Joint Exponential Window-Kalman Filter and Weight Normalized ReLU-Memristor Activation Function-LSTM-based Proactive Content Caching framework in IoV using Asynchronous Federated Learning
Sunitha Safavat, Danda B. Rawat · 2024
Advancements in wireless communication technolo-gies provide information interaction among vehicles, humans, and roadside infrastructure. As a result, the size and volume of content that must be streamed in real-time have grown rapidly, resulting in congested data traffic at content servers and a degradation in the user experience. Hence, to address these issues, an efficient Sliding Exponential Window-Kalman Filter (SEW-KF) and Weight Normalized ReLU-Memristor-like Activation Function-Long Short Term Memory (WNormRMAF-LSTM) based proactive content caching framework is proposed using Asynchronous Federated Learning (AFL). Primarily, the number of vehicles on the road is initialized and clustered using the Taylor Kernelized-Affinity Propagation Clustering (TK-APC) technique. Then, the cache vehicle is selected using HenonChebyshev Cosine-Honey Badger Optimization (HMCo-HBO). Next, the historical data (global model) from the AFL is given to cluster members by Cache Vehicle (CV) and trained using WNormRMAF -LSTM. After training the global model, each cluster member transfers the trained model to a CV, which in turn transfers to Road Side Unit (RSU). In RSU, content popularity prediction takes place using SEW-KF. After popularity prediction, more and less popular content is placed in Macro Base Station (MBS), whereas the popular content is placed in the RSU using Greedy Algorithm (GA). At last, data accessing takes place, where the availability of the data is checked and forwarded to the user. Finally, the results of the proposed method are compared to those of traditional algorithms.