In the process of temperature calibration of optical fiber displacement sensor, it is found that with the change of working environment, the measured value of the displacement sensor will deviate, which will reduce the accuracy of the displacement sensor with the change of ambient temperature. In this paper, the radial basis function (RBF) neural network is used to compensate the temperature of displacement sensor, and a self-adaptive design idea is used to find the center of radial basis function in order to reduce the drift deviation. By taking the displacement and ambient temperature as the input and the sensor output voltage as the output, the adaptive design idea is used to determine the center of the basis function, and a model based on RBF neural network is established. The results show that the training results of the model can reduce the relative error of the optical fiber displacement sensor by 9.23%, and the measurement accuracy is improved greatly, which verifies the feasibility of this method.