Construction of Sightseeing Vehicle Driving Cycles Using Autoencoder-based Dimensionality Reduction and Optimized K-clustering

To address the complexity and diversity of sightseeing vehicle driving cycles, which are challenging to accurately characterize using traditional methods, an innovative framework integrating autoencoder-based dimensionality reduction and optimized K-means clustering is proposed. The framework processes real-world driving data by segmenting it into kinematic segments. An autoencoder is employed to reduce dimensionality, mitigating the impact of non-linear features while preserving core information from the original data. The reduced-dimensional data is then clustered using K-means, with the number of clusters and distance metrics optimized via a cosine similarity algorithm. Based on experimental requirements, the driving cycle curves for sightseeing vehicles are constructed. The effectiveness and practicality of the proposed method are validated by comparing the feature parameters of the constructed driving cycles with those of actual driving data. The results show an average relative error of 3.74%, demonstrating the robustness and applicability of the innovative framework.