Application of improved BP neural network in predicting subsidence of buildings

With the increase of engineering construction, the safety of engineering became a realistic problem of great importance. As a major technical means, prediction of deformation monitoring was extremely important for construction and operation of buildings. In this paper, an improved BP neural network was adopted and a nonlinear coefficient was added in the response function of the neurons to enhance the approximation performance of the network. The cumulative subsidence of the 40th to the 43rd interval was predicted and compared with the measured data through training the subsidence monitoring data of total 35 intervals of 14 monitoring points of presupposition around Changchun Science & Technology Museum as sample data. The 1-5-4 network model based on improved BP network reached error demands by 4428 times training. The result showed that the prediction errors of 11 monitoring points were less than 0.5mm and the model performed well.