Brain networks have been widely used in the field of neuroimaging. In recent years, high order functional connection networks and hypernetworks have made great strides in brain diseases diagnosis. However, both types of networks have their owe disadvantages: high order functional connectivity networks consider the time varying characteristic of the network but does not deal with the problem of special multiple interactions in the network; hypernetworks can characterize relationship between multiple brain regions but the dynamic characteristics of the network are not accounted for. To slove the above problems, the characteristics of high order functional connection networks and hypernetworks was integrated, and a method for constructing brain function hypernetworks with time-varying characteristics was proposed. This method takes into account the time dynamics of the brain network, and builds the brain function hypernetwork based on the LASSO method, and applies the network to the brain diseases diagnosis. This network canceled out the disadvantages of the above two networks, and the results show that the classification accuracy of this method reaches 86.36%, which is significantly higher than the previously proposed method. It can effectively improve the classification performance of depression and has important theoretical significance and clinical value.