Enhanced PCI allocation in heterogeneous networks: A deep reinforcement learning approach with dynamic adjustments