Distributed Photovoltaic Power Forecasting Considering Perceived Data Quality

In distributed photovoltaic (PV) power generation systems, data quality plays a critical role in the accuracy of predictive models. However, the complexity of distributed PV sensor data, including issues such as missing values, noise and anomalous bias, often leads to data quality degradation, which seriously affects the reliability of the prediction. To address these challenges, this study proposes a novel forecasting method that integrates sensor data quality into the forecasting process to improve the reliability and robustness of forecasts. The method starts with a detailed analysis of distributed PV sensor data characteristics to identify patterns of data quality degradation and their impact on forecast accuracy. Data recovery algorithm based on adaptive mean drift clustering with multi-paradigm optimisation is presented to effectively deal with data loss and reduce noise. In addition, an adversarial graphical neural network (GNN) is used for knowledge migration and spatio-temporal correlation modelling to make accurate predictions even in the presence of perceived data quality degradation. The GNN framework exploits spatial dependence and temporal dynamics to create robust prediction models. Simulation results show that the proposed method significantly improves the data quality and prediction accuracy of distributed PV power generation, providing a reliable basis for decision making in energy management systems.