AU2020101613A4 - Collaborative federative learning approach for internet of vehicles traffic prediction - Google Patents

Abstract

COLLABORATIVE FEDERATIVE LEARNING APPROACH FOR INTERNET OF VEHICLES TRAFFIC PREDICTION ABSTRACT This invention is to deploy a collaborative federative learning approach for the internet of vehicle traffic prediction. Though there were many traditional and some intelligent technical approaches were deployed in traffic prediction, the concern about privacy and security, where the client is uncomfortable in sharing the information globally, creates the need to deploy a collaborative federative learning approach. In this application, the internet of drones is deployed to gather information regarding the vehicle location, which is configured with the internet of vehicle technology. Continuous image capturing in the area of traffic congestion is monitored. In this method, the training set is modeled in local devices itself. A deferred acceptance algorithm is deployed between local devices to model parameters and its associated device. The desired information or the model parameters alone is then sent to the cloud through the gateway. This is one of the easiest, fast computing methods to enable the prediction of traffic congestion with privacy and less storage consumption. 1| P a g e COLLABORATIVE FEDERATIVE LEARNING APPROACH FOR INTERNET OF VEHICLES TRAFFIC PREDICTION Drawings Collaborative Federative Learning with Matching technique Fig. 1 Collaborative Federative Learning with matching technique with associated drones. 1 P a g e

Description

COLLABORATIVE FEDERATIVE LEARNING APPROACH FOR INTERNET OF VEHICLES TRAFFIC PREDICTION

Drawings

Collaborative Federative Learning with Matching technique

Fig. 1 Collaborative Federative Learning with matching technique with associated drones.

1 Pa g e

COLLABORATIVE FEDERATIVE LEARNING APPROACH FOR INTERNET OF VEHICLES TRAFFIC PREDICTION

Description

Field of the Invention.

The Field of invention is related to collaborative federative learning approach.

The need for vehicle rise with the population as there is a necessity of a source for movement. As there is an enormous count of vehicles on road, especially during peak hours, it always ends up in traffic which affects the time management of busy schedule of the people. It creates an importance to design a traffic management of vehicles to assist people who are working on busy schedule and those keep travelling often in their business where time and money matters for them. So this invention brings about internet of things in vehicle traffic prediction. Internet of vehicles usage needs a server to hold the data to be processed. Considering the privacy of the clients, time management and storage space, a collaborative federative learning approach is deployed, to share the information that were the outcome of the matching approach of local devices.

Background of the invention.

For communication, especially involving trade and movement of people, transport plays a significant importance. It relates to the growth of economic and civilizations. Especially, the increase in land transport increases tremendously with the increase in population. In 2 0 ' century, there would be one vehicle at home but in this 21st century, each and every individual owns a vehicle and it leads to traffic congestion. So, there is a need to manage the traffic for both stationary and the moving.

1| P a g e

Initially, traffic controller centers were established to control the traffic in any location. But it involved manual mode of operation. But it needed the man power support working all day. Later, it was automatized with some timing set to enable the traffic flow without congestion. But even the automated traffic control has a drawback, since there may be excess delay where the signals follow a fixed time.

Later, intelligent systems designed to manage the traffic during the congestion. Especially, self-adaptive traffic management system designed to adjust the timing parameters based on the live traffic status in the junction. Since the time parameter is a fluctuating parameter very often continuously, there is a need to design a system to handle the varying timing parameter signal.

To gather all the necessary information regarding the road traffic status around the vehicle location area, sensors were used in all angles. But that has been improved with new technology of involving drone technology to capture all the information.

Earlier, vehicle to vehicle network with mobile internet was deployed. But to increase the efficiency and make it robust, internet of things was introduced. And now the internet of vehicles introduced, requires a server like cloud to store a large data to process and make predictions, so that the traffic can be controlled in minimum time consumption.

In a cloud, all the information that has been gathered will be moved and it consumes storage space. All the image data and the model data has to be moved to the storage space through a gateway. So, instead of moving all the datas, federal learning approach deployed which moves only the desired model parameters to the cloud.

More importantly, not all the clients will be comfortable in sharing their data and location globally. While considering their privacy, a collaborative federative learning approach is deployed, where the datas will be in the client devices only. The training data is modelled in the devices itself by 2|Page performing matching approach with associated devices. Only the matched desired dataset only transferred to cloud. This reduces the time and increases the efficiency of the system providing good trusted system.

Objects of the Invention

With the increase in demand of usage of vehicles for the increasing population, there is a necessity to design a management system to handle the traffic that will make ease of transportation for clients in frequent travel who are dealing with trade. For instance high designated person while travelling or involving any money movement between banks, require more concern about the privacy. So there is always a need to predict the traffic within certain time and make ease of the transportation. This invention uses an internet of vehicle with collaborative federative learning technique of storing the data within the device itself. By using the deferred acceptance algorithm, a matching method deployed with the associated client devices and the training set is modeled. Once the desired model training set obtained it is sent to the cloud so that it can be processed and used for prediction instead of moving all the datas and model parameters to the cloud.

Summary of the Invention

Though there are a number of techniques for traffic prediction, the internet of vehicle technologies has its own significance. Due to the increased traffic congestion, it is always recommended to design a process to predict the traffic and give proper information within a given duration with high security and privacy with minimum cost. The machine learning of modeling the parameters and processing the data through the cloud consumes excess data storage since all the model parameters and the continuous image data captured from many sources gets moved to the cloud in a centralized technique. But this has been overcome by deploying a federative learning, where the devices can model machine learning training sets locally. But considering the security and privacy of client devices that were not willing 3 1P a g e to share the datas, the internet of vehicles adopts a collaborative learning technique of performing the modeling of client devices with their associated client devices and then transfers only the desired data or the model parameters to the cloud. This deploys a deferred acceptance algorithm which can solve matching problem and can give optimized solution. Now from the results that were obtained, the end user can make predication for choosing the route which has traffic congestion free or less and the traffic management is done in minimal time and cost with reliable security.

Detailed Description of the Invention

Fig.1 shows the traffic prediction for internet of vehicles using collaborative federative learning approach. The figure shows that there are many sources of drones that capture the images of the traffic congestion. This approach uses drone sensors to capture the data. This is applicable for internet of vehicles where there is a need to get the update of the traffic in a location and find an alternate route to reach the destination. This is especially applicable for clients who do not want to reveal their location and require secure login to get the prediction of vehicle traffic. Usually, any image captured by sensors, it will be directed to the cloud server directly. Even it is a continuous monitoring and the values are repeated, it gets stored in the data storage location in cloud server through a gateway. But to reduce this data storage of huge information, each and every individual device if configured for collaborative federative learning technique, the training set are modeled with information captured in the device itself and the devices communicates with associated devices and compute using a matching method deploying deferred acceptance algorithm. The result of the matching method, which will be an update of the model parameter alone gets updated in the cloud server through the gateway. Only minimum model parameter from the obtained dataset is stored, that will occupy less storage space and performance of the system is greatly improved. It provides fast computation and better security with privacy as all the datas are not shared with the server. From the update of the desired parameter, the user can make prediction regarding the traffic and can plan the route of destination.

41Page

Fig.2 shows the source of capturing the images of the location of the traffic congestion. In any traffic signals, drones with sensors are placed to capture the images of the traffic congestion and the model parameters are obtained by modeling the training set with the test set of data in the client devices itself. Not only the parameters are modeled with the client devices, it also models parameters with the associated devices. In the figure.2 there are many drones to monitor a signal. There are two locations. One location is examined by drone 1, 2, 3 and 4. Another location monitored by drone 5, 6, 7 and 8. Only drone 4, 6 and 8 is connected to the server directly. Drone 1, 2, 3 and 4 perform modeling of parameters within their associated devices. Similarly, drone 5, 7 and 8 perform modeling of parameters within their associated devices. Only the update is set to the server.

1P a ge

6|Page

Claims (5)

COLLABORATIVE FEDERATIVE LEARNING APPROACH FOR INTERNET OF VEHICLES TRAFFIC PREDICTION CLAIMS I/We Claim:

1. High speed mobile wireless network to deploy a communication of client devices with the associated devices and also to move the desired model parameters to a gateway that was obtained from the matching method.

2. Appropriate sensors via the internet of drones (loD) deployed to capture the traffic congestion data.

3. A proper network topology designed to perform the collaborative federative learning technique.

4. A deferred acceptance algorithm to perform a matching method between client devices and the associated devices to determine the desired model parameter to be processed for prediction.

5. Data storage like cloud-deployed to collect the matched training set that has been modeled in local client devices with the associated client devices.

1| P a g e

COLLABORATIVE FEDERATIVE LEARNING APPROACH 02 Aug 2020

FOR INTERNET OF VEHICLES TRAFFIC PREDICTION

Drawings 2020101613

Fig. 1 Collaborative Federative Learning with matching technique with associated drones.

1|Page

Fig. 2 Drone capture in traffic congestion

2|Page