A collaborative fulfillment optimization method for omni-channel pharmaceutical retail orders

Abstract

The present invention belongs to the field of omni-channel operation management, and provides a collaborative fulfillment optimization method for omni-channel drug retail orders. Aiming at the joint decision-making problems of order allocation, collaborative multi-commodity network flow, and vehicle scheduling in omni-channel drug retail order fulfillment decisions, the present invention designs and implements a decomposition-iteration-search algorithm based on the idea of ​​decomposition and iteration, decouples the associated sub-problems in the complex joint decision-making problem, and obtains the decision plan through parameter transfer and iterative solution. The effect and benefit of the present invention is to effectively solve the omni-channel drug order fulfillment problem for multiple inventory locations with complete order delivery requirements by adopting the idea of ​​decomposition, make full use of the information between decision sub-problems, optimize the order processing process and vehicle path, reduce the order fulfillment cost, and improve the scientific nature of the order fulfillment plan decision and the efficiency of order processing, so as to provide decision support for the omni-channel drug order fulfillment operation practice.

Description

Technical Field

[0001] The present invention belongs to the field of omni-channel operation management and relates to a collaborative fulfillment optimization method for omni-channel drug retail orders. Background Art

[0002] The omni-channel drug retail model can combine the price advantages, category advantages, and service interaction of online pharmacies with the convenience of offline physical pharmacies. Through the form of "online pharmacy ordering and offline pharmacy fulfillment", it connects diverse customer needs with scattered drug inventories, breaking the time and space limitations of drug supply and demand. Through offline pharmacies, it provides additional pharmaceutical services such as insurance, pharmacist consultation, and on-time delivery, thereby improving consumers' service experience and satisfaction.

[0003] However, as the scale of omni-channel drug retail continues to expand, its daily operations are also facing great challenges. The primary difficulty is order fulfillment decisions, that is, which pharmacies in the order will ship the drugs from, when, and which vehicles will deliver them, when, and via which routes. Due to the special nature of drugs, omni-channel drug retail has high requirements for the timeliness and punctuality of delivery, and at the same time faces the contradiction between decentralized inventory and coordinated delivery. The above-mentioned fulfillment decisions face the following difficulties: In terms of inventory, China currently has up to 200,000 kinds of drugs available for online retail. The storage and transportation conditions of drugs are strict and diverse, while the types and quantities of drugs that can be stored in physical pharmacies are limited. Therefore, the omni-channel operation of drugs has the characteristics of multiple inventory locations and overlapping inventories; in terms of distribution, the treatment of a disease often requires the combination of multiple drugs. Drug compatibility is also a special attribute of drugs. Drug orders often have the attribute of "one order, multiple products", and order demand is highly random. Considering the special nature of drug commodities, drug distribution needs to ensure that the order is delivered in full. To sum up, order fulfillment optimization for omni-channel retail of medicines is a collaborative decision-making on order allocation, collaborative multi-commodity network flow, and vehicle scheduling problems. It is a new type of collaborative decision-making problem with a huge decision-making space and high requirements on solution speed and quality. It is necessary to build a new collaborative decision-making model and the corresponding efficient and fast optimization algorithm to ensure efficient and timely omni-channel drug order delivery.

[0004] In view of the characteristics of the pharmaceutical industry with a wide variety of products and scattered inventory, as well as the usage characteristics of drug compatibility and time requirements, the optimization of the fulfillment of omni-channel retail orders for drugs comprehensively considers multiple inventory locations, order time restrictions, and vehicle distribution in multiple parking lots. It includes three mutually coupled NP-hard problems: order allocation problem, multi-commodity network flow problem, and vehicle routing problem. This problem is a coupled integration of two scientific problems: the unpaired pickup and delivery vehicle routing problem (UPDVRP) and the multiple pickup-delivery vehicle routing problem with time windows (MPD-VRP-TW). It is a new type of scientific problem that has emerged with the popularization of omni-channel retail of drugs. The present invention combines these two problems for joint optimization, and the complexity of the problem is higher. At present, in the field of omni-channel operation optimization, there is no patent on the integrated problem of omni-channel collaborative order fulfillment of drugs with time constraints and related optimization methods. In terms of academic research, although there are results of optimization methods to solve the two scientific problems of UPDVRP and MPD-VRP-TW, there are no relevant research results for the new type of order fulfillment problem of UPDVRP and MPD-VRP-TW coupling considered in this patent. This type of new problem has the characteristics of large decision space and complex and irregular relationship between decision variables. Since the precise algorithm takes a long time to solve, it is difficult to meet the application practice; and the existing heuristic algorithm for solving this type of irregular relationship between decision variables has the defects of low search efficiency and poor solution effect, so the existing optimization method is difficult to use to solve this type of new problem.

[0005] In the context of omni-channel operation of medicines, for a batch of multi-product online orders received at a certain moment in the city, combined with the demand for medicines in the order and, considering the inventory distribution of physical pharmacies and the location of vehicles distributed in the urban road network, decisions are made on which pharmacies the medicines in the order are allocated to, and how to arrange vehicles for collaborative delivery of orders. The decision content includes decision-making order allocation, vehicle dispatch and collaborative delivery routes for pickup and delivery, etc., to reduce the operating cost of order fulfillment. The inventory of the omni-channel model has the characteristics of being dispersed in location and shared logically. There is a demand for inventory collaboration, which further increases the difficulty of omni-channel order fulfillment decisions. In this invention, due to inventory holding restrictions, multiple physical pharmacies may need to collaborate to complete the preparation of the same order, that is, different drug demands in the same order may be allocated to different physical pharmacies for preparation, and then the order is fully delivered through the sequential visits of the same vehicle, that is, the order demand is completed through one delivery. In corporate practice, when facing such problems, greedy rules are often used for decision-making. However, since the problem contains multiple decision sub-problems and there is a coupling relationship between the sub-problems, the decision variables of the problem have a complex and irregular structure, and the decision method based on greedy rules is less effective. Summary of the invention

[0006] The present invention is in the omni-channel operation mode, according to the inventory characteristics of multiple inventory locations and decentralized inventory of omni-channel drug retail and the order characteristics of "one order, multiple products" and complete order delivery, comprehensively considers inventory restrictions and time requirements, and makes decisions such as order allocation, collaborative multi-commodity network flow, and vehicle scheduling among multiple fulfillment entities, thereby reducing the fulfillment cost of omni-channel retail orders. Under the premise of considering the coupling relationship of multiple inventory locations, the present invention makes joint decisions on problems such as order allocation, multi-commodity network flow, and vehicle path problems, and proposes a decomposition-iteration-search omni-channel drug retail order collaborative fulfillment decision optimization method. Based on the idea of ​​decomposition and iteration, the present invention designs and implements a decomposition and iteration search algorithm, decouples and decomposes the associated sub-problems in the complex joint decision problem, and iteratively realizes the joint solution through the iterative parameter transfer obtained by the sub-problem solution, thereby obtaining the solution to the original problem, which can scientifically and effectively solve the difficult problem of decision sub-problems containing multiple couplings. The present invention can provide decision support for omni-channel operations or logistics park distribution scheduling with the same needs, and improve the scientificity, effectiveness and efficiency of the operation plan.

[0007] The basic principle of the present invention is to describe the collaborative fulfillment problem of omni-channel drug retail orders as an operational problem in which an omni-channel drug retailer in a certain area uses a drugstore network consisting of multiple physical drugstores and its own distribution fleet to meet the drug order needs of residents in the area. The retailer operates multiple varieties of drugs, and the inventory varieties between drugstores are partially overlapping, providing consumers with order delivery time commitments. Due to the strict storage and transportation conditions of drugs, temperature control must be carried out during the distribution process. Therefore, there are drug temperature control transportation costs related to transportation time during the distribution process, and this cost is related to the drug category. It is assumed that the order quantity of the same drug in a single order is at most one, because the processing method of multiple request orders is basically the same as that of a single request order. At the same time, the vehicles of the distribution fleet are scattered in the road network, and all vehicles have transportation volume restrictions. Under such conditions, the present invention adopts the research idea of ​​decomposition-iteration-search based on inventory information and order data, splits the problem into two sub-problems, and then designs an improved Hungarian-Dijkstra algorithm (MHDA) and a greedy search algorithm (GSA) to solve the two sub-problems respectively, adjusts the iterative process of the algorithm through the commodity flow variables until the commodity flow variables obtained from the two sub-problems can be matched, and finally uses the overall variable neighborhood search algorithm to further optimize the iterative solution to obtain the problem result.

[0008] The technical solution of the present invention:

[0009] A collaborative fulfillment optimization method for omni-channel drug retail orders, the adopted decision support system includes an initialization module, an information input module, a calculation module 1, a calculation module 2, a calculation module 3, an iterative coordination module, a calculation module 4 and a decision output module. The functions of each module are as follows: the initialization module is used to obtain entity network data and fleet data as parameters and resource constraint information of the decision model; the information input module is used to input order demand data of a certain batch; calculation module 1 is used to model the collaborative fulfillment problem of omni-channel drug retail orders, and decompose the original problem model according to the entity network data and order demand information to obtain two sub-models (model A and model B) and related parameters; calculation module 2 uses the improved Hungarian-Dijkstra algorithm (MHDA) to solve model A of calculation module 1, and obtain the distribution plan of drugs in the order, as well as the distribution information of orders and vehicles, as the input and iteration parameters of calculation module 3; calculation module 3 uses a search algorithm based on multiple neighborhoods to solve model B of calculation module 1 based on the distribution plan obtained by calculation module 2, and obtains a complete order fulfillment plan, and obtains new order and vehicle distribution information as iteration parameters; the iterative coordination module judges the distribution of calculation module 2 and calculation module 3. The iterative program is controlled by whether the order matches the allocation information of the vehicle; the calculation module 4 searches and optimizes the solution after the iteration through the neighborhood search; the decision output module is used to represent the decision result obtained by the calculation module 4 in a readable form and provide it to the decision maker as a decision support.

[0010] Based on the above ideas and system module description, the specific steps of the technical solution of the present invention are as follows:

[0011] (1) Initialization module: problem parameter setting

[0012] Step 1: The drugstore network in a certain area is represented as a network consisting of |P| physical drugstores scattered in the area, providing |M| kinds of drugs for the area. The drug inventory of each physical drugstore is partially overlapping but not exactly the same. At the same time, due to the strict requirements of drug transportation conditions, each drug has a corresponding temperature-controlled transportation cost. In addition, |K| homogeneous vehicles are distributed at different locations in the road network of the area. The vehicles have a volume limit V, and the vehicles have time parameters and cost parameters in the road network. These network settings constitute the decision problem parameters of the problem and serve as resource constraint information for subsequent models.

[0013] (2) Information input module

[0014] Step 2: Input the data of a batch of |O| order requirements, mainly including order location, order time, and drug requirements in the order.

[0015] (3) Computational Module 1: Problem Modeling and Problem Decomposition

[0016] Step 3.1: Problem Modeling

[0017] Based on the initialization module and the information input module, the collaborative fulfillment problem of omni-channel drug retail orders is described, and a mixed integer programming mathematical model TD-UMPD-VRPT is established. The main decisions include vehicle dispatch decision U, drug allocation decision Y in the order, commodity flow decision Z, and vehicle flow decision X.

[0018] Step 3.2: Problem decomposition

[0019] The original problem model is decomposed according to the entity network data such as pharmacy distribution and pharmacy inventory data and order demand information to obtain two sub-models (model A and model B) after decomposition, and effective constraints are added to the sub-models through corresponding coupling constraints and parameters to facilitate subsequent iterations;

[0020] (4) Calculation module 2: solving model A and corresponding iteration parameters

[0021] Step 4.1: Solution of Model A

[0022] The improved Hungarian-Dijkstra algorithm (MHDA) is obtained by embedding the Dijkstra algorithm in the Hungarian algorithm to solve model A. That is, the Dijkstra algorithm is used to solve the multi-vehicle shortest path problem to obtain the cost matrix corresponding to the multi-drug commodity flow, and the Hungarian algorithm is used according to the cost matrix to solve the drug allocation problem in the order, and the order allocation result including the multi-commodity network flow is obtained.

[0023] Step 4.2: Get the iteration parameters

[0024] Based on the solution of model A, the distribution plan of drugs in the order and the distribution information of orders and vehicles are obtained as input and iteration parameters for the next step;

[0025] (5) Calculation module 3: solving model B and corresponding iteration parameters

[0026] Step 5.1: Solution of Model B

[0027] Based on the allocation plan obtained in step 4.2, the greedy search algorithm (GSA) is used to obtain the initial path plan, and the initial path plan is improved through neighborhood search to obtain the solution result of model B and the complete order fulfillment plan.

[0028] Step 5.2: Solution of Model B

[0029] According to the solution of model B, the new order and vehicle allocation information is obtained as the iteration parameter;

[0030] (6) Iteration coordination module

[0031] Step 6: Determine whether the order and vehicle allocation information obtained in step 4.2 and step 5.2 match. If they match, the final iterative order fulfillment plan is obtained and the next step is performed; otherwise, the order and vehicle allocation information in step 5.2 is fed back to step 4 and the iterative procedure is executed again;

[0032] (7) Computation Module 4: Search Improvement

[0033] Step 7: Use the neighborhood structure to design a variable neighborhood search algorithm to further search and optimize the final iterative order fulfillment solution obtained in step 6.

[0034] (8) Decision output module

[0035] Step 8: Convert the vehicle dispatch decision U, order drug allocation decision Y, commodity flow decision Z, and vehicle flow decision X obtained in step 7 into a readable form and provide them to decision makers as decision support.

[0036] Effects and benefits of the present invention:

[0037] 1. Aiming at the omni-channel drug retail operation, the present invention proposes a new order fulfillment problem caused by the characteristics of omni-channel and drug retail, and formally describes the problem through modeling;

[0038] 2. The present invention takes into account the characteristics of inventory dispersion, drug compatibility, time requirements, etc., and adopts a decomposition research approach to effectively solve the problem of fulfilling omni-channel retail orders for drugs in multiple inventory locations with full order delivery requirements. It can make full use of the information between decision sub-problems and optimize the order processing process and vehicle routing;

[0039] 3. The present invention provides decision support for omni-channel drug retailers to carry out distribution practices, which can improve the scientific nature of order fulfillment plan decisions and the efficiency of order processing, thereby improving the efficiency of enterprise operations and resource utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic diagram of the structural relationship between various system modules of the decision support system in the collaborative fulfillment method of the omni-channel drug retail order of the present invention.

[0041] Figure 2 It is a framework diagram of the collaborative fulfillment optimization method for omni-channel drug retail orders of the present invention. DETAILED DESCRIPTION

[0042] In order to make the content and advantages of the present invention more clear, the specific implementation methods of the present invention are described in detail and completely in combination with the technical solutions and drawings below.

[0043] Figure 1 It is a schematic diagram of the structural relationship between the system modules of the decision support system for collaborative fulfillment of omni-channel drug retail orders of the present invention. It can be seen that the decision support system for collaborative fulfillment of omni-channel drug retail orders of the present invention is composed of an initialization module, an information input module, a calculation module 1, a calculation module 2, a calculation module 3, an iterative coordination module, a calculation module 4, and a decision output module. The initialization module is the information of the enterprise's operating resources, and the information input module receives the omni-channel order information. Then, for the order information, under the conditions of the existing operating resources, the calculation modules 1, 2, 3, iterative coordination module, and 4 calculate the allocation plan and delivery path plan of this part of the order to complete the decision of collaborative order fulfillment. Reference Figure 1 It can be seen that the collaborative fulfillment optimization method for omni-channel drug retail orders provided by the present invention includes the following steps:

[0044] Step 1: Initialize the module, input the operational resources into the system, determine the decision problem parameters and the corresponding resource constraints; the relevant parameters include the location of the physical pharmacy, the inventory status of the pharmacy, the current location of the vehicle, the time parameters and cost parameters of the vehicle's travel, and the temperature-controlled transportation cost parameters of the medicine.

[0045] Step 2: Information input module, input the data of a batch of order requirements, mainly including order location, order time, and drug requirements in the order.

[0046] Step 3: Calculation module 1, based on the initialization module and the information input module, describes the fulfillment problem of omni-channel drug retail order coordination, and establishes a mixed integer programming mathematical model. The main decisions include vehicle dispatch decision U, drug allocation decision Y in the order, commodity flow decision Z, and vehicle flow decision X. The original problem model is decomposed according to the entity network data such as drug store distribution and drug store inventory data and order demand information to obtain two sub-models (model A and model B) after decomposition, and effective constraints are added to the sub-models through corresponding coupling constraints and parameters to facilitate subsequent iterations;

[0047] Step 4: Calculation module 2 uses the improved Hungarian-Dijkstra algorithm (MHDA) to solve model A and obtain the order allocation result including the multi-commodity network flow, and then obtain the allocation plan of the drugs in the order, as well as the allocation information of the order and the vehicle, which are used as the input and iteration parameters of the next step;

[0048] Step 5: Calculation module 3, based on the allocation plan obtained in step 4, uses the greedy search algorithm (GSA) to obtain the initial path plan, and improves the initial path plan through neighborhood search, and then obtains the solution result of model B and the complete order fulfillment plan. Then, the new order and vehicle allocation information is obtained as the iteration parameter;

[0049] Step 6: The iterative coordination module determines whether the order and vehicle allocation information obtained in step 4 and step 5 match. If they match, the final iterative order fulfillment plan is obtained and the next step is performed; otherwise, the order and vehicle allocation information in step 5 is fed back to step 4 and the iterative procedure is re-executed;

[0050] Step 7: Calculation module 4 uses a suitable neighborhood structure to design a variable neighborhood search algorithm to further search and optimize the final iterative order fulfillment solution obtained in step 6.

[0051] Step 8: The decision output module converts the vehicle dispatch decision U, the drug allocation decision Y, the commodity flow decision Z, and the vehicle flow decision X obtained in step 7 into a readable form and provides them to the decision maker as decision support.

[0052] In the method provided by the present invention, a TD-UMPD-VRPT (time-dependent unpaired-multiple pickup and delivery vehicle routing problem with delivery time commitment) model with time commitment and time dependence can be established through operation resource information and order demand information, and a decomposition-iteration-search algorithm is designed to decompose the problem into two sub-problems for solving them separately, and then the iterative process of the algorithm is adjusted by the commodity flow variable until the iterative program ends, and then a suitable neighborhood structure is designed to further optimize the iterative results through variable neighborhood search, and finally a collaborative order fulfillment solution is obtained. Based on the method provided by the present invention, under the omni-channel drug retail model, the order fulfillment characteristics and drug demand characteristics of the omni-channel are fully considered, and an optimization solution is provided for physical drugstore stocking and vehicle distribution in omni-channel operations.

[0053] Combination Figure 2 The framework diagram of the collaborative fulfillment optimization method of the omni-channel drug retail order of the present invention is as follows: Taking an omni-channel drugstore network consisting of |P| physical drugstores, |K| vehicles, and |M| drug inventories, and deciding on a collaborative order fulfillment plan for |O| order requirements as an example, the specific implementation steps of the collaborative fulfillment optimization method of the omni-channel drug retail order proposed by the present invention are as follows:

[0054] Step 1: You need to first determine the operational resource allocation of the omni-channel retailer, where M is the drug set, K is the vehicle set, P is the physical drugstore set, and N is the P Represents a collection of pharmacy locations. In terms of parameter settings, inventory status s pm represents the inventory of drug m held by pharmacy p, p∈NP ,m∈M;v m represents the volume of drug m, V is the capacity of the vehicle; α is the fixed cost of the vehicle, β ij is the path cost of the vehicle from point i to point j, θ ij is the path time of the vehicle from point i to point j; γ m is the unit time temperature-controlled transportation cost of drug m.

[0055] Step 2: Determine the order demand information received, O is the order set, N o represents the order location set, N is the set of all points in the region, N = N P ∪N O , where i∈N is represented by plane coordinates (i a ,i b ) indicates; d o is the delivery deadline of order o, O∈O.

[0056] Step 3: Calculation module 1, based on the operational resource configuration of the initialization module and the order demand information of the information input module, describes the collaborative fulfillment problem of omni-channel drug retail orders, constructs the TD-UMPD-VRPT model, and obtains the order fulfillment plan including the commodity flow and vehicle flow according to the information such as the pharmacy distribution, pharmacy inventory data, and order demand. The objective function of the TD-UMPD-VRPT model is expressed by formula (1):

[0057]

[0058] Among them, M is the drug set, M o represents the set of drugs in order O, K is the set of vehicles, P is the set of physical pharmacies, N P represents the set of pharmacy locations, O is the set of orders, N O represents the order location set, N is the set of all points in the region, N = N P ∪N O , where i∈N is represented by plane coordinates (i a ,i b ) represents; α is the fixed cost of using the vehicle, β ij is the path cost of the vehicle from point i to point j, θ ij is the path time of the vehicle from point i to point j; γ m is the unit time temperature-controlled transportation cost of drug m; e k is the current position of vehicle k, e ′ is the virtual end point, and the distance from all points to the virtual end point is zero; min is the minimum function, and the 0-1 decision variable u k Indicates whether to use vehicle k, 0-1 decision variable x kijIndicates whether vehicle k passes through the path from point i to point j, a 0-1 decision variable Indicates whether vehicle k carries the medicine m in order o through the path from point i to point j.

[0059] The constraints of the TD-UMPD-VRPT model are expressed using formulas (2)-(20):

[0060] (a) Constraints on order splitting and allocation decisions

[0061]

[0062]

[0063] (b) Coupling constraints between order allocation and multi-product flow decisions

[0064]

[0065]

[0066]

[0067] (c) Constraints related to multi-commodity flow decisions

[0068]

[0069]

[0070]

[0071] (d) Coupling constraints between multi-commodity flows and vehicle routing decisions

[0072]

[0073] (e) Constraints related to delivery vehicle routing decisions

[0074]

[0075]

[0076]

[0077]

[0078]

[0079]

[0080]

[0081]

[0082]

[0083]

[0084] Among them, 0-1 type decision variables Indicates that drug m in order o is stocked by pharmacy p, and 0-1 decision variable b kij is an auxiliary variable, indicating whether node i is before node j in the path of vehicle k (not necessarily between adjacent nodes). L is a real number not less than 10^6, and the continuous decision variable t ki and They represent the time when vehicle k arrives at and leaves point i. Inventory status s pm represents the inventory of drug m held by pharmacy p, p∈N P , m∈M;v m represents the volume of the drug m, V is the capacity of the vehicle; d o is the delivery deadline of order o, o∈O.

[0085] By removing formula (10), the model can be decomposed into two sub-problems: model A and model B.

[0086] Model A:

[0087]

[0088] Constraints: including formula (2), formula (3), formula (8) and

[0089]

[0090]

[0091]

[0092]

[0093]

[0094] Among them, K O is the order-vehicle matching information set, k o Refers to the vehicle responsible for order o. By transforming the order-vehicle matching information set K O Incorporating it into model A, some constraints in the original model are updated to new constraints by taking the order-vehicle matching information set K OIncorporated into model A, some constraints in the original model are updated to new constraint formulas (22)-(26). In the first iteration, it can be set that all orders are assigned to a certain vehicle by default. For example, in the first iteration, set K1 O All orders are assigned to vehicle 1, which will not affect the solution of model A in the first iteration.

[0095] Model B:

[0096]

[0097] Constraints: Formulas (11)-(20) and

[0098]

[0099] Among them, N + is the set of physical pharmacies that have been assigned tasks. Formula (28) indicates that the physical pharmacies that have been assigned tasks have been visited at least once. This formula uses the solution of model A as the input of model B. The vehicle flow decision x of model B kij Corresponding to the corresponding commodity flow variable Therefore, solving model B can obtain the new order-vehicle matching pair information K O , and feed this information back to model A for further iteration.

[0100] The introduction of new constraints for Model A and Model B realizes the transfer and connection of the two sub-problem solving processes, ensuring that the solution to the problem is continuously corrected and optimized as the iteration process proceeds. Solve the two sub-problems separately, and control the iteration process through the iteration parameters obtained from the sub-problems until the iteration parameters match and complete the joint solution to obtain the solution to the original problem.

[0101] Step 4: Calculation module 2 uses the improved Hungarian-Dijkstra algorithm (MHDA) to solve model A. Model A mainly includes two decisions, namely order allocation decision and multi-commodity network flow decision. The objective function is to minimize the total commodity flow cost. The total commodity flow cost in model A is the sum of the flow costs of each drug m for each order o, that is, the assignment problem of each drug is independent of each other. Therefore, the corresponding assignment problem can be solved for each drug, that is, given the cost matrix, the drug allocation decision can be made by solving the standard allocation problem (AP). Similarly, given the commodity allocation plan, minimizing the commodity flow cost can be regarded as a multi-vehicle shortest path problem with fixed start and end nodes, so model A can be expressed as a combination of AP and MVSP. Therefore, model A can be transformed into a combination of the standard allocation problem (AP) of multiple drugs m and the multi-vehicle shortest path problem (MVSP), and the Dijkstra algorithm is embedded in the Hungarian algorithm. A polynomial time optimization algorithm (improved Hungarian-Dijkstra algorithm, MHDA) is constructed to solve this model. That is, the cost matrix composed of the commodity flow cost of each drug is obtained by solving the MVSP problem with the Dijkstra algorithm as the parameter of the AP problem, and the allocation plan of the drugs in the order is obtained by solving the AP problem with the Hungarian algorithm.

[0102] The drug demand in the order that each pharmacy holding each drug can provide is determined based on the inventory situation, and then the commodity flow cost of each drug assigned to all pharmacies holding the drug inventory is calculated. More specifically, the element d in the cost matrix of drug m ab It means that pharmacy a holding drug m is responsible for the commodity flow cost of order drug pair b, where the order drug pair is order o with demand for drug m. m The information pair composed of drug m is the task index information of each column in the cost matrix, and the execution subject index information of each row in the cost matrix is ​​the pharmacy a holding drug m. Cost parameter d ab is from pharmacy a to order o m The minimum commodity flow cost of the location node includes the commodity flow information and the responsible vehicle information. Through the solution Y1 of model A, the distribution plan M of the medicine in the order can be obtained P , and order and vehicle allocation information As input and iteration parameter for the next step.

[0103] The following is an example of how to obtain the iteration parameters from the solution of model A.

[0104] For example, the solution for model A is Among them, the first column is the order number, the second column is the drug number, and the third column is the pharmacy number. From Y1, we can conclude that order 1 is assigned to pharmacy 1 and pharmacy 2 for stocking, order 2 is assigned to pharmacy 3 for stocking, and order 3 is assigned to pharmacy 3 and pharmacy 4. In addition, the pharmacy allocation result of the drugs in the order is M P Pharmacy 1 is responsible for drug 5 in order 1; Pharmacy 2 is responsible for drug 2 in order 1; Pharmacy 3 is responsible for drug 1 and drug 4 in order 2, and drug 2 in order 3; Pharmacy 3 is responsible for drug 1 and drug 3 in order 3; These results will serve as input to model B in the next step.

[0105] Step 5: Calculate the distribution plan M of the medicines in the order obtained by module 3 and step 4 P is the input of model B. A greedy search algorithm (GSA) is designed. An effective neighborhood structure is designed for the multiple pickup characteristics of model B to improve the search efficiency of the algorithm. GSA includes two operations: path generation and local search. The former generates a path solution by inserting nodes in the path with the minimum insertion cost, and the latter generates a path solution by adding nodes to the decision variable u. k and Neighborhood search is used to improve the path plan.

[0106] Taking vehicle k as an example, first, we obtain the order and vehicle allocation information based on model A. Generate feasible vehicle paths for vehicles with order tasks. Based on the access sequence, vehicle loading and delivery time constraints, find the candidate nodes that can be inserted into the current path of vehicle k. Then, select the candidate node position with the lowest insertion cost to insert the node to be visited. Greedily insert the nodes to be visited one by one, and update the path content of vehicle k until all order requirements are met, that is, a feasible path plan is obtained; if some order tasks cannot be delivered by the corresponding vehicles, the order is assigned to the unused vehicles. After generating the initial routing plan, apply local search to optimize the path plan. Finally, the result of solving model B is obtained, that is, the complete order fulfillment plan is obtained. Based on the path plan, the new order and vehicle allocation information can be obtained, which is used as an iteration parameter and fed back to the model.

[0107] The following is an example of how to obtain the iteration parameters from the solution of model B.

[0108] For example, the solution for model B is The first column is the order number, the second column is the drug number, the third and fourth columns are the arcs in the path, and the fifth column is the vehicle. From Z, we can get the new order and vehicle allocation information Vehicle 1 is responsible for order 1, and vehicle 2 is responsible for order 2 and order 3. These results will be fed back into model A in the previous step as iterative parameters.

[0109] Step 6: Iterate the coordination module, and obtain the optimal solution allocation plan Y0 and commodity flow information Z0 by solving model A, and then obtain the allocation plan of each drug in each order, the pickup task of each pharmacy, and the vehicle information matched by each order. They are passed to model B as input information of model B. Then, the vehicle routing solution is obtained by solving model B, and then the new order-vehicle allocation information can be obtained.

[0110] judge and Is there a match? If so, the iteration process is completed to obtain the final iterative order fulfillment plan and proceed to the next step of search improvement; if not, the order-vehicle allocation information obtained from the results of model B is used. This is used as feedback information to input into model A and the iterative procedure is re-executed.

[0111] Step 7: Calculation module 4, in order to improve the order fulfillment plan obtained after the iteration, a suitable neighborhood structure is used to design a variable neighborhood search algorithm to further search and optimize the final iterative order fulfillment plan. The main idea is to destroy the final iterative order fulfillment plan through different neighborhood structures, obtain the set of order requirements to be met, and then apply the local search procedure based on the minimum insertion cost to reinsert the order requirements to be met into the plan to meet all order requirements and obtain the order fulfillment plan of the information. According to the characteristics of the problem, the designed neighborhood structure mainly includes three types: exchanging all the pick-up and delivery tasks of the two vehicles, adjusting some allocation results, and adjusting some path results.

[0112] Step 8: Decision output module, the final vehicle dispatch decision u k , Drug allocation decision in order Product flow decision Vehicle flow decision x kij The results are converted into readable form and provided to decision makers as decision support.

Claims

1. A collaborative fulfillment optimization method for omni-channel drug retail orders, characterized in that: The decision support system adopted includes an initialization module, an information input module, a calculation module 1, a calculation module 2, a calculation module 3, an iteration coordination module, a calculation module 4 and a decision output module; the functions of each module are as follows: the initialization module is used to obtain entity network data and fleet data as parameters and resource constraint information of the decision model; the information input module is used to input order demand data of a certain batch; The calculation module 1 is used to model the collaborative fulfillment problem of omni-channel drug retail orders, and decompose the original problem model according to the entity network data and order demand information to obtain two sub-models and related parameters; The calculation module 2 uses the improved Hungarian-Dijkstra algorithm MHDA to solve the model A of the calculation module 1, and obtains the distribution plan of the drugs in the order, as well as the distribution information of the order and the vehicle, as the input and iteration parameter of the calculation module 3; the calculation module 3 uses the search algorithm based on multiple neighborhoods to solve the model B of the calculation module 1 based on the distribution plan obtained by the calculation module 2, and obtains a complete order fulfillment plan, and obtains the new order and vehicle distribution information as the iteration parameter; the iteration coordination module determines whether the order and vehicle distribution information of the calculation module 2 and the calculation module 3 match, so as to control the iteration program; the calculation module 4 searches and optimizes the solution after the iteration is completed through the neighborhood search; the decision output module is used to represent the decision result obtained by the calculation module 4 in a readable form and provide it to the decision maker as a decision support; the specific steps are as follows: (1) Initialization module: problem parameter setting The drugstore network in a certain area is represented as a network consisting of |P| physical drugstores scattered in the area, providing |M| kinds of drugs for the area. The drug inventory of each physical drugstore is partially overlapping but not exactly the same. At the same time, due to the strict requirements of drug transportation conditions, each drug has a corresponding temperature-controlled transportation cost; in addition, |K| homogeneous vehicles are distributed at different locations in the road network of the area. The vehicles have a volume limit V, and the vehicles have time parameters and cost parameters in the road network; (2) Information input module Input the data of a batch of |O| orders, including order location, order time, and drug demand in the order; (3) Computational Module 1: Problem Modeling and Problem Decomposition Step 3.1: Problem Modeling Based on the initialization module and information input module, the collaborative fulfillment problem of omni-channel drug retail orders is described, and a mixed integer programming mathematical model TD-UMPD-VRPT is established. The main decisions include vehicle dispatch decision U, drug allocation decision Y in the order, commodity flow decision Z, and vehicle flow decision X. Step 3.2: Problem decomposition The original problem model is decomposed according to the entity network data such as pharmacy distribution and pharmacy inventory data and order demand information to obtain two sub-models, namely model A and model B. Effective constraints are added to the sub-models through corresponding coupling constraints and parameters to facilitate subsequent iterations. (4) Calculation module 2: solving model A and corresponding iteration parameters Step 4.1: Solution of Model A The improved Hungarian-Dijkstra algorithm MHDA solution model A is obtained by embedding the Dijkstra algorithm in the Hungarian algorithm, that is, the Dijkstra algorithm is used to solve the multi-vehicle shortest path problem to obtain the cost matrix corresponding to the multi-drug commodity flow, and the Hungarian algorithm is used according to the cost matrix to solve the drug allocation problem in the order, and the order allocation result including the multi-commodity network flow is obtained; Step 4.2: Get the iteration parameters Based on the solution of model A, the distribution plan of drugs in the order and the distribution information of orders and vehicles are obtained as input and iteration parameters for the next step; (5) Calculation module 3: solving model B and corresponding iteration parameters Step 5.1: Solution of Model B Based on the allocation plan obtained in step 4.2, the greedy search algorithm GSA is used to obtain the initial path plan, and the initial path plan is improved through neighborhood search to obtain the solution result of model B and obtain the complete order fulfillment plan; Step 5.2: Solution of Model B According to the solution of model B, the new order and vehicle allocation information is obtained as the iteration parameter; (6) Iteration coordination module Determine whether the order obtained in step 4.2 and step 5.2 matches the vehicle allocation information. If so, obtain the final iterative order fulfillment plan and proceed to the next step; Otherwise, the order and vehicle allocation information of step 5.2 is fed back to step 4, and the iterative procedure is executed again; (7) Computation Module 4: Search Improvement The variable neighborhood search algorithm is used to design the neighborhood structure to further search and optimize the final iterative order fulfillment solution obtained in step 6; (8) Decision output module Convert the vehicle dispatch decision U, order drug allocation decision Y, commodity flow decision Z, and vehicle flow decision X obtained in step (7) into a readable form and provide them to the decision maker as decision support; The objective function of the TD-UMPD-VRPT model is expressed by formula (1): Among them, M is the drug set, M o represents the set of drugs in order o, K is the set of vehicles, P is the set of physical pharmacies, N P represents the set of pharmacy locations, O is the set of orders, and N O represents the order location set, N is the set of all points in the region, N = N P ∪N O , where i∈N is represented by plane coordinates (i a ,i b ) represents; α is the fixed cost of using the vehicle, β ij is the path cost of the vehicle from point i to point j, θ ij is the path time of the vehicle from point i to point j; γ m is the unit time temperature-controlled transportation cost of drug m; e k is the current position of vehicle k, e ′ is the virtual end point, and the distance from all points to the virtual end point is zero; min is the minimum function, and the 0-1 decision variable u k Indicates whether to use vehicle k, 0-1 decision variable x kij Indicates whether vehicle k passes through the path from point i to point j, a 0-1 decision variable Indicates whether vehicle k carries the medicine m in order o through the path from point i to point j.

2. The method for optimizing the collaborative fulfillment of omni-channel drug retail orders according to claim 1, characterized in that: The constraints of the TD-UMPD-VRPT model are expressed using formulas (2)-(20): (a) Constraints on order splitting and allocation decisions (b) Coupling constraints between order allocation and multi-product flow decisions (c) Constraints related to multi-commodity flow decisions (d) Coupling constraints between multi-commodity flows and vehicle routing decisions (e) Constraints related to delivery vehicle routing decisions Among them, 0-1 type decision variables Indicates that drug m in order o is stocked by pharmacy p, and 0-1 decision variable b kij is an auxiliary variable, indicating whether node i is before node j in the path of vehicle k; L is a real number not less than 10^6; continuous decision variable t ki and They represent the time when vehicle k arrives at and leaves point i; the inventory status s pm represents the inventory of drug m held by pharmacy p, p∈N P , m∈M;v m represents the volume of the drug m, V is the capacity of the vehicle; d o is the delivery deadline of order o, o∈O; By removing formula (10), the model is decomposed into two sub-problems: model A and model B; Model A: Constraints: including formula (2), formula (3), formula (8) and Among them, K O is the order-vehicle matching information set, k o Refers to the vehicle responsible for order o; by taking the order-vehicle matching information set K O Incorporating it into model A, some constraints in the original model are updated to new constraints by taking the order-vehicle matching information set K O Incorporated into model A, some constraints in the original model are updated to new constraint formulas (22)-(26); in the first iteration, it is set that all orders are assigned to a certain vehicle by default; Model B: Constraints: Formulas (11)-(20) and Among them, N + is the set of physical pharmacies that have been assigned tasks. Formula (28) indicates that the physical pharmacies that have been assigned tasks have been visited at least once. This formula uses the solution of model A as the input of model B. The vehicle flow decision x of model B kij Corresponding to the corresponding commodity flow variable Therefore, solving model B can obtain the new order-vehicle matching pair information K O , and feed this information back to model A for further iteration.

Images