The subject matter described herein relates to delivery of vehicles to various destinations and, more specifically, to systems and methods to optimize loading of vehicle haulers and their interdepot transportation routes.
Vehicle logistics services (VLS) involve transporting vehicles to dealers or distribution centers for sales or subsequent processing. Considering the tens of millions of vehicles sold in the U.S., a conservative estimate of $500 per vehicle for the transport renders the approximated expenditure billions of dollars per year. The vehicles are transported using hauler trucks and the truck loading scheduling is a rather challenging task due to multiple restrictions in loading (dimensions, weight, time, vehicle destination proximity, etc.).
One major disadvantage of conventional loading operations is their heavy reliance on human operators' decisions at the scene. The use of human operators in loading decisions has led to frequent unutilized hauler bed spots and high variations in loading efficiency across multiple delivery partners. Further challenges arise when determining the routes to deliver the vehicles whose destinations sometimes are at different locations. As a result, these human driven routing choices have detrimental impacts on the distance and transit time to deliver the vehicles.
Disclosed is a computer-implemented method and system to optimize loading of vehicle haulers and their transportation routes. In a generalized method, the system obtains a request to deliver vehicles to a destination. The system obtains one or more constraints of the hauler that will be used to deliver vehicles. The constraints may be dimensions of the hauler, weight tolerances of the hauler, road height limits, load balancing requirements, layer spacing of the hauler, legally allowed driving hours, or other constraints. Traffic patterns associated with the destination are then determined. Based upon the constraints and traffic patterns, a loading plan or route for the vehicle hauler is determined. The system then determines an estimated time of arrival for the vehicles to the destination(s), according to the loading plan and/or route.
This Summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to limit the scope of the claimed subject matter. A more extensive presentation of features, details, utilities, and advantages of the disclosed system, as defined in the claims, is provided in the following written description of various embodiments of the disclosure and illustrated in the accompanying drawings.
Illustrative embodiments of the present disclosure will be described with reference to the accompanying drawings, of which:
Illustrative embodiments and related methods of the present disclosure are described below as they might be employed in a system and method to optimize loading of vehicle haulers and to optimize their transportation routes. In the interest of clarity, not all features of an actual implementation or methodology are described in this specification. It will of course be appreciated that in the development of any such actual embodiment, numerous implementation-specific decisions must be made to achieve the developers' specific goals, such as compliance with system-related and business-related constraints, which will vary from one implementation to another. Moreover, it will be appreciated that such a development effort might be complex and time-consuming, but would nevertheless be a routine undertaking for those of ordinary skill in the art having the benefit of this disclosure. Further aspects and advantages of the various embodiments and related methodologies of the disclosure will become apparent from consideration of the following description and drawings.
As will be described below, illustrative methods and embodiments of the present disclosure optimize the loading of vehicle haulers and their transportation routes for faster, more efficient delivery. The present disclosure provides a systematic, optimization-based truck loading and dynamic routing system to increase the vehicle loading efficiency and routing performance with standardized operational procedures. The illustrative systems will explicitly consider various loading constraints (e.g., dimension, weight, etc.), legally allowed daily drive hours and vehicle destination proximity, and generate a loading plan that minimizes the number of trucks to be used. The disclosed embodiments will also incorporate real-time and predicted traffic information and generate an optimal route to deliver the vehicles to multiple destinations.
The system also generates a modeling framework for a unified truck loading and dynamic routing problem by incorporating various loading constraints (e.g., maximum weight, load balancing, layer spacing), road height limits, and legal daily driving hours per day. The system then employs machine learning to develop a recurrent neural network (RNN) for accurate traffic forecasting by considering the spatio-temporal dynamics of the traffic. The traffic forecast will be exploited in a chance-constrained mixed integer programming (MIP) problem for the truck loading as well as in the dynamic routing optimization. To handle last-minute changes in vehicle deliver orders, the system also utilizes robust and efficient solvers using heuristics as well as the structural properties of the optimization problem. Accordingly, the disclosed methods will reduce the number of trucks used, which will consequently lower the transportation costs as well as greenhouse gas emissions.
Accordingly, there is a pressing need to develop a systematic approach to improve the truck loading efficiency and routing performance, which will reduce the number of truck haulers used and mileages during transit, and consequently lessen transport cost and green gas emissions. In view of this, the present disclosure provides methods and systems to improve the lead time (time between vehicle tendered and delivery to customer), as well as reduce the workload of site truck loading planners.
The illustrative embodiments described herein are innovative in at least the following aspects. First, certain illustrative systems consider the truck loading and dynamic routing in a unified optimization framework to maximize the loading efficiency, the utilization of drivers, and routing performance while explicitly considering various loading constraints (e.g., dimension, load, height etc.) and the legal daily driving hours. This unified framework provides optimal truck loading and driver assignment scheduling simultaneously. Secondly, the system utilizes a RNN framework to forecast road traffic during transit that incorporates the spatio-temporal dynamics of the road traffic. Factors such as time of the day, day of the week, weather, and traffic flow from neighboring traffic grids are inputs in the RNN model.
Unlike existing traffic predictions from commercial Apps (e.g., Google Maps) that use historical traffic statistics, the disclosed methods augment the forecast with dynamic traffic features such as weather conditions and recent traffic conditions from neighboring grids. This forecast will greatly facilitate the dynamic routing of the hauler truck to deliver the vehicles to different destinations. Further, the forecast will also be incorporated in the optimization problem to explicitly tackle the transit time uncertainty as a chance constraint. This, in turn, will result in less conservative results and better efficiency as compared to using average or worst-case transit time (as in conventional approaches). Last, the methods utilize heuristics to efficiently and robustly solve the high-dimensional chance-constrained MIP problem. Mixed integer quadratic programming (MIQP) is then used to reformulate the problem and decompose the problem into substructures so the algorithm can quickly adapt to local changes without resolving the entire problem every time. These local changes may be changes to delivery requests, changes in traffic patterns, etc. Collectively, these advances will result in a truck loading and dynamic routing system that standardizes the operation procedure with improved efficiency for significant reductions in transportation cost and green gas emissions.
In the following description is separated into five sections: 1) formulation of the unified truck loading and dynamic routing problem (Section I); 2) development of a machine-learning based transit time prediction based on spatio-temporal characteristics (Section II); 3) development of an optimal dynamic routing with a time-expanded network (Section III); 4) development of an efficient and robust MIP solver by exploiting heuristics and the structure of the optimization problem (Section IV); and 5) system prototype and case study on San Antonio to Houston (or other routes with similar volume) vehicle dispatch (Section V).
Consider a Toyota vehicle manufacturing plant which receives a set of orders for the upcoming days. For each day, the order consists of a list of vehicles, Vi, i=1, 2, . . . , N, which need to be transported to another depot or distribution center no later than a specific lead time. Each vehicle can be characterized by a tuple Vi={wi, li, hi, di,Di}, where wi, li, hi, di, and Di represent the weight, length, height, days to the lead time, and its destination ID, respectively. After the vehicles are manufactured, they will be first parked in a parking yard and wait to be loaded by the truck hauler. The vehicles are parked in lanes and each lane will be loaded into a hauler for transport. Each lane will be served by a truck hauler in a first-in-first-out (FIFO) manner. The vehicles in the same truck may have different destinations and it is desirable to load vehicles of close destinations in one truck to shorten the travel time.
In the optimization methods described herein, several constraints are respected in the truck loading. First of all, the truck has a maximum weight capacity W. Therefore, the total weight of the vehicles loaded in one truck cannot exceed W. The truck has two axles (front and rear) that can support the load and there are restrictions on the maximum weight each axle can support. In addition, for safety reasons the load has to be well spread out over the truck and must be appropriately distributed and/or balanced between the two axles. This constraint can be enforced by ensuring that the center of gravity of the load is between the front and the rear axles. In certain illustrative embodiments, this constraint may be a specified range of threshold balance tolerances or a specific load balance variable. Furthermore, each layer of the hauler has to provide enough spacing between vehicles to avoid damage due to severe vibrations. Also, each column is restricted to a height maximum to follow road height limits. After the truck loading plan is done, each hauler will be assigned to a driver to drive the truck. The drivers have to obey the law of maximum hours allowed to drive continuously as well as in a day.
The objective of the illustrative truck loading and driver allocation optimization methods described herein is to minimize the number of trucks used to transport the vehicles to the distribution center as well as minimize their overall travel distance while satisfying the aforementioned constraints.
min J=Σj=1Kyj+μΣj=1KyjDist(Vj)
s.t. Σ
i=1
NΣm=18xijmwi≤yjW, ∀j=1, 2, . . . , K (Gross weight)
Σi=1NΣm=14xijmwi≤yjW1, ∀j=1, 2, . . . , K (Front weight)
Σi=1NΣm=14xijmwi≤yjW2, ∀j=1, 2, . . . , K (Rear axle weight)
Σi=1NΣm=14xij(2m-1)li<yj,L1, ∀j=1, 2, . . . , K (Lower layer spacing)
Σi=1NΣm=14xij(2m)li≤yjL1, ∀j=1, 2, . . . , K (Upper layer spacing)
Σi=1N[xij(2m-1)hi+xij(2m)hi]<yjH, ∀j=1, 2, . . . , K, ∀m=1, 2, . . . , 4 (Height)
y
jσ1≤Σi=1NΣm=18xijmwiam/Σi=1NΣm=18xijmwi≤wi≤yjσ2, ∀j=1, 2, . . . , K (Balancing)
Pr[Ej=1KΣm=18xijmt0j+Tr≤di]≥p, ∀i=1, 2, . . . , N (Lead time)
Pr[Σj=18zjk(tk+T*Σi=1Nxijmdi)≥Tr]≥p, ∀k=1, 2, . . . , K (Allowable Hours)
Σi=1NΣm=18xijm=1, ∀i=1, 2, . . . , N (Vehicle Assignment)
Σk=1Kzjk=1, ∀j=1, 2, . . . ,K (Driver Assignment)
x
ijm∈{0,1}, ∀i=1, 2, . . . , N,j=1, 2, . . . , K,m=1,2, . . . ,8 (Binary)
y
j∈{0,1}, ∀j=1, 2, . . . ,K (Binary)
z
ik∈{0,1}, ∀j=1, 2, . . . ,K, k=1,2, . . . , M (Binary)
where xijm, yj, and zjk are binary decision variables and xijm, =1 if vehicle i is placed on hauler j, ramp position m, and yj=1 if hauler j is used for transport whereas zjk=1 if hauler j is assigned to driver k. So the system considers many variables including, for example, the position of the vehicles on ramps of the hauler, as well as loading and unloading sequences.
In this illustrative method, the objective is to minimize the number of truck haulers used as well as reduce the sum of distance traveled for the vehicles with μ>0 as the weighting factor and Vj representing the set of vehicles loading on hauler j. The first three constraints are the weight load constraints on the overall hauler, front axle, and rear axle, respectively, with W1 and W2 being the load limits of the front axle and the rear axle, respectively. The fourth and fifth constraints are the layer constraints with L1 and L2 being the layer spacing limits of the lower layer and upper layer, respectively. The Height constraints are imposed to make sure the loaded hauler does not exceed the road height limit H.
The balancing constraint is to restrict the center of gravity of the load to lie between the front and rear axles for safety reasons, where σ1 and σ2 are the axle distances to the front of the loading space from the front axle and the rear axle, respectively. The Lead time constraint is to ensure the vehicle is delivered to the destination no later than the specified lead time with a probability p, by considering the loading time in the FIFO parking yard t0*j and the stochastic transit time Tr. Note that the transit time has large variations due to uncertain traffic, resulting in a stochastic constraint. Section II describes the development of a data-driven method to accurately predict the transit time. One can tune the chance constraint threshold p to tradeoff the total cost and the lead time accuracy; larger p will place greater emphasis on shorter lead time.
The Allowable Hours constraint is to restrict the driver hours per day and the longest continuous hours (e.g., up to legal hour limit) while guaranteeing the lead time with a probability p. The Vehicle Assignment and Driver Assignment constraints are to make sure that each vehicle is assigned to exactly one hauler and that each hauler is assigned to only one driver, respectively. The last three constraints specify that the variables xijm, yj, and zjk are binary, which take values among 0 and 1, and xijm, =1 if vehicle i is placed on hauler j, ramp position m, and yj=1 if hauler j is used for transport whereas zjk=1 if hauler j is assigned to driver k.
Note the probabilistic constraints, Lead time and Allowable Hours, will produce less conservative results as compared to using the worst-case scenario. Assuming the transit time is Gaussian distributed, the illustrative method will transform the chance constraints using an error function method, which will translate the above chance-constrained MIP problem to a deterministic MIP problem. Several challenges need to be addressed to solve the induced MIP problem. Firstly, the problem is computationally demanding with a large number of vehicles. Computationally efficient algorithms must be used to solve the problem to be able to handle last-minute changes. Secondly, the transit time during transport has high variability and it is critical to incorporate predictive capabilities in the optimization to make the solutions less conservative. These two challenges will be addressed in the sections below.
This section describes an illustrative machine learning-based traffic forecasting method. The chance-constrained MIP problem formulated in Section I relies on the knowledge of the transit time, which has a direct influence on lead time and driver allocation plan. In addition, as the vehicles in a truck may have different destinations, the traffic information also has a great impact on the optimal routing choices. Therefore, it is crucial to accurately predict the traffic to improve the system efficiency.
The traffic prediction has long been a challenging task due to large variations in traffic flow. For instance, the distribution of transit times across haulers from Toyota San Antonio to the Houston distribution has shown a large variation. Commercial software such as Google Maps predict transit time, but is purely based on averaged historic traffic statistics without considering near-term traffic dynamics. The future traffic is also highly dependent on the traffic in the recent history as well as the traffic in the neighboring traffic grids. In this illustrative disclosure, the system applies a machine learning-based traffic prediction module that incorporates spatio-temporal characteristics which are metrics describing the traffic evolution in space and in time, such as traffic density and average vehicle speed. Specifically, illustrative embodiments apply a recurrent neural network (RNN)-based traffic prediction.
In certain illustrative methods, the training of the network involves two phases. First, the hybrid RNN model is trained offline using existing traffic databases. Then, the online data collected from the truck operations is applied to the RNN model for online adaptation. This online adaptation is necessary because the prediction performance may vary due to geographic variations. The online data collection and RNN model updates may be performed on a cloud platform. The model performance may be evaluated based on the prediction error in the root mean square sense.
Since the vehicles in one truck hauler can have different destinations, the disclosed methods also determine the best sequence to deliver the vehicles to the destinations, including a loading and unloading sequence. This can be referred to as the Traveling Salesman Problem (TSP), which seeks a minimum-cost route starting and ending at the depot, visiting each destination exactly once. In this embodiment, the dynamic routing of the system will extend the classical TSP by 1) assigning each destination a time window to guarantee acceptable delivery time; and 2) by explicitly incorporating the transit time prediction described in this Section III in the routing optimization.
To predict transit times in certain illustrative embodiments, more formally, let More formally, let G=(N,A) be a complete directed graph with nodes/destination set N and arc set A. Let the cost of traversing arc a EA be ca ER+ and the time to traverse arc a∈A be τa∈Z+. Each destination n EN can only be visited during the time interval [en, ln]. The method seeks a minimum-cost route starting and ending at the depot, visiting each destination n EN only once in its associated time window, [en, tn]. As the arc traverse time to can be predicted using the RNN developed in Section II, the network G is not static over time. Instead, the arc traverse time and costs are time varying.
To address this challenge, embodiments of the present disclosure exploit the concept of time-expanded graphs, in which a node encodes both a location and a time interval, and solutions prescribe dispatch time intervals for trucks and vehicles. An example of the time-expanded graph is shown in
In general, mixed integer programming (MIP) approaches use strategies such as branch and bound, branch and cut, branch and reduce, and outer approximation, to handle integer variables. Well-developed commercial solvers, such as Gurobi and CPLEX, exist for solving mixed integer linear programming (MILP) and mixed integer quadratic programming (MIQP) problems. Solvers such as AOA, BARON, Knitro, Bonmin have been developed that handle nonlinear mixed integer programming problems. The computational cost of solving MILPs, MIQPs and MINLPs can be very substantial especially for higher dimensional problems. Consequently, the treatment of practical problems, such as our unified truck loading and dynamic routing problem, invariably exploits problem-specific structure, heuristic approximations and simplifications.
In developing the methods described herein, these problems have been carefully analyzed and have resulted in the development of suitable approximations with the objective of reducing the problem to a convex MIQP or to a sequence of MIQPs. During that development, the structure of the problem was also carefully analyzed so that when minor changes occur (e.g., prioritizing certain in-demand vehicles), the system can effectively solve the new problem based on the old solution without solving the whole problem again. The disclosed systems then utilize available solvers from the perspective of accuracy, robustness and computational time, and enhance them with additional algorithmic modifications as needed to improve their speed and accuracy.
To demonstrate the effectiveness of the proposed truck loading and dynamic routing system, a software prototype was developed and employed in the San Antonio to Houston vehicle distribution route (or other distribution routes with sufficient volume). The disclosed software architecture 600 of an illustrative embodiment is shown in
The processor 960 may include a central processing unit (CPU), a digital signal processor (DSP), an ASIC, a controller, or any combination of general-purpose computing devices, reduced instruction set computing (RISC) devices, application-specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), or other related logic devices, including mechanical and quantum computers. The processor 960 may also comprise another hardware device, a firmware device, or any combination thereof configured to perform the operations described herein. The processor 960 may also be implemented as a combination of computing devices, e.g., a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration.
The memory 964 may include a cache memory (e.g., a cache memory of the processor 960), random access memory (RAM), magnetoresistive RAM (MRAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read only memory (EPROM), electrically erasable programmable read only memory (EEPROM), flash memory, solid state memory device, hard disk drives, other forms of volatile and non-volatile memory, or a combination of different types of memory. In an embodiment, the memory 964 includes a non-transitory computer-readable medium. The memory 964 may store instructions 966. The instructions 966 may include instructions that, when executed by the processor 960, cause the processor 960 to perform the operations described herein. Instructions 966 may also be referred to as code. The terms “instructions” and “code” should be interpreted broadly to include any type of computer-readable statement(s). For example, the terms “instructions” and “code” may refer to one or more programs, routines, sub-routines, functions, procedures, etc. “Instructions” and “code” may include a single computer-readable statement or many computer-readable statements.
The communication module 968 can include any electronic circuitry and/or logic circuitry to facilitate direct or indirect communication of data between the processor circuit 950, and other processors or devices. In that regard, the communication module 968 can be an input/output (I/O) device. In some instances, the communication module 968 facilitates direct or indirect communication between various elements of the processor circuit 950. The communication module 968 may communicate within the processor circuit 950 through numerous methods or protocols. Serial communication protocols may include but are not limited to US SPI, I2C, RS-232, RS-485, CAN, Ethernet, ARINC 429, MODBUS, MIL-STD-1553, or any other suitable method or protocol. Parallel protocols include but are not limited to ISA, ATA, SCSI, PCI, IEEE-488, IEEE-1284, and other suitable protocols. Where appropriate, serial and parallel communications may be bridged by a UART, USART, or other appropriate subsystem.
External communication (including but not limited to software updates, firmware updates, preset sharing between the processor and a central server, or readings from the sensors) may be accomplished using any suitable wireless or wired communication technology, such as a cable interface such as a USB, micro USB, Lightning, or FireWire interface, Bluetooth, Wi-Fi, ZigBee, Li-Fi, or cellular data connections such as 2G/GSM, 3G/UMTS, 4G/LTE/WiMax, or 5G. For example, a Bluetooth Low Energy (BLE) radio can be used to establish connectivity with a cloud service, for transmission of data, and for receipt of software patches. The controller may be configured to communicate with a remote server, or a local device such as a laptop, tablet, or handheld device, or may include a display capable of showing status variables and other information. Information may also be transferred on physical media such as a USB flash drive or memory stick.
Furthermore, any of the illustrative methods described herein may be implemented by a system comprising processing circuitry or a non-transitory computer readable medium comprising instructions which, when executed by at least one processor, causes the processor to perform any of the methods described herein.
Although various embodiments and methods have been shown and described, the disclosure is not limited to such embodiments and methods and will be understood to include all modifications and variations as would be apparent to one skilled in the art. Therefore, it should be understood that the disclosure is not intended to be limited to the particular forms disclosed. Rather, the intention is to cover all modifications, equivalents and alternatives falling within the spirit and scope of the disclosure as defined by the appended claims.