Eddy current sensors are widely used in railroad detection, but temperature variations can lead to sensitivity and zero drift, thus affecting sensors accuracy. For this reason, a temperature compensation model based on black widow optimization (BWO) -support vector machine (SVM) is proposed. The model overcomes the shortcomings of the traditional model that is easy to fall into local optimum by globally optimizing the parameters of the SVM kernel function and the penalty factor. The data of eddy current displacement sensors at different temperatures are obtained through two-dimensional calibration tests, and the temperature compensation models of BWO-SVM, firefly optimization algorithm-least squares support vector machine (FOA-LSSVM) and improved genetic algorithm-back propagation (GA-BP) are established and compared respectively. The experimental results show that the BWO-SVM temperature compensation model reduces the sensitivity temperature coefficient of the sensor from 1.58×10^-2/℃ to 3.28×10^-4/℃ and the zero temperature coefficient from 1.54×10^-2/℃ to 2.96×10^-4/℃. Compared with the FOA-LSSVM and GA-BP temperature compensation models, the BWO-SVM temperature compensation model shows more robustness and adaptability in complex environments. The research not only provides an effective and intelligent temperature compensation model for the health monitoring and maintenance of railroad infrastructure, but also provides a feasible idea for the other areas of nonlinear temperature compensation scenario.