Neural-Network-Based Temperature Drift Compensation for MEMS-IMU Systems

This study addresses the issue of temperature-induced variations in the output data of miniature inertial measurement units (MIMUs) caused by material properties and manufacturing inconsistencies. To mitigate this problem, a temperature drift compensation method based on an improved gated recurrent unit (GRU) neural network is proposed. Specifically, a GRU-based temperature compensation model optimized using the black-winged kite algorithm (BKA) is introduced. To validate the effectiveness of the BKA-GRU model, simulation experiments were conducted using MIMU data across the full temperature range, and the compensation results were compared with those obtained from the GRU and optimized monarch butterfly algorithm-GRU (OMBA-GRU). Experimental results demonstrate that the BKA-GRU model achieves higher accuracy and greater stability compared with the other two temperature compensation models.