Since market data is intrinsically noisy and nonlinear, it remains a big challenge to make an accurate prediction on financial time series. To improve the predictive performance of financial prediction, we adopt a hybrid model, which is the combination of PSO and MLP. To accelerate convergence and avoid trapping in a local optimum during training, PSO is employed to optimize the weight parameters and network structure of the MLP. The PSO-MLP model is contrasted with a conventional MLP and verified on real stock price data. Experimental results show that the PSO-optimized MLP learns considerably better, enhancing the R2 score from 0.9901 to 0.992 and reducing the values of Mean Absolute Error (MAE) from 1.9915 to 1.1581 and of Mean Squared Error (MSE) from 5.9062 to 1.8469. Furthermore, the MAPE value also drops from 272.20% to 78.32%, indicating that the improvement in relative accuracy is remarkable. Table 3 shows that the hybrid deep learning-swarm intelligence model can serve as a basis for enhanced financial time series prediction, opening a path for more accurate predictive models for the financial domain.