| With the continuous development of truck cranes,its applications in engineering projects are also increasing.Since the crane needs human control and decision-making during the hoisting process,it is inevitable that there will be operational errors or unreasonable plan planning.In response to the above problems,this paper proposes a multi-task path planning algorithm and a hoisting planning algorithm to assist hoisting personnel in hoisting operations.At the current stage,the truck crane still needs manual operation when hoisting,but in many other fields,deep reinforcement learning has been used to realize unmanned operation.However,as a major national heavy equipment,the hoisting of truck cranes is also developing in the direction of intelligence.Applying it to the hoisting plan of truck cranes has good significance for accelerating the pace of intelligent cranes.In the path planning of truck cranes,the traditional path planning method will be more efficient.For the crane’s action sequence planning,the reinforcement learning method is used to better fit the crane’s hoisting action sequence planning.Because the output of the Depth Deterministic Strategy Gradient Algorithm(DDPG)is a continuous action,and the process of hoisting the target by the crane is also a collection of a series of continuous actions,so there is theoretical feasibility.Therefore,this paper uses traditional path planning algorithms and deep reinforcement learning to carry out the following tasks:(1)Firstly,this paper analyzes the necessity and feasibility of the multi-task path planning algorithm for truck cranes.When designing a multi-task path planning algorithm,it is necessary to plan the traversal sequence and path between multiple tasks,and choose to plan the traversal sequence of the target point with the optimization algorithm.The main analysis is the ant colony algorithm,genetic algorithm and particle swarm algorithm.The genetic algorithm with the best performance is used as the optimization algorithm.By combining the JPS algorithm in the single-task path planning algorithm with the genetic algorithm,the multi-task path planning of the truck crane is realized;(2)Secondly,this paper studies the action sequence of a truck crane during hoisting,and designs a truck crane action sequence planning algorithm based on the improved DDPG algorithm.First,the hoisting environment of the crane is designed,the state and action model of the crane are established,and the reward function of the action sequence is designed.Then based on the hoisting environment of the truck crane,the network framework,hyperparameters and activation functions in the DDPG algorithm are designed and optimized.Based on the characteristics of the DDPG algorithm that generates samples during offline training,the sampling function in the DDPG algorithm is improved.The improved DDPG algorithm has better convergence effect.The improved DDPG algorithm is integrated with the no-load hoisting environment and the loaded hoisting environment of the truck crane,and the design of the truck crane action sequence planning algorithm is completed;(3)Finally,the multi-task path planning algorithm and the action sequence planning algorithm are combined to realize the multi-task hoisting planning of the truck crane.And the wind power plant and chemical plant are used as virtual environment to verify,the result shows the feasibility of the two algorithms in planning the multi-task hoisting scheme of truck crane. |
