基于改进的AT-QRCNN-BiLSTM模型的多变量时序短期光伏功率区间预测研究
Multivariable Time-sequential Short-term Photovoltaic Power Interval Prediction Based on Improved AT-QRCNN-BiLSTM Model
周文 1高强 1毛泽民2
作者信息
- 1. 天津理工大学电气工程与自动化学院,天津 300384
- 2. 天津理工大学聋人工学院,天津 300384
- 折叠
摘要
针对光伏发电的随机性、波动性和间歇性特征以及天气和环境的不确定性,导致其功率输出难以精准预测的问题,提出了一种改进的基于注意力机制的分位数回归卷积神经网络和双向长短时记忆神经网络模型(AT-QRCNN-BiLSTM)融合构建的多变量时序短期光伏功率区间预测方法.该方法由数据预估与校正、数据点预测、伴有优化目标函数的区间预测3个部分组成.首先考虑测量误差对历史数据的影响,将光伏功率及环境的历史数据进行预估处理,得到预估后的历史数据;然后将预估后的光伏功率和环境历史数据输入一个基于注意力机制的卷积神经和双向长短时记忆神经网络(AT-CNN-BiLSTM)融合构建的点预测模型进行训练,得到校正后的点预测结果;然后通过白鲸优化算法(BWO)改进后的AT-QRCNN-BiLSTM模型预测置信区间的上下界;最后通过区间预测的评价指标对预测结果进行分析.仿真结果表明,该融合网络模型预测性能良好,平均绝对误差和均方根误差分别为0.1636 kW和0.2511％.该方法在置信度95％的区间预测的覆盖率达到95.02％,区间宽仅有12.75％.
Abstract
To address the challenges of random variability,fluctuation,intermittency in photovoltaic(PV)power genera-tion,and the uncertainty in weather and environmental conditions,an improved quantile regression convolution neural net-work based on attention mechanism and bidirectional long-term memory neural network short-term photovoltaic power in-terval prediction method based on multivariable time sequence was established.The method consisted of three parts:data prediction and correction,data point prediction,and interval prediction with optimization objective function.First,consid-ering the influence of measurement error on historical data,the historical data of photovoltaic power and environment were estimated and processed,and the estimated historical data were obtained.Second the estimated photovoltaic power and environmental historical data were input into a point prediction model constructed by the fusion of convolution neural network based on attention mechanism and bidirectional short-term memory neural network,and the corrected point pre-diction results were obtained.Then the upper and lower bounds of the confidence interval were predicted by the improved beluga whale optimization algorithm.Finally the prediction results were analyzed by the evaluation index of interval predic-tion.The simulation results showed that the proposed hybrid network model can achieve high accuracy with fast conver-gence speed,and effectively solve the problem of serious impact of stochasticity,uncertainty and time-sequence on photo-voltaic power prediction by improving prediction accuracy.
关键词
光伏区间预测/不确定性/AT-CNN-BiLSTM/BWO/AT-QRCNN-BiLSTMKey words
photovoltaic interval prediction/uncertainty/AT-CNN-BiLSTM/BWO/AT-QRCNN-BiLSTM引用本文复制引用
出版年
2024
NETL
NSTL
万方数据