Design And Implementation Of Spatiotemporal Sequence Prediction Algorithm For Perceived Quality In Mobile Networks

With the rapid development of mobile internet technology,the number of mobile internet users has exploded.In the face of the ever-increasing number of access devices and the constantly increasing demand for service quality from users,it is of great significance to study how to effectively predict changes in the user-perceived quality of mobile networks.Predicting user perception of mobile networks can improve user experience,optimize network resources,improve network planning,and help business decision-making.Currently,most researches on predicting mobile network performance metrics consider it as time series prediction tasks and generally uses drive test data to predict a single or a few metrics,such as traffic.This research method does not consider the spatial relationship of the data and may result in incomplete evaluation results due to the prediction of a single metric.In addition,data collected through traditional methods such as drive tests have problems such as difficulty in obtaining,limited coverage,and inability to reflect actual user usage.To address the above problems,this paper considers the prediction of mobile network perceptual quality as a spatiotemporal sequence prediction problem and chooses cross-operator data instead of traditional road test data for experiments.This paper first proposes an Attention-based Vector Quantization Spatiotemporal Representation Model(Attn-VQST)for efficiently extracting spatial information from spatiotemporal sequence data.The algorithm is based on the vector quantization method combined with the convolutional attention module as the data space coder-decoder of the spatiotemporal sequence prediction model,which can achieve efficient extraction of spatial features from spatiotemporal sequence data.This method obtains enhancements in the image reconstruction task compared to baseline models.Next,this paper proposes a Discrete-Spatiotemporal Predictive Learning Model(D-STP),which can be used as a backbone network by using predictors with different architectures such as recurrent neural networks,transformers,etc.Experiments show that D-STP has smaller error and higher efficiency when performing spatiotemporal sequence prediction on different datasets.Based on the above algorithms,this paper implements a mobile network perceptual quality prediction system.The system provides a web application with a visualization interface,with real-time data collection,processing,model inference and online learning functions.System testing shows that the system is stable in function and performance,and can achieve basic mobile network perception prediction visualization and management functions.