Method and device for modeling a long-time-scale photovoltaic output time sequence

The patent segments the continuous photovoltaic output time sequence into discrete weather type states (clear sky, cloudy, overcast, changing weather). By dividing the prediction problem into discrete states and using Markov chains to model transitions between these states, the system can generate long-time-scale predictions while maintaining reliability through probabilistic modeling of weather patterns.

The patent introduces weather types as an intermediary between raw weather data and photovoltaic output prediction. The Markov chain models transitions between weather types, which then serve as the basis for generating photovoltaic output sequences. This intermediary layer enables long-time-scale prediction by capturing the probabilistic nature of weather patterns without requiring direct long-term weather forecasting.

The patent segments the photovoltaic output time sequence into discrete time intervals (15-minute resolution) and associates each interval with a weather type state. This segmentation allows the model to capture short-duration fluctuations by tracking weather type transitions at each time step, while the discrete state representation keeps the modeling complexity manageable through the use of transition probability matrices.

The patent changes the representation parameters by using weather type categories and transition probabilities instead of continuous weather variables. This parameter transformation enables the model to capture complex fluctuation characteristics through a simplified state-space representation, where the complexity is managed through probabilistic parameters rather than detailed physical models.