为有效地挖掘历史数据信息，提高短期负荷预测准确性，文章针对电力负荷时序性和非线性的特点,在原有一维卷积神经网络(Convolutional Neural Network, CNN)-长短期记忆网络(long short term memory，LSTM)模型的基础上，分别在CNN和LSTM侧嵌入通道注意力机制(Channel Attention，CA)和时序注意力机制(Temporal Attention，TA)，构建CA-CNN和TA-LSTM模块，然后结合CA-CNN和TA-LSTM模块构建TCA-CNN-LSTM的层级注意力机制短期负荷预测模型。同时，为提高训练数据的质量并加快模型训练速度，运用K-means和决策树模型选取相似日数据, 构建基于相似日数据的向量特征时序图，最后将时序图输入TCA-CNN-LSTM负荷预测模型完成预测。以澳大利亚某地真实数据集和2016电工杯数学建模竞赛电力负荷数据为算例，分别应用TCA-CNN-LSTM模型与支持向量机、多层感知机(Multilayer perceptron, MLP)、LSTM、CNN-LSTM和CNN-Attention-LSTM模型的预测结果进行对比，实验结果表明所提方法具有更高的预测精度。
英文摘要:
In order to accurately mine historical data information and improve the accuracy of short-term load forecasting. the article focuses on the time series and nonlinear characteristics of the power load, based on the original one-dimensional convolutional neural net-work(CNN)-long short term memory network(LSTM) model, channel attention(CA) and temporal attention mechanism(TA) were added to the CNN and LSTM respectively, combining CA-CNN and TA-LSTM modules, we construct the TCA-CNN-LSTM hierar-chical attention mechanism short-term load forecasting model. At the same time, In order to improve the quality of training data and speed up model training, K-means and decision tree model are used to extract similar days. construct a vector feature time series dia-gram based on similar day data, and finally input the time series diagram into TCA-CNN-LSTM load forecasting to complete the pre-diction. Taking the real data set of a certain place in Australia and the power load data of the 2016 Electrician Cup Mathematical Model-ing Competition as examples, the TCA-CNN-LSTM model is applied to compare with the prediction results of support vector ma-chine(SVM), Multilayer perceptron(MLP), LSTM, CNN-LSTM and CNN-Attention-LSTM models, respectively. The experimental results show that the proposed method has higher prediction accuracy.