With the growing complexity of integrated circuits and heterogeneous microsystem integration, effective thermal management and intelligent design are crucial for preventing thermal failures. This study proposes an adaptive deep reinforcement learning-based (ADRL) optimization method for pin-fin microchannel heat sinks, enhancing iterative optimization efficiency with a hybrid neural network (HNN).