Intelligent delivery trolley distribution sequence optimization method and device in three-dimensional space

Abstract

The application relates to a smart delivery trolley distribution sequence optimization method and device in a three-dimensional space, wherein the method comprises the following steps: S1, three-dimensional map information of a target distribution area is acquired, and an expression graph of the target area is generated; S2, distribution order data is acquired, virtual nodes are expanded based on the destination of the distribution order data and in combination with the three-dimensional map information of the target distribution area; S3, a distance complete graph of the target area is generated based on the generated virtual nodes and in combination with the expression graph of the target area; S4, a mathematical optimization model of the delivery service sequence corresponding to the park is established based on the distance complete graph; S5, linearization and simplification processing is performed on the mathematical optimization model; and S6, the mathematical optimization model is solved by introducing an improved ant colony algorithm with space-time clustering characteristics to obtain the distribution sequence. Compared with the prior art, the application combines height information, generates a distance complete graph, and models based on the distance complete graph, which is more suitable for a real three-dimensional distribution scene, so that the optimization effect is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of logistics optimization, in particular to a smart delivery trolley distribution sequence optimization method and device in a three-dimensional space. BACKGROUND

[0002] With the rapid development of the logistics industry, the demand for express delivery has increased significantly. However, rapid growth has brought many problems, especially in the "last mile" distribution link of logistics. The last mile refers to the final delivery step from the distribution center to the consumer, and is the most complex, highest cost and lowest efficiency part of the entire logistics supply chain. For customers, the quality of logistics services is often inconsistent because the last mile delivery is usually completed by temporary workers or outsourced personnel. As services closely related to residents' lives, such as express delivery and food delivery, their service quality will have a profound impact on customer satisfaction and pose new requirements on the supply side.

[0003] Therefore, the logistics system is faced with the contradiction between supply constraints and personalized demand. To some extent, the long-term inefficiency of the last mile logistics has become a bottleneck of the entire logistics system. The emergence of new technologies provides new solutions to solve the last mile problem. With the rapid development of artificial intelligence, big data and electrification, unmanned driving technology has become a global research hotspot. Intelligent delivery robots based on unmanned driving technology are widely believed to revolutionize the growing logistics network, as these robots not only reduce transportation costs, but also reduce carbon footprint and improve road safety.

[0004] With the rapid development of e-commerce, the logistics and distribution industry is facing unprecedented challenges. Traditional delivery modes mainly rely on human delivery and customer self-pickup at stores, but this mode has many limitations, such as high labor costs, time costs, and lack of portability. In order to cope with these challenges, the logistics industry has begun to explore new delivery modes. The expansion of the logistics network and the application of new technologies have made it possible to innovate delivery modes. Through the optimization of path planning algorithms, the efficiency of delivery trolleys can be significantly improved, reducing delivery time and cost. On the demand side, consumers have increasingly high requirements for delivery time windows, hoping to obtain faster and more accurate delivery services. By optimizing the delivery path, this demand of users can be better met. On the supply side, optimizing the delivery path not only improves delivery efficiency, but also reduces operating costs. By reducing unnecessary routes and repeated deliveries, fuel consumption and vehicle wear and tear can be reduced, thereby reducing logistics costs. At the same time, by improving operational efficiency, more delivery tasks can be completed in a shorter period of time, improving the competitiveness of logistics enterprises.

[0005] In addition, the traditional path planning model only considers the path optimization in the two-dimensional space, and does not consider the distance and time optimization of the delivery upstairs, and with the development of the delivery robot technology, the robot delivery upstairs becomes possible. Thus, the path optimization in the three-dimensional space considering the delivery upstairs needs to be carried out. SUMMARY

[0006] The purpose of the present application is to provide a three-dimensional space-oriented intelligent delivery trolley distribution sequence optimization method and device, which combines height information, generates a distance completion graph, and is based on the distance completion graph modeling, which is more suitable for the real three-dimensional distribution scene, thereby improving the optimization effect.

[0007] The purpose of the present application can be realized by the following technical solutions:

[0008] A three-dimensional space-oriented intelligent delivery trolley distribution sequence optimization method, comprising:

[0009] Step S1: Obtain the three-dimensional map information of the target distribution area, and generate an expression graph of the target area.

[0010] Step S2: Obtain the distribution order data, and based on the destination of the distribution order data, combine the three-dimensional map information of the target distribution area, and expand virtual nodes.

[0011] Step S3: Based on the generated virtual nodes, combine the expression graph of the target area, and generate a distance completion graph of the target area.

[0012] Step S4: Based on the distance completion graph, establish a mathematical optimization model of the delivery service sequence corresponding to the park.

[0013] Step S5: Linearize and simplify the mathematical optimization model.

[0014] Step S6: Solve the mathematical optimization model by introducing an improved ant colony algorithm with space-time clustering features to obtain the distribution sequence.

[0015] The expression graph of the target area includes nodes and plane distances between nodes.

[0016] The distribution order data includes order number, delivery user door number, order quantity, large item quantity, small item quantity, earliest arrival time and latest arrival time.

[0017] In step S2, if the destinations of multiple distribution orders are the same node, the node is expanded into multiple virtual nodes consistent with the number of distribution orders with the node as the destination.

[0018] The distance completion graph is:

[0019] G=(I, E, W)

[0020] wherein G is a distance completion graph, I is a set of augmented nodes, E is a set of planar distances between augmented nodes, and W is a set of three-dimensional distances including all augmented nodes;

[0021] The elements in the set of three-dimensional distances represent three-dimensional distances between nodes.

[0022] If two nodes are in the same building, the three-dimensional distance between the two nodes is:

[0023] w ij = τ × |f i -f j | + D(d i ) + D(d j )

[0024] wherein w ij is the three-dimensional distance between the two nodes, τ is a longitudinal distance coefficient, f i is the floor of node i, f j is the floor of node j, d i is the room type corresponding to the house number of node i, d j is the room type corresponding to the house number of node j, and D(·) is the distance from each room type to the exit of the corridor.

[0025] If two nodes are in different buildings, the three-dimensional distance between the two nodes is:

[0026] w ij = τ × (f i +f j ) + D(d i ) + D(d j ) + D ground (e i , e j )

[0027] wherein D ground (e i , e j ) is the planar distance between node i and node j.

[0028] The longitudinal distance coefficient is:

[0029]

[0030] wherein h is the height of a single-story building.

[0031] The planar distance between node i and node j is the planar distance from the exit of the unit building of node i to the exit of the unit building of node j.

[0032] The objective function of the mathematical optimization model is:

[0033]

[0034] where p is the sum of vehicle operating cost, vehicle fixed cost and window penalty, is a vehicle delivery parameter, which is 1 if the vth vehicle has passed the route from node i to node j, and 0 otherwise, V is the set of vehicles, C s is the cost of operating a single vehicle per unit time, g ij is the travel time from node i to node j, c f is the fixed cost of each vehicle, |·| is the number of elements in a set, w i is the window penalty for not arriving at node i by the target time;

[0035] The constraint conditions of the mathematical optimization model are:

[0036]

[0037]

[0038] where Y i v is the demand execution value, which is 1 if the vehicle v has executed the demand of node i, and 0 otherwise, vn is the number of started vehicles, ta j is the actual arrival time at node j, ta i is the actual arrival time at node i, ts i is the time required to serve node i, T max is a very large positive number used to relax the condition, β is the penalty coefficient of the vehicle arriving later than the latest arrival time, tl i is the amount of time allowed to be delayed for node i, is the amount of goods already delivered in the capacity window of size t before node i, is the goods status, which is 1 if node i needs to receive the goods of the trolley, and 0 otherwise, M is a very large positive number, is the amount of goods already received in the capacity window of size t before node i, is the goods status, S is the set of goods, n is the number of low-level goods that can be loaded in the high-level capacity window, F s is the number of trolleys with s-level capacity windows.

[0039] The model constraint conditions and the objective function are linearized by using the big M method in the step S5.

[0040] The step S6 includes:

[0041] Step S6-1: clustering from the earliest arrival time and the latest departure time dimension and the three-dimensional space dimension;

[0042] Step S6-2: initializing the initial data information of the ant colony algorithm: including setting the initial pheromone on the path, the number of ants in the population, and the maximum number of iterations;

[0043] Step S6-3: taking the objective function of the mathematical optimization model as the fitness function;

[0044] Step S6-4: selecting the path based on the taboo table;

[0045] Step S6-5: searching by long and short time pheromones to complete the solution to obtain the distribution sequence.

[0046] A kind of intelligent delivery trolley distribution sequence optimization device in three-dimensional space, including memory, processor, and program stored in the memory, characterized in that, when the processor executes the program, the method as described above is realized.

[0047] Compared with the prior art, the present application has the following beneficial effects:

[0048] 1, combined with height information, generate distance complete graph, and model based on distance complete graph, more fit real three-dimensional distribution scene, thereby improving the effect of optimization.

[0049] 2, combined with the capacity characteristics of intelligent delivery trolley modeling, considering the multi-level capacity window and soft time window characteristics of trolley, an improved ant colony search algorithm is proposed, and the solution in three-dimensional space is innovatively realized, to adapt to the real three-dimensional distribution scene. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 It is the main step flowchart of the method of the present application. DETAILED DESCRIPTION

[0051] The present application will be described in detail below in conjunction with the drawings and specific embodiments. The present embodiment is implemented on the premise of the technical solution of the present application, and detailed implementation and specific operation process are given, but the protection scope of the present application is not limited to the following examples.

[0052] A kind of intelligent delivery trolley distribution sequence optimization method in three-dimensional space, as shown in Figure 1 , comprising:

[0053] Step S1: obtaining the three-dimensional map information of the target distribution area, generating the expression graph of the target area;

[0054] The expression graph of the target area includes nodes and plane distances between nodes.

[0055] Step S2: Obtain the delivery order data, and based on the destination of the delivery order data, expand the virtual nodes in combination with the three-dimensional map information of the target delivery area;

[0056] The delivery order data includes order number, delivery user door number, order quantity, large item quantity, small item quantity, earliest arrival time and latest arrival time.

[0057] And if there are multiple delivery orders whose destinations are the same node, the node is expanded into multiple virtual nodes consistent with the number of delivery orders whose destinations are the node.

[0058] Step S3: Based on the generated virtual nodes, generate the distance completion graph of the target area in combination with the expression graph of the target area;

[0059] The distance completion graph is:

[0060] G=(I, E, W)

[0061] Wherein, G is the distance completion graph, I is the expanded node set, E is the plane distance set between the expanded nodes, and W is the three-dimensional distance set containing all the expanded nodes;

[0062] The elements in the three-dimensional distance set represent the three-dimensional distances between nodes;

[0063] If two nodes are in the same building, the three-dimensional distance between the two nodes is:

[0064] w ij =τ×|f i -f j |+D(d i )+D(d j )

[0065] Wherein, w ij is the three-dimensional distance between the two nodes, τ is the longitudinal distance coefficient, f i is the floor of node i, f j is the floor of node j, d i is the room type corresponding to the door number of node i, d j is the room type corresponding to the door number of node j, and D(·) is the distance from each room type to the exit,

[0066] If two nodes are in different buildings, the three-dimensional distance between the two nodes is:

[0067] w ij =τ×(f i +f j )+D(d i )+D(d j )+Dground (e i ,e j )

[0068] wherein, D ground (e i ,e j ) is the planar distance between node i and node j.

[0069] The longitudinal distance coefficient is:

[0070]

[0071] wherein, h is the height of the single-layer building.

[0072] The planar distance between node i and node j is the planar distance between the single-layer building entrance of node i and the single-layer building entrance of node j.

[0073] Step S4: Based on the distance complete graph, a mathematical optimization model of the delivery service sequence corresponding to the park is established.

[0074] The mathematical optimization model is based on the following assumptions:

[0075] 1. There is only one distribution center, and the distribution trolley must start from here and return here.

[0076] 2. The distribution trolley is of the same type, and the maximum load is known and the same.

[0077] 3. The vehicle transportation and climbing speed are uniform and known speed.

[0078] 4. The capacity of the large window allows to store the goods of multiple small windows.

[0079] The objective function of the mathematical optimization model is:

[0080]

[0081] wherein, p is the sum of vehicle operation cost, vehicle fixed cost and window penalty, is the vehicle distribution parameter, if the vth vehicle has passed the route from node i to node j, it takes the value of 1, otherwise it takes the value of 0, V is the vehicle set, C s is the unit time consumption cost of a single vehicle, g ij is the driving time from node i to node j, c f is the fixed cost of each vehicle, |·| is the number of elements in the set, w i is the window penalty for not arriving at node i within the target time.

[0082] The constraint condition of the mathematical optimization model is:

[0083]

[0084]

[0085] where Y i v is the demand execution value, which is equal to 1 if vehicle v has executed the demand of node i, and 0 otherwise, vn is the number of starting vehicles, ta j is the actual arrival time at node j, ta i is the actual arrival time at node i, ts i is the service time at node i, T max is a large positive number used to relax the condition, β is the penalty coefficient for the vehicle arriving later than the latest arrival time, tl i is the amount of time allowed to be delayed at node i, is the amount of goods already delivered by the capacity bin of size t at node i and before, is the goods status, which is equal to 1 if node i needs to receive the goods of the vehicle, and 0 otherwise, M is a large positive number, is the amount of goods already received by the capacity bin of size t at node i and before, is the number of goods of size k that can be contained in the capacity bin of size k+1, S is the set of goods, n is the number of goods of size k that can be contained in the capacity bin of size k+1, n s-t is the s-th power of n, F s is the number of capacity bins of size s owned by the vehicle.

[0086] Constraints (2) and (3) define the domain of the decision variables. Constraint (4) ensures that all vehicles start from a fixed depot. Constraint (5) requires all vehicles to return to the depot. Constraint (6) indicates that a vehicle cannot go to the same location twice. Constraint (7) computes the number of starting vehicles. Constraints (8) and (9) require that a customer can only be served once by one vehicle. Constraint (10) ensures that a vehicle must leave a node after visiting it. Constraint (11) limits the arrival time at a node. Constraint (12) defines the penalty when a vehicle fails to meet the time window of a customer. Constraints (13) and (14) compute the amount of goods that need to be received and sent before the node. Constraint (15) determines whether the capacity constraints of different sizes are met by calculating the amount of goods still to be sent and the amount of goods received before, and adding them to obtain the current load of goods.

[0087] Step S5: linearization and simplification of the mathematical optimization model;

[0088] The model constraints and objective function are linearized by using the big M method. First, the model is linearized and simplified to obtain the following constraints:

[0089]

[0090] The constraints (17) are replaced by constraints (19)-(24) to achieve model linearization.

[0091]

[0092] where is the binary variable involved in the linearization of the time penalty function.

[0093] First, the constraint (16) is divided into two sub-problems. When and , the constraint (21) is simplified as The constraint (22) is redundant. Together with the constraint (19), we have The constraint (20) is simplified as t ei ≤ t ai . It means that the vehicle arrives at node i later than the earliest arrival time, and the penalty for early arrival is equal to 0. When and , the constraint (22) becomes The constraint (21) becomes redundant. Together with the constraints (19) and (20), we have and t ei ≥ t ai . It means that the vehicle arrives at node i earlier than the earliest time, and the penalty for early arrival is equal to a(t ei -t ai ). When and , we have and It means that the vehicle arrives at the node just at the allowed earliest arrival time. For the constraint (18), by the same linearization method, the constraint (18) can be replaced by constraints (25)-(30).

[0094]

[0095] Step S6: solving the mathematical optimization model by introducing the improved ant colony algorithm with space-time clustering features to obtain the distribution sequence, comprising:

[0096] Step S6-1: clustering from the dimensions of the earliest arrival time and the latest departure time and the three-dimensional space dimension;

[0097] First step: Clustering in time dimension based on earliest arrival time and latest departure time. Focus on time feature of user demand, calculate distance matrix as clustering index by sending Euclidean distance of corresponding demand time between users, specific formula is:

[0098]

[0099] Where, T ij represents the distance of demand time between two users.

[0100] Second step: Clustering in three-dimensional space dimension, clustering based on three-dimensional space and category features, unit, floor and house number information, H ij represents the distance between two nodes, as a clustering index, specifically:

[0101]

[0102] Where e represents the unit, which can be T5A, T5B,..., T5D in this example. f represents the floor, which can be 1, 2,..., 6 in this example. d represents the house number, which can be 1, 2,..., 10 in this example. α, β are 0.5, 0.4 respectively.

[0103] Step S6-2: Initialize initial data information of ant colony algorithm: including setting initial pheromone on path, number of ants in population and maximum iteration number, specifically including:

[0104] First step, set the pheromone on all paths to a same initial value such as 1, to ensure that each path has the same attraction at the beginning of the algorithm.

[0105] Second step: Determine the number of ant colonies, which can be 100 in this example, which determines how many ants will construct solutions in each iteration.

[0106] Third step: Set the maximum iteration number, which can be 50 in this example, as one of the conditions for stopping the algorithm, to control the running time of the algorithm or the termination of the search process.

[0107] Step S6-3: Take the objective function of the mathematical optimization model as the fitness function;

[0108] Step S6-4: Path selection based on the taboo list, specifically in three steps:

[0109] First step: Introducing the search method of the tabu list: using roulette selection method to determine the next node. Check if the node is in the tabu list. If the node is not in the tabu list, continue and initialize a value on the corresponding edge, record the iteration number before the node is released from the tabu list. If the node is in the tabu list, the relevant value will decrease by 1. Then use the roulette selection method to determine the next node.

[0110] Second step: Determine the candidate subset. In the path generation process, a node is randomly selected from multiple clusters. Then select the nodes within the same cluster until all nodes within the cluster are selected, which is called intra-cluster node selection. After that, return to global node selection to determine the next node from different clusters.

[0111] Third step: Path termination mechanism. To solve the problem of capacity waste caused by discrete capacity containers, this mechanism introduces a forward-looking function. When the large-capacity container of the delivery robot cannot be fully utilized, the robot usually returns to the warehouse, resulting in underutilization of small-capacity containers. The forward-looking mechanism allows considering multiple steps when the capacity limit is not met, thereby improving the overall utilization of the robot's containers.

[0112] Step S6-5: Search by long and short time pheromones to complete the solution and obtain the delivery sequence.

[0113] Update the pheromone. The pheromone update mechanism of the ant colony algorithm includes pheromone evaporation and increase. Pheromone evaporation is to avoid excessive dependence on previous paths and maintain the exploratory nature of the algorithm, usually by setting a volatility coefficient to reduce the pheromone. Pheromone increase is to accumulate new pheromone through the path walked by the ant, and the shorter or more optimal solution will receive more pheromone, attracting subsequent ants to preferentially select it. The two work together to enable the algorithm to explore new paths while utilizing existing excellent paths.

[0114] Search by long and short time pheromones, the formula is as follows:

[0115]

[0116] The calculation formula of the time pheromone from node i to node j is as follows:

[0117]

[0118] The first part: capture the nodes with penalties under the current time window, called short time pheromone, the calculation formula is as follows:

[0119]

[0120] denotes the time of the ith ant from node i to node j.

[0121] Second part: introduce the knowledge in previous iteration, capture long time penalty relationship, called global time pheromone. To increase the flexibility of vehicle arrival time, enhance the ability to jump out of local optimal solution, the formula of vehicle arrival probability is:

[0122]

[0123] After deriving the arrival density function, the probability of a vehicle arriving at time t is given by: j to tl j Integrating, we have:

[0124]

[0125] The probability of a vehicle arriving on time, denoted by Δω j , can be calculated by this formula. The time pheromone also has an iterative update process, which is:

[0126]

[0127] Through this mechanism, the global time pheromone has a dynamic process of memory and forgetting.

[0128] If the above functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the parts that contribute to the prior art or parts of the technical solutions can be embodied in the form of software products. The computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk, and various program code storage media.

Claims

1. A method for optimizing the delivery sequence of an intelligent delivery trolley in a three-dimensional space, characterized in that, The method comprises the following steps: Step S1: obtaining three-dimensional map information of a target delivery area, and generating an expression graph of the target area; Step S2: obtaining delivery order data, and based on the destination of the delivery order data, combining the three-dimensional map information of the target delivery area to expand virtual nodes; Step S3: based on the generated virtual nodes, combining the expression graph of the target area to generate a distance complete graph of the target area; Step S4: based on the distance complete graph, establishing a mathematical optimization model of the delivery service order of the park; Step S5: linearizing and simplifying the mathematical optimization model; Step S6: solving the mathematical optimization model by introducing an improved ant colony algorithm with space-time clustering characteristics to obtain the delivery order; In step S2, if the destinations of multiple delivery orders are the same node, the node is expanded into multiple virtual nodes consistent with the number of delivery orders with the node as the destination; The distance complete graph is: Wherein, G is the distance complete graph, I is the expanded node set, E is the plane distance set between the expanded nodes, and W is the three-dimensional distance set containing all the expanded nodes; The elements in the three-dimensional distance set represent the three-dimensional distances between nodes; If two nodes are in the same building, the three-dimensional distance between the two nodes is: wherein, w ij is a three-dimensional distance between two nodes, If two nodes are in different buildings, the three-dimensional distance between the two nodes is: is a longitudinal distance coefficient, f i is a floor of a node i , f j is a floor of a node j , d i is a room type corresponding to a house number of a node i , d j is a room type corresponding to a house number of a node j , D(·) is a distance to an exit of a passage corresponding to each room type, The longitudinal distance coefficient is: wherein is the planar distance of the node i and the node j is the planar distance of the node The objective function of the mathematical optimization model is: wherein: h H is the height of the single-story building; The node i And the planar distance of the node j The planar distance from the unit building entrance of the node i To the unit building entrance of the node j The planar distance from the unit building entrance of the node The constraint condition of the mathematical optimization model is: in: p is the sum of vehicle operating cost, vehicle fixed cost and window penalty, is the vehicle distribution parameter, if v The car passes the slave node i To Node j If the route is , the value is 1, otherwise it is 0, V is the vehicle set, c s The unit time consumption cost of a single vehicle operation, g ij For nodes i To Node j driving time, c f is the fixed cost of enabling each vehicle, |·| is the number of elements in the set, w i If the node is not reached within the target time i Window penalty; Step S6 comprises: in, is the demand execution value. If vehicle v executes the demand of node i, the value is 1, otherwise it is 0. To start the number of vehicles, The actual arrival node j time, The actual arrival node i time, For service nodes i The time required, is a large positive number used to relax the conditions. is the penalty coefficient for vehicles arriving later than the latest arrival time, For nodes i The amount of time allowed for delay, For nodes i and before t The capacity of the size pane shows the number of goods that have been delivered. For the cargo status, if the node i If the goods need to be received by the cart, the value is 1, otherwise the value is 0. is a very large positive number, For nodes i and before t The capacity of the size pane is the number of goods that have been received, for, S For cargo collection, A higher capacity window can carry the same amount of cargo as a lower capacity window. The number of S-level capacity windows that the car has; Step S6-1: clustering in the dimensions of earliest arrival time and latest departure time and three-dimensional space; Step S6-2: initializing the initial data information of the ant colony algorithm, including setting the initial pheromone on the path, the number of ants in the population, and the maximum number of iterations; Step S6-3: taking the objective function of the mathematical optimization model as the fitness function; Step S6-4: selecting the path based on the taboo table; Step S6-5: searching by long-time and short-time pheromones to complete the solution and obtain the delivery order. The expression graph of the target area includes nodes and plane distances between nodes.

2. The method of claim 1, wherein, The delivery order data includes order number, delivery user door number, order quantity, large item quantity, small item quantity, earliest arrival time, and latest arrival time.

3. The method of claim 1, wherein, In step S5, the model constraint condition and the objective function are linearized by using the big M method.

4. The method of claim 1, wherein, The processor executes the program to implement the method of any one of claims 1-4. 5.A device for optimizing delivery sequence of an intelligent delivery trolley in a three-dimensional space, comprising a memory, a processor, and a program stored in the memory, characterized in that, ​

Images