<p>Early and accurate predictions of technology fusion can help explore technological innovation opportunities and play a significant role in promoting industrial change and development. However, technology fusion prediction suffers from insufficient data mining and poor prediction accuracy. To improve the accuracy of technology fusion prediction, this paper proposes a method based on patent multidimensional information extraction and deep neural networks. The proposed method first constructs an international patent classification (IPC) co-occurrence network based on existing technology fusion and then transforms the technology fusion prediction problem into a patent IPC co-occurrence network link prediction problem. Second, a fusion knowledge graph based on patent data is constructed and the technical characteristic information in the patent data, semantic information in the patent text, and link information in the IPC co-occurrence network are deeply mined. Third, deep neural networks are used to learn the features of historical data and train the technology fusion prediction model to achieve fine-grained and accurate predictions of future industrial technology fusion. An experimental study was conducted by evaluating the artificial intelligence and green technology of the high-speed rail industry, and the results demonstrate that all the indices of the proposed method are better than those of other comparative models, allowing us to realize the accurate prediction of the fusion trend of artificial intelligence and green technology in the high-speed rail industry. This study also has practical significance for the prediction of technological fusion and innovation in other industries.</p>