2D Time-varying functional modeling framework for multi-scale solar irradiance forecasting.

Area of Science:

• Renewable Energy Forecasting
• Time Series Analysis
• Machine Learning for Energy Systems

Background:

• Solar irradiance forecasting is crucial for energy management.
• Existing 1D time series methods struggle with multi-scale temporal dynamics.
• Accurate long-term solar irradiance prediction remains a challenge.

Purpose of the Study:

• To introduce a novel 2D Time-Varying Function Modeling (2D-TFM) framework.
• To overcome limitations of 1D representations in capturing multi-scale temporal dependencies.
• To enhance the accuracy and interpretability of solar irradiance forecasting.

Main Methods:

• Transforming 1D solar irradiance time series into 2D functional sequences using B-spline basis function expansion.
• Optimizing functional representation with an Adaptive Local Complexity (ALC) knot placement algorithm.
• Employing a functional Long Short-Term Memory (LSTM) network for learning parameter space mappings.

Main Results:

• The 2D-TFM framework demonstrated superior forecasting accuracy compared to existing methods like Seq2Seq-LSTM.
• Achieved significant reductions in Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) for both hourly and minutely forecasts.
• The model provides mesh-free predictions at arbitrary time resolutions, enhancing practical applicability.

Conclusions:

• The proposed 2D-TFM framework effectively captures short-term fluctuations and long-term trends in solar irradiance data.
• 2D-TFM offers improved computational efficiency and interpretability over traditional methods.
• This framework enhances solar irradiance prediction for energy management systems.