Deep Learning Caching for Edge Computing in Autonomous Vehicles

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing systems face delays in providing infotainment content to self-driving cars due to prolonged communication with data centers, which can lead to undesirable user experiences.

Innovation Solution

A deep learning-based caching system utilizing multi-access edge computing (MEC) that predicts content demand and optimally selects MEC servers for caching and content delivery, employing k-means algorithms and binary classification to recommend content based on user characteristics and probability of request, minimizing download delays.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of time

If content is provided from existing data centers, then content delivery is achieved, but communication delays are prolonged

Engineering Contradiction:
Improvecommunication delayVSAvoidsystem architecture
Core Design Contradiction:
Loss of timeVSDevice complexity

Solution Approach 1:

The system segments the centralized data center architecture into distributed edge computing nodes positioned along vehicle movement paths. Each edge server handles local content caching and delivery, dividing the monolithic data center function into multiple geographically distributed segments that reduce communication distance and latency.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system performs preliminary content caching at edge servers before vehicles arrive. Deep learning models predict which contents will be requested by vehicles on specific paths and pre-cache them at appropriate edge servers, so that when vehicles request content, it is already available locally rather than needing to be fetched from distant data centers.

Inventive Principle:
Principle #10Preliminary action

2Loss of time

If deep learning based caching is implemented, then content download delay is reduced, but computational complexity increases

Engineering Contradiction:
Improvecontent download delayVSAvoidcomputational complexity
Core Design Contradiction:
Loss of timeVSDevice complexity

Solution Approach 1:

The system introduces a deep learning-based prediction model as an intermediary between raw traffic data and caching decisions. This model processes vehicle movement patterns, content popularity, and path information to generate accurate predictions of future content requests, enabling intelligent caching without requiring complex real-time computation at edge servers.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The edge servers autonomously make caching decisions based on predictions from the deep learning model, without requiring continuous centralized control. The system self-adjusts by learning from historical data and automatically optimizing cache contents based on predicted vehicle behavior and content demand patterns.

Inventive Principle:
Principle #25Self-service

3Speed

If content is cached at edge servers, then real-time delivery is achieved, but cache management complexity increases

Engineering Contradiction:
Improvecontent delivery speedVSAvoidcache management complexity
Core Design Contradiction:
SpeedVSDevice complexity

Solution Approach 1:

The cache management system is dynamic rather than static. Cache contents, sizes, and locations are continuously adjusted based on changing vehicle traffic patterns, time of day, and predicted content demand. The system adapts cache configurations in real-time to match actual usage patterns, optimizing delivery speed while managing complexity through automated adaptation.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP3648435B1Deep learning based caching system and method for self-driving car in multi-access edge computing
Publication Date: 2020.10.28 UNIVERSITY INDUSTRY COOPERATION GROUP OF KYUNG HEE UNIVERSITY
  • EP3648435B1 patent drawingFigure 1
  • EP3648435B1 patent drawingFigure 2
  • EP3648435B1 patent drawingFigure 3

AI summary

A caching system based on the invention can include an object requiring a content and an MEC server configured to determine caching contents based on a first prediction value, which may include the probability of the content being requested by the object within an allotted area and a prediction rating of the content, and download and cache the determined caching contents from a content provider. The object can include a recommendation module configured to recommend a content from among the caching contents by applying a k-means algorithm and binary classification to the first prediction value and a second prediction value, which may include a prediction value associated with a characteristic of a user, and a deep learning based caching module configured to search available MEC servers on a movement path of the object, select an optimal MEC server, and download and cache the recommended content from the optimal MEC server.