| 甘志英,赵博轩,郝文鑫,尹浩哲,艾仕晨.基于DCA-CNN-LSTM的光伏功率预测模型[J].唐山学院学报,2026,39(3):37-43 |
| 基于DCA-CNN-LSTM的光伏功率预测模型 |
| Photovoltaic Power Prediction Model Based on DCA-CNN-LSTM |
| 投稿时间：2025-06-20 |
| DOI：10.16160/j.cnki.tsxyxb.2026.03.007 |
| 中文关键词: 光伏功率 预测功率 卷积神经网络 长短期记忆网络 多头注意力机制 |
| 英文关键词: photovoltaic power power prediction convolutional neural network(CNN) long-short-term memory network multi-head attention mechanism |
| 基金项目: |
|
| 摘要点击次数: 343 |
| 全文下载次数: 232 |
| 中文摘要: |
| 针对光伏功率具有高度随机性和波动性问题,提出结合卷积神经网络、长短期记忆网络、多头注意力机制双路架构的光伏功率预测模型(DCA-CNN-LSTM),即通过卷积神经网络提取数据空间模式和局部依赖关系,通过长短期记忆网络捕捉数据长期依赖和动态变化,通过多头注意力机制双路架构并行处理数据提高特征表示能力。实验采用皮尔逊相关分析来分析功率数据集,剔除无关变量,修正异常值,判断预测功率的输入时间窗口范围。结果表明,与其他模型预测功率相比,该模型的评估指标体现出更高的预测精度:在各模型最优输入时间窗口下,该模型的MAE,RMSE,MAPE比其他模型至少降低了11%,10%,4%,决定系数R2至少提高了0.3%,证明该模型在光伏功率预测方面具有良好的适用性。 |
| 英文摘要: |
| To address the high randomness and volatility of photovoltaic power, a hybrid photovoltaic power prediction model (DCA-CNN-LSTM) combining convolutional neural network, long-short-term memory networks, and a dual-path architecture with multi-head attention mechanism is proposed. Specifically, convolutional neural networks are used to extract spatial patterns and local dependencies of the data; Long-short-term memory networks capture long-term dependencies and dynamic changes; The dual-path architecture with multi-head attention processes data in parallel to improve feature representation capability. In the experiment, Pearson correlation analysis is used to analyze the power dataset, eliminate irrelevant variables, correct outliers, analyze input time window ranges, and predict power. The results show that compared to different models for power prediction, the evaluation indicators of this model demonstrate higher prediction accuracy. Under the optimal input time windows of each model, the MAE, RMSE and MAPE of this model are reduced by at least 11%, 10% and 4% lower, and the coefficient of determination R2 is increased by at least 0.3% higher than other models respectively, proving that this model has good applicability in photovoltaic power prediction. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|