Research On Deep Learning-based Caching Strategies In Industrial Applications

With the widespread application of Internet of Things technology in the industrial sector,issues concerning data transmission efficiency and latency in industrial Internet systems have become increasingly prominent.To address these challenges,emerging edge computing technologies have been widely adopted.However,edge computing nodes have limited computing and storage capabilities,necessitating the use of appropriate caching strategies.Against this backdrop,the development of artificial intelligence has propelled research and application of deep learning in edge computing.Deep learning models can analyze and learn from historical data to predict data access patterns,thereby improving the efficiency of caching strategies.This paper focuses on the research and application of content popularity prediction and edge caching strategies,with the following main research areas:(1)For the application scenario of industrial edge caching,a content popularity prediction model based on the Seq2 Seq neural network is proposed.In conjunction with this prediction model,a caching replacement algorithm called Simulated Optimal Page Replacement(S-OPT)is designed.This algorithm utilizes historical content request characteristics to predict the popularity features in different future time periods.By replacing the content with the lowest predicted popularity,the caching efficiency is optimized.(2)For scenarios where the popularity of content dynamically changes in industrial applications,a Dual-policy Adaptive Cache Replacement Algorithm Based on Popularity Prediction(DPAPP)is proposed.This algorithm combines two caching replacement strategies: S-OPT and reactive policy.By predicting content popularity and real-time tracking the hit rate,the algorithm automatically selects the optimal replacement strategy to mitigate the decrease in cache hit rate caused by the fluctuation in popularity predictions.Experimental results demonstrate that the proposed DPAPP algorithm achieves optimal performance across different user request models and effectively improves the cache hit rate of edge caching.(3)For scenarios where the dynamic popularity of content in industrial applications is unknown,a Cache Replacement Algorithm Based on Ensemble Popularity Prediction Models(CEPPM)is proposed.This algorithm is designed to learn new content request patterns from newly obtained data extracted from the evolving distribution over time.This mechanism allows the cache replacement algorithm to adapt to the constantly changing production environment by employing content request pattern classifiers.If a certain content request pattern reoccurs in the cyclical production environment,the existing popularity prediction models will be reused.Experimental results demonstrate that the proposed CEPPM algorithm achieves optimal performance in scenarios with unknown content popularity compared to the DPAPP algorithm and traditional cache replacement algorithms.