Design Of QoS Time Series Prediction Model Based On Deep Learnin

Predicting Quality of Service(QoS)is an important problem in the field of service recommendation that has received extensive research attention in recent years.Many existing QoS prediction methods rely on the historical invocation records of services to predict unknown QoS.Most of these methods use collaborative filtering to perform static prediction based on information from similar users or services.Although these methods are largely applicable to services in static environments,there are still three major challenges in dynamic environments based on time-varying changes.First,QoS of services in dynamically complex environments changes frequently over time,and the traditional static computing mode cannot meet the new requirements for QoS prediction in dynamic environments,which require frequent and repetitive calculations.Second,the inputs of existing QoS prediction methods lack other information that can be helpful for prediction.Preprocessing methods that are needed to add additional computation costs still have room for improvement.Finally,with the rapid development of deep learning and the continuous updating of computing devices,deep learning has achieved significant results in many fields.It has also begun to be widely used in the QoS prediction field,achieving more accurate prediction results.Based on the above considerations,how to capture the time-varying characteristics of QoS and combine them with deep learning methods to accurately predict service QoS values has become a key issue.To address these issues,this thesis starts from deep learning and uses deep learning for QoS prediction of services.The main research contents are as follows:1)A time-aware feature optimization approach for deep learning-based QoS prediction is proposed: With the increase in computing power,QoS prediction methods in complex dynamic environments have gained widespread attention.Although these methods combine time-related features to achieve better prediction results,the lack of reference auxiliary information during feature extraction makes it difficult to obtain better quality features.To address this issue,a time-aware feature optimization approach for deep learning-based QoS prediction is proposed.This approach employs a Deep Gate Recurrent Network(DGRN)to capture time-aware QoS time series and extract user and service features based on time awareness.Additionally,a novel Gate Recurrent Generative Adversarial Network(GRGAN)is designed to optimize the extracted user and service features and improve the accuracy of QoS prediction.2)Time-aware and feature fusion based end-to-end deep learning model for QoS prediction is proposed: Existing deep learning-based QoS prediction methods generally use one-hot encoding as input features for users and services,which limits the network’s ability to learn other information beneficial for prediction.Additionally,prediction using the sparse matrix decomposition method requires data preprocessing,increasing computational costs.To address these issues,a dual feature fusion-based deep learning QoS prediction method is proposed.This method constructs a user-service encoding conversion module that can convert one-hot encoding to more suitable user-service latent features during model training.A time feature extraction module is used to extract time features,and the user-service latent features are fused with the time features to conduct QoS prediction.