Abstract: Distributed photovoltaic systems are widely distributed, with high local penetration rates and complex and ever-changing installation environments. Reliable measurement data is the basis for performance analysis, output prediction, and operation and maintenance control. However, factors such as sensor failures and communication blockages can lead to missing measurement values, deteriorating the quality of raw data, and thus affecting the accuracy of distribution network operation decision-making. Traditional data repair methods only consider the distribution characteristics of a single measurement value, ignoring the coupling relationship of multidimensional time series data, resulting in low repair accuracy. A coupled data enhancement method based on a bidirectional multi-stage recurrent imputation network is proposed to address this issue. Experimental results demonstrate that the proposed method exhibits good repair performance even under high levels of missing data, effectively enhancing the quality of fundamental data for distributed photovoltaic clusters, and improving the fine-grained perception capability of grid operators towards photovoltaic clusters.