Research On Dynamic Wi-Fi Localization Algorithm In Underground Mine Environment

The intelligent and informative construction of coal mines is becoming more and more mature,and the precise positioning technology of underground moving targets has become a hot spot and difficult research point in the academic field.Various underground wireless positioning technologies have been developed rapidly,and Wi-Fi wireless positioning technology has been commonly used in large state-owned coal mines.The strong time-varying nature of underground wireless communication environment,the real-time movement of various underground infrastructures and operators and other positioning targets,and the serious multipath propagation phenomenon of electromagnetic waves caused by large amount of dust,fog droplets and narrow and irregular tunnel walls can all lead to the decrease of positioning model accuracy.Wi-Fi-based downhole wireless positioning technology can conveniently achieve high accuracy downhole positioning using explosion-proof mobile intelligent terminals and fully meet the accuracy requirements of downhole positioning.The wireless communication environment in underground coal mines is changing in real time,and the highly dynamic underground environment reduces the accuracy of positioning models,and wireless positioning technologies cannot meet the demand for positioning accuracy in underground coal mines.Among them,Wi-Fi fingerprint-based underground positioning technology has been commonly used in underground positioning due to its unique positioning advantages.It is a current research hotspot to maintain or improve the model positioning accuracy in the highly dynamic changing underground environment.In this thesis,the downhole localization algorithm of Online Sequential Extreme Learning Machine(OSELM)is used to update the model online by adding new data,reducing the workload of experimental data collection and model training,and improving the problem of model accuracy degradation due to dynamic changes of downhole environment.At the same time,the OSELM algorithm is improved to address the shortcomings of the OSELM algorithm.The experimental study shows that the proposed OSELM and its improved algorithm can effectively improve the problem of model positioning accuracy degradation due to dynamic changes of downhole environment.Based on the OSELM online incremental localization algorithm,this thesis addresses the problem of the decrease in accuracy of the localization model due to high dynamic changes in the wireless communication environment in underground coal mines,and the main research elements can be summarized as follows:(1)Using the OSELM online incremental learning algorithm for downhole locationConsidering the time-varying wireless communication environment in underground coal mines,using OSELM algorithm with online learning capability for downhole location.In practice,due to the complex and changing environment of underground coal mines,there are many disturbances,and the collected data are usually real-time,it is difficult to maintain the original positioning accuracy of the positioning model without real-time updating.Therefore,the OSELM algorithm is proposed based on the Extreme Learning Machine(ELM)to realize online real-time updating of the localization model.Experimental validation shows that OSELM algorithm can improve model localization accuracy more effectively than BP and ELM batch-type learning methods.(2)Assigning timeliness and coverage weights to the OSELM algorithm and proposing improvements to the OSELM algorithmThe OSELM online model update method can effectively solve the problem of model accuracy degradation caused by dynamic changes in the downhole environment.However,the model only completes the process of online updating and does not consider aspects such as the validity of the added data.In this thesis,the OSELM algorithm is improved in terms of the timeliness of the added data and the coverage of the reference points for collecting the added data as well as the comprehensive consideration of these two factors,and the weights are used to indicate the degree of updating of the original model by the added data.The experimental validation shows that the introduction of the weight term for the OSELM algorithm can effectively improve the positioning model accuracy compared with the unimproved OSELM positioning algorithm.(3)Introducing regularization and forgetting factor mechanisms for the OSELM algorithm and proposing improvements to the OSELM algorithmTo address the shortcomings of the OSELM algorithm in terms of sick matrix inversion and treating all new data equally,we propose the OSELM based on Regularization(R-OSELM)and the OSELM based on Forgetting factor(F-OSELM),and fuse the two mechanisms to propose OSELM based on Regularization and Forgetting factor(FR-OSELM).Experimental validation shows that the localization accuracy of both R-OSELM and F-OSELM algorithms is higher than that of OSELM algorithm after the change of experimental environment;within 3m error distance,the localization accuracy of FR-OSELM algorithm,which fuses the two mechanisms,is higher than that of OSELM,R-OSELM and F-OSELM algorithms,and the proposed fusion improvement algorithm can better improve the problem that the accuracy of the localization model decreases due to the highly dynamic changes of the downhole wireless communication environment,and the proposed improvement algorithm is effective.