An Optimized Hybrid Framework Based on Long-Short Term Memory Neural Networks and Fourier SynchroSqueezed Transform for Photovoltaic Power Forecasting
Accurate prediction of photovoltaic power is crucial to optimize its integration into the power grid. However, this task is highly complex due to the inherently stochastic nature of photovoltaic power generation. To address this challenge, this paper proposes a novel hybrid framework that combines a...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11080022/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Accurate prediction of photovoltaic power is crucial to optimize its integration into the power grid. However, this task is highly complex due to the inherently stochastic nature of photovoltaic power generation. To address this challenge, this paper proposes a novel hybrid framework that combines a Long Short-Term Memory network with the Fourier SynchroSqueezed Transform, and Bayesian Optimization. The model integrates the Fourier SynchroSqueezed Transform with the Long Short-Term Memory network to enhance the identification of very short-term patterns in photovoltaic power production. Additionally, Bayesian Optimization is used to determine the most relevant hyperparameters of the Long Short-Term Memory network. The model was evaluated using real-world data from the Desert Knowledge Australian Solar Centre project, considering different data segmentation sizes, ranging from 15 days to four months, and input types, including univariate and multivariate data, with prediction horizons ranging from five minutes to three days. Significant improvements in prediction accuracy were observed. For example, the Root Mean Squared Error for the 15-day data segmentation decreased by 19.48% when using multivariate inputs and by 29.59% for univariate inputs. For a four-month data segmentation, improvements reached 40.12% and 64.47%, respectively. Furthermore, the model demonstrated robust performance across various prediction horizons. For the 15-day segmentation, the Root Mean Squared Error was approximately 0.707 kW with an average power of 8.540 kW, while for the four-month segmentation, the error ranged around 0.467 kW with an average power of 4.733 kW. These results demonstrate the effectiveness and consistency of the proposed model in enhancing photovoltaic power forecasts across different time scales and data segmentations. The demonstrated ability to provide consistent and accurate predictions across multiple time horizons and data segmentations represents a significant advancement toward seamless integration of solar energy into the electricity grid. |
|---|---|
| ISSN: | 2169-3536 |