<p>Despite the rapid progress of network modeling, understanding, controlling and predicting structures and dynamics of temporal network models are still lagging. Although activity driven network models as the significant paradigm to construct temporal networks have made substantial advertisements in the past decade, an unanswered fundamental question is how to break through the limitations imposed by the homogeneous connectivity patterns on the structural relevance of network models and realistic networks, especially in the higher-order perspective. Moreover, heterogeneous connectivity patterns can be more wisely applied to real-world networked systems such as social networks and biological networks. Here we address this fundamental question by modeling a Heterogeneous Probabilistic Activity Driven (HPAD) network model and a framework for predicting and controlling the evolution adjacency of heterogeneous connectivity patterns generated by the HPAD network model. We then develop the corresponding epidemic threshold framework to discuss the dynamical processes on HPAD network model. Extensive numerical experiments validate the effectiveness and practicality of the framework, and we have discovered some meaningful results. Firstly, the clustering-based optimization algorithm not only dramatically improves the feasibility and efficiency of the framework, but also has great potential to be applied to various control problems in network science, particularly those related to large-scale networks. More importantly, we find that the distinct evolution adjacency of heterogeneous connectivity patterns markedly affect dynamical processes of HPAD network model, which is significant for controlling and predicting norms like disease and fake news.</p>