In order to address the challenges encountered in intelligent ship global path planning using the DQN algorithm, such as paths being planned too close to obstacles, excessive turning points, large turning angles, and slow algorithm convergence, a method based on Noisy DQN (NoisyNet-DQN) for global path planning is proposed. Firstly, to maintain a safe distance between intelligent ships and obstacles, and to reduce path turning points and large turning angles, additional reward functions including heading reward, time reward, turning point reward, and safety reward are incorporated on top of the traditional reward function. Secondly, to tackle the slow convergence issue in complex navigation scenarios, parameter noise is introduced into the output layer of the DQN neural network, thereby enhancing the convergence speed of the DQN network. Simulation studies are conducted in the actual maritime environments of Dalian and Zhoushan. The simulation results indicate that compared to the traditional DQN algorithm, the proposed Noise-DQN algorithm significantly improves the convergence speed and, greatly enhances the safety and economy of the planned global path, better aligning with the actual navigation requirements of ships. The research results can provide a certain reference for global path planning in intelligent ship navigation.