Inventory allocation optimization method and system based on genetic algorithm

Abstract

The invention provides an inventory allocation optimization method and system based on a genetic algorithm, and relates to the technical field of inventory allocation, and the method comprises the steps: obtaining a commodity standard data set, extracting key features in the standard data set, inputting the key features into a pre-trained inventory demand prediction model, and obtaining a demand prediction result of each inventory unit in a period of time in the future; calculating the safety inventory level of each inventory unit by adopting a safety inventory formula, and constructing a multi-target inventory allocation optimization model; solving the multi-target inventory allocation optimization model to obtain an optimal solution set, selecting a solution with an optimal comprehensive evaluation index in the optimal solution set as an optimal inventory allocation scheme, and generating an inventory allocation plan and a distribution path on the basis of the obtained optimal inventory allocation scheme in combination with the category attributes and the distribution priority of each inventory unit. And according to a comparison result of the actual inventory state and the inventory allocation plan, dynamic optimization of inventory allocation is realized in combination with a demand prediction result and a replenishment strategy.

Description

Technical Field

[0001] The present invention relates to inventory allocation technology, and in particular to an inventory allocation optimization method and system based on genetic algorithm. Background Art

[0002] With the rapid development of the modern logistics industry, enterprises are facing increasingly fierce market competition and rapid changes in customer demand. In order to improve the response speed and service level of the supply chain, enterprises need to allocate inventory efficiently and reasonably. Traditional inventory allocation methods mainly rely on manual experience and simple mathematical models, which are difficult to adapt to the complex and changing market environment, resulting in inventory backlogs or shortages, reducing the operational efficiency and profitability of enterprises. At present, some studies at home and abroad have explored the problem of inventory allocation. Some scholars use mathematical programming methods, such as linear programming and integer programming, to establish inventory allocation optimization models and solve the optimal inventory allocation strategy. However, the existing inventory allocation optimization methods still have some shortcomings. On the one hand, most methods are aimed at the inventory allocation of a single variety or a single warehouse, which is difficult to meet the needs of multi-variety and multi-warehouse collaborative allocation in the actual supply chain. On the other hand, the existing methods do not take into account the uncertainty of input parameters such as demand forecasting and inventory status, and the robustness and adaptability of the optimization results need to be improved. In addition, inventory allocation involves multiple business departments and links, and requires an integrated optimization system that supports full-process management such as data collection, solution generation, optimization solution, and allocation execution. However, existing research mainly focuses on the algorithm level and lacks a systematic solution.

[0003] Therefore, there is an urgent need for an inventory allocation method and system that can coordinate multi-variety and multi-warehouse allocation, adapt to demand uncertainty, and integrate optimization to improve the efficiency and accuracy of inventory allocation and enhance the company's market responsiveness and competitiveness. Summary of the invention

[0004] The embodiment of the present invention provides an inventory allocation optimization method and system based on a genetic algorithm, which can solve the problems in the prior art.

[0005] According to a first aspect of the embodiments of the present invention,

[0006] A method for optimizing inventory allocation based on a genetic algorithm is provided, comprising:

[0007] Obtain the sales data, inventory data, market data and customer data of the goods and perform normalization processing to obtain a standard data set, extract the key features in the standard data set, input them into the pre-trained inventory demand forecasting model, and obtain the demand forecast results of each inventory unit in the future period;

[0008] According to the demand forecast results and the preset service level, the safety stock formula is used to calculate the safety stock level of each inventory unit. At the same time, the optimization goals of minimizing total inventory cost, maximizing customer satisfaction and minimizing out-of-stock rate are set. The safety stock level and service level of each warehouse unit are used as constraints to build a multi-objective inventory allocation optimization model.

[0009] The multi-objective solution method based on genetic algorithm is used to solve the multi-objective inventory allocation optimization model to obtain the optimal solution set. The solution with the best comprehensive evaluation index is selected from the optimal solution set as the optimal inventory allocation plan. Based on the obtained optimal inventory allocation plan, combined with the category attributes and distribution priorities of each inventory unit, an inventory allocation plan and distribution path are generated. The real-time inventory status and sales data of each inventory unit are collected in real time through sensors. According to the comparison results between the actual inventory status and the inventory allocation plan, combined with the demand forecast results and the replenishment strategy, dynamic optimization of inventory allocation is achieved.

[0010] In an optional embodiment,

[0011] Extract key features from the standard data set and input them into the pre-trained inventory demand forecasting model to obtain the demand forecast results for each inventory unit in the future, including:

[0012] Based on the standard data set, a weighted attribute-based heterogeneous commodity association graph is constructed, where nodes represent commodity entities, edges represent multi-dimensional associations between commodities, and edge weights are assigned based on commodity attributes.

[0013] Based on the heterogeneous commodity association graph, a graph attention network is designed. The original features of each node are mapped to a low-dimensional feature vector through an embedding layer, and the neighbor node information is aggregated through a multi-head attention mechanism to update the node hidden representation. In the multi-head attention mechanism, each attention head calculates the attention weight between nodes through an attention vector and a weight matrix, and the node hidden representation is obtained through a feedforward neural network according to the output of each attention head;

[0014] The historical demand sequence is input into the pre-trained inventory demand forecasting model, and the long-term and short-term dependencies of the demand sequence are captured by the long-term and short-term memory network through the gating mechanism, and the hidden state of the historical demand sequence is output;

[0015] The hidden state of the historical demand sequence is concatenated with the node hidden representation, the causal attention score is calculated by a multi-layer perceptron, and the causal attention score is used to perform weighted summation on the node hidden representation to obtain an explanation vector of the demand;

[0016] Based on the hidden state of the historical demand sequence and the explanation vector of the demand, the demand forecast results of each inventory unit in the future period are obtained through feedforward neural network decoding.

[0017] In an optional embodiment,

[0018] Based on the demand forecast results and the preset service level, the safety stock formula is used to calculate the safety stock level of each inventory unit, including:

[0019]

[0020] Among them, SS i represents safety stock, i represents inventory unit, z αi (Vi, Ci) represents the service level α i The corresponding standard normal quantile, α i Indicates the service level, V i represents the coefficient of variation of demand, C i Indicates loss due to shortage of stock. Represents the feature vector of the future time period, μ i represents the mean demand vector of inventory unit i, T represents the transpose, Σ i represents the covariance matrix of inventory unit i, L i Indicates lead time.

[0021] In an optional embodiment,

[0022] The multi-objective solution method based on genetic algorithm is used to solve the multi-objective inventory allocation optimization model to obtain the optimal solution set. The solution with the best comprehensive evaluation index is selected from the optimal solution set as the optimal inventory allocation solution, including:

[0023] A certain number of initial individuals are randomly generated, each of which represents a feasible inventory allocation plan, consisting of the safety stock level and service level of each warehouse unit;

[0024] For each individual, calculate its fitness value under each optimization objective of the multi-objective inventory allocation optimization model, and use the non-dominated sorting genetic algorithm to perform non-dominated sorting on the population, combine with the evolutionary algorithm to calculate the crowding distance of the individual, and select the target individual based on the obtained non-dominated sorting results and crowding distance;

[0025] Perform crossover and mutation operations on the selected target individuals, and adaptively adjust the crossover probability and mutation probability according to the population diversity and convergence state to generate new individuals;

[0026] Divide the population into multiple sub-populations, perform inter-population migration operations regularly, and copy the target individuals of each sub-population to other sub-populations;

[0027] The preferred Bayesian optimization algorithm is introduced to construct a Gaussian process model to predict the good and bad areas of the solution space. The preference function is constructed based on the decision maker's preference information to calculate the expected improvement value of the candidate solution. The candidate solution with the highest expected improvement value is added to the next generation population. The iteration is repeated until the preset termination condition is met to obtain the target optimal solution set.

[0028] The solution with the best comprehensive evaluation index is selected from the final target optimal solution set as the optimal inventory allocation plan.

[0029] In an optional embodiment,

[0030] The preferred Bayesian optimization algorithm is introduced to construct a Gaussian process model to predict the good and bad areas of the solution space. The preference function is constructed by combining the decision maker's preference information. The formula for calculating the expected improvement value of the candidate solution is as follows:

[0031]

[0032] Where EI(a) represents the expected improvement value of candidate solution a, The prediction function representing the objective function value of the candidate solution, a * represents the current optimal solution, a represents the candidate solution, Φ(·) represents the cumulative distribution function, represents the predicted standard deviation of the objective function value of the candidate solution, φ(·) represents the kernel density function, λ represents the weight coefficient, and d(a, r) ​​represents the distance measure between the candidate solution a and the reference point r, which is used to introduce the decision maker's preference information.

[0033] In an optional embodiment,

[0034] Based on the optimal inventory allocation plan, combined with the category attributes and distribution priorities of each inventory unit, the inventory allocation plan and distribution route are generated, and the real-time inventory status and sales data of each inventory unit are collected in real time through sensors. According to the comparison results between the actual inventory status and the inventory allocation plan, combined with the demand forecast results and replenishment strategy, the dynamic optimization of inventory allocation is realized, including:

[0035] Based on the optimal inventory allocation plan, the category attribute matrix is ​​established in combination with the category attributes of each inventory unit. The category attributes of each inventory unit are quantitatively evaluated and the inventory units are divided into different category attribute combinations. The category attributes of the inventory units include product life cycle, demand volatility and product importance.

[0036] According to the combination of the classified inventory unit category attributes, the inventory tension, commodity importance and time sensitivity are comprehensively considered to establish a distribution priority evaluation model, quantitatively calculate the distribution priority score of each inventory unit, sort the inventory units according to the distribution priority score, and generate a distribution priority list;

[0037] Combined with the category attribute matrix and the delivery priority list, the inventory allocation plan is automatically generated to determine the allocation quantity and allocation order of high-priority inventory units, and output the delivery time, delivery quantity and corresponding receiving inventory unit of each inventory unit;

[0038] Input the outputted receiving inventory unit information into a pre-built vehicle routing problem model, obtain real-time road condition information, and use an optimization algorithm to solve the vehicle routing problem model to obtain an optimized delivery routing solution;

[0039] Through the intelligent sensors installed in the inventory units, the inventory status data of each inventory unit is collected in real time and compared with the inventory allocation plan to identify the deviation between the actual inventory status and the plan. Based on the identified deviation between the actual inventory status and the plan and the demand forecast results, the dynamic optimization of the inventory allocation plan is automatically triggered. According to the preset replenishment strategy, the inventory allocation plan is adjusted, the allocation quantity and priority in the inventory allocation plan are updated, and the re-optimization of the distribution route is triggered at the same time. The dynamically optimized inventory allocation plan and distribution route plan are generated to achieve dynamic inventory allocation optimization.

[0040] In an optional embodiment,

[0041] According to the output of the receiving inventory unit information, the vehicle routing problem model is input into the pre-built one, and the real-time road condition information is obtained at the same time. The vehicle routing problem model is solved by the optimization algorithm, and the optimized distribution routing solution is obtained, including:

[0042] Preprocess the acquired real-time traffic information, input the preprocessed real-time traffic information into the trained traffic flow prediction model, obtain the traffic flow prediction value of each future time step through model reasoning, and post-process the prediction value to obtain the traffic flow prediction result;

[0043] According to the traffic flow prediction results, the predicted congested sections and time periods are identified by calculating the congestion index value of each section at each future time step and comparing it with the preset threshold. The identified congested sections and time periods are clustered using the spatiotemporal clustering algorithm to obtain the predicted congested areas and time periods.

[0044] The traffic flow prediction results and the predicted congested areas and congested periods are used as the dynamic travel costs of the road section to build a vehicle path optimization model. By solving the vehicle path optimization model, the dynamically adjusted vehicle optimal path planning and departure time plan are obtained;

[0045] The delivery task is executed according to the obtained vehicle optimal route planning and departure time plan, and the latest traffic flow prediction value and congestion index prediction value are obtained in real time. When the deviation between the traffic flow prediction value and the congestion index prediction value and the corresponding actual value is higher than the preset deviation threshold, the vehicle route re-optimization is triggered, the dynamic travel cost of the road section in the optimization model is updated, and the updated optimization model is solved to obtain the optimized delivery route plan.

[0046] According to a second aspect of the embodiments of the present invention,

[0047] Provided is an inventory allocation optimization system based on genetic algorithm, comprising:

[0048] The first unit is used to obtain the sales data, inventory data, market data and customer data of the goods and perform normalization processing to obtain a standard data set, extract the key features in the standard data set, input them into the pre-trained inventory demand forecasting model, and obtain the demand forecast results of each inventory unit in the future period;

[0049] The second unit is used to calculate the safety stock level of each inventory unit using the safety stock formula according to the demand forecast results and the preset service level. At the same time, the optimization goals of minimizing the total inventory cost, maximizing customer satisfaction and minimizing the out-of-stock rate are set. The safety stock level and service level of each warehouse unit are used as constraints to build a multi-objective inventory allocation optimization model.

[0050] The third unit is used to solve the multi-objective inventory allocation optimization model through a multi-objective solution method based on a genetic algorithm to obtain an optimal solution set, select the solution with the best comprehensive evaluation index in the optimal solution set as the optimal inventory allocation plan, and generate an inventory allocation plan and distribution path based on the obtained optimal inventory allocation plan in combination with the category attributes and distribution priorities of each inventory unit, and collect the real-time inventory status and sales data of each inventory unit in real time through sensors, and realize dynamic optimization of inventory allocation based on the comparison results between the actual inventory status and the inventory allocation plan, combined with the demand forecast results and replenishment strategy.

[0051] According to a third aspect of the embodiments of the present invention,

[0052] An electronic device is provided, comprising:

[0053] processor;

[0054] a memory for storing processor-executable instructions;

[0055] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0056] A fourth aspect of the embodiments of the present invention is:

[0057] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0058] In this embodiment, by acquiring multi-source heterogeneous data and extracting key features to input into the pre-trained demand forecasting model, combined with advanced technologies such as heterogeneous graph neural networks, sequence models and attention mechanisms, the demand for each inventory unit in the future period can be more accurately predicted. According to the demand forecast results and the preset service level, the safety stock level of each unit is calculated using the safety stock formula, and a multi-objective optimization model is constructed to minimize the total inventory cost, maximize customer satisfaction and minimize the out-of-stock rate. The optimal inventory allocation plan can be obtained by solving, thereby optimizing the overall inventory management. Based on the optimal inventory allocation plan, the allocation plan and path are generated in combination with the category attributes and distribution priority. At the same time, by collecting the inventory status and sales data of each unit in real time, according to the difference between the actual situation and the plan, combined with demand forecasting and replenishment strategies, dynamic optimization of inventory allocation is achieved. Optimizing inventory management and dynamic allocation can reduce inventory backlogs and out-of-stock rates, and reduce enterprise operating costs. At the same time, improve customer satisfaction and order fulfillment capabilities, thereby improving the operational efficiency of the overall supply chain. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a flowchart of an inventory allocation optimization method based on a genetic algorithm according to an embodiment of the present invention;

[0060] Figure 2 It is a structural diagram of an inventory allocation optimization system based on a genetic algorithm according to an embodiment of the present invention. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0063] Figure 1 FIG. 1 is a flow chart of an inventory allocation optimization method based on a genetic algorithm according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0064] S101. Obtain the sales data, inventory data, market data and customer data of the goods and normalize them to obtain a standard data set, extract the key features in the standard data set, input them into the pre-trained inventory demand forecasting model, and obtain the demand forecast results of each inventory unit in the future period.

[0065] First of all, it is necessary to obtain sales data, inventory data, market data and customer data from various channels and links of the enterprise. Sales data includes information such as sales volume, sales amount, discount strength, etc. of each SKU in different sales channels; inventory data includes real-time inventory, in-transit inventory, available inventory, etc. of each warehouse and store; market data includes information such as sales of competing products, market share, promotional activities, etc.; customer data includes information such as basic attributes, purchase history, preferences, etc. of customers. These data are usually scattered in multiple business systems such as the enterprise's ERP system, WMS system, CRM system, etc., and it is necessary to integrate multi-source heterogeneous data into a unified data platform through data collection and integration technologies such as ETL tools and data integration middleware. There are inevitably quality problems such as noise, missing values, and outliers in the original data, so data cleaning is required before data analysis and mining. Common data cleaning methods include missing value processing, outlier detection, and data deduplication. After completing data cleaning, due to the large differences in the dimensions and numerical ranges of different data attributes, in order to eliminate the impact of such differences on subsequent analysis, the data needs to be normalized. Commonly used data normalization methods include: Min-Max normalization, Z-score normalization, Decimal scaling normalization, etc. After normalization, a standardized standard data set can be obtained.

[0066] Feature engineering is the process of converting the original features in the standard data set into new features that are more conducive to inventory demand forecasting and optimization. In the feature extraction stage, it is necessary to make full use of domain knowledge to construct new features that reflect inventory change trends, seasonality, and correlation from different angles such as time dimension, category dimension, and customer dimension. In the feature selection stage, it is necessary to select the feature subset that is most relevant to the target variable and has the least redundancy from many candidate features to reduce the feature dimension. Common feature selection methods include filtering methods (such as variance selection method, chi-square test method), wrapping methods (such as recursive feature elimination method), embedding methods (such as L1 regularization), etc. After completing data preparation and preprocessing, the standard data set can be used for inventory demand forecasting modeling. Inventory demand forecasting belongs to the problem of time series forecasting. Its goal is to predict the demand of a SKU in the future based on its sales sequence in the past period of time. Formally, assuming that the sales sequence of a SKU in the past T days is, the task of inventory demand forecasting is to learn a prediction function to approximate the demand sequence in the next N days. Optionally, an LSTM network can be selected as the inventory demand forecasting model. LSTM introduces three gating mechanisms: input gate, forget gate, and output gate.

[0067] In an optional embodiment,

[0068] Extract key features from the standard data set and input them into the pre-trained inventory demand forecasting model to obtain the demand forecast results for each inventory unit in the future, including:

[0069] Based on the standard data set, a property-weighted heterogeneous product association graph is constructed, where nodes represent product entities, edges represent multi-dimensional association relationships between products, and edge weights are assigned based on product attributes.

[0070] Based on the heterogeneous commodity association graph, a graph attention network is designed. The original features of each node are mapped to a low-dimensional feature vector through an embedding layer, and the neighbor node information is aggregated through a multi-head attention mechanism to update the node hidden representation. In the multi-head attention mechanism, each attention head calculates the attention weight between nodes through an attention vector and a weight matrix, and the node hidden representation is obtained through a feedforward neural network according to the output of each attention head;

[0071] The historical demand sequence is input into the pre-trained inventory demand forecasting model, and the long-term and short-term dependencies of the demand sequence are captured by the long-term and short-term memory network through the gating mechanism, and the hidden state of the historical demand sequence is output;

[0072] The hidden state of the historical demand sequence is concatenated with the node hidden representation, the causal attention score is calculated by a multi-layer perceptron, and the causal attention score is used to perform weighted summation on the node hidden representation to obtain an explanation vector of the demand;

[0073] Based on the hidden state of the historical demand sequence and the explanation vector of the demand, the demand forecast results of each inventory unit in the future period are obtained through feedforward neural network decoding.

[0074] Exemplarily, all commodities are stored as nodes in a dictionary, with the key being the node ID and the value being the commodity attribute. An adjacency matrix is ​​constructed based on the association between commodities. If two commodities are of the same type, a value of 1 is assigned to indicate an edge connection. An edge weight function is defined to calculate the weight score of the edge between nodes based on the commodity attributes.

[0075] Design a graph attention network, which includes an embedding layer for traversing the node dictionary, mapping the original features of each node to a low-dimensional vector representation through the query embedding matrix, and a multi-head attention mechanism for constructing a query / key / value matrix. The dot product score of the query matrix and the key matrix is ​​calculated for each attention head through the low-dimensional vector of the node and the low-dimensional vector of the neighboring node, and the attention score is obtained by dividing the dimension under the square root for normalization. Apply softmax to the attention score to obtain the weight distribution, multiply the weight distribution with the value matrix to obtain the weighted feature representation, and concatenate the weighted feature representations of all attention heads as the attention output of the current node. The feedforward neural network maps the multi-head attention output to the final hidden representation of the node.

[0076] Construct an LSTM network and input the historical demand sequence in time steps. LSTM captures long-term and short-term temporal dependencies through gating units and outputs the hidden state vector of each time step.

[0077] The historical demand hidden state vector is concatenated with the hidden representation of the product node, and the attention weight score is output through the feedforward network to reflect the degree of influence of the demand on each product. The attention weight score is weighted with the hidden representation of the product node to obtain the demand explanation vector. Finally, the final hidden state of the demand sequence and the demand explanation vector are concatenated, decoded through the feedforward network, and the predicted future demand sequence is output.

[0078] In this embodiment, the combination of multi-source heterogeneous data, commodity graph association information, sequence modeling and attention mechanism technologies can fully mine the patterns and features in the data, thereby improving the accuracy of demand forecasting. Through heterogeneous graph neural networks and causal attention mechanisms, the associated commodities and historical demand patterns that affect the demand for a certain commodity are discovered, revealing the driving factors of demand and providing a basis for decision-making. Using long short-term memory networks to process historical demand sequences can simultaneously capture the impact of long-term cyclical laws and short-term emergencies on demand, improve the robustness of the forecast, and thus greatly improve the accuracy and interpretability of inventory demand forecasts. Compared with traditional forecasting models, it has great advantages and potential, and is an important technical support for supply chain intelligence.

[0079] S102. Based on the demand forecast results and the preset service level, the safety stock formula is used to calculate the safety stock level of each inventory unit. At the same time, the optimization goals of minimizing total inventory cost, maximizing customer satisfaction and minimizing out-of-stock rate are set. With the safety stock level and service level of each warehouse unit as constraints, a multi-objective inventory allocation optimization model is constructed.

[0080] For example, the result of inventory demand forecast is the expected demand for inventory units in various time periods in the future, but due to the uncertainty of market changes, actual sales often deviate from the expected value. In order to cope with demand fluctuations and prevent out-of-stock and shortages, it is necessary to prepare a certain amount of safety stock on the basis of the expected inventory. The size of the safety stock depends on the uncertainty of demand and the company's target service level.

[0081] The service level refers to the probability that demand can be met within a replenishment cycle. For example, a service level of 95% means that out of every 100 replenishment cycles, demand can be fully met in 95 cycles, and there may be a shortage of stock in the other 5 cycles. The setting of the service level requires a balance between inventory cost and shortage cost: the higher the service level, the lower the risk of shortage, but the larger the safety stock will be, resulting in increased inventory cost; and vice versa. Therefore, enterprises need to determine a balance point as the target service level based on their own strategic positioning, operating costs, competitive pressure and other factors.

[0082] After determining the safety stock level of each inventory unit, the next step is to make the optimal inventory allocation decision, that is, how to reasonably allocate inventory between warehouses to minimize the global inventory cost while maximizing customer satisfaction and distribution service level. Assuming that a company has m inventory units and n warehouses, the goal is to determine the n×n warehouse transfer matrix, with the optimization goals of minimizing the total inventory management cost (including storage cost and out-of-stock cost) and maximizing customer satisfaction (reflected in minimizing the out-of-stock rate), and construct a multi-objective optimization problem.

[0083] In an optional embodiment,

[0084] Based on the demand forecast results and the preset service level, the safety stock formula is used to calculate the safety stock level of each inventory unit, including:

[0085]

[0086] Among them, SS i represents safety stock, i represents inventory unit, z αi (Vi, Ci) represents the service level α i The corresponding standard normal quantile, α i Indicates the service level, V i represents the coefficient of variation of demand, Ci Indicates loss due to shortage of stock. Represents the feature vector of the future time period, μ i represents the mean demand vector of inventory unit i, T represents the transpose, Σ i represents the covariance matrix of inventory unit i, L i Indicates lead time.

[0087] In this embodiment, by calculating a reasonable safety stock level, sufficient safety stock can be prepared for each inventory unit, so as to better cope with the uncertain fluctuations in market demand, reduce the risk of out-of-stock, and improve supply capacity. By using service level as the basis for calculating safety stock, the balance between service level and inventory cost can be weighed according to the company's own strategy, the optimal safety stock value can be determined, and the best balance between cost and service can be achieved. A multi-objective allocation optimization model is constructed with minimizing the total inventory management cost as one of the goals. By solving the model, the optimal allocation plan can be obtained, and the overall inventory cost expenditure can be reduced. The demand forecast, safety stock calculation and allocation optimization are closely combined, which can significantly improve the flexibility of the supply chain and customer service capabilities, while reducing the overall operating costs, which is of great significance to the sustainable development of the enterprise.

[0088] S103. Solve the multi-objective inventory allocation optimization model through a multi-objective solution method based on a genetic algorithm to obtain an optimal solution set, select the solution with the best comprehensive evaluation index from the optimal solution set as the optimal inventory allocation plan, and generate an inventory allocation plan and distribution path based on the obtained optimal inventory allocation plan in combination with the category attributes and distribution priorities of each inventory unit. Collect the real-time inventory status and sales data of each inventory unit through sensors in real time, and realize dynamic optimization of inventory allocation based on the comparison results between the actual inventory status and the inventory allocation plan in combination with the demand forecast results and replenishment strategy.

[0089] Based on the optimal allocation plan, combined with actual transportation capacity, time constraints and other constraints, a detailed inventory allocation plan and distribution path planning are generated. The generation of the allocation plan can be further refined into two sub-problems: inventory allocation problem and vehicle routing problem.

[0090] The inventory allocation problem is that after determining the transfer quantity between warehouses, further decisions need to be made on which downstream sales outlets to allocate the available inventory of each warehouse. The vehicle routing problem is that after the inventory allocation is completed, further decisions need to be made on the optimal driving route of the delivery vehicles of each warehouse, so that the number of vehicles is minimized and the total mileage is shortest.

[0091] The purpose of inventory allocation optimization is to make the best inventory allocation decision in the future (such as 1 week, 2 weeks). However, in the actual implementation process, due to sales forecast errors, suppliers' untimely delivery, etc., the inventory level and demand forecast may deviate from the original plan, resulting in the original allocation plan no longer being optimal. Therefore, it is necessary to establish a dynamic optimization mechanism for the allocation plan to obtain the real-time inventory status and sales data of each outlet in a timely manner, detect forecast deviations, and adjust and correct the original allocation plan accordingly.

[0092] In an optional embodiment,

[0093] The multi-objective solution method based on genetic algorithm is used to solve the multi-objective inventory allocation optimization model to obtain the optimal solution set. The solution with the best comprehensive evaluation index is selected from the optimal solution set as the optimal inventory allocation solution, including:

[0094] A certain number of initial individuals are randomly generated, each of which represents a feasible inventory allocation plan, consisting of the safety stock level and service level of each warehouse unit;

[0095] For each individual, calculate its fitness value under each optimization objective of the multi-objective inventory allocation optimization model, and use the non-dominated sorting genetic algorithm to perform non-dominated sorting on the population, combine with the evolutionary algorithm to calculate the crowding distance of the individual, and select the target individual based on the obtained non-dominated sorting results and crowding distance;

[0096] Perform crossover and mutation operations on the selected target individuals, and adaptively adjust the crossover probability and mutation probability according to the population diversity and convergence state to generate new individuals;

[0097] Divide the population into multiple sub-populations, perform inter-population migration operations regularly, and copy the target individuals of each sub-population to other sub-populations;

[0098] The preferred Bayesian optimization algorithm is introduced to construct a Gaussian process model to predict the good and bad areas of the solution space. The preference function is constructed based on the decision maker's preference information to calculate the expected improvement value of the candidate solution. The candidate solution with the highest expected improvement value is added to the next generation population. The iteration is repeated until the preset termination condition is met to obtain the target optimal solution set.

[0099] The solution with the best comprehensive evaluation index is selected from the final target optimal solution set as the optimal inventory allocation plan.

[0100] For example, first, a certain number of initial individuals are randomly generated, each of which represents a feasible inventory allocation plan, consisting of the safety inventory level and service level of each warehouse unit. These initial individuals constitute the starting point of the solution process and provide a basis for subsequent optimization.

[0101] Then, for each individual, calculate its fitness value under each optimization target of the multi-objective inventory allocation optimization model. The fitness value reflects the performance of the individual under different optimization targets and is an important indicator for evaluating the quality of the individual. The non-dominated sorting genetic algorithm is used to perform non-dominated sorting on the population. According to the fitness value of the individual under each optimization target, the population is divided into multiple non-dominated frontiers, which reflects the quality relationship between individuals. At the same time, the crowding distance of the individual is calculated in combination with the evolutionary algorithm to measure the distribution density of the individual in the same non-dominated frontier in order to maintain the diversity of the population. According to the obtained non-dominated sorting results and crowding distance, the target individual with excellent performance is selected as the parent individual to provide high-quality genetic material for subsequent genetic operations.

[0102] Perform crossover and mutation operations on the selected target individuals to generate new offspring individuals by exchanging and changing some of the individual's genes. The crossover operation promotes the transmission and combination of excellent genes, while the mutation operation introduces new genes and increases the diversity of the population. In order to balance the local search and global exploration capabilities, the crossover probability and mutation probability are adaptively adjusted. When the population diversity is low or the convergence speed is slow, the crossover probability and mutation probability are appropriately increased to speed up the search process; when the population diversity is high or the convergence speed is fast, the crossover probability and mutation probability are appropriately reduced to prevent premature convergence to the local optimal solution.

[0103] In order to further improve the search efficiency and solution quality, the population is divided into multiple sub-populations, and each sub-population is evolved and optimized independently. Inter-population migration operations are performed regularly to copy the target individuals with excellent performance in each sub-population to other sub-populations, thus realizing information exchange and gene exchange between sub-populations. This strategy promotes global optimization while maintaining the independence of sub-populations, improving the convergence speed of the algorithm and the quality of the solution.

[0104] In the iterative process of the genetic algorithm, the preferred Bayesian optimization algorithm is introduced to intelligently search and optimize the solution space. By constructing a Gaussian process model, the good and bad areas in the solution space are predicted, and the areas that are expected to contain high-quality solutions are identified. At the same time, the preference function is constructed in combination with the decision maker's preference information to quantitatively evaluate the expected improvement value of the candidate solution. The candidate solution with the highest expected improvement value is added to the next generation population, guiding the search process to optimize in the direction preferred by the decision maker, thereby improving the practicality and acceptability of the solution.

[0105] The above iterative process is repeated until the preset termination conditions are met, such as reaching the maximum number of iterations, the quality of the solution reaching the threshold, etc. During the iterative process, the population continues to evolve and optimize, the quality of the solution gradually improves, and finally a set of target optimal solutions is obtained.

[0106] From the obtained optimal solution set, the weights of each optimization objective and the decision maker's preferences are comprehensively considered, and the solution with the best comprehensive evaluation index is selected as the optimal inventory allocation plan.

[0107] In an optional embodiment,

[0108] The preferred Bayesian optimization algorithm is introduced to construct a Gaussian process model to predict the good and bad areas of the solution space. The preference function is constructed by combining the decision maker's preference information. The formula for calculating the expected improvement value of the candidate solution is as follows:

[0109]

[0110] Where EI(a) represents the expected improvement value of candidate solution a, The prediction function representing the objective function value of the candidate solution, a * represents the current optimal solution, a represents the candidate solution, Φ(·) represents the cumulative distribution function, represents the predicted standard deviation of the objective function value of the candidate solution, φ(·) represents the kernel density function, λ represents the weight coefficient, and d(a,r) represents the distance measure between the candidate solution a and the reference point r, which is used to introduce the decision maker's preference information.

[0111] In this embodiment, while balancing various optimization objectives, the needs and preferences of decision makers are met to the greatest extent, which has high practical value. By applying the multi-objective solution method based on genetic algorithm, the multi-objective inventory allocation optimization model is effectively solved and the optimal inventory allocation plan is obtained. This method integrates multiple optimization strategies and technologies, has strong global search capabilities and convergence performance, can handle complex multi-objective optimization problems, and provides a scientific basis and support for inventory allocation decisions.

[0112] In an optional embodiment,

[0113] Based on the optimal inventory allocation plan, combined with the category attributes and distribution priorities of each inventory unit, the inventory allocation plan and distribution route are generated, and the real-time inventory status and sales data of each inventory unit are collected in real time through sensors. According to the comparison results between the actual inventory status and the inventory allocation plan, combined with the demand forecast results and replenishment strategy, the dynamic optimization of inventory allocation is realized, including:

[0114] Based on the optimal inventory allocation plan, the category attribute matrix is ​​established in combination with the category attributes of each inventory unit. The category attributes of each inventory unit are quantitatively evaluated and the inventory units are divided into different category attribute combinations. The category attributes of the inventory units include product life cycle, demand volatility and product importance.

[0115] According to the combination of the classified inventory unit category attributes, the inventory tension, commodity importance and time sensitivity are comprehensively considered to establish a distribution priority evaluation model, quantitatively calculate the distribution priority score of each inventory unit, sort the inventory units according to the distribution priority score, and generate a distribution priority list;

[0116] Combined with the category attribute matrix and the delivery priority list, the inventory allocation plan is automatically generated to determine the allocation quantity and allocation order of high-priority inventory units, and output the delivery time, delivery quantity and corresponding receiving inventory unit of each inventory unit;

[0117] Input the outputted receiving inventory unit information into a pre-built vehicle routing problem model, obtain real-time road condition information, and use an optimization algorithm to solve the vehicle routing problem model to obtain an optimized delivery routing solution;

[0118] Through the intelligent sensors installed in the inventory units, the inventory status data of each inventory unit is collected in real time and compared with the inventory allocation plan to identify the deviation between the actual inventory status and the plan. Based on the identified deviation between the actual inventory status and the plan and the demand forecast results, the dynamic optimization of the inventory allocation plan is automatically triggered. According to the preset replenishment strategy, the inventory allocation plan is adjusted, the allocation quantity and priority in the inventory allocation plan are updated, and the re-optimization of the distribution route is triggered at the same time. The dynamically optimized inventory allocation plan and distribution route plan are generated to achieve dynamic inventory allocation optimization.

[0119] For example, first, a category attribute matrix is ​​established by combining the category attributes of each inventory unit. By quantitatively evaluating the category attributes of each inventory unit, the inventory units are divided into different category attribute combinations. Among them, the category attributes of the inventory unit include product life cycle, demand volatility and product importance. The product life cycle reflects the timeliness and obsolescence risk of the product, demand volatility characterizes the stability and predictability of product demand, and product importance measures the impact of the product on corporate profits and customer satisfaction. By constructing a category attribute matrix, the characteristics and risks of the inventory unit can be comprehensively evaluated, providing a basis for subsequent distribution priority evaluation and inventory allocation decisions.

[0120] According to the combination of the classified inventory unit category attributes, the distribution priority evaluation model is established by comprehensively considering the inventory tension, commodity importance and time sensitivity. By quantitatively calculating the distribution priority score of each inventory unit, the urgency and importance of the inventory unit can be objectively evaluated. The inventory tension reflects the gap between the inventory level of the inventory unit and the safety inventory level. The higher the inventory tension, the greater the inventory risk and the higher the distribution priority should be. The commodity importance takes into account the contribution of the commodity to the company's profits and customer satisfaction. The more important the commodity is, the higher the distribution priority should be. Time sensitivity measures the life cycle and demand timeliness of the commodity. For commodities with short life cycles and strong demand timeliness, higher distribution priority should be given. By comprehensively evaluating these factors, the distribution priority score of each inventory unit is obtained, and the inventory units are sorted according to the distribution priority score to generate a distribution priority list.

[0121] Combine the category attribute matrix and the distribution priority list to automatically generate an inventory allocation plan. According to the distribution priority list, determine the allocation quantity and allocation order of high-priority inventory units, and give priority to meeting urgent and important inventory needs. At the same time, output the delivery time, delivery quantity and corresponding receiving inventory unit of each inventory unit to clarify the specific arrangements for inventory allocation. The automatically generated inventory allocation plan can reduce manual intervention, improve allocation efficiency, and ensure the rationality and feasibility of allocation decisions.

[0122] In order to optimize the delivery route, the output of the received inventory unit information is input into the pre-built vehicle routing problem model, and the real-time traffic information is obtained at the same time. The vehicle routing problem model is solved by the optimization algorithm to obtain the optimized delivery route plan. The vehicle routing problem model takes into account the constraints such as the number of vehicles, capacity restrictions, and time windows. With the goal of minimizing the delivery cost or maximizing customer satisfaction, the optimal delivery route and scheduling plan are solved. By introducing real-time traffic information, the delivery route can be dynamically adjusted to avoid congested sections and improve delivery efficiency and punctuality.

[0123] In order to achieve dynamic inventory allocation optimization, intelligent sensors are set in the inventory units to collect the inventory status data of each inventory unit in real time, compare it with the inventory allocation plan, and identify the deviation between the actual inventory status and the plan. When there is a large deviation between the actual inventory status and the plan, it indicates that the original inventory allocation plan may no longer be applicable and needs to be dynamically optimized. Combined with the identified deviation between the actual inventory status and the plan and the demand forecast results, the dynamic optimization of the inventory allocation plan is automatically triggered. According to the preset replenishment strategies, such as replenishment point strategy, replenishment quantity strategy, etc., the inventory allocation plan is adjusted, and the allocation quantity and priority in the inventory allocation plan are updated to adapt to the changes in the actual inventory status and the forecast of future demand. At the same time, the re-optimization of the distribution path is triggered to generate the dynamically optimized inventory allocation plan and distribution path plan. By monitoring the inventory status in real time and dynamically adjusting the inventory allocation plan and distribution path, it is possible to quickly respond to inventory changes, improve the flexibility and adaptability of inventory management, reduce inventory backlogs and out-of-stock risks, and achieve dynamic inventory allocation optimization.

[0124] In this embodiment, the characteristics of inventory units, delivery priorities and real-time inventory status are fully considered, and efficient and accurate inventory allocation and delivery management are achieved by automatically generating inventory allocation plans and delivery path plans, and dynamically optimizing according to actual conditions. This method comprehensively uses technologies such as category management, priority evaluation, vehicle path optimization and dynamic adjustment, improves the intelligence level and adaptability of inventory allocation, and is of great significance to improving supply chain efficiency and customer satisfaction.

[0125] In an optional embodiment,

[0126] According to the output of the receiving inventory unit information, the vehicle routing problem model is input into the pre-built one, and the real-time road condition information is obtained at the same time. The vehicle routing problem model is solved by the optimization algorithm, and the optimized distribution routing solution is obtained, including:

[0127] Preprocess the acquired real-time traffic information, input the preprocessed real-time traffic information into the trained traffic flow prediction model, obtain the traffic flow prediction value of each future time step through model reasoning, and post-process the prediction value to obtain the traffic flow prediction result;

[0128] According to the traffic flow prediction results, the predicted congested sections and time periods are identified by calculating the congestion index value of each section at each future time step and comparing it with the preset threshold. The identified congested sections and time periods are clustered using the spatiotemporal clustering algorithm to obtain the predicted congested areas and time periods.

[0129] The traffic flow prediction results and the predicted congested areas and congested periods are used as the dynamic travel costs of the road section to build a vehicle path optimization model. By solving the vehicle path optimization model, the dynamically adjusted vehicle optimal path planning and departure time plan are obtained;

[0130] The delivery task is executed according to the obtained vehicle optimal route planning and departure time plan, and the latest traffic flow prediction value and congestion index prediction value are obtained in real time. When the deviation between the traffic flow prediction value and the congestion index prediction value and the corresponding actual value is higher than the preset deviation threshold, the vehicle route re-optimization is triggered, the dynamic travel cost of the road section in the optimization model is updated, and the updated optimization model is solved to obtain the optimized delivery route plan.

[0131] Exemplarily, first, the acquired real-time traffic information is preprocessed. By cleaning, filtering and converting the real-time traffic information, noise and redundant data are removed, key features are extracted, and the data is converted into a format suitable for input into the traffic flow prediction model. The preprocessed real-time traffic information includes road network structure, road section attributes, traffic flow historical data, etc., providing accurate and complete input data for subsequent traffic flow prediction.

[0132] The pre-processed real-time traffic information is input into the pre-trained traffic flow prediction model. The traffic flow prediction model uses deep learning algorithms, such as long short-term memory network (LSTM), convolutional neural network (CNN), etc., to establish an accurate traffic flow prediction model by learning the spatiotemporal correlation of historical traffic flow data and road condition information. Through model reasoning, the traffic flow prediction value of each future time step is obtained, that is, the traffic flow prediction result of each road section in different time periods in the future. The predicted value is post-processed, such as smoothing and outlier processing, to improve the stability and reliability of the prediction result.

[0133] According to the traffic flow prediction results, the congestion index value of each road section at each time step in the future is calculated. The congestion index can be used to quantitatively evaluate the congestion degree of the road section by using indicators such as the ratio of traffic flow to road capacity and the ratio of average speed to free flow speed. The calculated congestion index value is compared with the preset threshold to identify the predicted congested sections and time periods. For sections and time periods whose predicted values ​​exceed the congestion threshold, it indicates that these sections may experience traffic congestion at a specific time in the future. Spatiotemporal clustering algorithms such as DBSCAN and ST-DBSCAN are used to cluster the identified congested sections and time periods to obtain predicted congested areas and time periods. The spatiotemporal clustering algorithm takes into account the spatial adjacency and temporal continuity of congested sections, and aggregates adjacent congested sections and continuous congested time periods into a congested area and time period, which is convenient for subsequent path planning and adjustment.

[0134] The traffic flow prediction results and the predicted congestion areas and congestion periods are used as the dynamic travel costs of the road section to build a vehicle path optimization model. The vehicle path optimization model aims to minimize the distribution cost or maximize the distribution efficiency, taking into account constraints such as vehicle capacity, time window, and road section travel time to solve the optimal distribution path and departure time plan. The dynamic travel cost reflects the difficulty of the road section in different time periods. By mapping the traffic flow prediction results and congestion conditions to the road section travel cost, the path optimization model can dynamically adjust the distribution path according to the real-time road conditions and future predictions, avoid the predicted congestion areas and time periods, select unobstructed paths, and improve distribution efficiency and punctuality.

[0135] The delivery task is performed according to the optimal vehicle route planning and departure time plan obtained by solving the vehicle route optimization model. During the delivery process, the latest traffic flow forecast value and congestion index forecast value are obtained in real time and compared with the actual traffic flow and congestion conditions. When there is a large deviation between the predicted value and the actual value, that is, the deviation exceeds the preset deviation threshold, it indicates that the original delivery route planning may no longer be applicable and the route needs to be re-optimized. Trigger vehicle route re-optimization, update the dynamic passage cost of the road section in the optimization model, feed back the real-time traffic flow and congestion conditions into the road section passage cost, and re-solve the updated optimization model to obtain the optimized delivery route plan.

[0136] In this embodiment, real-time traffic data and prediction technology are fully utilized to dynamically adjust the delivery path and departure time to avoid congested areas and time periods, thereby improving delivery efficiency and punctuality. This method comprehensively uses technologies such as traffic flow prediction, spatiotemporal clustering, and vehicle path optimization to achieve dynamic optimization of the delivery path and enhance the adaptability and robustness of the delivery plan. At the same time, by real-time monitoring of traffic conditions and triggering path re-optimization, the real-time and reliability of the delivery plan are further improved, providing strong support for the efficient execution of delivery tasks.

[0137] Figure 2 FIG. 1 is a schematic diagram of the structure of an inventory allocation optimization system based on a genetic algorithm according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0138] The first unit is used to obtain the sales data, inventory data, market data and customer data of the goods and perform normalization processing to obtain a standard data set, extract the key features in the standard data set, input them into the pre-trained inventory demand forecasting model, and obtain the demand forecast results of each inventory unit in the future period;

[0139] The second unit is used to calculate the safety stock level of each inventory unit using the safety stock formula according to the demand forecast results and the preset service level. At the same time, the optimization goals of minimizing the total inventory cost, maximizing customer satisfaction and minimizing the out-of-stock rate are set. The safety stock level and service level of each warehouse unit are used as constraints to build a multi-objective inventory allocation optimization model.

[0140] The third unit is used to solve the multi-objective inventory allocation optimization model through a multi-objective solution method based on a genetic algorithm to obtain an optimal solution set, select the solution with the best comprehensive evaluation index in the optimal solution set as the optimal inventory allocation plan, and generate an inventory allocation plan and distribution path based on the obtained optimal inventory allocation plan in combination with the category attributes and distribution priorities of each inventory unit, and collect the real-time inventory status and sales data of each inventory unit in real time through sensors, and realize dynamic optimization of inventory allocation based on the comparison results between the actual inventory status and the inventory allocation plan, combined with the demand forecast results and replenishment strategy.

[0141] According to a third aspect of the embodiments of the present invention,

[0142] An electronic device is provided, comprising:

[0143] processor;

[0144] a memory for storing processor-executable instructions;

[0145] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0146] A fourth aspect of the embodiments of the present invention is:

[0147] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0148] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0149] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optimizing inventory allocation based on genetic algorithm, characterized in that: include: Obtain the sales data, inventory data, market data and customer data of the goods and perform normalization processing to obtain a standard data set. Extract the key features in the standard data set and input them into the pre-trained inventory demand forecasting model to obtain the demand forecast results of each inventory unit in the future period. According to the demand forecast results and the preset service level, the safety stock formula is used to calculate the safety stock level of each inventory unit. At the same time, the optimization goals of minimizing total inventory cost, maximizing customer satisfaction and minimizing out-of-stock rate are set. The safety stock level and service level of each warehouse unit are used as constraints to build a multi-objective inventory allocation optimization model. The multi-objective solution method based on genetic algorithm is used to solve the multi-objective inventory allocation optimization model to obtain the optimal solution set. The solution with the best comprehensive evaluation index is selected from the optimal solution set as the optimal inventory allocation plan. Based on the obtained optimal inventory allocation plan, combined with the category attributes and distribution priorities of each inventory unit, an inventory allocation plan and distribution path are generated. The real-time inventory status and sales data of each inventory unit are collected in real time through sensors. According to the comparison results between the actual inventory status and the inventory allocation plan, combined with the demand forecast results and the replenishment strategy, dynamic optimization of inventory allocation is achieved.

2. The method according to claim 1, characterized in that Extract key features from the standard data set and input them into the pre-trained inventory demand forecasting model to obtain the demand forecast results for each inventory unit in the future, including: Based on the standard data set, a property-weighted heterogeneous product association graph is constructed, where nodes represent product entities, edges represent multi-dimensional association relationships between products, and edge weights are assigned based on product attributes. Based on the heterogeneous commodity association graph, a graph attention network is designed. The original features of each node are mapped to a low-dimensional feature vector through an embedding layer, and the neighbor node information is aggregated through a multi-head attention mechanism to update the node hidden representation. In the multi-head attention mechanism, each attention head calculates the attention weight between nodes through an attention vector and a weight matrix, and the node hidden representation is obtained through a feedforward neural network according to the output of each attention head; The historical demand sequence is input into the pre-trained inventory demand forecasting model, and the long-term and short-term dependencies of the demand sequence are captured by the long-term and short-term memory network through the gating mechanism, and the hidden state of the historical demand sequence is output; The hidden state of the historical demand sequence is concatenated with the node hidden representation, the causal attention score is calculated by a multi-layer perceptron, and the causal attention score is used to perform weighted summation on the node hidden representation to obtain an explanation vector of the demand; Based on the hidden state of the historical demand sequence and the explanation vector of the demand, the demand forecast results of each inventory unit in the future period are obtained through feedforward neural network decoding.

3. The method according to claim 1, characterized in that Based on the demand forecast results and the preset service level, the safety stock formula is used to calculate the safety stock level of each inventory unit, including: Among them, SS i represents safety stock, i represents inventory unit, z αi (Vi, Ci) represents the service level α i The corresponding standard normal quantile, α i Indicates the service level, V i represents the coefficient of variation of demand, C i Indicates loss due to shortage of stock. Represents the feature vector of the future time period, μ i represents the mean demand vector of inventory unit i, T represents the transpose, Σ i represents the covariance matrix of inventory unit i, L i Indicates lead time.

4. The method according to claim 1, characterized in that: The multi-objective solution method based on genetic algorithm is used to solve the multi-objective inventory allocation optimization model to obtain the optimal solution set. The solution with the best comprehensive evaluation index is selected from the optimal solution set as the optimal inventory allocation solution, including: A certain number of initial individuals are randomly generated, each of which represents a feasible inventory allocation plan, consisting of the safety stock level and service level of each warehouse unit; For each individual, calculate its fitness value under each optimization objective of the multi-objective inventory allocation optimization model, and use the non-dominated sorting genetic algorithm to perform non-dominated sorting on the population, combine with the evolutionary algorithm to calculate the crowding distance of the individual, and select the target individual based on the obtained non-dominated sorting results and crowding distance; Perform crossover and mutation operations on the selected target individuals, and adaptively adjust the crossover probability and mutation probability according to the population diversity and convergence state to generate new individuals; Divide the population into multiple sub-populations, perform inter-population migration operations regularly, and copy the target individuals of each sub-population to other sub-populations; The preferred Bayesian optimization algorithm is introduced to construct a Gaussian process model to predict the good and bad areas of the solution space. The preference function is constructed based on the decision maker's preference information to calculate the expected improvement value of the candidate solution. The candidate solution with the highest expected improvement value is added to the next generation population. The iteration is repeated until the preset termination condition is met to obtain the target optimal solution set. The solution with the best comprehensive evaluation index is selected from the final target optimal solution set as the optimal inventory allocation plan.

5. The method according to claim 4, characterized in that The preferred Bayesian optimization algorithm is introduced to construct a Gaussian process model to predict the good and bad areas of the solution space. The preference function is constructed by combining the decision maker's preference information. The formula for calculating the expected improvement value of the candidate solution is as follows: Where EI(a) represents the expected improvement value of candidate solution a, The prediction function representing the objective function value of the candidate solution, a * represents the current optimal solution, a represents the candidate solution, Φ(·) represents the cumulative distribution function, represents the predicted standard deviation of the objective function value of the candidate solution, φ(·) represents the kernel density function, λ represents the weight coefficient, and d(a,r) represents the distance measure between the candidate solution a and the reference point r, which is used to introduce the decision maker's preference information.

6. The method according to claim 1, characterized in that Based on the optimal inventory allocation plan, combined with the category attributes and distribution priorities of each inventory unit, the inventory allocation plan and distribution route are generated, and the real-time inventory status and sales data of each inventory unit are collected in real time through sensors. According to the comparison results between the actual inventory status and the inventory allocation plan, combined with the demand forecast results and replenishment strategy, the dynamic optimization of inventory allocation is realized, including: Based on the optimal inventory allocation plan, the category attribute matrix is ​​established in combination with the category attributes of each inventory unit. The category attributes of each inventory unit are quantitatively evaluated and the inventory units are divided into different category attribute combinations. The category attributes of the inventory units include product life cycle, demand volatility and product importance. According to the combination of the classified inventory unit category attributes, the inventory tension, commodity importance and time sensitivity are comprehensively considered to establish a distribution priority evaluation model, quantitatively calculate the distribution priority score of each inventory unit, sort the inventory units according to the distribution priority score, and generate a distribution priority list; Combined with the category attribute matrix and the delivery priority list, the inventory allocation plan is automatically generated to determine the allocation quantity and allocation order of high-priority inventory units, and output the delivery time, delivery quantity and corresponding receiving inventory unit of each inventory unit; Input the outputted receiving inventory unit information into a pre-built vehicle routing problem model, obtain real-time road condition information, and use an optimization algorithm to solve the vehicle routing problem model to obtain an optimized delivery routing solution; Through the intelligent sensors installed in the inventory units, the inventory status data of each inventory unit is collected in real time and compared with the inventory allocation plan to identify the deviation between the actual inventory status and the plan. Based on the identified deviation between the actual inventory status and the plan and the demand forecast results, the dynamic optimization of the inventory allocation plan is automatically triggered. According to the preset replenishment strategy, the inventory allocation plan is adjusted, the allocation quantity and priority in the inventory allocation plan are updated, and the re-optimization of the distribution route is triggered at the same time. The dynamically optimized inventory allocation plan and distribution route plan are generated to achieve dynamic inventory allocation optimization.

7. The method according to claim 6, characterized in that According to the output of the receiving inventory unit information, the vehicle routing problem model is input into the pre-built one, and the real-time road condition information is obtained at the same time. The vehicle routing problem model is solved by the optimization algorithm, and the optimized distribution routing solution is obtained, including: Preprocess the acquired real-time traffic information, input the preprocessed real-time traffic information into the trained traffic flow prediction model, obtain the traffic flow prediction value of each future time step through model reasoning, and post-process the prediction value to obtain the traffic flow prediction result; According to the traffic flow prediction results, the predicted congested sections and time periods are identified by calculating the congestion index value of each section at each future time step and comparing it with the preset threshold. The identified congested sections and time periods are clustered using the spatiotemporal clustering algorithm to obtain the predicted congested areas and time periods. The traffic flow prediction results and the predicted congested areas and congested periods are used as the dynamic travel costs of the road section to build a vehicle path optimization model. By solving the vehicle path optimization model, the dynamically adjusted vehicle optimal path planning and departure time plan are obtained; The delivery task is executed according to the obtained vehicle optimal route planning and departure time plan, and the latest traffic flow prediction value and congestion index prediction value are obtained in real time. When the deviation between the traffic flow prediction value and the congestion index prediction value and the corresponding actual value is higher than the preset deviation threshold, the vehicle route re-optimization is triggered, the dynamic travel cost of the road section in the optimization model is updated, and the updated optimization model is solved to obtain the optimized delivery route plan.

8. An inventory allocation optimization system based on genetic algorithm, used to implement the method described in any one of claims 1 to 7, characterized in that: include: The first unit is used to obtain the sales data, inventory data, market data and customer data of the goods and perform normalization processing to obtain a standard data set, extract the key features in the standard data set, input them into the pre-trained inventory demand forecasting model, and obtain the demand forecast results of each inventory unit in the future period; The second unit is used to calculate the safety stock level of each inventory unit using the safety stock formula according to the demand forecast results and the preset service level. At the same time, the optimization goals of minimizing the total inventory cost, maximizing customer satisfaction and minimizing the out-of-stock rate are set. The safety stock level and service level of each warehouse unit are used as constraints to build a multi-objective inventory allocation optimization model. The third unit is used to solve the multi-objective inventory allocation optimization model through a multi-objective solution method based on a genetic algorithm to obtain an optimal solution set, select the solution with the best comprehensive evaluation index in the optimal solution set as the optimal inventory allocation plan, and generate an inventory allocation plan and distribution path based on the obtained optimal inventory allocation plan in combination with the category attributes and distribution priorities of each inventory unit, and collect the real-time inventory status and sales data of each inventory unit in real time through sensors, and realize dynamic optimization of inventory allocation based on the comparison results between the actual inventory status and the inventory allocation plan, combined with the demand forecast results and replenishment strategy.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images