During the construction of non-symmetric double-arch rock highway tunnel, the complicated geological condition may affect the safety of the constructor and the engineering quality. In this paper, two treatment methods are put forward. At first, the site monitoring of surrounding rock displacement must be carried out, then, BP neural network is applied in predicting the displacement of surrounding rock based on the learning sample of measured value, so the stability of surrounding rocks may be analyzed and forecasted. During the analysis of BP neural network, the effects of joint and fracture of surrounding rock on displacement can be comprehensively considered, comparing the predicted values of displacement with those by FEM. The results show that not only the predicted error of BP neural network is relative small, but the predicted values of surrounding rock displacement are close to measured ones. So, the predicted values of BP neural network are reliable and may guide the engineering construction in site.