Game software automatic testing method based on multi-objective optimization and deep reinforcement learning

Abstract

The invention discloses a game software automatic testing method based on multi-objective optimization and deep reinforcement learning. The method comprises the following steps: 1, constructing an anomaly detection mechanism oriented to a game scene and an evaluation index of a game test strategy; 2, carrying out game software automatic test design; and 3, based on a multi-objective optimization algorithm, according to the fitness value FitnesValue (pi) of the strategy pi, measuring the performance of the strategy, selecting high-quality offspring, and further improving the efficiency and effect of game testing. Each strategy in the strategy group has two performance indexes including the winning rate and the exploration capability; and based on a measurement result, a test strategy on a Pareto optimal plane is searched as an excellent test strategy to be reserved, and meanwhile, strategies which cannot be expressed on the two optimization targets are eliminated, so that more effective test strategy optimization is realized. Compared with the prior art, the method effectively improves the game test efficiency and effectiveness, and has great practical application significance.

Description

technical field [0001] The invention relates to the technical field of reinforcement learning and software testing, in particular to a software automatic testing method combining a multi-objective evolutionary algorithm and a reinforcement learning method. Background technique [0002] Game testing has long been considered a challenging task. In the industry, game testing generally uses a combination of script testing and manual testing. Today, the research on automated game testing is still in its infancy. One of the main reasons is that playing the game itself is a continuous decision-making process, and game defects (Bugs) are often hidden deep, only when some difficult intermediate tasks are completed. , it is possible to be triggered, which requires the game testing algorithm to have human-like intelligence. In recent years, the extraordinary success of Deep Reinforcement Learning (DRL), especially in the field of game control, has even shown intelligence beyond human...

AI Technical Summary

Problems solved by technology

However, the existing DRL algorithm mainly focuses on how to win the game, rather than game testing, so it may not be able to widely cover the branching scenarios that need to be tested

[0003] On the other hand, the existing game software testing mainly relies on manual writing of test scripts, which not only requires a large cost of manpower and time, but also has certain limitations in the understanding of the game software itself by the engineers who write the scripts, resulting in the It is not possible to comprehensively test the game software, so there are still certain hidden dangers when the software goes online

In addition, scripts written by humans contain a wealth of human prior knowledge, which leads to a large amount of labor costs in current game testing. Although a method similar to fuzzing (Fuzzing) has been proposed for testing game software, it lacks With effective prior knowledge guidance, there is still a big gap between its effect and that of manually written scripts.

[0004] To sum up, existing game testing technologies either rely heavily on manual writing, cannot be fully automated, and have low testing efficiency; or use heuristic testing algorithms, which have large randomness, and the algorithm effect cannot be guaranteed. For modern game software, It is difficult to guarantee the validity of the test

Images