The present study attempts to promote the quality of differential pressure casting component by two-stage optimization, orthogonal virtual casting and BP neural network. In the first stage, the critical parameters determined by the numerical model of casting procedure indicate that the qualities of castings, including casting solidification time, secondary dendrite spacing and porosity, are highly affected by the die temperature, pouring temperature and cooling medium temperature. In the second stage, the input–output relationship developed by utilizing BP neural network is found to be statistically adequate and yielded better prediction accuracy. Artificial fish swarm algorithm (AFSA) performs multi-directional search in multi-dimensional space simultaneously and performs desirability function approach. The results of nonlinear neural network-based models and the performance of artificial fish swarm algorithm optimization technique are summarized.

本研究试图通过两阶段优化，正交虚拟铸造和BP神经网络来提高压差铸造部件的质量。在第一阶段，由浇铸过程数值模型确定的关键参数表明，铸模质量，浇铸温度和冷却介质温度对铸件质量（包括铸件凝固时间，二次枝晶间距和孔隙率）的影响很大。在第二阶段，发现利用BP神经网络建立的投入产出关系在统计上是适当的，并且产生了更好的预测精度。人工鱼群算法（AFSA）在多维空间中同时执行多方向搜索，并执行期望函数方法。