Based on the research on the dependence of the components in the deep Web integration (executive partial order and knowledge dependency), a knowledge-based method is given to process the changes in such integration, which includes environmental changes processing model, a self-adaptive software architecture and algorithm. This method can provide a reference to the further research or toward application for the large-scale deep Web integration. The experimental results show that the method can not only process the changes, but also highly improve the performance of the integrated system.