An Analysis On The Application Of Machine Learning In Road Maintenance Policy

In the science of machine learning (Machine Learning), the issue of designing machines that learn by using the examples given to them and their own experiences is discussed. In fact, in this science, an attempt is made to design a machine in such a way that it can learn and act without explicitly planning and dictating each and every action by using algorithms. Road maintenance operations are very important in preserving and preventing their premature deterioration and using the maximum capacity of the road in its expected service life which needs to be done according to a clearly defined plan and according to the needs and problems of the road. In this paper, a framework for finding optimal maintenance policies in a road network was proposed. This framework included: identification of effective factors in policy making, network clustering based on these factors, identification of criteria influencing optimal policies and determining optimal policies and periods using these criteria. To test the applicability of the framework, it was applied step by step in the road network of Iran. To quantify the cost of policies, seven machine learning algorithms including gradient boosting regression, lasso, ridge, random forest regression, elastic network, neural network and multiple linear regression were tested. Using the coefficient of determination as an accuracy metric, it was found that in all subnets, the gradient boosting regression has the highest accuracy in the test set while the other algorithms are between ۵۰% and ۹۰%. The conditions of the sub-grids were modeled using the Markov chain model and measured by the average pavement condition index (PCI). Having the cost of the policies and the PCI of the subnets, the optimal policy was selected using the technique of priority ordering based on similarity to the ideal solution (TOPSIS). It was concluded that the maintenance period of four years was optimal in all sub-networks. Roads in warm regions require the most stringent policies, followed by roads in cold and wet regions. The same applies to arterial roads followed by local roads.

کلیدواژه ها:

Maintenance Cost Optimization ، Network PCI ، Gradient Boosting Regression ، Markov Chain Model