Aiming at the problem of UAV path planning, a 3D path planning method is proposed based on the parallel neural network structure and the dynamic adjustable step size strategy.Firstly, different strategies are adopted according to the distance between the UAV and the threat.When the UAV is outside the risk area, the method of taking large steps is adopted to achieve the purpose of rapid generation of the path.When the UAV is inside the risk area, the method of taking the adjustable step is used to achieve the fine search of the path.Then, the neural network of the obstacles penalty function and the energy function of the path are constructed.By combining the gradient descent method with the Newton Downhill Method, the motion equation of the path is established.According to different path points, learning rates with different adaptive learning factors are used to realize rapid escaping from the threats.The simulation results show that the proposed algorithm not only guarantees the safety of UAV to bypass the threat, but also improves the convergence speed of the algorithm.