Inventory management optimization method for preventing fast moving consumer goods shortage in retail supply chain

Abstract

The invention discloses an inventory management optimization method for preventing fast moving consumer goods shortage in a retail supply chain, and belongs to the field of inventory management optimization, and the method comprises the following steps: S1, building a fast moving consumer goods replenishment optimization model by taking the minimization of replenishment cost as a target function; s2, introducing a deep learning model combining reinforcement learning and a Q network, modeling the decision process of the fast moving consumer goods into a Markov decision process, training the deep learning model by using a deep Q network algorithm, and solving the fast moving consumer goods replenishment optimization model by using the deep learning model to obtain an optimal replenishment strategy; and S3, executing the optimal replenishment strategy to replenish the fast moving consumer goods. According to the invention, the replenishment decision can be optimized, the overall inventory cost can be reduced, the occurrence rate of inventory shortage can be significantly reduced, the replenishment strategy can be dynamically adjusted according to the actual situation through online learning and real-time monitoring, and the real-time performance and effectiveness of the replenishment decision can be ensured.

Description

Technical Field

[0001] The present invention belongs to the field of inventory management optimization, and in particular relates to an inventory management optimization method for preventing shortages of fast-moving consumer goods in a retail supply chain. Background Art

[0002] In the retail supply chain, fast moving consumer goods (FMCG) refer to products with high consumption frequency and high demand, such as food, beverages, daily necessities, etc. Due to the large fluctuations in demand for these products, the difficulty of inventory management also increases. Traditional inventory management methods, such as an inventory management method based on daily sales statistics disclosed in the invention patent application with application publication number CN108446876A, include: step S1, counting the average daily sales volume DMS of the previous day; step S2, counting the actual daily sales volume of the day; step S3, calculating the DMS of the day based on the average daily sales volume DMS of the previous day and the inventory of the day, judging the type of the DMS of the day, and managing the inventory according to the type of the DMS.

[0003] Traditional inventory management methods mainly rely on experience and simple statistical models, which cannot effectively cope with complex and changing market demands, resulting in high inventory costs and frequent inventory shortages. Inventory shortages also directly affect retailers' sales and customer satisfaction. When customers cannot buy the fast-moving consumer goods they need, they may turn to other competitors, resulting in customer loss. In addition, frequent inventory shortages will increase retailers' replenishment pressure and costs, and reduce the operating efficiency of the supply chain. Summary of the invention

[0004] The purpose of the present invention is to provide an inventory management optimization method for preventing shortages of fast-moving consumer goods in a retail supply chain, so as to solve the problems of high inventory costs and frequent inventory shortages in traditional inventory management methods.

[0005] In order to achieve the above object, the technical solution of the present invention is as follows:

[0006] The present invention relates to an inventory management optimization method for preventing shortage of fast-moving consumer goods in a retail supply chain, which comprises the following steps:

[0007] S1. Taking minimizing the replenishment cost as the objective function, a fast-moving consumer goods replenishment optimization model is constructed;

[0008] S2. Introduce a deep learning model that combines reinforcement learning and Q-network, and model the decision-making process of fast-moving consumer goods as a Markov decision process. Use the deep Q-network algorithm to train the deep learning model, and use the deep learning model to solve the fast-moving consumer goods replenishment optimization model to obtain the optimal replenishment strategy.

[0009] S3. Execute the optimal replenishment strategy to replenish fast-moving consumer goods.

[0010] Preferably, the expression of the objective function in S1 is:

[0011]

[0012] Among them, α i is the weight parameter of fast-moving consumer goods i, p i (t) is the replenishment cost of fast-moving consumer product i at time t, x i (t) is the replenishment quantity of fast-moving consumer product i at time t, α i p i (t)x i (t) is the purchase cost of fast-moving consumer goods, v i (t) is the inventory of fast-moving consumer goods i at time t, r i (t) is the storage cost of fast-moving consumer product i at time t, R i (v i (t),r i (t)) is the storage cost function of fast-moving consumer product i, is the shortage penalty cost function for FMCG product i;

[0013] Preferably, the S1 also constructs the inventory capacity constraint of the fast-moving consumer goods replenishment optimization model, and the expression of the inventory capacity constraint is:

[0014]

[0015] Among them, C i (t) is the minimum inventory capacity of fast-moving consumer product i, v i (t) and v i (t-1) are the inventory of fast-moving consumer product i at time t and time t-1, respectively, x i (t) is the replenishment quantity of fast-moving consumer product i at time t, is the maximum inventory capacity of fast-moving consumer goods i, λ i (t) is the demand for fast-moving consumer product i, e i (t) is the expired quantity of fast-moving consumer goods i at time t, It means that for any fast-moving consumer product i, T is the set of time steps and L is the set of fast-moving consumer products.

[0016] Preferably, S1 also constructs a budget constraint of the fast-moving consumer goods replenishment optimization model; the budget constraint is expressed as:

[0017]

[0018] Among them, x i (t) is the replenishment quantity of fast-moving consumer product i at time t, p i(t) is the replenishment cost of FMCG product i at time t, B is the budget threshold, T is the set of time steps, and L is the set of FMCG products.

[0019] Preferably, the S1 further constructs a replenishment range constraint of the fast-moving consumer goods replenishment optimization model, and the expression of the replenishment range constraint is:

[0020]

[0021] in, p i (t) and are the minimum and maximum replenishment quantities of fast-moving consumer product i at time t, x i (t) is the replenishment quantity of fast-moving consumer product i at time t, It means that for any fast-moving consumer product i, T is the set of time steps and L is the set of fast-moving consumer products.

[0022] Preferably, when S2 models the decision-making process of fast-moving consumer goods as a Markov decision process, states, actions and rewards are defined; the states are used to represent the current inventory, user demand and purchase and storage costs of fast-moving consumer goods, the actions are used to represent the replenishment decisions of each fast-moving consumer goods, and the rewards are used to evaluate the effectiveness of each replenishment decision.

[0023] Preferably, the reward is expressed as:

[0024]

[0025] Among them, L i (x i (t)) represents the total cost of fast-moving consumer product i at time t.

[0026] Preferably, the specific steps of using the deep Q network algorithm to train the deep learning model in S2 are:

[0027] S2.1. Initialize the experience samples in the experience replay pool, each of which contains the initial state of the FMCG, the decision action, the decision reward, and the state of the FMCG after the decision action;

[0028] S2.2. Initialize Q network and target Q network;

[0029] S2.3. Use the epsilon-greedy strategy to select actions: select a random action with probability epsilon, and select the optimal action estimated by the Q network with probability 1-epsilon;

[0030] S2.4. Randomly extract a number of experience samples from the experience replay pool, calculate the target Q value, and then train the deep Q network model. The calculation formula of the target Q value is:

[0031]

[0032] Among them, y i is the target Q value, a' is the action taken by the agent, r i is the immediate reward, γ is the discount factor, Q' is the target Q network, θ is the parameter of the target Q network, The inventory status of fast-moving consumer goods;

[0033] S2.5. Update the parameters of the Q network, adjust the network weight to θ by minimizing the loss function, and complete the model training.

[0034] Preferably, the step S2 of solving the optimal replenishment strategy is to input the current inventory of fast-moving consumer goods and user demand into a trained deep learning model to generate a replenishment strategy for fast-moving consumer goods.

[0035] Preferably, after S3 executes the optimal replenishment strategy to replenish fast-moving consumer goods, the replenishment decision is evaluated, and the steps S2.1-S2.5 are used to update the optimized deep learning model, thereby updating the replenishment strategy.

[0036] Compared with the prior art, the technical solution provided by the present invention has the following beneficial effects:

[0037] 1. The inventory management optimization method for preventing shortages of fast-moving consumer goods in the retail supply chain involved in the present invention constructs a fast-moving consumer goods replenishment optimization model with minimizing the replenishment cost as the objective function, and introduces a deep learning model combining reinforcement learning and Q network. The deep learning model is trained using existing historical data, and the decision-making strategy is gradually optimized by providing feedback on the status and reward after each replenishment decision. The fast-moving consumer goods replenishment optimization model is solved. Compared with traditional methods that rely on experience and simple statistical models, the present invention can optimize replenishment decisions while ensuring that demand is met, reduce overall inventory costs, and significantly reduce the incidence of inventory shortages.

[0038] 2. The inventory management optimization method for preventing shortages of fast-moving consumer goods in the retail supply chain involved in the present invention evaluates the replenishment decision after executing the optimal replenishment strategy to replenish the fast-moving consumer goods, updates the optimized deep learning model, and then updates the replenishment strategy. Then, through online learning and real-time monitoring, the replenishment strategy is dynamically adjusted according to the actual situation, thereby ensuring the real-time and effectiveness of the replenishment decision. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A flowchart of an inventory management optimization method for preventing shortages of fast-moving consumer goods in a retail supply chain;

[0040] Figure 2 Schematic diagram of Markov decision process;

[0041] Figure 3Flowchart for obtaining and executing the optimal replenishment strategy using a deep learning model. DETAILED DESCRIPTION

[0042] In order to further understand the content of the present invention, the present invention is described in detail in conjunction with examples. The following examples are used to illustrate the present invention but are not used to limit the scope of the present invention.

[0043] Example: Refer to the attached Figure 1 The present invention relates to an inventory management optimization method for preventing shortage of fast-moving consumer goods in a retail supply chain, which comprises the following steps:

[0044] S1. Taking minimizing replenishment cost as the objective function, a fast-moving consumer goods replenishment optimization model is constructed. Specifically, considering the demand for fast-moving consumer goods, inventory capacity, replenishment cost and shortage penalty, the minimization of replenishment cost is established as the objective function and constraint, thereby forming a fast-moving consumer goods replenishment optimization model.

[0045] The expression of the objective function is:

[0046]

[0047] Among them, α i is the weight parameter of fast-moving consumer goods i, p i (t) is the replenishment cost of fast-moving consumer product i at time t, x i (t) is the replenishment quantity of fast-moving consumer product i at time t, α i p i (t)x i (t) is the purchase cost of fast-moving consumer goods, v i (t) is the inventory of fast-moving consumer goods i at time t, r i (t) is the storage cost of fast-moving consumer product i at time t, R i (v i (t),r i (t)) is the storage cost function of fast-moving consumer product i, is the shortage penalty cost function for fast-moving consumer product i.

[0048] The constraints established include inventory capacity constraints, budget constraints, and replenishment scope constraints:

[0049] The expression of the inventory capacity constraint is:

[0050]

[0051] Among them, C i (t) is the minimum inventory capacity of fast-moving consumer product i, v i (t) and v i (t-1) are the inventory of fast-moving consumer product i at time t and time t-1, respectively, x i(t) is the replenishment quantity of fast-moving consumer product i at time t, is the maximum inventory capacity of fast-moving consumer goods i, λ i (t) is the demand for fast-moving consumer product i, e i (t) is the expired quantity of fast-moving consumer goods i at time t, It means that for any FMCG product i, T is the set of time steps and L is the set of FMCG products;

[0052] The budget constraint is expressed as:

[0053]

[0054] Among them, x i (t) is the replenishment quantity of fast-moving consumer product i at time t, p i (t) is the replenishment cost of FMCG product i at time t, B is the budget threshold, T is the set of time steps, and L is the set of FMCG products.

[0055] The expression of the replenishment range constraint is:

[0056]

[0057] in, p i (t) and are the minimum and maximum replenishment quantities of fast-moving consumer product i at time t, x i (t) is the replenishment quantity of fast-moving consumer product i at time t, It means that for any fast-moving consumer product i, T is the set of time steps and L is the set of fast-moving consumer products.

[0058] S2. Introduce a deep learning model that combines reinforcement learning and Q-network, and model the decision-making process of fast-moving consumer goods as a Markov decision process. Use the deep Q-network algorithm to train the deep learning model, and use the deep learning model to solve the fast-moving consumer goods replenishment optimization model to obtain the optimal replenishment strategy.

[0059] Refer to the attached Figure 2 and attached Figure 3 As shown, modeling the decision-making process of fast-moving consumer goods as a Markov decision process means establishing experience samples of the interaction between the agent and the environment, defining states, actions, and rewards, wherein the states are used to represent the current inventory of fast-moving consumer goods, user demand, and purchase and storage costs, the actions are used to represent the replenishment decision of each fast-moving consumer goods, and the rewards are used to evaluate the effect of each replenishment decision;

[0060] The function expression of the reward is:

[0061]

[0062] Among them, L i (xi (t)) represents the total cost of fast-moving consumer product i at time t.

[0063] Refer to the attached Figure 3 As shown in the figure, the specific steps of using the deep Q network algorithm to train a deep learning model are:

[0064] S2.1. Initialize the experience samples in the experience replay pool, each of which contains the initial state of the FMCG, the decision action, the decision reward, and the state of the FMCG after the decision action;

[0065] S2.2. Initialize Q network and target Q network;

[0066] S2.3. Use epsilon-greedy strategy to select actions: select random actions with probability of epsilon, and select the optimal action estimated by Q network with probability of 1-epsilon. This step is common knowledge in this field and will not be described in detail in this invention.

[0067] S2.4. Use experience replay training model: randomly extract a number of experience samples from the experience replay pool, calculate the target Q value, and then train the deep Q network model. The calculation formula of the target Q value is:

[0068]

[0069] Among them, y i is the target Q value, which is used to evaluate the expected long-term return value of the current replenishment decision, a' is the action taken by the agent, and r i is the immediate reward, γ is the discount factor, Q' is the target Q network, θ is the parameter of the target Q network, The inventory status of fast-moving consumer goods;

[0070] S2.5. Update the parameters of the Q network, adjust the network weight to θ by minimizing the loss function, and complete the model training;

[0071] S2.6. Generate replenishment strategy based on deep learning model: Input the current inventory of fast-moving consumer goods and user demand into the trained deep learning model to generate the replenishment strategy of fast-moving consumer goods.

[0072] S3. Execute the optimal replenishment strategy to replenish fast-moving consumer goods, evaluate the replenishment decision, and use steps S2.1-S2.5 to update and optimize the deep learning model, and then update the replenishment strategy to generate the optimal replenishment decision.

[0073] Effect example: In this effect example, a simulation experiment was conducted using the fast-moving consumer goods supply chain data of an industrial company. Table 1 records the parameter settings of the simulation experiment.

[0074] Table 1 Parameter settings of simulation experiments

[0075] parameter set up Storage requirements for fast moving consumer goods <![CDATA[[0.001-0.005]ft 3 ]]> Total inventory capacity <![CDATA[40ft 3 ]]> FMCG replenishment costs [5-100] Yuan / unit Storage costs <![CDATA[[2 - 5] yuan / ft 3 > Fast Moving Consumer Goods Demand [5-20]unit

[0076] Tables 2 and 3 respectively show the comparative effects of replenishment cost and shortage rate of fast-moving consumer goods replenishment strategies generated by the deep learning model (DRL) involved in the present invention compared with other traditional replenishment strategies.

[0077] Table 2 Replenishment cost evaluation

[0078]

[0079] Table 3 Shortage assessment

[0080]

[0081] It can be seen from Tables 2 and 3 that the deep reinforcement learning model is superior to other traditional methods in the table in terms of replenishment cost and shortage rate. The DRL method maintains low replenishment cost and shortage rate throughout the entire time period, while the traditional oversupply, ski rental and maximum and minimum methods have relatively high replenishment costs and shortage rates.

[0082] The present invention is described in detail above in conjunction with the embodiments, but the contents described are only preferred embodiments of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made according to the scope of application of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. An inventory management optimization method for preventing shortages of fast-moving consumer goods in a retail supply chain, characterized by: It includes the following steps: S1. Taking minimizing the replenishment cost as the objective function, a fast-moving consumer goods replenishment optimization model is constructed; S2. Introduce a deep learning model that combines reinforcement learning and Q-network, and model the decision-making process of fast-moving consumer goods as a Markov decision process. Use the deep Q-network algorithm to train the deep learning model, and use the deep learning model to solve the fast-moving consumer goods replenishment optimization model to obtain the optimal replenishment strategy. S3. Execute the optimal replenishment strategy to replenish fast-moving consumer goods.

2. The inventory management optimization method for preventing shortage of fast-moving consumer goods in a retail supply chain according to claim 1, characterized in that: The expression of the objective function in S1 is: Among them, α i is the weight parameter of fast-moving consumer goods i, p i (t) is the replenishment cost of fast-moving consumer product i at time t, x i (t) is the replenishment quantity of fast-moving consumer product i at time t, α i p i (t)x i (t) is the purchase cost of fast-moving consumer goods, v i (t) is the inventory of fast-moving consumer goods i at time t, r i (t) is the storage cost of fast-moving consumer product i at time t, R i (v i (t),r i (t)) is the storage cost function of fast-moving consumer product i, is the shortage penalty cost function for fast-moving consumer product i.

3. The inventory management optimization method for preventing shortage of fast-moving consumer goods in a retail supply chain according to claim 1, characterized in that: S1 also constructs the inventory capacity constraint of the fast-moving consumer goods replenishment optimization model, and the expression of the inventory capacity constraint is: Among them, C i (t) is the minimum inventory capacity of fast-moving consumer product i, v i (t) and v i (t-1) are the inventory of fast-moving consumer product i at time t and time t-1, respectively, x i (t) is the replenishment quantity of fast-moving consumer product i at time t, is the maximum inventory capacity of fast-moving consumer goods i, λ i (t) is the demand for fast-moving consumer product i, e i (t) is the expired quantity of fast-moving consumer goods i at time t, It means that for any fast-moving consumer product i, T is the set of time steps and L is the set of fast-moving consumer products.

4. The inventory management optimization method for preventing shortage of fast-moving consumer goods in a retail supply chain according to claim 1, characterized in that: S1 also constructs the budget constraint of the fast-moving consumer goods replenishment optimization model; the expression of the budget constraint is: Among them, x i (t) is the replenishment quantity of fast-moving consumer product i at time t, p i (t) is the replenishment cost of FMCG product i at time t, B is the budget threshold, T is the set of time steps, and L is the set of FMCG products.

5. The inventory management optimization method for preventing shortage of fast-moving consumer goods in a retail supply chain according to claim 1, characterized in that: The S1 also constructs the replenishment range constraint of the fast-moving consumer goods replenishment optimization model, and the expression of the replenishment range constraint is: in, p i (t) and are the minimum and maximum replenishment quantities of fast-moving consumer product i at time t, x i (t) is the replenishment quantity of fast-moving consumer product i at time t, It means that for any fast-moving consumer product i, T is the set of time steps and L is the set of fast-moving consumer products.

6. The inventory management optimization method for preventing shortage of fast-moving consumer goods in a retail supply chain according to claim 2, characterized in that: When S2 models the decision-making process of fast-moving consumer goods as a Markov decision process, it defines states, actions and rewards; the states are used to represent the current inventory of fast-moving consumer goods, user demand, and purchase and storage costs; the actions are used to represent the replenishment decisions of each fast-moving consumer goods; and the rewards are used to evaluate the effectiveness of each replenishment decision.

7. The inventory management optimization method for preventing shortage of fast-moving consumer goods in a retail supply chain according to claim 6, characterized in that: The reward expression is: Among them, L i (x i (t)) represents the total cost of fast-moving consumer product i at time t.

8. The inventory management optimization method for preventing shortage of fast-moving consumer goods in a retail supply chain according to claim 6, characterized in that: The specific steps of using the deep Q network algorithm to train the deep learning model in S2 are: S2.

1. Initialize the experience samples in the experience replay pool, each of which contains the initial state of the FMCG, the decision action, the decision reward, and the state of the FMCG after the decision action; S2.

2. Initialize Q network and target Q network; S2.

3. Use the epsilon-greedy strategy to select actions: select a random action with probability epsilon, and select the optimal action estimated by the Q network with probability 1-epsilon; S2.

4. Randomly extract a number of experience samples from the experience replay pool, calculate the target Q value, and then train the deep Q network model. The calculation formula of the target Q value is: Among them, y i is the target Q value, a' is the action taken by the agent, r i is the immediate reward, γ is the discount factor, Q' is the target Q network, θ is the parameter of the target Q network, The inventory status of fast-moving consumer goods; S2.

5. Update the parameters of the Q network, adjust the network weight to θ by minimizing the loss function, and complete the model training.

9. The inventory management optimization method for preventing shortage of fast-moving consumer goods in a retail supply chain according to claim 8, characterized in that: The S2 method for solving the optimal replenishment strategy is to input the current inventory of fast-moving consumer goods and user demand into the trained deep learning model to generate a replenishment strategy for fast-moving consumer goods.

10. The inventory management optimization method for preventing shortage of fast-moving consumer goods in a retail supply chain according to claim 8, characterized in that: After S3 executes the optimal replenishment strategy to replenish fast-moving consumer goods, it evaluates the replenishment decision, uses steps S2.1-S2.5 to update the optimized deep learning model, and then updates the replenishment strategy.

Images