Research On Geomagnetic/Wi-Fi Fusion Positioning Algorithm In Complex Underground Environment

With the update and iteration of technology,underground coal mine positioning technology is slowly transitioning from traditional wired communication technology to wireless communication technology.Among them,the use of wireless signals can not only communicate,but also perform daily activities such as ranging and positioning,which has become a very important resource in our lives.It has the characteristics of good flexibility,fast signal propagation,wide coverage and strong robustness.It can realize positioning through the cooperation of various portable mobile hardware,and is one of the important means to realize underground positioning.Its environmental uncertainty is greater,the irregular activities of underground workers and various coal mining mechanical tools will have a great impact on the propagation of wireless signals,because these non-negligible effects allow efficient underground positioning.The system is full of challenges and puzzles.In this thesis,funded by the Science and Technology Program of Inner Mongolia Autonomous Region,the research on the wireless signal to achieve high-precision positioning algorithm in coal mines.Focusing on the goals of reducing the influence of the dynamic changes of the environment on the RSSI signal in the coal mine,study simple and efficient data collection methods,establish a real-time and effective fingerprint database,and use more advanced positioning algorithms to further improve the positioning accuracy of underground personnel.In the process of achieving the above goals,this paper adopts the feature analysis and selection of underground wireless signals.,data resampling and segmentation,deterministic fingerprint matching algorithms,LSTM neural network,Attention mechanism,gradient boosting tree fusion localization algorithm and other theories,the research results show that the use of geomagnetic signals to realize underground positioning the algorithm proposed in this paper has certain feasibility,and it shows that the fusion positioning algorithm does have a better positioning effect in complex environments.Based on the powerful learning ability of neural network,this thesis conducts localization research on the characteristics of geomagnetic signals in the special environment of coal mines.The main research contents are as follows:(1)The proposed LSTM geomagnetic localization algorithm based on attention mechanism,and a hierarchical LSTM(HLSTM)geomagnetic localization model is further proposed.Aiming at the instability and inaccuracy of traditional wireless signals for indoor positioning,in this thesis,based on the geomagnetic sequence signal,the long-short-term memory neural network model is selected as the matching algorithm.In order to speed up the convergence of the model,an attention mechanism is introduced,and a hierarchical long-shortterm memory network localization model is further proposed.The model further improves the positioning accuracy by combining more historical geomagnetic signals to increase the variability of different locations.(2)Study the data acquisition method and characteristic for the underground.The physical environment of coal mines is completely different from the ground.The particularity of the environment makes it difficult to collect underground data.We found that as long as the underground environment does not change greatly,the geomagnetic signal is very stable.Therefore,we use the sequence enhancement algorithm to generate underground coal mines.A large number of geomagnetic trajectories are used for model training and testing methods,rather than manual extensive collection.(3)A multi-source fusion gradient boosting tree(GBDT)localization model is proposed.Due to the particularity of underground coal mines,it is difficult for a single signal to achieve high-precision positioning.This paper uses the Stacking strategy to combine the singlemode positioning results to obtain a new training set.Because of the stability and robustness of the gradient boosting tree(GBDT),This subject uses GBDT to learn the weight of each base station RF signal positioning result.Build a new training set based on single-mode positioning results(Wi-Fi/mobile network,geomagnetic).