Parameter auto-selection for hemispherical resonator gyroscope's long-term prediction model based on cooperative game theory

As a new vibration gyro with features of high accuracy, long lifespan, no wear-out, and great reliability, the hemispherical resonator gyroscope's (HRG's) lifespan prediction without whole lifetime test is a tough task. Dai et al, based on data driven, proposed a residual modified autoregressive grey model ARGM to predict HRG's lifespan, in which the parameters however are selected by expert experience. In order to enhance the predictive lifetime, we propose a novel approach to auto-select parameters for the multi-parametric long-term prediction model ARGM based on cooperative game theory that we call CoG-ARGM. Our idea is to map parameter auto-selection of the prediction model to coalition formation in a combined cooperative game, which is proofed convex, where each parameter is respectively considered as a sub-coalition in its own pure cooperative game. In addition, we also bring failure mode originally derived from FMEA to evaluate the real-time prediction reliability. The experiments indicate that CoG-ARGM with real-time reliability evaluation yields high-quality prediction results. Furthermore, we also demonstrate the superiority of CoG-ARGM over state-of-the-art prediction methods through detailed experiments using evaluation criteria such as MAPE, Ln(Q) and time consumption on real HRG drift data.