Construction and optimization of representative actual driving cycles based on the improved autoencoder

Area of Science:

• Engineering
• Artificial Intelligence
• Transportation Science

Background:

• Developing representative driving cycles is crucial for electric vehicle (EV) research.
• Existing methods may not capture localized driving patterns effectively.

Purpose of the Study:

• To propose a systematic method for constructing representative actual driving cycles from raw EV road test data.
• To enhance the accuracy and efficiency of driving cycle development.

Main Methods:

• Collected EV road test data using a manual driving method.
• Applied five-scale wavelet analysis for noise reduction and data smoothing.
• Utilized Gaussian Kernel Principal Component Analysis (KPCA) for dimensionality reduction.
• Developed the Changsha Driving Cycle Construction (CS-DCC) method.

Main Results:

• Noise reduction resulted in more stable and smoother driving data.
• KPCA selected 5 principal components, achieving an 85.99% cumulative contribution rate.
• The average error of characteristic parameters decreased from 13.6% to 6.1% (55.1% reduction).
• The constructed CS-DCC cycle exhibited distinct local characteristics compared to standard cycles.

Conclusions:

• The CS-DCC method effectively creates representative driving cycles reflecting localized driving patterns.
• The study highlights the necessity of localized driving cycle construction for accurate EV analysis.
• Artificial intelligence techniques offer powerful applications in advancing transportation engineering.