DTCMMA: Efficient Wind-Power Forecasting Based on Dimensional Transformation Combined with Multidimensional and...

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Energy
• Time Series Forecasting

Background:

• Accurate wind-power forecasting is crucial for stable power grid operation amid growing clean energy demand.
• Existing models like RNNs and LSTMs struggle with long-term dependencies, while Transformers face efficiency issues with long sequences and capturing local periodic features.
• Wind power is inherently uncertain and fluctuates across multiple scales, influenced by meteorological conditions.

Purpose of the Study:

• To propose an efficient and accurate wind-power forecasting method addressing limitations of existing models.
• To enhance the modeling of long-term temporal dependencies and local periodic features in wind-power data.
• To improve both prediction accuracy and computational efficiency for long-sequence wind-power forecasting.

Main Methods:

• Developed a dimension-transformed collaborative multidimensional multiscale attention (DTCMMA) method for wind-power forecasting.
• Utilized fast Fourier transform (FFT) to reconstruct 1D time series into 2D spatiotemporal representations, encoding periodic features.
• Designed a collaborative multidimensional multiscale attention (CMMA) mechanism integrating channel, spatial, and pixel attention with asymmetric convolution kernels.

Main Results:

• DTCMMA demonstrated superior performance over Transformer, iTransformer, and TimeMixer in long-sequence forecasting tasks.
• Achieved significant improvements in Mean Squared Error (MSE) performance: 34.22% over Transformer, 2.57% over iTransformer, and 0.51% over TimeMixer.
• The model's training speed was 300% faster than the quickest baseline, indicating enhanced computational efficiency.

Conclusions:

• DTCMMA offers an efficient and accurate solution for wind-power forecasting, outperforming current state-of-the-art methods.
• The method effectively captures complex spatiotemporal dependencies and periodic features, crucial for wind energy prediction.
• This advancement supports the integration and application of wind energy in the global energy mix.