Multi-sensor Data Fusion Method Based on ARIMA-LightGBM for AGV Positioning

This paper proposes a novel AGV multi-sensor data fusion method. First, for the problem of multi-sensor error changing with time, use ARIMA time series to process the positioning error data, and establish a time series model of each sensor error as a new feature input of the subsequent regression model. Then, the widely used machine learning regression model LightGBM is used to fuse the positioning data, train two LightGBM models for the coordinate data in two directions, and output the positioning results at the same time. Finally, through comparison experiments, the positioning accuracy can reach 6.5mm, which verifies the superiority of this method.