Research On Multi-source Fusion Location Algorithm Based On PDR/Wi-Fi In Underground Coal Mine

Coal is the main production and living energy in our country,and the development of coal industry is directly related to the economic development of the whole country.Due to the complex and harsh working environment of the coal industry,safety issues have always affected the development of the coal industry,and the overall production of coal mines in my country has transitioned to an intelligent stage.With the acceleration of the construction of intelligent coal mines,it is of great significance to realize the precise positioning of mine personnel,such as underground mine rescue and dangerous area early warning,to ensure the safety of underground personnel.In order to achieve accurate positioning of underground personnel,this thesis takes smart phones as a platform,makes full use of the underground Wi-Fi base stations and smart phone inertial sensors,and proposes an underground PDR/Wi-Fi multi-source fusion positioning algorithm.The main research contents are as follows:(1)A heading estimation algorithm based on denoising auto-encoder is proposedAiming at the problem that there is a large cumulative error in the calculation of the heading angle of the smart phone gyroscope in the coal mine,this thesis proposed a heading estimation algorithm based on the denoising auto-encoder.It is composed of a denoising auto-encoder for inertial sensor data noise.The noise reduction process is performed on the downhole inertial sensor data by training the denoising auto-encoder,and the Kalman filter is used to integrate the gyroscope integral heading solution and the nineaxis sensor heading solution to obtain The heading angle of the miner’s movement.The motion data of miners was collected in the Gaotouyao Coal Mine in Ordos.The test results show that the algorithm in this thesis has stronger anti-interference ability than the nineaxis sensor heading solution in a complex mine with strong magnetic field,which can meet the PDR heading estimation needs of underground personnel.(2)A PDR algorithm for underground personnel location estimation based on LSTM personalized step size estimation is proposedThis thesis studied the dead reckoning technology for underground workers in coal mines.First,offline training of the personalized step size estimation LSTM model for underground workers.Second,in the online stage,real-time data of miners such as acceleration,gyroscope,and geomagnetism are collected through mining intrinsically safe smartphones.Step detection algorithm and personalized step size estimation model are used to obtain the miner’s movement steps and the step size of each step,combined with the KF-DAE heading estimation algorithm to obtain the heading angle;finally,the current position of the miner is predicted according to the PDR algorithm.Experiments are carried out in the real underground environment.The results show that the PDR algorithm based on LSTM personalized step size estimation proposed in this paper has a detection accuracy of 96.5% and a step size estimation accuracy of 90% in the movement of underground miners,both of which meet the requirements of the PDR positioning algorithm for underground miners accuracy requirements.The relative error of the PDR algorithm based on LSTM personalized step size estimation for the real underground environment is 2.33 %,which effectively improves the positioning accuracy of underground miners and meets the needs of underground positioning and navigation.(3)A classification model based on CNN-LSTM miner behavior recognition is proposedAiming at the problem of unsatisfactory dead reckoning positioning results in single mode due to the changeable movement patterns and postures of miners when using smartphones,a CNN-LSTM classification model was proposed to classify and identify the daily behavior of underground personnel.First,the acceleration and gyroscope inertial data of the miners in the working state are collected through the smartphone,then the data is preprocessed and feature extraction is performed,and finally the CNN-LSTM model is used to classify the daily behavior of underground personnel.The test results show that the CNN-LSTM classification model proposed in this paper has an accuracy rate of 92.56% in the recognition of miners’ behavior patterns in the test set.According to the classification results,the interference behavior of the positioning caused by shaking such as talking while standing still is excluded,and the PDR positioning algorithm is assisted to update the miners’ positions.In this way,the anti-interference and practicability of the PDR positioning algorithm are improved.(4)A PDR/Wi-Fi multi-source fusion localization algorithm based on extended Kalman filter is proposedAiming at the complex underground environment of coal mine,the PDR/Wi-Fi fingerprint positioning algorithm has certain limitations,and a single positioning algorithm cannot achieve high-precision positioning for underground personnel.In this thesis,a PDR/Wi-Fi multi-source fusion positioning algorithm based on extended Kalman filter is proposed.The test results show that PDR/Wi-Fi multi-source fusion positioning algorithm based on extended Kalman filter uses the Wi-Fi fingerprint positioning results to correct the offset of the PDR positioning algorithm.The positioning trajectory basically reflects the real walking route of the tester,and obtains high-precision,stable and continuous positioning results.