Dynamic Financial Risk Monitoring Using Deep Reinforcement Learning with Temporal Attention and Hypergraph Networks

Existing monitoring methods rely on static data and are difficult to cope with the dynamic changes and uncertainties of the financial market, especially in the face of sudden and nonlinear events. To solve the problem of financial risk, this paper applies deep reinforcement learning to build an intelligent dynamic decision-making model. A large amount of risk data is collected through the financial market data interface, and the variables are simplified by principal component analysis to establish core risk indicators. A deep Q-network (DQN) model is constructed, and convolution long short-term memory (ConvLSTM) and hypergraph attention network are integrated to capture the complex relationship between temporal dependencies and risks. Long short-term memory network and temporal attention mechanism are used to capture market preferences, and reward functions and optimization algorithms are designed accordingly. End-to-end policy gradient optimization is used to train the model, and real-time data feedback is applied to adjust internal parameters and structure, so that the model can quickly adapt to market changes. Experiments show that in 10 market conditions, the average accuracy and recall of the DQN-based model are 94.25% and 91.38%, respectively, and the average response time and standard deviation are 1.33 s and 0.064, respectively; the average accuracy and recall of the Graph Neural Network (GNN) model are 82.23% and 75.51%, respectively, and the average response time and standard deviation are 2.84[Formula: see text]s and 0.260, respectively. The accuracy, real-time performance, and stability of the proposed model are significantly better than those of the GNN model.