematically explored decision tree techniques for both forward and inverse problems of the RODT. Through extensive experimentation, we integrated advanced techniques such as Bayesian optimization, GPU acceleration, and parallel processing to enhance the decision tree’s training efficiency and reduce its memory footprint. Our findings reveal that Gradient Boosting Decision Trees (GBDT) outperform KNN in accuracy for forward problems, while Adaboost, though slightly less accurate, offers comparable stability with respect to noisy measurements for inverse problems. Despite a slight dip in performance under noisy conditions, decision trees still hold promise in digital twin modeling. This research not only bridges the application gap of decision tree algorithms in digital twin modeling but also significantly improves the overall performance of the RODT. The insights from our experiments, particularly the synergy between GBDT and Bayesian optimization, offer valuable contributions to a broad spectrum of applications.