基于lookahead的交互式动态影响图的DMU改进算法

区别模型更新(discriminative model update，DMU)是一种常用的求解交互式动态影响图(interactive dynamic influence diagrams, I-DIDs)问题的算法。结合lookahead思想提出了一种判断模型近似行为等价的改进DMU方法。所提方法首先将满足近似行为等价的模型聚类形成代表模型集合，然后自上而下对代表模型进行更新，在模型更新过程中，只更新那些与其他模型预测行为不同的模型。结合lookahead思想提出了一种判断模型近似行为等价的方法。与DMU算法相比，该算法能迅速有效地减少模型的数量，从而减少了计算机的存储空间和运行时间，提高了算法的效率。最后通过对多Agent老虎问题及机器维修问题实验来验证所提方法的有效性。

Improved look ahead based DMU algorithm for interactive dynamic influence diagrams

The discriminative model update (DMU) is a common algorithm for solving interactive dynamic influence diagrams (I-DIDs). The look ahead method is used to give an improved discriminative model update algorithm which determines approximate behavior equivalence. Firstly, the models that are approximately behavior equivalent are clustered into a representative model set. Then the models within the representative model set are updated from top to bottom. In the updating process, only the models whose predictive behavior is different from others are updated. Compared with the DMU algorithm, the proposed algorithm can quickly and effectively reduce the model’s number, thus reducing the storage space and the running time of the computer, and improving the efficiency of the algorithm. The effectiveness of the proposed method is verified through experiments on the multi agent tiger and multi agent machine maintenance problems.