Synthesis and Feature Selection-Supported Validation of Multidimensional Driving Cycles

Abstract

This paper presents the synthesis and validation of multidimensional driving cycles represented by vehicle velocity, vehicle acceleration, and road slope profiles. For this purpose, a rich set of city bus driving cycles has been recorded. First, a Markov chain model is established based on velocity, acceleration, road slope and road slope time derivative states. Next, a large set of synthetic driving cycles is generated by using a corresponding 8D transition probability matrix, which is implemented in a sparse form based on a dictionary of keys to improve computational efficiency and reduce memory requirements. In support of synthetic driving cycles validation, a number of time- and frequency-domain statistical features are considered, including unique cross-correlation velocity–acceleration–road slope indicators. To predict fuel consumption related to synthetic driving cycles, an accurate neural network model is introduced which uses a fixed 3D histogram of counted discrete velocity, acceleration, and road slope inputs. The significance of each nominated statistical feature and its impact on fuel consumption is revealed by means of linear regression modelling and least absolute shrinkage and selection operator (LASSO) feature selection method. A model having only several most significant features as inputs and fuel consumption as output is proposed to be used for unambiguous single-criterion validation of synthetic driving cycles with respect to recorded ones. Finally, the proposed validation approach is verified against a widely used method relying on minimization of statistical feature deviations with respect to true values. The results point out that, by applying the proposed synthesis and validation method, it is possible to extract most representative synthetic driving cycles in a straightforward and computationally efficient way. The main anticipated applications include various simulation-based analyses that require representative synthetic driving cycles and/or accurate vehicle energy consumption predictions.