A method for solving the shortest path problem based on an improved Q-learning algorithm

Abstract

The invention discloses a method for solving the shortest path problem based on an improved Q-learning algorithm. The method specifically comprises the following steps of: establishing a position-action value table, completely initializing the Q table to be 0, and selecting an appropriate updating step length, greedy parameters, reward discount and smooth parameters; setting an initial position tobe recorded as x0 and a target position to be recorded as xtf according to the requirement of the shortest path problem; The current position is marked as X, the initial position is assigned to the current position, and epson-is used for marking the current position as X; selecting an action according to a greedy method, and recording the action as A; Taking an action A at the X position, recording the obtained earnings r and the position X'transferred after the action A is taken, if X 'is equal to the target position xtf, assigning the initial position x0 to X, and otherwise assigning the X'to X; Updating the Q table; All positions adjacent to the position X are marked as X1, X2,..., Xn, and a Q table is updated; And repeating the above steps until the Q table converges. According to the invention, the existing Q-use at present is relieved; And the problem that the Q table convergence time is long when the problem of the shortest path is solved by the leader is solved.

Description

technical field [0001] The invention relates to a method for solving the shortest path problem based on an improved Q-learning algorithm, and belongs to the technical field of reinforcement learning. Background technique [0002] The shortest path problem widely exists in navigation, urban planning, video games and other fields. Traditional solutions to shortest path problems generally rely on shortest path algorithms, such as Floyd's algorithm, Dijkstra's algorithm, or A * Algorithms and other search algorithms. In recent years, with the rise of reinforcement learning, the use of reinforcement learning in the shortest path problem has become more and more extensive. The Q-Learning algorithm is a classic reinforcement learning algorithm, which was first proposed in the early 1990s. The earliest Q-Learning is based on the Q table, that is, a table that records any action taken in any state. This value is generally recorded as Q(x,a), where x represents the state and a repr...

AI Technical Summary

Problems solved by technology

[0004] The technical problem to be solved by the present invention is to provide a solution to the shortest path problem based on the improved Q-learning algorithm, which solves the problem that the Q table is difficult to converge in the process of using the Q-learning algorithm to find the shortest path

Images