A Triple-Optimized Extreme Learning Machine Model for Power Load Forecasting
Electricity load forecasting constitutes a pivotal task in achieving an equilibrium between supply and demand within the power system, facilitating effective power grid dispatching, and ensuring the safe and stable operation of the grid. The ELM model, characterized by its high efficiency and expedi...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Open Access Journal of Power and Energy |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11071898/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Electricity load forecasting constitutes a pivotal task in achieving an equilibrium between supply and demand within the power system, facilitating effective power grid dispatching, and ensuring the safe and stable operation of the grid. The ELM model, characterized by its high efficiency and expeditious training, has become a prevalent approach in the domain of electricity load forecasting. The model’s architecture comprises a front end, a core, and a back end. However, the optimization scheme of the model is optimized for a specific aspect, namely single-objective optimization. This approach disregards the pathological characteristics and overfitting that arise from the simultaneous optimization of the three, the challenges of calculation, and the deviation of the prediction results. This paper proposes a seamless enhanced incremental ELM triple optimization model (SBOA-SEI-MRU-ELM) based on the Secretary bird optimization algorithm and the MINres regularization under the U-curve method to solve the above problem. The optimal input weight matrix and threshold vector can be selected through the front-end module, incremental iteration can be performed through the core, and pathological problems and overfitting can be eliminated through the back-end module. A comparison of the proposed method with traditional single-weight optimization reveals a twofold reduction in MSE and a more than 20% decrease in MAPE. When evaluated against LSTM, SVM, and RBF methods, the proposed method exhibits a one-to-two-order magnitude reduction in MSE and a 1% to 16% decrease in MAPE. The findings demonstrate a competitive edge over research conducted within a specialized branch that utilizes metaheuristic algorithms. |
|---|---|
| ISSN: | 2687-7910 |