Operation information prediction method, model training method and related device

Abstract

The invention discloses an operation information prediction method. The operation information prediction method comprises the following steps: acquiring image data to be predicted; determining N characters to be predicted in a first character set according to image data to be predicted; acquiring a feature set to be predicted of each role to be predicted in the image data to be predicted; and acquiring first operation information corresponding to each role to be predicted through a target joint model, wherein the target joint model is used for generating second operation information accordingto the feature set to be predicted, the target joint model is also used for generating first operation information according to the second operation information, the first operation information represents information related to operation contents, and the second operation information represents information related to the operation intention. The invention also discloses a model training method anda related device. According to the invention, the cooperative capability of the micro-operation level and the macro-operation level can be obtained simultaneously by utilizing the target joint model,so that the prediction capability of the model is enhanced, and the rationality of information prediction is improved.

Description

technical field [0001] The present application relates to the technical field of artificial intelligence, and in particular to a method for predicting operation information, a method for model training, and related devices. Background technique [0002] Artificial intelligence (AI) programs are already beating top professionals at board games with well-defined rules. In contrast, the operations of multiplayer online battle arena (MOBA) games are more complex and closer to real-world scenarios. Overcoming AI problems in MOBA games will help us explore and solve complex problems in the real world. [0003] At present, for multi-person cooperation in MOBA games, multi-agent reinforcement learning methods are usually used to learn cooperation through self-play. see figure 1 , figure 1 It is a schematic diagram of a model structure of multi-agent reinforcement learning in the existing scheme. As shown in the figure, the input of the model is attribute information (such as her...

AI Technical Summary

Problems solved by technology

[0004] However, for MOBA games, the cooperation between multiple people not only needs to be reflected in the micro-operation of objects, but also the cooperation and interaction between different objects is very important in practical applications. Therefore, the multi-intelligence The prediction of operation information by the body reinforcement learning model is not comprehensive, which is not conducive to the rationality of information prediction

Images