Abstract: The presence of numerous stochastic elements in distributed energy resources (DERs) leads to frequent changes in Multi-Virtual Power Plant (MVPP) when it comes to the strategy of individual VPPs. For a given entity, understanding the trend of the impact on its own returns when perceiving sudden changes in the strategies of other entities and rapidly adjusting its own optimization strategy is a critical issue that urgently needs to be addressed. This paper proposes a self-trending optimization strategy for MVPPs based on second-order stochastic dynamics, aiming to enhance the autonomy of VPPs in responding to changes in the strategies of other entities. First, addressing the heterogeneous operational characteristics of DERs, the paper focuses on the adjustable space of resources to construct a clustered operational model for VPP resources. Next, the stochastic nature of VPP strategy transitions is depicted based on the theory of random graphs. Then, second-order stochastic dynamic equations are used to explore its spontaneous evolutionary information to adjust the comprehensive profit of VPPs with the change of other entities' strategies. Moreover, the adjusted profit is used as the true reward function for the Integrated Soft Actor–Critic (ISAC) deep reinforcement learning decision model to establish a multi-agent distributed solution framework. Finally, multiple algorithm comparison experiments are designed to validate the self-trending performance of the proposed strategy in this paper.