Search

Improving the positioning accuracy of underground personnel can not only strengthen mine safety monitoring, but also increase the speed of rescue, thus ensuring the life safety of underground personnel to the maximum extent. This paper proposes a positioning model based on IPSO-LSTM for position prediction of underground moving targets in response to the problem of existing ranging algorithms which are affected by the on-site environment, resulting in insufficient positioning accuracy. This article uses LSTM to build a fingerprint positioning model. It collects distance information through the UWB wireless module to build a distance-position fingerprint relationship database, which is used to train the PSO-LSTM model. Then we use the trained model to predict target trajectories. We compared four improvement strategies on PSO including random initialization of population position by chaotic mapping, nonlinear inertia weight reduction and fitness function optimization. Experiments show that the improved PSO optimization algorithm in this paper exhibit fast convergence speed and good robustness. In order to verify the positioning effect of IPSO-LSTM, we compared the IPSO-LSTM model with the Chan algorithm, PSO-LSTM model, LSTM neural network, SSA-LSTM model and GWO-LSTM. The average positioning error is used as the evaluation index. The results show that the average positioning error of the IPSO-LSTM positioning model proposed in this study is 30mm, which is 76% higher than the traditional Chan algorithm, 49% higher than the LSTM, and 24% higher than the PSO-LSTM model. In order to reduce large local errors, we used median filtering to process input information, further improving positioning accuracy. This study offers references for improving the accuracy and stability of the existing underground moving target positioning system. [ABSTRACT FROM AUTHOR]