Data-driven power flow calculation has emerged as an effective approach for real-time analysis of massive scenarios under the high uncertainty of renewable energy. However

existing data-driven power flow models suffer from error amplification in branch power computation and lack effective trustworthiness assessment

limiting their applicability in practical engineering scenarios. To address this

this paper focuses on improving both the accuracy and trustworthiness of data-driven power flow methods and proposes a multiple task data-driven power flow framework equipped with trustworthiness self-validation. The proposed framework introduces a topology-aware graph Transformer (TAGT) module tailored for power flow tasks

enabling effective feature extraction under varying topologies. This study designs a parallel multi-task architecture containing bus voltage and branch power prediction to avoid error propagation in branch power estimation. In addition

it also constructs a self-validation mechanism based on power balance and angle inversion to enhance the physical consistency and trustworthiness of model outputs. Extensive training and testing are conducted on the IEEE 39-bus system and a provincial-level 333-bus transmission network. Through comparative experiments

model scaling tests

and attention visualization analyses

the proposed method is comprehensively validated in terms of accuracy

scalability

and trustworthiness.