Data Load Prediction Based On Spatio-temporal Characteristics

As an information processing infrastructure model and business model,cloud computing has gradually gained wide acceptance.The execution of cloud services is inseparable from the computing resources and data support provided by the service execution environment.As an important support of the service execution environment,data and its load also affect the quality of service.The stability of the service execution environment is a key part of ensuring the quality of service when the data load is constantly fluctuating.To ensure the stability of the service execution environment,a popular system management approach is to actively allocating resources based on data load trends.In this paper,we aim to improve the quality of service by predicting the spatio-temporal data load trend based on the multidimensional characteristics.The specific research work is as follows.1.Statistical analysis of measured values of the data load.We analyse the measured values for the different characteristics of the data load.In this paper,the measured values are aggregated with a granularity time of 5 minutes.The measured load values are divided into short-time data sets and long-time data sets with a time range of 5minutes and 24 hours.We analyze the short-term and long-term data sets with spatiotemporal features and other dimensional features such as weather and date types.By analyzing the characteristics of the obtained datasets and combining the periodicity of data load and the relationship between different influencing factors,we provide the data basis for the data load trend prediction model.2.Trend prediction for data load.Trend prediction can anticipate data load fluctuations,which can be used to guarantee the stability of the service execution environment.According to the data load of different time granularity,and using the network structure characteristics of measurement points,this paper proposes a hybrid deep learning model based on convolutional network and long-short-term memory network.According to the data load of different characteristics,the data load trend can be predicted by changing the model parameters.It not only improves the model adaptability,but also proves the prediction accuracy through experiments.