目前国内出现了一种新型云轨, 云轨有着造价低、能耗小以及施工周期较短等优点。然而云轨的各项指标要求很高, 其中轨道定位尤为重要。为了实现了云轨检测的精确定位, 本文设计了一种新型轨道检测车, 并开发了基于双层双向长短期记忆模型(LSTM)的云轨SIN-GPS定位算法。首先, 介绍了轨道检测车的机械结构和各项传感器参数。接着, 分析了传统的SIN-GPS定位算法及其缺点, 在GPS信号消失后会出现误差积累。然后, 引出双层双向长短期记忆模型, 说明了该模型对GPS信号消失时的误差动态学习和补偿。最后, 通过3组实验分析算法在云轨检测车的不同运动状态下的准确率。证明了长短期记忆模型均优于传统算法模型和其他智能算法模型。实验结果表明: 在运动状态下LSTM算法比SINS误差小79.8%, 静止状态下SINS误差最小。设定速度阈值为0.2 m/s, 大于此阈值采用LSTM算法, 小于此阈值直接用SINS的数据, 可以得到最准确的位置预测结果。

At present, a new type of cloud track has emerged in China. Cloud track has the advantages of low cost, low energy consumption and short construction period. This new type of cloud track requires precise positioning of track detection. In order to eliminate the position error of cloud track detection, a new type of track detection vehicle was designed, and a new SIN-GPS positioning algorithm based on double-layer bidirectional LSTM network was developed. Firstly, the construction and sensor parameters of the track detection vehicle was inod uced. Then, the traditional SIN-GPS positioning algorithm and its shortcomings was analyzed. If GPS signal disappeared, the positioning error was very large. a bi-directional LSTM algorithm was proposed to illustrate the dynamic learning and compensation of errors when GPS signals disappeared. Finally, the accuracy of the algorithm in different motion states of the cloud track detection vehicle with three groups of experiments was analyzed. The results of experiments show that LSTM algorithm is superior to traditional algorithms and other intelligent algorithms. It reveals that the error of LSTM is 79.8% smaller than that of SINS when the vehicle is moving. The error of SINS is smallest when the vehicle is static. When setting the speed threshold of 0.2 m/s, using LSTM algorithm when it is larger than this threshold, and directly using SINS when it is smaller, the most accurate location results can be obtained.