Game control method and device and storage medium

Abstract

The invention discloses a game control method and a device and a storage medium, and the method comprises the steps: obtaining a current video frame when the start of a target video game is detected; based on the current video frame and a preset online value network, obtaining a current grey-scale map and a current action, controlling the target video game to execute the current action, obtaining a current reward value and a next video frame, obtaining a current quintuple, and storing the current quintuple in a preset database; when the number of the obtained current quintuple is greater than or equal to the preset sample number and the preset total training round number is greater than zero, performing parameter updating on a preset online value network based on a preset target value network and a preset database; when the number of the current quintuple is an integral multiple of a preset training total step length, replacing a parameter of a preset target value network by using a parameter of a preset online value network, and subtracting one from a preset training total round number; and taking the next video frame as the current video frame, and continuing the process. According to the invention, the event control effect can be improved.

Description

technical field [0001] The invention relates to dynamic optimization technology, in particular to a game control method and device, and a storage medium. Background technique [0002] At present, deep reinforcement learning is a hot field of artificial intelligence research. It combines the perception ability of deep learning and the decision-making ability of reinforcement learning, and provides a new control idea for dynamic optimization related events (such as video games). [0003] Due to the problems of instability and overestimation in deep reinforcement learning, related technologies use an improved algorithm of deep Q network in the control of video games to try to alleviate the overestimation problem of deep reinforcement learning algorithms. However, since the Q-Learning algorithm is an off-strategy algorithm, this itself will increase the instability of the deep reinforcement learning algorithm, and each time a greedy strategy (greedy strategy) is used to select t...

AI Technical Summary

Problems solved by technology

However, since the Q-Learning algorithm is an off-strategy algorithm, this itself will increase the instability of the deep reinforcement learning algorithm, and each time a greedy strategy (greedy strategy) is used to select the actions performed in the video game, each time Both use the Max operator to select the optimal action, but it will amplify the overestimation problem of the deep reinforcement learning algorithm

Therefore, the instability and overestimation problems in deep reinforcement learning still exist, thereby reducing the control effect of events

Images