In the context of digital transformation, in order to explore the optimization path of AI technology empowering financial market risk management, solve the problems of improving accuracy and model interpretability of machine learning in stock price prediction, and achieve better financial market risk control. This article proposes a Full-Scale Channel Simple Graph Convolutional Transformer Prediction Model (FSC-SGC Transformer), which combines SGCT blocks with the advantages of SGC and Transformer to mine channel relationships; Realize multi-scale feature fusion through Lightweight Channel Cross Fusion; Utilize channel cross attention to fuse inconsistent features. Construct a multi-objective portfolio optimization model (M-SV-S), considering factors such as transaction costs, and solve it using differential evolution algorithm. The experiment uses a “rolling window” to simulate investment, calculates indicators such as cumulative return rate and annual Sharpe ratio, and compares them with the proportional model and Shanghai Securities Composite Index. The results show that under different transaction costs and upper bound constraints, the cumulative rate of return of the proposed model is significantly higher than that of the comparative model, and the overall level of annual Sharpe ratio is higher. On the research surface, the proposed model performs better in financial market risk management, effectively improving investment returns and reducing risks.
