Delay Prediction for Real-Time Video Based on Improved LSTM
Zongshuai Zhang, Jiaying Huang, Lin Tian, Chunjing Yuan, Yuanyuan Wang · 2022
The scale of mobile video users increases year by year. Ensuring the quality of mobile video user experience can not only improve user satisfaction, but also further promote the process of informatization. The key to ensuring the quality of mobile video user experience is to be able to predict the time of mobile video delay and prevent the occurrence of video delay; for the prediction of video delay, this paper proposes an improved algorithm based on LSTM network with Dropout optimization. By merging multiple models according to their accuracy to balance the errors generated by the models, overfitting can be prevented, and at the same time, the influence of the model trained by some special eigenvalues on the whole can be avoided, and eliminate the model's dependence on some special eigenvalues. It solves the problems of the original model's low prediction accuracy, weak generalization ability, and overfitting for video delay prediction. At the same time, according to the idea of optimizing the part first and then the whole, to reduce the time complexity of model training. The experimental verification shows that under the same conditions, compared with the original LSTM model, the improved model's prediction accuracy is significantly improved for predicts the video delay, which provides a factual basis for the guarantee of video user experience quality.