Research of neural network-based model for nonlinear temperature drift compensation of MEMS accelerometers.

Downhole instrumentation requires more and more accuracy of MEMS inertial sensors. However, in measurement while drilling, temperature drift phenomenon of the sensor will have a cumulative impact on the drill pipe attitude solution. After experimental testing, the output response of the accelerometer had strong local linear and global nonlinear characteristics. In this paper, we proposed a temperature compensation model based on tent chaotic mapping and sparrow search algorithm optimized back propagation (BP) neural network (Tent-SSA-BPNN). Sparrow search algorithm (SSA) was optimized by tent chaotic mapping, which was utilized to improve the uniformity and search ability of SSA populations. Then, the improved SSA was used to optimize the weight and bias parameters of the BP neural network for constructing the temperature compensation model. Finally, the trained compensation model is integrated into the microprogram control unit for real-time compensation testing. The experimental results show that after sacrificing a small amount of sampling frequency, the compensation model proposed in this article has good global compensation performance, and the mean absolute percentage error is reduced from 2% to 0.2% compared to the original output. The mean absolute error and root mean square error of the improved compensation model are all reduced compared to the pre-improved BP compensation model. This temperature-compensated modeling method has a reference value for low-cost and high-precision modeling in high temperature environments, while greatly saving time cost and measurement costs.