Rapid fluctuations of large-scale distributed generation (DG) can easily cause issues such as voltage violation in distribution networks, affecting their safe and stable operation. Additionally, distribution clusters may belong to different stakeholders, making inter-cluster data privacy increasingly important. To address the problem of DG cluster control under multiple stakeholders, an adaptive and trustworthy collaborative decision-making method that considers data privacy protection is proposed. First, a multi-cluster trusted collaborative framework for distribution networks is constructed based on split federated learning approach, and a deep learning model is established for voltage decision-making. This framework enables data fusion and collaboration among multiple clusters with privacy protection between stakeholders. Then, a reward mechanism is incorporated into the deep learning model to evaluate the quality of measurement data, allowing for adaptive model updates. Finally, the feasibility and effectiveness of the proposed method are verified using a case study of the Jiaomen distribution network in Guangzhou. The results demonstrate that the proposed method offers strong privacy protection capabilities, improves voltage quality, and effectively mitigates voltage violation issues in the distribution network.