LU503043B1

LU503043B1 - Smart city dynamic cold-chain logistics scheduling method based on ant colony optimization algorithm

Info

Publication number: LU503043B1
Application number: LU503043A
Authority: LU
Inventors: Lin Shi; Zhi-Hui Zhan
Original assignee: Univ South China Tech
Priority date: 2021-03-11
Filing date: 2021-09-18
Publication date: 2023-03-13
Also published as: WO2022188388A1; CN113011644A; CN113011644B

Abstract

The present invention discloses a smart city dynamic cold chain logistics scheduling method based on an ant colony optimization algorithm, and relates to the field of smart logistics and intelligent computing technology. The present invention establishes a dynamic scene-oriented cold chain logistics scheduling model, which is different from the traditional cold chain logistics scheduling model in that the model considers factors such as order information changes and refrigerated vehicle state changes in dynamic scenes. Besides, the model also considers the requirement of picking up goods before delivery, which is closer to the realistic scheduling scene. In order to solve the model, the present invention designs a scheduling method based on the ant colony optimization algorithm. The method divides the scheduling process into two stages of order allocation and path planning, and integrates dual-pheromone strategy, pre-pruning strategy and memory learning strategy, which can effectively narrow the search space and utilize the historical experience. Experiments show that this method, compared with scheduling methods such as first-come, first-served, has better performance, that is, the obtained scheduling scheme has lower cost.

Description

SMART CITY DYNAMIC COLD-CHAIN LOGISTICS SCHEDULING METHOD 7503063

BASED ON ANT COLONY OPTIMIZATION ALGORITHM

FIELD OF THE INVENTION

The present invention relates to the field of smart logistics and intelligent computing technology, specifically to a smart city dynamic cold chain logistics scheduling method based on an ant colony optimization algorithm.

BACKGROUND OF THE INVENTION

As an important branch of the logistics industry, cold chain logistics plays an important role in modern life and smart city construction. The main transportation objects of the cold chain logistics are items that have high requirements on the temperature and humidity of the transportation environment, including but not limited to fresh food, electronic components and pharmaceutical products. Due to the limitations of refrigeration technology, the quality of goods gradually declines over time during transportation. The longer the goods are transported, the more serious the quality loss will be. The quality of goods is an important factor affecting customer satisfaction, and thus has a great impact on the competitiveness of logistics companies. Therefore, compared with ordinary logistics scheduling, cold chain logistics scheduling should not only ensure the timely delivery of goods, but also minimize the quality loss caused by transportation. As a new logistics scheduling model, the logistics scheduling model with the added quality loss factor also brings new challenges.

Relevant data show that the waste caused by quality loss during cold chain logistics transportation is quite serious. To alleviate this phenomenon, in addition to improving refrigeration technology, it is also needed to improve transportation efficiency by designing an efficient scheduling method. Besides, most of the current work on cold chain logistics scheduling is scheduling based on static environment. Here, the cold chain logistics scheduling based on static environment means that at the beginning of the scheduling, all order information 503043 known, and all the refrigerated vehicles are located at the initial position and in the initial state. In contrast, the cold chain logistics scheduling based on dynamic environment means that at the beginning of the scheduling, only part of the order information is known, and the location and status (e.g. the remaining capacity) of the refrigerated vehicles are different. Moreover, considering the large time span of dynamic cold chain logistics scheduling, in order to avoid that orders cannot be delivered in time due to excessive waiting, it is necessary to schedule at regular intervals the orders that are received in this period. This approach is equivalent to dividing the entire dynamic problem into multiple sub-problems, and then using a scheduling algorithm to optimize each sub-problem in turn. In fact, the cold chain logistics scheduling based on dynamic environment is closer to the actual scheduling scene. In addition, the cold chain logistics scheduling based on dynamic environment is also more challenging.

As an important and effective swarm intelligence optimization algorithm, the ant colony optimization algorithm mainly constructs solutions by simulating the foraging behavior of ant colonies in nature, having good global search ability.

Besides, because the ant colony optimization algorithm gradually completes the solution construction through the movement of ants, the algorithm has natural advantages in solving discrete combinatorial optimization problems, and is thus widely used to solve complex discrete combinatorial optimization problems.

Moreover, different from other swarm intelligence optimization algorithms, the ant colony optimization algorithm contains problem-related knowledge in its heuristic information. By designing suitable heuristic information, the ant colony optimization algorithm can quickly find high-quality solutions. The main task of dynamic cold chain logistics scheduling is to assign order delivery tasks to the refrigerated vehicles, and meanwhile to plan delivery paths for the refrigerated vehicles, which is still a discrete combinatorial optimization problem in essence, so the ant colony optimization algorithm is also suitable for this problem. In addition, considering the complexity of dynamic cold chain logistics scheduling,

in order to achieve a better scheduling effect, a new scheduling optimizatiqnjspzp43 method can be designed based on the ant colony optimization algorithm.

CONTENTS OF THE INVENTION

The purpose of the present invention is to expand the research on the cold chain logistics scheduling in dynamic scenes (i.e., to establish a corresponding scheduling model for dynamic cold chain logistics scheduling scenes), and to provide a smart city dynamic cold chain logistics scheduling method based on an ant colony optimization algorithm. Considering the characteristics such as order information changes and refrigerated vehicle state changes (e.g. real-time location and load changes) in dynamic scenes, this method divides the optimization process of each sub-problem of the dynamic cold chain logistics scheduling problem into two stages of order allocation and path planning, and then, on the premise of satisfying the constraints of the model, a reasonable order allocation scheme is designed, and then the delivery path is further planned, so as to reduce the delivery cost as much as possible. In order to further improve the scheduling performance, the method also integrates a dual-pheromone strategy, a pre-pruning strategy, and a memory learning strategy.

The purpose of the present invention can be achieved through the follow technical solution:

A smart city dynamic cold chain logistics scheduling method based on an ant colony optimization algorithm is provided, comprising the following steps:

S1. According to the division of a sub-problem 7, finding out a set of orders participating in allocation and a set of orders participating in path planning in the i-th scheduling, whereini=1,2, 3, ...;

S2. implementing a dual-pheromone strategy by defining two pheromones and initializing them, one pheromone vr being set between the orders participating in the i-th allocation and all the refrigerated vehicles, the other pheromone or being set between all the orders participating in the i-th path planning;

S3. implementing a pre-pruning strategy by dividing an appropriate503043 refrigerated vehicle selection range Vset(o) for each order according to heuristic information weight(o, v) before order allocation, wherein weight(o, v) is calculated as follows: jeht(o.v)= dis(depart p(v.).depot(o)) + 0, if load order(v,i)|=0 (1) wel 0,V) = dis(depar V,1 epot(o 8H: POI PAIE ECP dis(midpos(v),dest(0)), otherwise where depart p(v, i) represents the location of the refrigerated vehicle, depot(o) represents the refrigeration storage where goods of order o are located, dis(depart p(v, i), depot(o)) is the distance between depart p(v, i) and depot(o), load order(v, i) represents a set of orders that have been loaded on a refrigerated vehicle v and are to be delivered before the i-th scheduling, |:| is the number of elements in a set -, dis(midpos(v), dest(0)) is the distance between midpos(v) and dest(o), midpos(v) represents the center of the distination of the order in load order(v, i), and dest(o) represents the destination of order o;

S4. implementing a memory learning strategy by using an optimal scheduling result s of the (i-1)-th scheduling to update the pheromone, with the update method as follows: vr(o,v)=vr(o,v)+rand(0,0.1)x(F,)" (2) ot(o,w) = or(o,w) + rand(0,0.1)x(F;)" (3) where vr(o, v) represents the pheromone between order o and refrigerated vehicle v, ot(o, w) represents the pheromone between order o and order w, rand(0, 0.1) represents a random decimal between 0 and 0.1, and Fis the fitness value of s; and

S5. using the ant colony optimization algorithm to construct a solution, wherein the ant colony optimization algorithm first allocates the orders involved in the allocation in the i-th scheduling to the refrigerated vehicle in turn according to a roulette selection method, and then each ant constructs a delivery path, i.e.

determines an order delivery sequence, for each refrigerated vehicle in tumy503043 according to a unified order allocation scheme.

Further, in step S3, when allocating order o, first calculating weight(o, v) for all the refrigerated vehicles, in which only the refrigerated vehicles with a ranking 5 less than or equal to [srxV] can be classified into the refrigerated vehicle selection range Vset(o) of order o, where V is the number of the refrigerated vehicles, sr is set to 0.4, and [-] represents a round up operation.

Further, in step S4, if i = 1, that is, the current scheduling is the first scheduling, since there is no historical experience, the memory learning strategy is not executed; if / > 1, sorting the solutions in memory(i-1) according to the fitness values from small to large, selecting the solutions with a ranking less than or equal to [lrxpop} to form the set Sol, and learning from each solution in Sol, where {7 is a parameter between [0, 1], pop represents the population size, [] represents a round up operation, and the variable memory(i-1) stores the solution ranked within pop in the (i-1)-th scheduling process; for a solution s in the set Sol, if order o to be allocated in the current sub-problem appears in s and is allocated to the refrigerated vehicle v by s, the pheromone vr(o, v) between o and v is randomly increased according to formula (2); similarly, if the destinations of orders o and v participating in scheduling in the current sub-problem are continuously accessed by a refrigerated vehicle in s, the pheromone or(o, w) between o and w is randomly increased according to formula (3).

Further, in step S5, in the order allocation stage, the probability p1(o, v) of order o being allocated to the refrigerated vehicle v 1s as follows: _ _ve(o,v)xweight(0,v) © A v —, if velset(o) p(o,v)= 2,70 <veis t(o,r) 4) 0, otherwise where Vset(o) represents the refrigerated vehicle selection range of order o, weight(o, v) represents the heuristic information between order o and the refrigerated vehicle v, and the exponential coefficient à is 2; besides, when an ant completes the path construction, the pheromone or is locally updated; after all the;s503043 ants complete the path construction, a global optimal solution is found, and the pheromones vr and or are globally updated.

Compared with the prior art, the present invention has the following advantages and beneficial effects: 1. The present invention is a scheduling method for cold chain logistics scheduling in dynamic scenes. The dynamic scenes optimized by this method consider not only the order information changes and the position and status changes of the refrigerated vehicle, but also the pickup situation. If the order goods to be delivered are not loaded by the corresponding refrigerated vehicle, the refrigerated vehicle needs to go to the corresponding refrigeration storage to pick up the goods before delivery. Therefore, the dynamic scenes optimized by this method are closer to the real scheduling scenes. 2. The present invention divides the optimization of sub-problems into two stages of order allocation and path planning, and sets the dual-pheromone strategy.

By allowing multiple ants to construct a delivery path according to an allocation scheme at the same time, not only the allocation scheme can be reasonably evaluated, but also the pheromones can be prevented from being incorrectly updated due to unreasonable evaluation. 3. When dividing the refrigerated vehicle selection range for an order, the pre-pruning strategy proposed by the present invention considers not only the corresponding pickup distance but also the existing delivery tasks of the refrigerated vehicle, which is conducive to guiding the method proposed by the present invention to allocate the refrigerated vehicle with the delivery range close to the order destination to the order, thereby improving the performance of the scheduling method. 4. The memory learning strategy proposed by the present invention converts the experience of historical scheduling into the pheromone of the current scheduling, and improves the convergence speed of the scheduling method.

BRIEF DESCRIPTION OF THE DRAWINGS

Fig. 1 1s a schematic diagram of the dynamic cold chain logistics scheduling in an example of the present invention; and

Fig. 2 1s a flowchart of the smart city dynamic cold chain logistics scheduling method based on an ant colony optimization algorithm in the example of the present invention.

DETAILED DESCRIPTION OF THE EMBODIMENTS

In order to make the purposes, technical solutions, and advantages of the examples of the present invention clearer, the technical solutions in the examples of the present invention will be described clearly and completely in combination with the drawings in the examples of the present invention. Obviously, the described examples are some, but not all, examples of the present invention. Any other examples obtained by those of ordinary skill in the art according to the examples of the present invention without making any inventive effort shall fall within the scope of protection of the present invention.

Examples

In this example, a smart city dynamic cold chain logistics scheduling method based on an ant colony optimization algorithm is introduced in detail with reference to Figs. 1 and 2. First, in the dynamic cold chain logistics scheduling scene, the order service platform continuously receives orders and submits them to a scheduling platform, each order including information such as order destination, goods demand, and service time window. The scheduling platform triggers scheduling at regular intervals to schedule all the orders collected but not delivered at the current moment, that is, the i-th scheduling corresponds to the i-th sub-problem. The task of the scheduling platform is to allocate all the orders that are not loaded by refrigerated vehicles to suitable refrigerated vehicles, and plan the delivery path for each refrigerated vehicle in turn. After the planning is completed, the scheduling platform releases the delivery tasks and delivery paths to the corresponding refrigerated vehicles. After the refrigerated vehicles recely@503043 the delivery tasks, they will deliver the goods to each customer in turn according to the delivery paths. To be noted, when planning the delivery paths, if the goods required by a customer are not loaded in the refrigerated vehicle, the refrigerated vehicle needs to go to the corresponding refrigeration storage to take out the goods, and then deliver the goods to the customer. Fig. 1 shows a schematic diagram of the dynamic cold chain logistics scheduling.

For the i-th sub-problem, the scheduling method of the present invention comprises the following execution steps:

S1. In the i-th scheduling, the order to be delivered can have two states, the order having not been loaded, or having been loaded, by the refrigerated vehicle.

The order that has not been loaded needs to be allocated to the refrigerated vehicle first, and then participates in path planning together with the orders that have been loaded; and the order that has been loaded can directly participate in path planning.

Therefore, it is first necessary to find out a set AS of orders participating in allocation and a set RS of orders participating in path planning in the i-th scheduling. For example, in the sub-problem 2 of Fig. 1, AS = {3, 6}, and RS = {3, 5, 6}, that is, the orders that need to participate in allocation include orders 3 and 6, and the orders that need to participate in path planning include orders 3, 5 and 6.

S2. The experience obtained by the ant colony in the search process can be converted into pheromone, and then guide the ant colony to find a better solution.

Therefore, for the order allocation stage and the path planning stage, this method sets a dual-pheromone strategy. vr is defined to be an MxM matrix, and or is an

NXN matrix, where M is the number of elements in AS, and N is the number of elements in RS. For the sub-problem 2 of Fig. 1, vr is a 2x2 matrix, and or is a 3x3 matrix. An initial solution needs to be constructed before initializing vr and or. In the process of constructing the initial solution, the orders to be allocated are firstly allocated to the refrigerated vehicles randomly, and then the order delivery sequence is determined for each refrigerated vehicle according to the order receiving time and the principle of first-come, first-served. To be noted, in the process of constructing the delivery path, if the refrigerated vehicle is overloadad;503043 with the currently selected order, the current order will be temporarily skipped, and the next order that meets the capacity limit will be selected for delivery. After the construction, the initial solution needs to be evaluated, and the pheromones vr and or are initialized with the fitness value of this solution.

S3. Implementing the pre-pruning strategy can divide an appropriate subset of refrigerated vehicles for the orders in AS, narrow the search space, and avoid considering all the refrigerated vehicles when allocating orders, thereby improving the scheduling performance. In the dynamic cold chain logistics scheduling, if the refrigerated vehicle v is too far from the refrigeration storage depot(o) where goods of the order to be allocated (recorded as order 0) are located, or the delivery range of the delivery task the refrigerated vehicle already has is too far from the destination dest(o) of order o, the refrigerated vehicle v can be ignored when order o 1s allocated. Here, the midpoint midpos(v) of the destination of the order to be delivered by the refrigerated vehicle v is used to represent the delivery range of v.

When implementing the pre-pruning strategy, the heuristic information weight(o, v) between order o and the refrigerated vehicle v needs to be obtained first according to formula (5): 0, if [load order(v,i)|=0 weight(o,v) = dis(depart p(v,i), depot(o)) + Bt otherwise (5) and then the refrigerated vehicles are sorted in ascending order according to weight(o, v). The refrigerated vehicles with a ranking less than or equal to [s<V1] are selected to form the set Vset(o), where V is the number of refrigerated vehicles, sr is set to 0.4, and Vset(o) represents the subset of refrigerated vehicles that only need to be considered when allocating order o.

S4. The memory learning strategy 1s implemented, which can make full use of the historical experience existing in the past scheduling to speed up convergence. For example, if order o exists in AS, and order o is allocated to the refrigerated vehicle v by a certain solution with a better fitness value in the (i-1)-th scheduling, it means that order o is suitable to be allocated to the refrigerated vehicle v, so the pheromone ot(o, v) between order o and the refrigerated vehicle yspz043 can be appropriately increased, thus guiding the scheduling method to allocate order o to the refrigerated vehicle v. vr(o, v) is increased as follows: vr(o,v)=vr(o,v)+rand(0,0.1)x(F,)" (6) where F, is the fitness value of the solution s, and s is the solution with the fitness value ranking less than or equal to [/rxpop! in memory(i-1), where /r is a parameter between 10, 11, pop represents the population size, and rand(0, 0.1) represents a random decimal between 0 and 0.1. The variable memory(i-1) stores the solution ranked within pop in the (i-1)-th scheduling process.

Similarly, in the (i-1)-th scheduling, if orders o and w participating in path planning in the i-th scheduling are continuously accessed by a refrigerated vehicle, it can be considered that the refrigerated vehicle is suitable to deliver order w after delivering order o, so the pheromone vr(o, w) between order o and order w can be appropriately increased. ot(o, v) is increased as follows: ot(o,w) = or(o,w) + rand(0, 0.1) x (F}) (7)

It is worth noting that if / = 1, that is, the current scheduling is the first scheduling, since there is no historical experience, the memory learning strategy 1s not executed.

S5. Using the ant colony optimization algorithm to construct a solution.

First, the current scheduling is divided into two stages, the first one being order allocation, the second one being path planning. In the first stage, the ant colony optimization algorithm first selects a refrigerated vehicle for the order in

AS, 1.e. allocates the order to the refrigerated vehicle, in a roulette manner according to the probability p1(0, v). p1(0, v) is calculated as follows: vz(o,v) x weight(o,v) * if veVser(o) p(o,v)= = vr(o,r)xweight(o,r) (8) 0, otherwise where a 1s 2.

By allowing multiple ants to construct a delivery path according to an allocation scheme at the same time, not only the allocation scheme can h@503043 reasonably evaluated, but also the pheromones can be prevented from being incorrectly updated due to unreasonable evaluation. Therefore, in the second stage, according to the allocation scheme obtained in the first stage, the order to be delivered by each refrigerated vehicle is determined, and the ants in the ant colony plan the delivery path for the refrigerated vehicles according to the delivery task of each refrigerated vehicle. Every step the ants move, they will determine the next delivered order, i.e. determine the destination of the next delivered order, for the refrigerated vehicle according to the state transition rule. To be noted, if the goods corresponding to the next order are not loaded, the next destination of the refrigerated vehicle is modified to the refrigeration storage where the goods of the next order are located, and then the destination of the current order. The movement 1s repeated until the delivery sequence of all the orders has been planned.

When an ant completes the path planning, it needs to locally update or. To be noted, since the purpose of the local update is to guide different ants to explore different solutions, and here the ants in the ant colony share the same allocation scheme in the same iteration, no local update of vr is required. After all the ants complete the path planning, in order to improve the convergence speed, it is necessary to find the historical optimal solution, and globally update or and vr according to the historical optimal solution. Step S5 is repeatedly executed until termination conditions are met.

The above examples are preferred embodiments of the present invention, but the embodiments of the present invention are not limited thereto, and any other alterations, modifications, replacements, combinations and simplifications shall be equivalent substitutions and fall within the scope of protection of the present invention.

Claims

CLAIMS LU503043

1. A smart city dynamic cold chain logistics scheduling method based on an ant colony optimization algorithm, characterized in that the scheduling method comprises the following steps:

S1. according to the division of a sub-problem i, finding out a set of orders participating in allocation and a set of orders participating in path planning in the i-th scheduling, whereini=1,2, 3, ...;

S2. implementing a dual-pheromone strategy by defining two pheromones and initializing them, one pheromone vr being set between the orders participating in the i-th allocation and all refrigerated vehicles, the other pheromone ot being set between all the orders participating in the i-th path planning;

S3. implementing a pre-pruning strategy by dividing an appropriate refrigerated vehicle selection range Vset(o) for each order according to heuristic information weight(o, v) before order allocation, wherein weight(o, v) is calculated as follows: jeht(o.v)= dis(depart p(v.).depot(o)) + 0, if load order(v,i)|=0 (1) wel 0,V) = dis(depar V,1 epot(o 8H: POI PAIE ECP dis(midpos(v),dest(0)), otherwise where depart p(v, i) represents the location of the refrigerated vehicle, depot(o) represents the refrigeration storage where goods of order o are located, dis(depart p(v, i), depot(o)) is the distance between depart p(v, i) and depot(o), load order(v, i) represents a set of orders that have been loaded on a refrigerated vehicle v and are to be delivered before the i-th scheduling, || is the number of elements in a set -, dis(midpos(v), dest(0)) is the distance between midpos(v) and dest(o), midpos(v) represents the center of the destination of the order in load order(v, i), and dest(o) represents the destination of order o;

S4. implementing a memory learning strategy by using an optimal scheduling result s of the (i-1)-th scheduling to update the pheromone, with the updat@503043 method as follows: vr(o,v)=vr(o,v)+rand(0,0.1)x(F,)" (2) ot(o,w) = or(o,w) + rand(0,0.1)x(F;)" (3) where vi(o, v) represents the pheromone between order o and the refrigerated vehicle v, ot(o, w) represents the pheromone between orders o and w, rand(0, 0.1) represents a random decimal between 0 and 0.1, and Fis the fitness value of s; and

S5. using the ant colony optimization algorithm to construct a solution, wherein the ant colony optimization algorithm first allocates the orders involved in the allocation in the i-th scheduling to the refrigerated vehicle in turn according to a roulette selection method, and then each ant constructs a delivery path, i.e. determines an order delivery sequence, for each refrigerated vehicle in turn according to a unified order allocation scheme.

2. The smart city dynamic cold chain logistics scheduling method based on an ant colony optimization algorithm according to claim 1, characterized in that: in step S3, when allocating order o, first calculating weight(o, v) for all the refrigerated vehicles, in which only the refrigerated vehicles with a ranking less than or equal to [srxV] can be classified into the refrigerated vehicle selection range Vset(o) of order o, where V is the number of the refrigerated vehicles, sr is an adjustment parameter, and [| represents a round up operation.

3. The smart city dynamic cold chain logistics scheduling method based on an ant colony optimization algorithm according to claim 1, characterized in that: in step S4, if i = 1, that is, the current scheduling is the first scheduling, since there is no historical experience, the memory learning strategy is not executed; if i > 1, then sorting the solutions in memory(i-1) according to the fitness values from small to large, selecting the solutions with a ranking less than or equal to [/rxpop! to form a set Sol, and learning from each solution in Sol, where /r is a parameter between [0, 1], pop represents population size, [-] represents a round up operation,

and the variable memory(i-1) stores the solution ranked within pop in the (i-1)-thys03043 scheduling process; for a solution s in the set Sol, if order o to be allocated in the current sub-problem appears in s and is allocated to the refrigerated vehicle v by s, the pheromone vr(o, v) between o and v is randomly increased according to formula (2); similarly, if the destinations of orders o and v participating in scheduling in the current sub-problem are continuously accessed by a refrigerated vehicle in s, the pheromone ot(o, w) between o and w is randomly increased according to formula (3).

4. The smart city dynamic cold chain logistics scheduling method based on an ant colony optimization algorithm according to claim 1, characterized in that: in step S5, in the order allocation stage, the probability pi(o, v) of order o being allocated to the refrigerated vehicle v 1s as follows: vz(o,v) x weight(o,v) * if veVser(o) p(o,v)= = vr(o,r)xweight(o,r) (4) 0, otherwise where Vset(o) represents the refrigerated vehicle selection range of order o, weight(o, v) represents the heuristic information between order o and the refrigerated vehicle v, and a is an exponential coefficient; besides, after an ant completes the path construction, the pheromone or 1s locally updated; after all the ants complete the path construction, a global optimal solution is found, and the pheromones vr and or are globally updated.

5. The smart city dynamic cold chain logistics scheduling method based on an ant colony optimization algorithm according to claim 2, characterized in that the adjustment parameter sr is set to 0.4.

6. The smart city dynamic cold chain logistics scheduling method based on an ant colony optimization algorithm according to claim 4, characterized in that the exponential coefficient a is set to 2.