Research On Online Displacement Forecast Of Mine Dump Based On Flicker Firefly Algorithm And Fuzzy Parameter Self-tuning

The safety and stability of mine dumps have always been an important issue related to the mining company's "safety and production".Compared with other industries,the monitoring and alarming research on mine dumping grounds are still in its infancy,but frequent occurrences of dumping landslides,landslides,mudslides and other disasters have caused people's property losses and terrible social influence which is alarming the bells for Today's mining companies.With the development of monitoring technology based on the Internet of Things,the multi-index visual real-time monitoring system for mine dumps has been developed in some of China's advanced mining companies.However,this alarm mechanism based on "real-time monitoring-feedback" does not meet the requirements of the enterprise.Therefore,realizing the short-term prediction and alarm mechanism of the safety status of the mine dump based on the real-time monitoring system has a very important value to further enhance the enterprise's initiative in handling disaster accidents and reduce or even eliminate the casualties and property losses which is also deserves more attention.In this paper,the status quo of safety pre-warning for mine waste dumps is studied.Based on real-time monitoring data of waste dumps,a support vector regression prediction model is established which is based on surface displacement vectors and other indicators of dumping sites.At the same time,the concept of chaotic flicker factor ? is generalized to the firefly algorithm to realize the optimization in a complex multi-maximum nonlinear SVM model.The convergence performance is analyzed using the classical function test set and the actual dump data set.Aiming at the surge of monitoring data of mine dumps over a long period of time,the support vector regression machine has low processing efficiency,and cannot satisfy the needs of real-time forecast,an online learning based on feature sample reduction and insensitivity loss function fuzzy self-tuning is proposed.This method effectively solved the problem of rapid model training and accurate prediction of large-scale datasets.At the end of this paper,we try to use distributed strategy to solve the parallel cross-training problem of support vector regression and build a distributed offline model training method for batch data sets based on Hadoop distributed cluster,which provides a more effective bottom layer solution for real-time forecasting.