4. Research background
Background:
As accurate and timely traffic flow information is extremely important for traffic management,
traffic flow prediction has become a vital component of intelligent transportation systems. However,
existing traffic flow prediction methods based on centralized machine learning need to gather raw
data for model training, which involves serious privacy exposure risks.
8. 1. Fedrate Learning
The specific implementation of ideas
The goal of this paper is to achieve accurate and
timely TFP through a lightweight federated learning
scheme. Using GRU neural network as a lightweight
technology to predict future traffic flow and
implementing the application of FL in TFP systems.
9. Establish a dedicated consortium
blockchain and implement a decentralized
TFP system using FL. For the convenience of
system functionality, we first select a certain
number of RSUs as authorized miners. To this
end, upgrade the hardware configuration of
these RSUs to have powerful computing,
storage, and communication capabilities to
verify local model updates submitted by
distributed vehicles and identify low accuracy
and unreliable updates. Through carefully
designed consensus algorithms, they
generate a new block containing qualified
local model update records.
2. Distributed TFP Systems and Federated Learning
The specific implementation of ideas
10. Use a consensus algorithm called dBFT to execute the consensus process among miners; Only
qualified local model updates are aggregated to generate the latest global model.In this consensus
process, false or low-quality local model updates are not integrated into the global model training,
which improves the FL performance and reduces poisoning attacks. Therefore, the proposed
consortium blockchain with a dBFT-based consensus algorithm ensures the secure, reliable, and
privacy-preserving characteristics of the FL-based TFP method for ITS.
The specific implementation of ideas
3. Design of consortium blockchain
11.
12. step 1:
Initialization
Step 2:
Leader Selection
Step 3:
Local model training
Step 4:
Model update verification
The specific implementation steps of consensus algorithm
Step 5:
Candidate block
verification
Step 6:
Global model
training
13. The specific implementation of ideas
4.Local Differential Privacy Technology for Protecting Vehicle Communication Privacy in TFP
Definition 1: Participants should add random perturbations to the data. In order to protect the privacy of
vehicle location information in TFP tasks, a Gaussian mechanism is used to add carefully designed noise to
interfere with the location information. Firstly, define the Gaussian mechanism as defined in Definition 2:
Definition 2:Gaussian mechanism
Definition 3:Vehicle information after disturbance
17. Introduced a federated learning framework to protect the privacy of
vehicle data.
Applying GRU model to FL framework to obtain accurate TFP.
In order to prevent malicious attackers from damaging the urban traffic
flow management system, a decentralized FL framework has been
implemented using blockchain to defend against poisoning attacks.
Contrib
utions
Contribution of the paper
Using local differential privacy technology to protect privacy in vehicle
location sharing.
19. Conduct experiments and improve ideas, with a preliminary logic.
The second semester of the first year of graduate school
Complete your own experiment and prepare the preliminary draft of the paper.
The first semester of the second year of graduate school
Successfully issued the paper.
The second semester of the second year of graduate school
Successfully graduated.
Third year of graduate school
My own graduate plan
20. References
[1] Y. Lv, Y. Duan, W. Kang, Z. Li, F.Y. Wang, Traffic flow prediction with big data: a deep learning approach, IEEE
Trans. Intell. Transp. Syst. 16 (2014) 865–873.
[2] R. Vinayakumar, M. Alazab, S. Srinivasan, Q. Pham, S.K. Padannayil, K. Simran, A visualized botnet detection
system based deep learning for the internet of things networks of smart cities, IEEE Trans. Ind. Appl. 56 (2020)
4436–4456.
[3] X. Yang, et al., Deep relative attributes, IEEE Trans. Multimed. 18 (2016) 1832–1842.
[4] M.S. Hossain, M. Al-Hammadi, G. Muhammad, Automatic fruit classification using deep learning for
industrial applications, IEEE Trans. Ind. Inf. 15 (2019) 1027–1034.
[5] L. Lyu, J. Yu, K. Nandakumar, Y. Li, X. Ma, J. Jin, H. Yu, K.S. Ng, Towards fair and privacy-preserving federated
deep models, IEEE Trans. Parallel Distrib. Syst. 31 (2020) 2524–2541.
[6] K. Cho, B. Van Merriënboer, C. Gulcehre, D. Bahdanau, F. Bougares, H. Schwenk, Y. Bengio, Learning phrase
representations using rnn encoder– decoder for statistical machine translation, 2014, arXiv preprint arXiv:
1406.1078.
21. TH
A
N
KS
2 0 2 4 年 03月
T h a n k y o u f o r w a t c h i n g