A noise-immune LSTM network for short-term traffic flow forecasting.

Area of Science:

• Intelligent Transportation Systems
• Machine Learning
• Traffic Engineering

Background:

• Accurate short-term traffic flow forecasting is crucial for intelligent transportation systems and traffic control.
• Existing forecasting methods face challenges due to the stochastic and evolutionary nature of traffic data.
• Conventional models often struggle with non-Gaussian noise, impacting prediction accuracy.

Purpose of the Study:

• To propose a novel noise-immune long short-term memory (NiLSTM) network for enhanced short-term traffic flow forecasting.
• To improve the robustness of traffic forecasting models against non-Gaussian noise.
• To demonstrate the superior performance of the proposed NiLSTM network.

Main Methods:

• Developed a noise-immune long short-term memory (NiLSTM) network.
• Embedded a noise-immune loss function, derived from maximum correntropy, into the LSTM architecture.
• Utilized a local similarity metric (maximum correntropy induced loss) robust to non-Gaussian noises.

Main Results:

• The NiLSTM network demonstrated superior performance compared to conventional and state-of-the-art models.
• Experiments on four benchmark datasets validated the effectiveness of the proposed method.
• The noise-immune loss function significantly improved forecasting accuracy in the presence of noise.

Conclusions:

• The proposed NiLSTM network offers a robust and accurate solution for short-term traffic flow forecasting.
• The integration of maximum correntropy induced loss enhances LSTM's resilience to noise.
• This advancement contributes to more reliable intelligent transportation systems.