With the continuous increase in the penetration rate of photovoltaic power generation, data quality have become a key factor affecting the intelligent operation and grid connection research of photovoltaic power plants. The existence of bad data not only affects the accuracy of predictions, but may also lead to deviations in photovoltaic system status monitoring and fault diagnosis. To improve the integrity and reliability of photovoltaic power plant data, this paper proposes a method for identifying and reconstructing bad data in photovoltaic power plants based on multi-source heterogeneous data correlation. Firstly, analyze the data characteristics and correlation between multisource parameters of the photovoltaic system under normal operation, and select the historical data with the most similar characteristics to the data to be reconstructed as input. Secondly, the multi density clustering algorithm based on relative density is used to identify and clean power poor data. Finally, based on the correlation of environmental data, a photovoltaic system combination long short-term memory data reconstruction model is established to achieve high-precision reconstruction of the data. The calculation results show that the proposed method can effectively identify the bad data of photovoltaic power station output and accurately reconstruct it.