随着容器云技术的不断深入发展，通过预测分析云资源请求的整体趋势及高峰期，对于容器云资源的高效利用和合理分配具有重要意义。利用深度学习技术进行负载预测已经成为解决容器云资源利用率不平衡的关键技术。针对目前负载预测的单一模型和组合模型所存在的预测精度低以及捕获序列特征不充分问题，提出基于时间卷积和长短期记忆网络（temporal convolutional network-long short-term memory, TCN-LSTM）的短期云资源组合预测模型，组合模型中的空洞卷积在不减少特征尺寸的情况下增加感受野获取更长久的时间序列特征，其中残差网络可以跨层传递信息以加快网络的收敛，所获取的时间序列特征可有效提高LSTM的预测精度。利用阿里巴巴公开数据集的进行预测，实验表明所提出的模型与单一的预测模型以及其他组合模型进行对比分析，误差指标-平均绝对误差（mean absolute error, MAE）降低8% ~ 13.7%，均方根误差（root mean squared error, RMSE）降低9.8% ~ 13.1%，证明所提模型的有效性。

With the continuous development of container cloud technology, it is of great significance to predict and analyze the overall trend and peak of cloud resource requests for efficient utilization and reasonable allocation of container cloud resources. Deep learning technology for load prediction has become a key technology to solve the unbalanced utilization of container cloud resources. Aiming at the problems of low prediction accuracy and insufficient capture sequence features existing in the current single model and combination model of load prediction, a cloud resource combination prediction model based on temporal convolutional network-long short-term memory(TCN-LSTM) was proposed. The hollow convolution in the combination model increased the sensitivity field without reducing the feature size to obtain longer time series features. The residual network could transfer information across layers to accelerate the convergence of the network, and the obtained time series features could effectively improve the prediction accuracy of LSTM. Useing Alibaba's publicly available dataset to make predictions, the experiment shows that the proposed model is compared with the single prediction model and other combined models, and the error index-mean absolute error(MAE) is reduced by 8% ~ 13.7% and root mean squared error (RMSE) by 9.8% ~ 13.1%, which proves the effectiveness of the proposed model.