| 摘 要: 在能源结构低碳转型背景下，风电和光伏发电占比持续提高，其功率波动为电网调度的稳定运行带来挑战。针对风光发电功率序列非线性、非平稳性强，以及单一时序模型难以兼顾长期趋势与短时波动的问题，提出一种融合时间卷积网络与长短期记忆网络的双分支神经网络预测方法(TCN-LSTM hybrid network, TLNet)。该方法通过LSTM分支提取长期依赖信息，利用TCN分支捕捉局部波动特征，并结合特征自适应加权机制与残差融合机制提升预测精度。实验结果表明，所提方法在四个数据集上的均方误差(MSE)、均方根误差(RMSE)、平均绝对误差(MAE)对比其他方法平均降幅为18.83%、9.90% 和13.69%，验证了TLNet在风电与光伏功率预测任务中的有效性。 |
| 关键词: 发电功率预测 双分支神经网络 LSTM TCN |
|
中图分类号:
文献标识码:
|
| 基金项目: 基金项目:陕西省教育服务厅地方专项计划项目(22JC019) |
|
| Research on Power Generation Forecasting Method Based on a Dual-Branch Neural Network |
|
qiyong, guoyumeng
|
陕西科技大学
|
| Abstract: Against the background of low-carbon energy transition, the proportion of wind and photovoltaic power generation continues to increase, and the resulting power fluctuations pose challenges to the stable operation of power grid dispatching. To address the strong nonlinearity and non-stationarity of wind and photovoltaic power sequences, as well as the limitation of a single temporal model in capturing both long-term trends and short-term fluctuations, this paper proposes a dual-branch neural network forecasting method integrating Temporal Convolutional Network and Long Short-Term Memory, named the TCN-LSTM Hybrid Network (TLNet). In this method, the LSTM branch is used to extract long-term dependency information, while the TCN branch captures local fluctuation features. Meanwhile, a feature adaptive weighting mechanism and a residual fusion mechanism are introduced to improve forecasting accuracy. Experimental results on four datasets show that the proposed method achieves average reductions of 18.83%, 9.90%, and 13.69% in Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE), respectively, compared with other methods, demonstrating the effectiveness of TLNet in wind and photovoltaic power forecasting tasks. |
| Keywords: Power generation forecasting Dual-branch neural network LSTM TCN |
