To address the issues of existing regional distributed photovoltaic (PV) power forecasting, such as heavy reliance on meteorological data, high operation and maintenance costs, poor data quality and insufficient result credibility, a joint credible forecasting method for PV power and energy is proposed. First, power measurements from smart meters and daily frozen energy data are jointly filtered, fused, and normalized to enhance data set quality. Second, a multi-time-scale, high-accuracy sequence-to-sequence (Seq2Seq) forecasting framework is developed, integrating historical and forecast data from centralized regional PV plants; a multi-time-scale loss function that jointly accounts for both power and energy is employed to optimize prediction accuracy. Finally, a model integrity verification scheme based on commit-and-prove succinct non-interactive argument of knowledges (cp-SNARKs) is designed to ensure result credibility while preserving model confidentiality. Experimental validation using real-world data from a city in North China demonstrates that the proposed method significantly reduces forecasting errors for both power and energy, thereby improving PV power prediction accuracy. Requiring no meteorological inputs or system modifications, the approach features high data quality, superior prediction accuracy, low operational cost, and strong verifiability, making it readily extensible to other time-series forecasting tasks such as load forecasting and wind power prediction.