Predictive analytics of energy economy for electric mobility

• Autores: Román Michael Sennefelder
• Directores de la Tesis: Ramón González Carvajal (dir. tes.), Rubén Martín Clemente (dir. tes.)
• Lectura: En la Universidad de Sevilla ( España ) en 2024
• Idioma: inglés
• Número de páginas: 157
• Enlaces
  • Tesis en acceso abierto en: Idus
• Resumen

  • Mobility is one of the most fundamental necessities in society. Especially in urban environment, transportation based on conventional technologies causes traffic jam, pollution, greenhouse gas emissions and many other negative effects. To meet the demand for mobility and environmental respectively health goals alike, anticipating alternative drive systems is inevitable.

    Both private and public transportation still rely mainly on combustion engine vehicles. The adoption to electrified systems is complex, since they operate differently. Uncertainty about energy consumption and thus operating time and range, leads to conservative vehicle system design, which lasts in inefficiency, high costs and of the lack of trust into novel technologies.

    Optimized vehicles, a tailored infrastructure and a robust range prediction, reduce costs, save resources, encourage for new technology and thus can be a solution to this socioeconomic problem.

    There is a broad consense to use real-world data whenever vehicles energetics are to be calculated.

    Either for running vehicle driving simulations, for speed profile synthesis or for novel big data driven methods. Within this dissertation I introduce several different data driven approaches to improve accuracy and robustness of calculating the energy demand and it's prediction. Predictive analytics increase transparency in this field and facilitates design and operation planning of alternative fleets.

    Especially public transport is often organized heuristically making it prone to error and inefficiency.

    In the first study (chapter 2), a feasibility assessment for retrofitting a conventional diesel engine bus fleet in a hot country application, a methodology to innovatively synthesize speed profiles, auxiliary power demand and passenger variation to determine the energy demand of BEBs is introduced. We decribe a novel approach using bootstrapping, inverse transform sampling and empirical equations to synthesize these profiles and utilize them in a massive simulation framework. The confidence interval width of 95% likelihood for key parameters of interest such as operating time and range, overall energy consumption, recuperated energy or spatial rated consumption, is reduced and thus uncertainty from 10% at the beginning based on the real measurement data to less than 1% using the extended data.

    In the second study (chapter 3), we split speed profiles by a smart segmentation algorithm and introduce a feature space characterizing these profiles. The extracted meta data is used in a second-stage regression model to predict the vehicles' energy demand by statistical means from these driving characteristics - in particular using multiple linear regression. Achieving a prediction accuracy of more than 85% this simple approach turns out to be robust and precise alike.

    Starting from this baseline, in the next study (chapter 4), we analyze the postulated feature space more deeply. The extracted meta data is used in several machine learning models and a comprehensive sensitivity analysis regarding their relevance and predictive value is done. Considering features, like Shannon Entropy, that respect the temporal evolution of the velocity, we reach an outstanding prediction accuracy of up to 94% by Gaussian Process Regression.

    The last study (chapter 5) completes this preliminary framework as it is about predictive classification of vehicles' energy economy. We investigate a data driven approach based on a set of descriptive features to be used in different machine learning algorithms for exact identification and predictive classification.

    This generic approach targets the highest, strategic planning level for alternatively powered fleets and sustainable public transport. It supports decision makers to design and plan the operation of electric buses in urban environment properly on a solid base with a prediction accuracy of up to 90% and identification of up to 95%.