Pin-Wise Convolutional Neural Network for MOC 3D Pin Power Prediction

摘要

SymNet which is a variant of ResNet and PinNet were applied to 3D MOC PPPF prediction. From the previous architecture, we added MOC calculation functionality. The results show that it can predict 3D MOC PPPF around 1% error. In order to improve the performance of the final full cycle depletion model, data collection must be improved. Therefore, we will soon incorporate "reinforcement learning". Finally, considering the accuracy and speed, both networks are now ready to be used as an evaluator in the LP optimization process.