Research On A Method Of Gross Error Elimination For Slope Monitoring Data Based On Machine Learning

Safety monitoring in open-pit mine is a significant part in measurement field.In the process of open-pit mining,the stability of steep slopes shaped by excavation will directly affect the safety of mining.Therefore,it is particularly important for detection and elimination the gross errors in monitoring data to the safety of open-pit mining.Most of gross error elimination methods have their own shortcomings,some traditional methods such as pauta criterion can only remove one univariate gross error at a time with a large amount of calculation,the relation between variates is not considered in these methods,at the same time,the excessive gross error data will cause the shift of mean,which will affect the accuracy of pauta criterion;The wavelet analysis method mainly eliminates the noise in monitoring data,but it can't detect the existence of gross error,and the selection of wavelet base and threshold function will affect the results;The BP neural network model is too simple to get a good effect in gross error elimination,and it is easy to fall into a local optimal solution.This paper proposes a Particle Swarm Optimization Two-hidden-layer Extreme Learning Machine based on kernel function.The new method adds a hidden layer to the structure of ELM network.Specifically,the monitoring data will be processed by kernel function before input into the model which the parameters of input layer and the first layer are optimized by PSO.The simulation results show that the use of kernel function can weaken the coupling of data,the use of PSO can calculate the optimal solution of ELM network parameters.Therefore,the test accuracy of the model is further improved.In practical engineering applications,the distance of monitoring data is different,and the coupling is so stronger that the PSO-TELM can't eliminate the influence well enough.Therefore,this paper proposes a mean variance model based on PSO-TELM improved by deep belief network.This algorithm reduces the influence of coupling and improve the accuracy mainly by changing the input layer of the TELM network into 60 Restricted Boltzmann Machines models and optimizing these models with minimum error by PSO method.In this algorithm,monitoring data will be calculated by kernel function,then they will be processed by RBMs and generate a data matrix,this data matrix will be integrated processed by TELM model,and get the gross error elimination result by mean variance.The simulation results show that the new method has the ability to eliminate gross error quickly with a high accuracy compared with traditional methods,the robustness is also strong enough to adapt the application in gross error elimination in open-pit mining,which is a highly practical algorithm.