Research On Multi-type Vehicle Scheduling Problem In Open-pit Mine Based On Reinforcement Learning

Logistics transportation accounts for a high cost in the production of open-pit mine enterprise.As the main mode of logistics transportation,vehicle scheduling usually has many characters of limited vehicle allocation,complex transportation tasks and strict transportation requirements.And the comprehensive scheduling of multi-type transport vehicles with different loads and different speeds is complex,resulting in extremely difficult challenges for optimizing the scheduling of open-pit mine vehicles.Thus,there are some problems open-pit mine enterprises urgently need to solve,including how to plan proper driving routes of vehicles under current resource conditions;improve the logistics scheduling ability of enterprises;enhance efficiency;reduce cost and increase profits.In this thesis,the research on multi type vehicle scheduling problem in open pit production is carried out.Based on the analysis of practical problems,the mathematical model of vehicle scheduling optimization is established,and different improved algorithms are designed to solve the model,so as to improve the vehicle scheduling capacity of open pit mines,reduce transportation costs and improve transportation efficiency.The main contents,contributions and salient features are summarised as follows:(1)According to the characteristics analysis of the multi-type open pit vehicle scheduling problem,in view of the difficulty of multi-variable and complex constraints,a mathematical model aiming at minimizing the traveling cost of trucks is established,which is solved by solver.The validity and rapidity of the model are verified by numerical experiments on different scales.It is concluded that the model can get the optimal solution on a small scale,but it takes a long time to solve the large and medium scale cases,and even cannot get the optimal solution.In contrast,when solving the vehicle scheduling problem of the same type of open-pit mine,the optimal solution can also be obtained in a relatively short time for large-scale examples,which highlights the difficulty of multi-type trucks in the vehicle scheduling problem of open-pit mine.(2)Analyzing the characteristics of real-time dispatching of mining shovel path in open-pit mine vehicle dispatching in actual production,designing the state space,action space,state transition and reward-punishment function of the system,establishing a continuous-time model,and solving the model by using the linear value function approximation method based on Q-learning.Comparing the results with the greedy algorithm,the experiment shows that the linear function approximation method based on Q learning can effectively solve the vehicle scheduling problem in large-scale open-pit mines.(3)The deterministic strategy gradient method is used to solve the multi-type open pit vehicle scheduling problem,and the improved deterministic strategy algorithm is constructed by combining the sampling method based on priority experience pool.The numerical experiments verify the superiority of the improved deterministic strategy algorithm.Compared with the linear approximation function,the improved deterministic strategy algorithm has faster convergence speed and can obtain a better scheduling scheme for large-scale multi-type open-pit mine vehicle scheduling problem.(4)This thesis designs and develops a multi-type truck dispatching decision support system for Open-pit mines.The system can meet the needs of enterprises for information management of production equipment,dispatching information allocation and vehicle dispatching optimization,formulate reasonable dispatching schemes,and realize efficient and low-cost use of trucks for ore transportation.