基于负载特征与注意力的容器负载预测

With the rapid development of China's Internet, more and more applications have been transformed into microservice architectures. Microservice container load prediction is of great significance. Accurate load prediction helps complete tasks such as resource scheduling and service monitoring. Based on the convolutional neural network and the LSTM Encoder-Decoder with attention mechanism, this paper proposes an algorithm that can predict the single load metricusing multivariate load data. The algorithm captures the pattern characteristics between the load metrics through a convolutional neural network, and completes the mapping of the pattern characteristics to the prediction results through the LSTM Encoder-Decoder with the attention mechanism. The verification on the open source dataset shows that the container load prediction algorithm proposed in this paper has higher prediction accuracy than traditional algorithms such as LSTM.