In recent years, patent citations have play a very important role in the process of S&T evaluation, and have attached much attentions. However, as the foundation of patent citation research, what’s determinants of patent citation formation are not solved satisfactorily. With embedded study, scholars have formed an amount of research outputs on patent citation network. These outputs reflected that the formation of patent citation was influenced by the structure characteristics of patent citation network, However, the current framework of existing statistical inference methods based on logistical regression was failing to incorporate the above factors. This article uses an ERG (exponential random graph) model approach, which is a tie-focused branch consists of models that aims to explain or predict ties and their patterns. Under the framework of patent citation formation, attribute-based processes, node-level covariates processes, and self-organizing network processes are incorporated into ERG models. The usefulness of ERG models is illustrated by an empirical study on the structure of wind energy patent citation network originated from the PATSTAT database. The results show that the self-organized level determinants have a more important impact in explaining formation of patent citation, especially “transitivity closure”, and probably the influence of attributes-based factors were overestimated. At last, this paper discusses the direction for the next step work.