Data-driven solutions for transport policy interventions using machine learning and optimisation methods

Abstract

The use of data-driven approaches, machine learning techniques, and optimisation methods in transport policy-making and the subsequent implementation of policy commitments has seen a substantial growth in recent years. The potential benefits of big data in transportation are significant, but the challenge lies in extracting knowledge from data to inform policy design, implementation, and validation. There is a significant gap in the literature on the application of machine learning and optimisation methods applied for policy validation and its implementation in transportation. The aim of this PhD project it to study the potential of data-driven techniques for analysing and validating the objectives of policy interventions, and implementing policy commitments in the transport arenas. To achieve the aim of this PhD research, the following research questions are specifically addressed: (a) Given the large volume of data gathered from the transportation network, how to find data types that are relevant to a policy objective? (b) What machine learning techniques are suitable for combining large datasets, processing the data, and validating a policy objective? (c) Can these large dataset techniques be integrated in the implementation of policy commitments? The study’s methodology involves identifying relevant data types for the proposed policy objectives, selecting appropriate machine learning techniques for processing data and validating the policy objectives, and determining the potential use of these techniques for policy commitment implementation. Two frameworks have been designed to tackle the specific challenge of finding datasets related to the policy objective and validating policy interventions using machine learning techniques. A third framework have been designed for finding the best implementation of policy commitments using multi-objective optimisations. The term ‘framework’ is used since the proposed approaches are high level and flexible, and can be applied to different policy objectives. The details and the choice of machine learning models can be decided depending on the specifics of the policy objective. The study focuses on two case studies aimed at improving air quality and reducing greenhouse gas emissions, which are essential components for meeting the UK’s target of achieving net-zero emissions by 2050. Datasets from the Newcastle Urban Observatory and open-source datasets gathered from the industrial Case-funding partner of this PhD, Arup, is electric vehicles (EVs). The objective of the clean air zone policy is to reduce exposure to harmful levels of NO2, while the most important policy commitment of transitioning towards electric vehicles is to expand the electric vehicle charging infrastructure, ensuring that the EV charging infrastructure meets the demand of users. Two data-driven approaches are employed, including machine learning models for policy objectives and the use of simulation in combination with optimisation for implementing policy commitments. By leveraging these advanced techniques, this research aims to provide valuable insights for policymakers, helping them make more informed decisions when planning and implementing transportation policies. In the first case study, common machine learning classifiers are used, which include Decision Tree, K-Nearest Neighbours, Gradient-Boosted Decision Trees, and Light Gradient- Boosting Machine (LGBM). It is shown that the constructed models share common conclusions about the importance of features in predicting NO2 concentrations with LGBM performing best in capturing the relations in the dataset with accuracy 88%. Subsequently, historical data is used to model air quality in Newcastle upon Tyne both assuming with and without the implementation of the clean air zone. The long short-term memory model is used to predict the NO2 concentration with root mean square error of 0.95. The approach shows the use of machine learning methods in analysing and validating the objectives of interventions in transportation systems. The role of machine learning can be summarised as predicting what is going to happen in the future if the policy is not implemented (using available historical data), and predicting the air quality and other related variables using transport behaviour changes in response to the implemented policy. The second case study is the expansion of the EV charging infrastructure of Newcastle upon Tyne, UK. An optimisation model is developed to estimate and optimise the charging points types, charging points quantity, charging points locations, total expenditures, and utilisation of charging points for four different future energy scenarios. Quantitatively, the optimal solutions recommend installing higher number of faster charging points to reduce the percentage of slower charging points from the current 60% to around 25% in the four scenarios. Still, the optimal solutions put priority on the slower charging points (around 25%), with faster charging points having smaller portions each around 10%-13%. The optimisation shows that while 7kW charging dominates the market currently, it is more beneficial to improve charging efficiency and reduce investment costs by having a higher percentage of installations from other types of charging points in the future installations. The results also illustrate the spatial distribution of charging points, with higher concentrations in urban areas and near major roads. This PhD research contributes to the body of knowledge on using quantitative methods to validate the objectives of policy interventions and implement policy commitments in transportation. The findings will provide policymakers with valuable insights to make more informed choices to improve transportation systems. This research provides an opportunity to explore the benefits and challenges of using data-driven approaches, machine learning techniques, and optimisation methods to improve transportation planning and policy-making. The study’s methodology and results will be significant for policymakers, stakeholders, and researchers interested in using quantitative methods to improve transportation systems.