In modern complex systems,real-time regression prediction plays a vital role in performance evaluation and risk warning.Nevertheless,existing methods still face challenges in maintaining stability and predictive accur...In modern complex systems,real-time regression prediction plays a vital role in performance evaluation and risk warning.Nevertheless,existing methods still face challenges in maintaining stability and predictive accuracy under complex conditions.To address these limitations,this study proposes an online prediction approach that integrates event tracking sensitivity analysis with machine learning.Specifically,a real-time event tracking sensitivity analysis method is employed to capture and quantify the impact of key events on system outputs.On this basis,a mutualinformation–based self-extraction mechanism is introduced to construct prior weights,which are then incorporated into a LightGBM prediction model.Furthermore,iterative optimization of the feature selection threshold is performed to enhance both stability and accuracy.Experiments on composite microsensor data demonstrate that the proposed method achieves robust and efficient real-time prediction,with potential extension to industrial monitoring and control applications.展开更多
基金financial support from the National Natural Science Foundation of China(Grants No.U2330206,No.U2230206,and No.62173068)the Natural Science Foundation of Guangxi Province(Grant No.AB24010157)+1 种基金the Sichuan Forestry and Grassland Bureau(Grant Nos.G202206012 and G202206012-2)Sichuan Science and Technology Program(Grant Nos.2024NSFSC1483,2024ZYD0156,2023NSFC1962,and DQ202412).
文摘In modern complex systems,real-time regression prediction plays a vital role in performance evaluation and risk warning.Nevertheless,existing methods still face challenges in maintaining stability and predictive accuracy under complex conditions.To address these limitations,this study proposes an online prediction approach that integrates event tracking sensitivity analysis with machine learning.Specifically,a real-time event tracking sensitivity analysis method is employed to capture and quantify the impact of key events on system outputs.On this basis,a mutualinformation–based self-extraction mechanism is introduced to construct prior weights,which are then incorporated into a LightGBM prediction model.Furthermore,iterative optimization of the feature selection threshold is performed to enhance both stability and accuracy.Experiments on composite microsensor data demonstrate that the proposed method achieves robust and efficient real-time prediction,with potential extension to industrial monitoring and control applications.
