41 / 2023-03-30 09:30:20

A robust positioning method for personnel with UWB in the mine based on CNN-LSTM

Ultra Wide Band; Location; CNN-LSTM neural network; Polynomial function; Exponential function

Abstract: In view of the complex geological environment of the mine, the dense personnel and electromechanical equipment, there are serious safety hazards. This paper proposes a UWB mine positioning algorithm based on CNN-LSTM. Its idea is to take advantage of CNN's short-sequence feature abstraction ability, and then use LSTM to combine short-sequence high-dimensional features for prediction. This algorithm is suitable for processing data with local correlation . In this paper, the ultra-wideband system is used to collect mine positioning data, including the ranging value between the base station and the tag, and the initial coordinates of the tag solved by the Chan algorithm are used as the input of the CNN-LSTM network for training to reduce the impact of the ranging error on the positioning accuracy. Then using the Gauss-Newton iterative algorithm to solve the final positioning result. At the same time, a polynomial function model and an exponential function model taking into account ranging error modeling are established, and a comparative analysis is carried out. The static and dynamic two actual measurement experiments are compared; the experimental results show that in the static positioning, the four test label points T1, T2, T3, and T4 have ranging errors of 0-30cm under the four models, of which The ranging error based on the CNN-LSTM model is the smallest, and the ranging error is basically less than 5cm; the ranging error based on the PFREC model and the EFREC model is next, and the ranging error ranges from 5-10cm. The RMSE is also significantly reduced. Compared with the real value, after the CNN-LSTM model correction, the positioning accuracy of T1, T2, T3, and T4 points increased by 87%, 72%, 80%, and 82% respectively; after the PFREC model correction, The positioning accuracy of T1, T2, T3, and T4 points increased by 77%, 66%, 70%, and 74% respectively; after the correction of the EFREC model, the positioning accuracy of T1, T2, T3, and T4 points increased by 81%, 66%, 67%, and 71% respectively. In the dynamic experiment, the accuracy of the results based on the CNN-LSTM network model is better than that of the other two models, which shows that the algorithm effectively reduces the ranging error and improves the mine positioning accuracy of UWB. The UWB mine positioning technology based on CNN-LSTM has basically met the original design intention after testing, effectively completed the mine positioning work, and can meet the daily production needs of coal mines. Ultra-wideband coal mine positioning technology is necessary and has market demand.