Benefits of Mathematical Optimization and Machine Learning-Based Deterioration Modeling: A City of Calgary Case Study

Abstract
Calgary’s road network constitutes a major investment over many generations and plays a crucial role in the City’s well-being by guaranteeing its citizens with full accessibility, ensuring safe travel, and providing a strong business competitiveness through an efficient movement of goods and services. This study identifies key limitations in current pavement network life cycle cost analysis processes by comparing the results of traditional prioritization approaches to a true multi-year multi-constraint optimization analysis. The results shows that the optimization solutions outperformed prioritization at all years showing an average 5.3% improvement over the planning horizon and 9.3% by the end of the plan. Monetization methods also arrived at significant cost savings via added performance over a 10-year planning horizon by switching to a mathematically optimized solution. To further improve modeling accuracy and reliability of results, this study investigates the quality of performance models used within the pavement management system and discusses the development of machine learning-based deterioration models using decision tree regression. The effects of more modern performance modeling methods on investment planning is examined by comparing various optimization scenarios using both the ML-based and the traditional age-based deterioration models. The paper shows the importance of condition-based predictive modeling and integrating accurate performance models into the current asset management system to provide more accurate information on monitoring the network's life expectations, capital investment plans, and vulnerable communities with accelerated pavement deterioration patterns.