Abstract: According to the complex process and low accuracy of goaf area risk discrimination， the improved TOPSIS coupled with neural network was proposed. Firstly， the TOPSIS method was improved and used on goaf area in order to enhance the objection and accuracy of sample goafs’ risk discrimination. Absolute ideal point and improved fuzzy analytic hierarchy process (IFAHP) objective were respectively used to avoid the reverse phenomenon caused by the change of ideal point and weights. And risk grade division was realized through different instability degree interval threshold value of discrimination index. The improved TOPSIS method was used to risk discrimination of 100 groups goaf area samples in a certain mine and results were validated. Then， in order to simplify the discrimination process， the improved TOPSIS and GA-BP neural network were combined effectively. Calculating neural network model was trained by the 100 groups sample data which were discriminated by improved TOPSIS， the 5 groups were discriminated by the model， and the discrimination results agree with the facts. The study results not only enhance the objection of goaf area risk discrimination， but also provide a new thought for simplifying the discrimination process and expand the engineering application in field.

Key words: goaf, risk discrimination, improved TOPSIS, IFAHP(improved fuzzy analytic hierarchy process), GA-BP neural network