A genetic algorithm and backpropagation neural network based temperature compensation method of spin-exchange relaxation-free co-magnetometer

This paper presents a temperature compensation method based on the genetic algorithm (GA) and backpropagation (BP) neural network to reduce the temperature induced error of the spin-exchange relaxation-free (SERF) co-magnetometer. The fluctuation of the cell temperature results in the variation of the optical rotation angle and the probe light absorption. The temperature fluctuation of the magnetic field shielding layer induces the variation of the magnetic field. In addition, one of the causes of light power variation is temperature fluctuation of the optical element. In summary, temperature fluctuations cause a variety of SERF co-magnetometer errors, and the relationship between these errors and temperature fluctuations has the characteristics of time-variance and non-linearity. There are two kinds of methods to suppress these errors. One way is to reduce temperature fluctuations of the SERF co-magnetometer. However, this method requires additional hardware and high cost, which are not suitable for miniaturization and low cost applications. Another effective method to suppress nonlinear and time-varying errors is to utilize intelligent algorithms for temperature compensation. In this paper, the BP neural network is applied for temperature compensation, and the GA is utilized to overcome the disadvantages of the BP neural network. The training data were obtained by changing the ambient temperature of the SERF co-magnetometer. The experimental results show that the method proposed in this work can significantly improve the accuracy of the co-magnetometer at complex ambient temperatures, and the stability of the SERF co-magnetometer at room temperature can be improved by at least 45%.