Efficient Deterministic Renewable Energy Forecasting Guided by Multiple-Location Weather Data

TL;DR

This work tackles deterministic day-ahead forecasting for multiple wind and solar sites by leveraging weather data from multiple locations without requiring geolocation information. It combines a UTCAE for multi-scale temporal representation with MKST-Attention, a spatio-temporal attention mechanism that transfers weather-time patterns to energy-time series using a learned spatial relation matrix. The method demonstrates state-of-the-art performance across five datasets, outperforming a broad set of baselines and exhibiting robustness to variations in weather-data availability. By enabling effective cross-site information transfer and temporal processing, the approach enhances grid integration of renewables and reduces forecast uncertainty in operational contexts.

Abstract

Electricity generated from renewable energy sources has been established as an efficient remedy for both energy shortages and the environmental pollution stemming from conventional energy production methods. Solar and wind power are two of the most dominant renewable energy sources. The accurate forecasting of the energy generation of those sources facilitates their integration into electric grids, by minimizing the negative impact of uncertainty regarding their management and operation. This paper proposes a novel methodology for deterministic wind and solar energy generation forecasting for multiple generation sites, utilizing multi-location weather forecasts. The method employs a U-shaped Temporal Convolutional Auto-Encoder (UTCAE) architecture for temporal processing of weather-related and energy-related time-series across each site. The Multi-sized Kernels convolutional Spatio-Temporal Attention (MKST-Attention), inspired by the multi-head scaled-dot product attention mechanism, is also proposed aiming to efficiently transfer temporal patterns from weather data to energy data, without a priori knowledge of the locations of the power stations and the locations of provided weather data. The conducted experimental evaluation on a day-ahead solar and wind energy forecasting scenario on five datasets demonstrated that the proposed method achieves top results, outperforming all competitive time-series forecasting state-of-the-art methods.

Figures (8)

• Figure 1: Abstract example of the Multi-sized Kernels convolutional Spatio-Temporal Attention (MKST-Attention) mechanism in the inference stage of a wind energy forecasting scenario, for a single wind power station (site #3) and two weather data locations (sites #1 and #2). The size of the future-time temporal window $T_f$ is 1 and the size of past-time temporal window $T_h$ is 4.
• Figure 2: (a): Scaled Dot-Product Attention, (b): the novel Multi-sized kernels convolutional scaled dot-product attention. $c_i$ denotes the size of the $i$-th convolutional kernel, employed in the temporal domain whereas $\Upsilon$ denotes the number of convolutional kernels.
• Figure 3: Illustration of the architecture of the U-shaped Temporal Convolutional Auto-Encoder, where the number of pyramid levels is 3.
• Figure 4: Architecture of the Joint Processing Block.
• Figure 5: Architecture of the proposed wind/solar energy forecasting method.