Accurate photovoltaic power probability forecasting assists dispatch departments in making rational decisions. Joint probability distributions constructed using Copula functions can flexibly characterize complex nonlinear correlations and tail dependencies among random variables. However, existing research has not thoroughly explored the multivariate dynamic coupling characteristics related to forecasting errors, nor has it sufficiently considered the complementary advantages among different Copula functions. To address this, we propose a method for forecasting photovoltaic power output probabilities days in advance, integrating multivariate dynamic Copula functions with reinforcement learning. First, to capture the time-varying structure of photovoltaic power-related variables, we introduce a sliding time window for segmented modeling of historical data, fitting marginal probability distributions for predicted irradiance, forecasting power, and forecasting error. Second, a joint probability distribution of dynamic Gaussian Copula and t-Copula is constructed based on historical samples within the time window, generating a probabilistic prediction interval for the target time. Finally, reinforcement learning is employed to adaptively combine the probability prediction intervals derived from both Copula types, yielding the final photovoltaic power probability forecast. Simulations using actual operational data from a photovoltaic power plant in Shanxi Province validate the effectiveness of the proposed method.

The deterministic evaluation framework for traditional virtual power plants (VPP) overlooks the capacity squeeze caused by day-ahead forecasting errors in real-time settlement, resulting in a significant overestimation of the adjustable capacity committed to external parties; simultaneously, conventional distributed PV forecasting methods struggle to capture the dynamic spatiotemporal coupling characteristics arising from meteorological time-varying factors. To address these issues, this paper proposes a comprehensive evaluation method for VPP adjustable capacity that integrates front-end dynamic spatiotemporal forecasting with back-end risk-averse scheduling. Firstly, a two-step dynamic sub-region partitioning strategy based on a sliding time window is proposed. By combining the Spacetimeformer self-attention model with K-means error mining, high-precision scenario-based available power boundaries are generated. Secondly, conditional value at risk (CVaR) is introduced to construct a two-stage risk-averse stochastic programming (RA-SP) model that accounts for expected physical default penalties. Finally, a parallel control evaluation system is established based on the physical baseline anchoring without response of internal flexible resources and real-time post-event verification. Simulations using real operational data demonstrate that the proposed dynamic spatiotemporal forecasting architecture significantly improves regional power forecasting accuracy; compared with traditional deterministic models, the RA-SP model effectively eliminates artificially inflated committed capacity exceeding physical limits (reducing the assessed tracking root-mean-square error by 32.42%), and by proactively setting aside a physical risk-resistance margin, the cumulative physical default electricity consumption throughout the day was reduced by 99.99%. This not only significantly enhanced the reliability of physical delivery but also achieved a flexible and reliable settlement rate with high confidence.