Privacy-Preserving Data Processing Framework Using Distributed Computing in Vehicle-to-Everything (V2X) Systems

Abstract

ABSTRACT Privacy-Preserving Data Processing Framework Using Distributed Computing in Vehicle-to-Everything (V2X) Systems The present invention provides a privacy-preserving data processing framework for Vehicle-to-Everything (V2X) communication systems, utilizing distributed computing and edge processing. It leverages federated learning to enable localized data computation while preserving user privacy by limiting the transmission of sensitive data to central servers. The framework integrates edge computing units to collect, analyze, and process vehicular data in real time, facilitating low latency and secure data handling. A scalable traffic management system is supported, utilizing local model aggregation and predictive modeling for optimal traffic control. The framework reduces network bandwidth consumption and enhances privacy by processing data close to its source, making it adaptable for future intelligent traffic systems with increasing vehicular connectivity. Fig.1

Specification

Description:FIELD OF THE INVENTION
The present invention relates to the field of intelligent traffic management systems, focusing on data security and privacy within distributed computing environments.
BACKGROUND OF THE INVENTION
The modern world is experiencing a gradual yet significant shift towards the automation epoch, a period where virtually all entities and systems are being automated to carry out various tasks efficiently without constant human intervention. This widespread adoption of automation has undeniably enhanced people's lives by increasing convenience and overall comfort. The influence of automation extends across all aspects of computing and has now started to permeate into other spheres beyond just technology.

The automotive industry, for instance, has seen remarkable advancements over the years with the continuous development and integration of cutting-edge technologies. This progress has led to a surge in the utilization of automobiles, resulting in a noticeable rise in traffic complexity and the stress levels of drivers, consequently leading to a significant increase in road accidents.

In the context of data transmission within this automated ecosystem, a plethora of data is generated by vehicles and connected infrastructure. This data often contains highly sensitive information such as the precise locations, driving routes, and personal details of drivers. This data privacy concern has highlighted the importance of deploying privacy-preserving artificial intelligence methods, with federated learning emerging as a promising solution to protect personal information from unauthorized exposure or misuse.

Furthermore, handling the massive data generated by vehicles and infrastructure poses a challenge in terms of network bandwidth capacity. The conventional method of transferring all data from network edges to centralized cloud data centres for processing may result in latency issues.

These challenges necessitate a shift towards privacy-preserving data handling solutions. The use of distributed computing techniques offers a pathway to manage these issues effectively, by processing data closer to the data sources and reducing reliance on central servers.

OBJECTS OF THE INVENTION
It is the primary object of the invention to provide a system for efficient and privacy-preserving data processing within V2X communication frameworks.

It is another object of the invention to minimize latency in traffic data processing by enabling decentralized computation near data sources.
It is another object of the invention to enhance data security and privacy in vehicular communication by reducing the exposure of sensitive information.

It is another object of the invention to reduce network bandwidth requirements by implementing local data processing.

It is yet another object of the invention to facilitate scalability in V2X systems with a growing number of interconnected vehicles and devices.

SUMMARY OF THE INVENTION
To meet the objects of the invention, it is disclosed here a privacy-preserving data processing system for Vehicle-to-Everything (V2X) communication, comprises: a plurality of edge computing units; a federated learning unit; a real-time data manager module; and a predictive modeling module, wherein the edge computing units are positioned near data sources and configured to collect and process vehicular data including but not limited to vehicle speed, location, and route information in real time; the federated learning unit is connected to the edge computing units and configured to perform localized model training on collected data while aggregating model parameters in a centralized server to generate a global model, wherein the model aggregation is performed without transmitting raw vehicular data to the central server, thereby preserving privacy; the real-time data manager module is configured to organize and store data within the edge computing units, facilitating model training and data processing locally; and the predictive modeling module is integrated within each edge computing unit to optimize traffic control by analyzing processed vehicular data and generating output for traffic management.

Further disclosed here a method for privacy-preserving data processing in V2X communication systems, comprising steps of: collecting vehicular data in real-time at edge computing units located near data sources, training local models at each edge computing unit using federated learning to preserve data privacy by restricting raw data transfer, aggregating model parameters from local models at a centralized server to form a global model, and utilizing predictive modeling on processed data at each edge computing unit for optimized traffic control outputs.

BRIEF DESCRIPTION OF THE FIGURES
Fig.1 illustrates the federated learning framework in traffic management systems.
Fig.2 depicts the Conceptual paradigm for edge cloud-centric IoT-based smart traffic management.
Fig.3 shows Privacy-preserving AI using Edge computing in V2X framework.

DETAILED DESCRIPTION OF THE INVENTION
The invention provides a distributed data processing system for V2X communication, utilizing privacy-preserving computational techniques at the network edge. It leverages decentralized processing units to handle vehicle-generated data in real time, ensuring low latency and enhanced data security. A federated learning framework is implemented to allow individual devices to compute locally while contributing to global model improvements, thus preserving privacy. Edge computing units collect and process mobility data, including vehicle speed, location, and travel routes, to optimize traffic management through predictive modeling, while protecting sensitive information.

Fig 1 architecture gives exposure on how the basic FL works with Every five runs, the FL in the machine learning loop is triggered. The FL loop is triggered after one FL cycle, which is represented by this number of simulations. Consequently, following each FL cycle, the FL dedicated server generates a global model. Every client receives one as they show up in that specific simulation. Every FL cycle, a global model is updated in addition to the local ones. In order to allow clients to have a long enough local learning period in between global model updates, the repeating FL cycle with a predetermined number of simulations is implemented. The sensitivity of local models for global control patterns is decreased by the lengthy FL cycles that need an excessive number of simulations. After every FL cycle, the clients send the average performance and trainable variables of their local models to the FL dedicated server. Furthermore, the local model aggregation procedure is carried out by the FL server. The weighted average of the trainable variables of a subset of the best-performing local models is used to aggregate the models locally.

Once models of particular clients are obtained, it is completed in three phases. The weight scaling factor is first calculated using the best-performing customers that have been chosen. It is calculated as the ratio of a client's local performance to the total performance attained by all of the clients that were chosen. This makes it possible for the top-performing clients from the chosen group to have a bigger influence on the finished global model. Based on the scaling factors that were calculated in the previous step, the second stage scales each of the local model's weights. In the third and final stage, a global model is created by adding up all of the customers' scaled weights. The updated global model is therefore the end product of weighted average over the top-performing models among clients. Every FL cycle is transmitted to every client in the highway system.

Inadequate in real time operations. Centralized servers find it difficult to match user demands for real-time performance because of the vast number of computer resources needed to process the massive amounts of data generated in ITS. The speed of response is further decreased by the delay brought on by sending data to a distant server for processing and then sending the user's results back. Time-sensitive applications like autonomous driving and real-time traffic flow prediction may be severely impacted by this kind of delay.

The road network in the urban region, where the entire system is based, consists of four-leg intersections with two lanes on each road: a left lane for traffic traveling upstream and a right lane for traffic traveling downstream. RSUs, loop detectors, and traffic signals pointing down the lane at the intersection are installed in each lane. Every intersection has a light Controlling System (SCS) that modifies the length of time that different traffic phases take at different intersection traffic light locations. Based on the corresponding projected green phase time span that was obtained from the cloud, the phase time adaptation is performed. The user, cloud, and action subsystems make up the system (Fig.2). The RSUs in the user subsystem get real-time mobility data from vehicles, while the loop detectors gather the number of vehicles. The cloud storage platform's data manager (DM) module organizes collected data from loop detectors and RSUs into distinct datasets according to their respective sources. The cloud's time calculator (TC) component determines each vehicle's time to reach (TTR) the signal point down the lane. The list of shortest routes to the vehicles' destinations is determined by the cloud's navigator (NG) component. The cloud subsystem's data analysis (DA) component examines the relevant datasets to forecast the waiting line length and waiting time at the appropriate signal locations, which allows it to forecast the green phase time span.

RSUs obtain the following mobility data in the user subsystem: vehicle id, velocity, acceleration/deceleration, vehicle location, and travel destination. The data analysis (DA) component of the cloud subsystem predicts the waiting queue lengths, waiting times, and adaptive green phase time spans at the signal sites based on the corresponding datasets in the pre-decision data (PDD) and traffic flow data (TFD) repositories. Data security and privacy are issues that are brought up by the gathering and processing of traffic data in real time. Users' privacy may be at danger if sensitive data is accessed without authorization and is misused or exploited.

In Vehicle-to-Everything (V2X) applications, privacy-preserving [12] AI and edge computing provide several noteworthy benefits. Low latency is one of the key advantages of edge computing since it lowers the need for data transmission to centralized cloud servers, resulting in faster reaction times, which are critical for real-time applications. Furthermore, by processing data locally and utilizing privacy-preserving methods, these technologies improve data security and privacy by shielding sensitive information from unwanted access. Another significant benefit is scalability, which is made possible by edge computing's capacity to support distributed data processing. This enables the V2X system to grow effectively in response to the growing number of connected cars and gadgets. Additionally, processing data at the edge results in a significant reduction in bandwidth usage, which in turn reduces the quantity of data transferred via networks, thereby lowering bandwidth consumption and related expenses.

Data protection to the sensitive data, easy flow of traffic with less congestion and Original Equipment Manufacturers (OEMs) to the autonomous vehicles. The system optimizes traffic management through real-time predictive modeling, reducing network bandwidth and enhancing security by limiting the exposure of sensitive vehicular data. This framework supports scalable and efficient traffic management in future autonomous vehicle ecosystems.
 
, Claims:We Claim:

1. A privacy-preserving data processing system for Vehicle-to-Everything (V2X) communication, comprises:
a plurality of edge computing units;
a federated learning unit;
a real-time data manager module; and
a predictive modeling module,
wherein the edge computing units are positioned near data sources and configured to collect and process vehicular data including but not limited to vehicle speed, location, and route information in real time; the federated learning unit is connected to the edge computing units and configured to perform localized model training on collected data while aggregating model parameters in a centralized server to generate a global model, wherein the model aggregation is performed without transmitting raw vehicular data to the central server, thereby preserving privacy; the real-time data manager module is configured to organize and store data within the edge computing units, facilitating model training and data processing locally; and the predictive modeling module is integrated within each edge computing unit to optimize traffic control by analyzing processed vehicular data and generating output for traffic management.

2. The privacy-preserving data processing system as claimed in Claim 1, wherein the federated learning unit comprises:

a local model training module for each edge computing unit, configured to train local models on collected data independently; and
a model aggregation module within the centralized server, configured to compute the weighted average of selected top-performing local models based on their performance metrics, thereby updating the global model.

3. The privacy-preserving data processing system as claimed in Claim 1, wherein the predictive modeling module at each edge computing unit utilizes real-time data inputs to predict traffic flow, waiting times, and vehicle queuing lengths, enabling adaptive adjustments to traffic signals.

4. The privacy-preserving data processing system as claimed in Claim 1, wherein each edge computing unit is configured to limit data transmission to only aggregated model parameters, thus minimizing network bandwidth usage and reducing latency in data processing.

5. The privacy-preserving data processing system as claimed in Claim 1, wherein the system comprises a communication interface for data exchange between the edge computing units and the centralized server, wherein the communication interface enables secure transmission of model parameters and aggregated data through encrypted channels to ensure data security.

6. The privacy-preserving data processing system as claimed in Claim 1, wherein each edge computing unit is configured to perform data filtering to exclude unnecessary or redundant data, and local data storage management to retain only relevant data required for model training and traffic optimization.

7. The privacy-preserving data processing system as claimed in Claim 1, wherein the real-time data manager module organizes vehicular data into datasets categorized by their respective sources, enabling efficient handling and processing within each edge computing unit.

8. The privacy-preserving data processing system as claimed in Claim 1, wherein the federated learning unit is configured to perform iterative cycles of model aggregation, wherein each cycle includes computation of scaling factors for selected models based on relative local performance metrics, weighted scaling of each selected local model's parameters, and creation of a global model by aggregating scaled parameters, which is then redistributed to each edge computing unit.

9. A method for privacy-preserving data processing in V2X communication systems, comprising steps of:
collecting vehicular data in real-time at edge computing units located near data sources,
training local models at each edge computing unit using federated learning to preserve data privacy by restricting raw data transfer,
aggregating model parameters from local models at a centralized server to form a global model, and
utilizing predictive modeling on processed data at each edge computing unit for optimized traffic control outputs.

10. The method as claimed in Claim 9, comprising the steps of:
scaling model parameters based on local model performance before aggregation,
securing model parameter transfers between edge computing units and the centralized server using encrypted communication, and
implementing bandwidth conservation by transmitting only aggregated data across the network.

Documents