Research On Intelligent Urban Resource Optimization Methods And Applications

With the rapid development of urbanization,more and more people choose to work and live in big cities.In contrast to the massive citizens in a city,the resources in a city,however,are very limited.With the limited resources,how to provide better work and life services for citizens has become a big challenge.Urban resource intelligent optimization is an effective way to tackle this challenge,which aims to improve the efficiency of the limited urban resources such that they can provide better services for citizens.To this end,this thesis aims at studying the urban resource intelligent optimization problem.Specifically,based on the spatio-temporal properties of the urban big data,this thesis studies the spatio-temporal operations optimization,graph theory,data mining,machine learning and reinforcement learning,and designs urban resource intelligent optimization methods such that the efficiency of the limited urban resources can be maximized.According to the intrinsic properties of urban resource intelligent optimization problems,they are divided into two classes,one as static urban resource intelligent optimization and the other as dynamic urban resource intelligent optimization.The designed urban resource intelligent optimization methods are successfully applied to four specifical problems:? Static urban resource optimization methods for mobile crowdsensing: Mobile crowdsensing is an important way of collecting urban big data.Thus,in this thesis the static urban resource intelligent optimization problem in mobile crowdsensing is firstly studied.That is,given the limited budget(resource)and users' mobile information,how to recruit users and how to design proper data collection tasks for recruited users such that the quality of the data collected is best? To this end,a novel mobile crowdsensing framework based on human mobility is proposed,which can collect balancing urban data even if the budget is limited and human mobility is unbalancing.Specifically,the proposed framework contains three components: a hierarchical entropy-based data coverage index to measure the quality of the data collected,a data point graph-based data collection task design method,and an effective participant recruitment method.Based on the real mobility information of 34 users,extensive experiments are conducted.Comparing with existing methods,the proposed framework can collect data with higher quality under the limited budget.? Static urban resource optimization methods for food delivery task grouping: Ordering food online has become a new lifestyle for people living in cities.With only limited food carriers,how to efficiently deliver the massive food orders is a vital problem to food ordering platforms.Thus,based on the real food order data in history,this thesis designs an effective graph-cut-based food delivery task grouping method,aiming at improving the delivery efficiency of the limited food carriers.First,according to the historic food delivery data,a food delivery task graph is constructed,in which a food delivery task is an edge.The problem can be then formulated as a graph-cut problem.Next,based on the analysis of the problem,an objective function is developed to better guide the food delivery task grouping.Finally,an effective graph-cut algorithm is designed to do the food delivery task grouping.Based on the data from a real-world food ordering platform,extensive experiments are conducted.Comparing with existing methods,the proposed method can reduce 16% of the average delivery time for each food order.As a result,the proposed method can significantly improve the food delivery efficiency of the limited food carriers.? Dynamic urban resource optimization methods for ambulance redeployment: Ambulances are important resources for the Emergency Medical Services(EMS)center in a city,which protect citizens' lives by quickly transporting them to hospitals.Dynamic ambulance redeployment is able to improve the efficiency of ambulances in a city.This thesis provides a data-driven dynamic ambulance redeployment method,which has taken five types of data into consideration.The proposed method has two steps.First,a safety time-based urgency index is designed to measure the urgency degree of each ambulance station.Next,based on each station's urgency degree and the status of other occupied ambulances,a two-stage operations optimization method is proposed to do the dynamic redeployment for each ambulance.Experimental results based on data from the EMS center in Tianjin city show that comparing with prior methods,the proposed method can reduce 35% of the avarage pickup time for each patient,and improve the ratio of patients picked up within 10 minutes from 68.4% to 80.3%.Thus,the proposed method can significantly improve the efficiency of the limited ambulances in a city.? Dynamic urban resource optimization methods for taxi route recommendation: Taxis are one of the most important and most frequently used public transportation modes in real life.With the massive riding demand of passengers,improving the transportation efficiency of the limited taxis is important.Dynamic taxi route recommendation,i.e.,recommending routes to vacant taxis such that by following the routes taxis can quickly find next passengers,provides an effective way.Thus,in this thesis,the dynamic taxi route recommendation problem is modeled as a sequential decision making problem,and an effective method is designed in order to improve the efficiency of the limited taxis in a city.First,multiple related spatio-temporal features are considered and extracted,which can well indicate the probablity of taxis finding passengers on each route.Next,a deep reinforcement learning method is designed to learn a deep policy network,which is carefully designed to fuse the extracted spatio-temporal features of each route such that the best route can be recommended.Experiments based on data in San Francisco and New York demonstrate that comparing with prior methods,the proposed method can improve 42.8% of average revenue for taxi drivers and reduce 44.4% of average waiting time for passengers.That is,the proposed method has significantly improved the efficiency of the limited taxis in a city.