The formation of financial risk involves the interaction of many factors, showing a highly nonlinear characteristic relationship. In this study, an intelligent financial risk assessment system based on deep learning (DL) is constructed to solve the key problems of the traditional pre-alarm mechanism, such as insufficient identification accuracy and poor timeliness. This model gives full play to the advantages of Convolutional Neural Network (CNN) in complex feature extraction, and automatically captures the deep-seated related features in financial data through end-to-end learning. The research shows that, compared with traditional methods, this system shows excellent prediction performance on three independent data sets, while maintaining high operational efficiency. This innovative scheme, which integrates the functions of intelligent pre-alarm and dynamic assessment, provides an accurate and efficient decision support tool for modern enterprise risk management. The theoretical contribution and practical value of this study are mainly reflected in: first, a new paradigm of financial risk analysis is put forward; second, an intelligent risk control system with practical application potential is developed.