Path Planning Algorithm For Substation Inspection Robot Based On Reinforcement Learning

The work requirement of the inspection robot is to traverse the detection points without collision in the whole process,which relies on effective and reasonable path planning.Existing power station inspection robots rely too much on the accuracy of the environment map.When the environment changes,they need to be remodeled and recalibrated,which makes the subsequent maintenance work arduous.Therefore,this paper proposes a method that does not require environmental modeling.This method uses a reinforcement learning control strategy to enable the inspection robot to complete the path planning tasks of avoiding obstacles and traversing the detection points without collisions when the scene of the substation changes.,The specific work is as follows:(1)Using neural networks for a large amount of training can find the characteristics of the logical relationship between input and output.The distance information detected by the sensor,the signal reached by the detection point and the historical action of the robot are used as the input of the neural network.The reward function induces the desired neural network action output,and then trains the induced output and input.After the training is completed,the logical relationship is retained by fixing the neural network weight to guide the inspection robot to complete the inspection based on the acquired environmental information.Check the target.This method enables the robot to complete the path planning task of no collision and detection point traversal without the need of environment modeling.(2)Substation inspection needs to realize 28 types of equipment inspection points traversal on the basis of ordinary path planning.In order to enable the inspection robot to realize the collision-free inspection point traversal path planning in the whole process with the optimal path,a reward function with a number of scoring rules is designed.For the inspection task of the substation,there is a tendency for collisions and the tendency to approach new inspection points.Various possible situations such as collision,arrival at new detection point,traversal detection point,detour,timeout,etc.,were evaluated in detail.(3)The convergence and generalization analysis of reinforcement learning was carried out on the Open AI Gym platform.The convergence analysis proved that the algorithm in this paper enables the inspection robot to achieve the task of traversing detection points without collision as the number of reinforcement learning increases.The goal and generalization analysis proves that the algorithm in this paper can also make the inspection robot respond to changes in the substation scene and still have the ability to traverse detection points without collision.