Aimed at the problems such as pattern recognition, cluster analysis, effective use of unlabeled samples, etc. in the time-varying space, a two-dimensional self-organizing process neural network with self-adaptive structure based on semisupervised learning is proposed. By building the generalized Fréchet distance which is used to measure the similarity among time-varying function samples, using the class information and process features of partial labeled dynamic samples, adopting reward-punishment update rule, and according to the network learning objective function, the dynamic reconstruction of the network structure is realized with splitting and mergering competitive nodes in two-dimensional plane layer, and then effective clustering is implemented. Experimental results verify the effectiveness of the proposed model and algorithm.