| 摘 要: 针对电力负荷复杂的特征关系以及神经网络超参数选取困难的问题,提出贝叶斯优化(Bayesian Optimization,BO)驱动变分自编码器(Variational Autoencoder,VAE)和双向长短期记忆网络(Bidirectional Long Short-Term Memory,BiLSTM)的短期负荷预测模型VBiLBO-STLF。通过VAE的编码器-解码器结构提取电力数据的潜在表示,将其以时序序列的形式构建为BiLSTM的输入进行短期负荷预测,采用贝叶斯优化对VAE和BiLSTM 进行超参数优化达到最佳网络模型。以中国某市区真实电力数据进行日负荷预测实验,结果表明本文模型的预测精度达到约96.8%,与其他模型相比具有更好的预测效果。 |
| 关键词: 短期负荷预测 变分自编码器 双向长短期记忆网络 贝叶斯优化 |
|
中图分类号:
文献标识码: A
|
| 基金项目: 陕西省自然科学基金项目(2024JC-YBQN-0724) |
|
| Bayesian Optimization Driven VAE-BiLSTM Short-Term Load Forecasting Model |
|
ZHANG Xiuxian, WANG Xiaoxia, LI Xiang, CHEN Xiao
|
(School of Electronic In formation and Artificial Intelligence, Shaanxi University of Science & Technology, Xi’an 710021, China)
Zhangxiuxian1201@163.com; wangxiaoxia@sust.edu.cn; lixiangzj@sust.edu.cn; chenxiao@sust.edu.cn
|
| Abstract: A shor-t term load forecasting model VBiLBO-STLF is proposed to address the complex feature relationships of power loads and the difficulty in selecting hyperparameters for neural networks. The model is driven by Bayesian optimization (BO), variational autoencoder (VAE), and bidirectional long shor-t term memory network (BiLSTM). Extracting potential representations of power data through the encode-r decoder structure of VAE, and constructing these potential representations in the form of time-series sequences as inputs for BiLSTM for shor-t term load forecasting. Bayesian optimization is used to hyperparameter VAE and BiLSTM to achieve the optimal network model. The daily load forecasting experiment was conducted using real electricity data from a certain urban area in China, and the results showed that the prediction accuracy of our model reached about 96. 8% , which has better prediction performance compared to other models. |
| Keywords: short term load forecasting variational autoencoder bidirectional long short term memory network Bayesian optimization |
