Based on the boundary binary sampling information of the heat equation, a support vector regression machine(SVR) algorithm is proposed. Given the problem that the SVR algorithm cannot actively select the optimal parameters, we use the particle swarm optimization (PSO) algorithm to optimize it. Besides, an improved PSO-SVR model is constructed to estimate the actual boundary value of the heat equation. The results show that the improved PSO-SVR model can effectively predict the boundary value of the heat equation. Also, the forecast accuracy is excellent, and the data collected can be used for more in-depth research.