Hospital medicine inventory management method and system based on multi-objective collaborative optimization

Abstract

The invention relates to a hospital medicine inventory management method and system based on multi-objective collaborative optimization, and belongs to the technical field of artificial intelligence and data processing. The method comprises the following steps: defining optimization parameters, an optimization objective function and inventory constraint conditions of hospital medicine inventory management; constructing a space-time coding vector based on the optimization parameters; the space-time coding vector is input into an improved NSGA-II algorithm, and multi-objective optimization is carried out; monitoring an environmental change comprehensive index of the standard deviation of the three-day consumption change rate of all drugs and the storage unit volume change rate, and when the index exceeds a threshold value, executing local optimization iteration by taking the current optimal solution as a starting point; and generating a Pareto frontier solution set through an improved NSGA-II algorithm, calculating a comprehensive utility value of each cluster, selecting a cluster center solution with the highest utility value as a final output result, and realizing four-dimensional target equilibrium optimization. The adaptive capacity of the model to the dynamic change of the actual operation environment can be improved.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence and data processing technology, specifically relating to a hospital drug inventory management method and system based on multi-objective collaborative optimization. Background Technology

[0002] With the continuous expansion of hospital scale and the increasing complexity of drug types, drug inventory management faces practical challenges such as multi-factor coupling, strong dynamic fluctuations, and diverse management constraints. Hospitals have a wide variety of drugs, and inventory levels must meet the continuity and timeliness of clinical medication while also considering cost control, shelf-life management, and limited storage space. Chinese invention patent CN119849715A proposes an artificial intelligence-based method for multi-dimensional supply chain fluctuation prediction and inventory optimization, relating to the field of supply chain management technology. This method involves acquiring and preprocessing multi-dimensional data to generate standardized time-series datasets and knowledge graphs; using wavelet transform for multi-scale decomposition to extract fluctuation features; constructing a state matrix by combining graph attention networks and multi-head self-attention mechanisms; predicting risk transmission paths based on historical fluctuation pattern matching and graph convolutional networks; and generating risk prevention and control schemes. This invention can accurately identify fluctuation characteristics and risk transmission paths in the supply chain system, improving the accuracy of early warning and the timeliness of response. The aforementioned technical solutions suffer from the following problems that require further resolution: Most existing hospital drug inventory management models focus solely on cost minimization or employ static weighting to linearly combine multiple objectives. This fails to reflect the changing priorities of optimization needs at different stages and ignores the dynamic nature of actual operational goals in healthcare scenarios. Consequently, the models struggle to flexibly adjust their optimization direction under varying conditions, leading to issues such as unreasonable resource allocation or conflicting objectives. Existing methods typically set a fixed upper limit for drug storage capacity, neglecting capacity adaptability issues arising from fluctuations in actual demand. Rigid capacity constraints can easily result in limited procurement or wasted inventory space, undermining inventory security and hindering efficient utilization of inventory resources, lacking flexible scheduling capabilities. Traditional inventory optimization techniques largely rely on static features such as inventory levels and consumption rates for modeling, neglecting the evolutionary trends of drug inventory over time and the collaborative relationships in spatial layout. This makes it difficult to characterize the structural information within the inventory system, limiting the optimization algorithm's responsiveness to complex states and making the models ill-suited for handling multi-source heterogeneous data and dynamic evolutionary processes in reality. Existing optimization methods employ general-purpose genetic algorithms, such as the standard NSGA-II, but they lack policy support that is coupled with the business structure in terms of population initialization, crossover strategy, and mutation method, resulting in low efficiency in solution space search and slow convergence speed. Summary of the Invention

[0003] To address the aforementioned problems, this invention provides a hospital drug inventory management method and system based on multi-objective collaborative optimization.

[0004] To achieve the above objectives, the present invention employs the following technical solution: This invention provides a hospital drug inventory management method based on multi-objective collaborative optimization, comprising the following steps: S1. Define the optimization parameters, objective function, and inventory constraints for hospital drug inventory management; the optimization parameters include inventory level, average daily consumption, remaining shelf life in days, and supplier response time; the objective function includes inventory cost, stockout risk, shelf-life loss, and storage space utilization rate. S2. Construct a spatiotemporal encoding vector based on optimized parameters, including inventory quantity, average change rate of inventory quantity in the past 3 days, correlation coefficient of drug inventory quantity in the same storage area, coefficient of variation of consumption in the past 7 days, and cumulative shortage probability. S3. Input the spatiotemporal encoding vector into the improved NSGA-II non-dominated sorting genetic algorithm for multi-objective optimization; S4. Trigger dynamic environment re-optimization operation: Monitor the comprehensive environmental change index of the standard deviation of the 3-day consumption change rate of all drugs and the change rate of storage unit volume. When the comprehensive environmental change index exceeds the threshold, execute local optimization iteration starting from the current optimal solution. S5. A Pareto front solution set is generated by the improved non-dominated sorting genetic algorithm NSGA-II. Fuzzy C-means clustering is used to group the solution set and calculate the comprehensive utility value of each cluster. The cluster center solution with the highest utility value is selected as the final output result to achieve four-dimensional objective equilibrium optimization.

[0005] Furthermore, the optimization parameters for hospital drug inventory management defined in step S1 specifically include: Definition of the first The inventory of this type of drug is , No. The average daily consumption of this type of drug is , No. The remaining shelf life of this type of drug is [number] days. , No. The supplier response time for this type of drug is ; , This indicates the total number of drug categories managed by the hospital pharmacy.

[0006] Furthermore, a multi-objective collaborative optimization algorithm is adopted for hospital drug inventory management, which dynamically balances multiple objectives such as inventory cost, stockout risk, shelf-life loss, and storage space utilization. Inventory cost objective function The stockout risk objective function is the sum of holding costs and ordering costs, where the number of orders is calculated by rounding up the ratio of the maximum safety stock difference to the purchase quantity; A logical function mapping is used to represent the relationship between inventory and consumption; the objective function for shelf-life loss is... Modeling the inverse relationship between inventory level and remaining shelf life sensitivity; space utilization objective function. Calculated by the average of the volume deviations of each storage cell; By dynamically adjusting the inventory of each type of drug, a multi-objective optimization is performed. The Pareto optimal solution is found among the four objective functions, yielding a weighted summation of the overall objective function, expressed as follows: , in, Represent the overall objective function; Indicates the first Dynamic weights of each objective function; Indicates the first An adaptive normalization factor for each objective function; Indicates the first The mean of the objective functions; Represents a constant; Indicates the constraint penalty coefficient; Indicates the total number of storage units; Indicates the first Non-negative truncation for overcapacity of a storage cell, when Punishment should be imposed at the appropriate time; Indicates the first Capacity constraints for each storage unit; No. The formula for the dynamic weights of the objective function is expressed as follows: , in, This represents the weighted smoothing factor; Indicates the first The rate of change of the objective function; Indicates the first The rate of change of the objective function; This represents the natural exponential function.

[0007] Furthermore, the first The specific capacity constraints for each storage cell are as follows: An adaptive relaxation strategy is adopted, which combines storage capacity constraints with dynamic relaxation variables. When the rate of change of storage unit demand exceeds the upper threshold, the capacity limit is relaxed proportionally in the positive direction. The relaxation amount is determined by the product of a preset coefficient and the rate of change of demand. When the rate of change of storage unit demand is lower than the lower threshold, the capacity limit is contracted in the reverse direction based on the volume ratio. The relaxation amount is determined by the product of a preset coefficient and the volume benchmark value.

[0008] Further, in step S2, the correlation coefficient of drug inventory within the same storage area is calculated by the ratio of the covariance to the standard deviation of inventory within the sliding time window, and the cumulative shortage probability is derived from the joint probability of shortage probabilities at each time point. Spatiotemporal encoding vector of drug class The formula is expressed as follows: , in, This indicates the average rate of change in inventory over the past three days. Indicates the first Class of drugs and the first in the same storage area Correlation coefficient of inventory levels for this type of drug; Indicates the first Coefficient of variation of consumption of this type of drug over the past 7 days; Indicates the first The cumulative out-of-stock probability of this type of drug.

[0009] Furthermore, step S3 specifically includes: S31. Population initialization based on Pareto front: The average inventory of historical Pareto optimal solutions is used as the initial population center point. An initial solution is generated by superimposing a Gaussian distribution perturbation that follows the variance of historical solutions. The population distribution is guided to the high-quality solution region by the statistical properties of the reference solution set. S32. Neighborhood Crossover and Mutation Strategy: Design a neighborhood crossover and mutation strategy based on inventory correlation. Parent selection adopts an exponential probability distribution based on solution ranking, prioritizing the selection of high-quality individuals. Meanwhile, the crossover operation is limited to gene exchange at positions where the spatial correlation coefficient exceeds a preset threshold. The mutation operation applies perturbation according to the direction of recent inventory change trends, and the perturbation amplitude is proportional to the consumption mutation coefficient. S33. Dynamic Weight Adjustment of the Objective Function: Based on the rate of change of the objective function in the current iteration, the dynamic weights of the objective function are adjusted and updated. The calculation method is expressed as follows: , in, Indicates the first The first generation of the population The rate of change of the objective function; Indicates the first The first generation of the population The mean of the objective functions; Indicates the first The first generation of the population The mean of the nth objective function; The first generation of the population The rate of change of the first objective function is substituted into the formula for calculating the dynamic weight of the objective function, replacing the first... The rate of change of each objective function is used to update the dynamic weights of the objective function, and the updated dynamic weights of the objective function are used as the dynamic weights of the objective function in the next iteration. S34. Iteration Stopping Criteria Judgment: The iteration stopping criteria adopt a multi-index joint criterion. If any of the following stopping criteria are met, the iteration will stop. The stopping criteria include Pareto front hypervolume stagnation criteria, population diversity threshold criteria, and maximum number of iterations criteria. S341. Criterion for hypervolume stagnation at the Pareto front: Calculate the nearest... The relative improvement rate of the Pareto front hypervolume is considered to be converged if the relative improvement rate is less than a preset threshold, and the iteration is stopped. The algebraic window threshold for determining supervolume stagnation; S342. Population diversity threshold judgment condition: Calculate the crowding distance mean of the current generation population. ,like If the population diversity is less than a preset threshold, the iteration is considered insufficient and the iteration stops. S343. Maximum iteration count determination condition: Set a hard upper limit and stop iteration.

[0010] Furthermore, the comprehensive environmental change indicators in step S4: The formula for the comprehensive environmental change index is as follows: , in, Indicates a comprehensive index of environmental change; Indicates the first The demand change rate of each storage unit over the past 3 days; The standard deviation of all drug consumption; Indicates the first The change rate of storage unit volume demand over the past 3 days; Indicates the first The maximum volume of each storage unit; if If the value exceeds a preset threshold, a local search is performed starting from the current optimal solution for a preset number of iterations.

[0011] Furthermore, in step S5... The formula for calculating the overall utility value of a cluster is as follows: , in, For the first The overall utility value of a cluster; For the first Cluster in the The weights on the objective are obtained by learning the cluster centers of the objective function weights from historical decision data, and the calculation method is expressed as follows: ; For the first The first historical solution Each target weight; For the historical Pareto solution set, it belongs to the first The set of indices of the solutions to the cluster; For the first The cluster in the th The average function value over each objective; Indicates the first The maximum value of the objective function at the current frontier; Indicates the first The minimum value of the objective function at the current frontier.

[0012] Furthermore, the inventory level, average daily consumption, remaining shelf life, and supplier response time in step S1 are dynamically weighted to obtain standardized optimized parameters. These standardized optimized parameters can replace the optimized parameters involved in subsequent steps S2 to S5. The inventory level is standardized based on the mean and standard deviation, and a dynamic weight reflecting its historical fluctuation frequency is added. The average daily consumption is logarithmically scaled and then adjusted with dynamic weights. The remaining shelf life is mapped to sensitivity using an S-curve function and then multiplied with the dynamic weights. The supplier response time is weighted after linear scaling. The dynamic weights are calculated based on the percentage of times the coefficient of variation of each indicator exceeds a preset threshold over the past 7 days.

[0013] This invention also provides a hospital drug inventory management system based on multi-objective collaborative optimization, comprising: The hospital drug inventory management definition module is used to define the optimization parameters, objective function, and inventory constraints for hospital drug inventory management. The optimization parameters include inventory level, average daily consumption, remaining shelf life in days, and supplier response time. The objective function includes inventory cost, stockout risk, shelf-life loss, and storage space utilization. Spatiotemporal encoding vector construction module: used to construct spatiotemporal encoding vectors based on optimization parameters, including inventory quantity, average change rate of inventory quantity in the past 3 days, correlation coefficient of drug inventory quantity in the same storage area, coefficient of variation of consumption in the past 7 days, and cumulative shortage probability. Multi-objective optimization module: used to input the spatiotemporal encoded vector into the improved NSGA-II non-dominated sorting genetic algorithm for multi-objective optimization; Trigger dynamic environment re-optimization operation module: Used to monitor the comprehensive environmental change index of the standard deviation of the 3-day consumption change rate of all drugs and the change rate of storage unit volume. When the comprehensive environmental change index exceeds the threshold, local optimization iteration is performed starting from the current optimal solution. The results generation module is used to generate Pareto front solution sets using the improved NSGA-II non-dominated sorting genetic algorithm. Fuzzy C-means clustering is used to group the solution sets, and the comprehensive utility value of each cluster is calculated. The solution center of the cluster with the highest utility value is selected as the final output result, thus achieving four-dimensional objective equilibrium optimization.

[0014] The advantages of this invention are: This invention employs a multi-objective collaborative optimization framework for hospital drug inventory management, encompassing four key optimization objectives: inventory cost, stockout risk, shelf-life loss, and storage space utilization. Unlike traditional fixed-weight summation strategies, this invention adaptively adjusts the weight distribution based on the relative rate of change of each objective function. This allows the optimization process to dynamically focus on different objectives based on fluctuations during the actual operating cycle, achieving a balance and coordination among multiple objectives and significantly enhancing the generalization ability and practicality of the optimization model. Addressing the lack of flexibility caused by the fixed maximum capacity in traditional inventory management, this invention proposes an adaptive capacity relaxation strategy based on the rate of demand change. It automatically determines whether to expand or contract capacity based on the demand changes of each storage unit over the past seven days. The relaxation amount is determined by historical demand trends and a preset coefficient. By flexibly adjusting the capacity limit range, the system can dynamically adjust the inventory control strategy according to actual demand, effectively avoiding stockouts or resource waste under sudden demand. A spatiotemporal feature fusion inventory state encoding mechanism is adopted to express the behavioral characteristics of a single drug class during dynamic evolution. It combines information from multiple dimensions, including inventory change trends, spatial distribution correlation with other drugs, consumption fluctuation intensity, and stockout probability, effectively overcoming the limitations of traditional methods that only focus on the current inventory snapshot. A structural improvement is made to the traditional NSGA-II model. An initial population is generated based on the distribution characteristics of historical Pareto solutions, and a neighborhood crossover strategy is designed in conjunction with inventory spatial correlation. This further integrates inventory change trends and consumption fluctuation intensity to construct a direction-aware individual mutation mechanism. Furthermore, environmental change monitoring indicators are used during operation. Local optimization restarts are triggered by inventory fluctuations and storage capacity changes, effectively improving the model's adaptability to dynamic changes in the actual operating environment, shortening convergence time, and enhancing the practicality and stability of the solution. Attached Figure Description

[0015] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0016] Figure 1 This is a flowchart of the steps of the method of the present invention; Figure 2 The solution space distribution corresponding to the dynamic weight adjustment mechanism; Figure 3The solution space distribution corresponding to the traditional fixed-weight strategy; Figure 4 For adaptive constraint strategy requirements response; Figure 5 Response to fixed-constraint strategy requirements; Figure 6 This is a comparison of the convergence process of the method of the present invention with that of the traditional NSGA-II; Figure 7 This is a comparison diagram of the Pareto front of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1 In this embodiment, as Figure 1 As shown, this invention provides a hospital drug inventory management method based on multi-objective collaborative optimization, the specific steps of which include: S1. Define the optimization parameters, objective function, and inventory constraints for hospital drug inventory management; the optimization parameters include inventory level, average daily consumption, remaining shelf life days, and supplier response time; the objective function includes inventory cost, stockout risk, shelf-life loss, and storage space utilization.

[0019] Specifically, define the first The inventory of this type of drug is , No. The average daily consumption of this type of drug is , No. The remaining shelf life of this type of drug is [number] days. , No. The supplier response time for this type of drug is ; , This indicates the total number of drug categories managed by the hospital pharmacy; Right now, Indicates the first Inventory of this type of drug Corresponding to a single drug variety, such as: This may represent "enteric-coated aspirin tablets"; This may represent "ibuprofen extended-release capsules"; For example: This indicates that the current inventory of Category 5 drugs is 120 boxes; This indicates that the average daily consumption of Category 5 drugs is 15 boxes; This indicates that the remaining shelf life of the Class 5 drug is 30 days. This indicates that the average response time for Category 5 drug suppliers is 2 days.

[0020] Therefore, the optimization problem is defined as: finding the optimal inventory level for each drug, thereby optimizing the cost, risk, and resource utilization of hospital drug inventory management.

[0021] Specifically, a multi-objective collaborative optimization algorithm is used for hospital drug inventory management, dynamically balancing multiple objectives such as inventory cost, stockout risk, shelf-life loss, and storage space utilization. Inventory cost objective function The stockout risk objective function is the sum of holding costs and ordering costs, where the number of orders is calculated by rounding up the ratio of the maximum safety stock difference to the purchase quantity; A logical function mapping is used to represent the relationship between inventory and consumption; the objective function for shelf-life loss is... Modeling the inverse relationship between inventory level and remaining shelf life sensitivity; space utilization objective function. Calculated by the average of the volume deviations of each storage unit; details are as follows: Define the objective function for inventory cost The calculation method is expressed as follows: ,in, Indicates the first The unit holding cost of this type of drug, for example, the daily storage cost per box is 0.5 yuan; Indicates the first The cost per order for this type of drug, for example, a fixed cost of 50 yuan per purchase; This represents the round-up operator; For the first Maximum safety stock level for this type of drug; Indicates the first Purchase volume of this type of drug, for example, a minimum purchase of 50 boxes each time; Define the stockout risk objective function The calculation method is expressed as follows: ,in, This parameter represents the sensitivity to stockout risk, for example, a value of 0.1. Indicates the first Minimum safety stock level for this type of drug; Indicates the first Coefficient of variation of consumption of this type of drug over the past 7 days; Define the objective function for shelf-life loss. The calculation method is expressed as follows: ,in, This parameter represents the sensitivity to shelf-life loss; for example, a value of 0.05. Define the objective function for space utilization. The calculation method is expressed as follows: ,in, Indicates the total number of storage units; Indicates the first A collection of drugs within a storage unit; Indicates the first The volume of a single unit of a drug class; Indicates the first Maximum volume of each storage unit; Represents the absolute value symbol.

[0022] Perform a weighted summation of the objective function; Conventional methods use weighted summation, but fixed weights can lead to conflicting objectives. By dynamically adjusting the inventory of each type of drug, a multi-objective optimization is performed. The Pareto optimal solution is found among the four objective functions, yielding a weighted summation of the overall objective function, expressed as follows: , in, Represent the overall objective function; Indicates the first Dynamic weights of each objective function; Indicates the first The adaptive normalization factor of each objective function is calculated as follows: It dynamically scales the values ​​of each objective function to the same dimension, avoiding the weight invalidation problem caused by differences in the absolute values ​​of the objective functions. For example, if inventory costs... The absolute value is 10,000 yuan, while the risk of stockouts... If the sum is 0.1, direct weighted summation will lead to... If ignored, normalization can balance the contributions of each objective; To represent a constant and prevent division by zero errors, for example, a value of ; This represents the constraint penalty coefficient, used to penalize violations of storage capacity constraints. By amplifying the penalty, the optimization is forced away from infeasible solutions; for example, it can be set to 1000. Indicates the first Non-negative truncation for overcapacity of a storage cell, when Punishment should be imposed at the appropriate time; Indicates the first Capacity constraints for each storage unit.

[0023] Based on the relative rate of change of each objective function's value in the current optimization cycle with its historical average, the weight adjustment factor for each objective function is calculated. Objective functions with a high rate of change will receive a larger weight allocation. Exponential function normalization is used to ensure that the total weight is 1, so that the optimization direction can adaptively track the changes in the importance of each objective function. No. The formula for the dynamic weights of the objective function is expressed as follows: , in, This represents the weight adjustment smoothing factor, for example, a value of 1.0; Indicates the first The rate of change of the objective function ; Indicates the first The rate of change of the objective function; Indicates the th in the current optimization cycle One objective function value; Indicates the historical average The objective function value.

[0024] Hospital drug demand is highly uncertain, such as due to sudden illnesses and seasonal changes. Traditional inventory models typically employ hard storage capacity constraints, such as a fixed maximum volume, but this can lead to two extreme problems: 1> When demand surges: the inability to expand inventory forces a reduction in procurement, further increasing the risk of stockouts; 2> When demand drops sharply: Idle storage space leads to reduced space utilization, which in turn leads to resource waste.

[0025] This invention employs an adaptive relaxation strategy, combining storage capacity constraints with dynamic relaxation variables. When the rate of change in storage unit demand exceeds an upper threshold, the capacity constraint is relaxed proportionally in the positive direction, with the relaxation amount determined by the product of a preset coefficient and the rate of change in demand. When the rate of change in storage unit demand is below a lower threshold, the capacity constraint is contracted in the reverse direction based on the volume ratio, with the relaxation amount determined by the product of a preset coefficient and a volume baseline value. The formula is as follows: , , in, Indicates the first Capacity constraints for each storage unit; Indicates the first A collection of drugs within a storage unit; Indicates the first The volume of a single unit of a drug class; Indicates the first Maximum volume of each storage unit; Indicates the first Capacity slack per storage unit; Indicates the first relaxation factor; Indicates the second relaxation coefficient; Indicates the first The demand change rate of each storage unit over the past 7 days is calculated as follows: ; Indicates the first Average demand for each storage unit over the past 7 days; Indicates the first Average demand for each storage unit over the past 14 days; This represents the threshold for the rate of change in demand, for example, set to 0.25; This represents the threshold for the rate of change in demand, for example, set to -0.15.

[0026] Therefore, the adaptive relaxation strategy, by incorporating capacity relaxation, allows storage capacity constraints to be flexibly adjusted under certain conditions, thereby achieving: 1> Flexible capacity expansion: When demand increases, capacity restrictions are relaxed proportionally, i.e. ; 2> Intelligent contraction: When demand declines, capacity constraints are proactively tightened, i.e. .

[0027] In one embodiment, the triggering condition for capacity relaxation is: 1> Surge in Demand: The 7-day change rate of demand for storage units Exceeding the upper threshold ,but To increase capacity; 2> Demand Decline: The rate of change in demand for storage units over the past 7 days Below the lower threshold ,but To reduce capacity.

[0028] Therefore, the first Capacity constraints of each storage cell That is, to represent the first The total volume of all medicines within each storage unit must not exceed the adjusted capacity limit. ,Right now .

[0029] S2. Construct a spatiotemporal encoding vector based on optimized parameters, including inventory quantity, average change rate of inventory quantity in the past 3 days, correlation coefficient of drug inventory quantity in the same storage area, coefficient of variation of consumption in the past 7 days, and cumulative shortage probability.

[0030] Specifically, the correlation coefficient of drug inventory within the same storage area is calculated by the ratio of the covariance to the standard deviation of inventory within the sliding time window, and the cumulative shortage probability is derived from the joint probability of shortage probabilities at each time point.

[0031] No. Spatiotemporal encoding vector of drug class The formula is expressed as follows: in, Indicates the first Inventory levels of this type of drug; This represents the average change rate of inventory over the past 3 days, calculated as follows: ; Indicates the current moment; Indicates the first Class 1 drugs Inventory level at any given time; Indicates the first Class 1 drugs Inventory level at any given time; Indicates the first Class of drugs and the first in the same storage area The correlation coefficient of inventory levels for drugs of this class is calculated as follows: ; Indicates the length of the observation time window; Indicates the first Class 1 drugs Average inventory level within a time window; For the first Inventory levels of this type of drug; Indicates the first Class 1 drugs Average inventory level within a time window; Indicates the first observation within the observation time window Standard deviation of drug inventory levels; Indicates the first observation within the observation time window Standard deviation of drug inventory levels; Indicates the first The coefficient of variation of consumption of this type of drug over the past 7 days is calculated as follows: ; Indicates the first Standard deviation of consumption of this type of drug over the past 7 days; Indicates the first The average consumption of this type of drug over the past 7 days; Indicates the first The cumulative shortage probability of this type of drug, incorporating spatiotemporal dynamic characteristics, is calculated as follows: ; Indicates the first Class 1 drugs The probability of being out of stock at any given time is calculated as follows: ; This indicates taking the maximum value; Indicates the first Class 1 drugs Average daily consumption at any given moment.

[0032] S3. Input the spatiotemporal encoding vector into the improved NSGA-II non-dominated sorting genetic algorithm for multi-objective optimization.

[0033] Specifically, S31. Population initialization based on Pareto front: Traditional NSGA-II random initialization of the population is inefficient, which can lead to long optimization processes and difficulty in quickly converging and finding the optimal solution. This invention uses the average inventory of historical Pareto optimal solutions as the initial population center, and superimposes a Gaussian distribution perturbation following the variance of historical solutions to generate the initial solution. The statistical properties of the reference solution set guide the population distribution to the high-quality solution region, as expressed by the formula: in, Indicates the first The initial stock of this type of drug; Represents the historical Pareto optimal reference solution set; Indicates the first solution in the reference solution set The inventory quantity corresponding to each solution; This indicates that the expression follows a pattern with a mean of 0 and a variance of . Gaussian distribution; The variance of the historical solutions is expressed as follows: ; This represents the average historical inventory level.

[0034] S32. Neighborhood crossover and mutation strategies: Conventional simulations of binary crossover and polynomial mutations destroy high-quality gene fragments, resulting in low optimization accuracy. This invention designs a neighborhood crossover and mutation strategy based on inventory correlation. Parent selection adopts an exponential probability distribution based on solution ranking to prioritize high-quality individuals. Meanwhile, the crossover operation is limited to gene exchange at positions where the spatial correlation coefficient exceeds a preset threshold. The mutation operation applies perturbation according to the recent inventory change trend, and the perturbation amplitude is proportional to the consumption mutation coefficient. Specifically, the probability of individual selection and the determination of gene crossover positions are expressed as follows: in, Indicates the first The probability that an individual is selected as a parent; This represents the ranking decay factor, for example, set to 0.5; Indicates the first The non-dominated ranking order of a solution in the population; Indicates the first The non-dominated ranking order of a solution in the population; Indicates the location of gene crossing over; Indicates the first Spatiotemporal encoding vectors of drug class; Indicates the first Spatiotemporal encoding vectors of drug class; This means taking the expression that minimizes the expression. The value; Indicates the first The first individual corresponding to the first Inventory of similar drugs; This represents the preset cosine similarity threshold. Characterization and Cosine similarity; Indicates the first The first individual corresponding to the first Inventory of similar drugs; Indicates the first Class and the Spatial correlation coefficient of drugs of the same class Characterizing the first Class and the The set of drug categories whose spatial correlation coefficient exceeds 0.7 and whose spatiotemporal encoding vector cosine similarity is greater than a preset cosine similarity threshold takes into account both spatial location and drug consumption trends.

[0035] The mutation operation perturbs the inventory according to the trend. Specifically, for the selected drug inventory, the direction of the mutation perturbation is determined based on the positive or negative direction of the average rate of change of its inventory over the past 3 days. If the inventory is trending upward, the perturbation is applied in the direction of increasing inventory; if it is trending downward, the perturbation is applied in the direction of decreasing inventory. The perturbation amplitude is dynamically adjusted by the coefficient of variation of the drug's consumption over the past 7 days. The greater the consumption fluctuation, the greater the perturbation amplitude. The mutation operation is expressed as follows: in, Indicates the mutated first Inventory of similar drugs; This represents the coefficient of variation, for example, set to 0.1; The sign function returns the sign of the input value. This indicates the average rate of change in inventory over the past three days. This represents the coefficient of variation of consumption over the past 7 days.

[0036] S33. Dynamic weight adjustment of the objective function: The dynamic weights of the objective function are adjusted and updated based on the rate of change of the objective function in the current iteration. The calculation method is expressed as follows: , in, Indicates the first The first generation of the population The rate of change of the objective function; Indicates the first The first generation of the population The mean of the objective functions; Indicates the first The first generation of the population The mean of the objective functions.

[0037] Optionally, to discover as many diversity requirements as possible and avoid premature convergence of the optimization algorithm, the rate of change of the objective function can be further updated by combining the population diversity index, i.e.: , in, This indicates a parameter update operation; Indicates the first The first generation of the population The population diversity index of the objective function is calculated as follows: ; Indicates the first The first generation of the population The standard deviation of each objective function; Indicates the first The first generation of the population The mean of the objective functions; This represents the diversity impact coefficient, used to control the degree of influence of population diversity indicators on the dynamic weight updates of the objective function. For example, it can be set to 0.2.

[0038] Furthermore, the first The first generation of the population The rate of change of the first objective function is substituted into the formula for calculating the dynamic weight of the objective function, replacing the first... The rate of change of each objective function is used to update the dynamic weights of the objective function. The updated dynamic weights of the objective function are then used as the dynamic weights of the objective function in the next iteration.

[0039] It needs to be explained that "gene segment" is a term from the field of genetic algorithms, used to describe the encoded structure of a solution in an optimization problem. That is, a gene segment represents a certain dimension parameter of the solution (individual), corresponding to the inventory quantity of a single drug in this invention, such as the... The inventory of a drug class is a gene fragment, and the entire solution vector (The set of inventory) constitutes a complete chromosome (individual).

[0040] S34. Iteration stopping condition judgment: The iteration stopping criteria adopt a multi-index joint criterion to comprehensively evaluate convergence, diversity and computational efficiency, avoid misjudgment by a single index, and prevent infinite loops by the maximum number of iterations. The iteration stops when any of the following stopping conditions are met: Pareto front hypervolume stagnation condition, population diversity threshold condition, and maximum number of iterations condition. Pareto front hypervolume stagnation criterion: calculate the nearest Relative improvement rate of Pareto front hypervolume ,like If the value is less than the preset threshold, the frontier is considered to have converged, and the iteration stops. The algebraic window threshold for determining supervolume stagnation; Population diversity threshold judgment condition: Calculate the crowding distance mean of the current generation population. ,like If the population diversity is less than a preset threshold, the iteration is considered insufficient and the iteration stops. Maximum iteration count determination condition: Set a hard upper limit, then stop iteration; (1) Criteria for judging hypervolume stagnation at the Pareto front: Calculate the most recent generation (e.g.) The relative improvement rate of the Pareto front hypervolume is expressed as: in, Indicates the first generation and The relative improvement rate of supervolume between generations; The algebraic window threshold for determining supervolume stagnation is set to 10. Indicates the first Hypervolume indices of Pareto frontier in generational populations; Indicates the first Hypervolume indices of the Pareto front of a generation; if Less than a preset threshold (e.g.) If the frontier converges, then the iteration is stopped.

[0041] (2) Criteria for determining the population diversity threshold: The mean crowding distance of the current generation population is calculated and expressed as: in, Indicates the first Crowding distance mean of the generation population; Indicates the population size; for example, setting it to 100 indicates that the population has 100 individuals. Indicates the first The crowding distance between solutions is calculated as follows: ; Indicates the first The solution is at the th solution. The values ​​of each objective function; Indicates the first The solution is at the th solution. The values ​​of each objective function; Indicates the first The maximum value of the objective function at the current frontier; Indicates the first The minimum value of the objective function at the current frontier.

[0042] like Less than a preset threshold (e.g.) If the population diversity is insufficient, the iteration will stop.

[0043] (3) Criteria for determining the maximum number of iterations Set a hard limit (e.g., maximum number of iterations). Stop iterating.

[0044] S4. Trigger dynamic environment re-optimization operation: Monitor the comprehensive environmental change index of the standard deviation of the 3-day consumption change rate of all drugs and the change rate of storage unit volume. When the comprehensive environmental change index exceeds the threshold, execute local optimization iteration starting from the current optimal solution.

[0045] Specifically, the formula for the comprehensive environmental change index is as follows: , in, Indicates a comprehensive index of environmental change; Indicates the first The demand change rate of each storage unit over the past 3 days; The standard deviation of all drug consumption; Indicates the first The change rate of storage unit volume demand over the past 3 days; Indicates the first The maximum volume of each storage unit; if If the value exceeds a preset threshold, a local search is performed starting from the current optimal solution for a preset number of iterations, instead of a global restart, thus reducing computational overhead.

[0046] S5. A Pareto front solution set is generated by the improved non-dominated sorting genetic algorithm NSGA-II. Fuzzy C-means clustering is used to group the solution set and calculate the comprehensive utility value of each cluster. The cluster center solution with the highest utility value is selected as the final output result to achieve four-dimensional objective equilibrium optimization.

[0047] Specifically, no. The formula for calculating the overall utility value of a cluster is as follows: , in, For the first The overall utility value of a cluster; For the first Cluster in the The weights on the objective are obtained by learning the cluster centers of the objective function weights from historical decision data, and the calculation method is expressed as follows: ; For the first The first historical solution Each target weight; For the historical Pareto solution set, it belongs to the first The set of indices of the solutions to the cluster; For the first The cluster in the th The average function value over each objective is calculated as follows: ; For the first A cluster is a set of indices for solutions. For the first The solution is at the th solution. The values ​​of each objective function; Indicates the first The maximum value of the objective function at the current frontier; Indicates the first The minimum value of the objective function at the current frontier.

[0048] Preferably, the inventory, average daily consumption, remaining shelf life days, and supplier response time in step S1 are dynamically weighted to obtain standardized optimized parameters; the standardized optimized parameters can replace the optimized parameters involved in subsequent steps S2 to S5. Inventory levels are standardized based on mean and standard deviation, and then weighted with dynamic weights reflecting historical fluctuations. Daily consumption is logarithmically scaled and adjusted for dynamic weights. Remaining shelf life is mapped to sensitivity using an S-curve function and then multiplied by the dynamic weights. Supplier response time is weighted after linear scaling. Dynamic weights are calculated based on the percentage of times each indicator's coefficient of variation exceeds a preset threshold over the past 7 days. This achieves multi-dimensional inventory data fusion and standardization, represented as follows: , , , , in, Represents the standardized first Inventory levels of this type of drug; This represents the current average inventory level. This represents the standard deviation of the current inventory level; This indicates the dynamic weighting of inventory levels, reflecting the indicators over the past 7 days. The frequency of abnormal fluctuations is calculated as follows: ; Indicates the indicator for day d The coefficient of variation is calculated as follows: ,in Indicates the indicator for day d standard deviation Indicates the indicator for day d The mean; Indicates function substitution ; Indicates the indicator category, These correspond to inventory level indicators. Daily average consumption index Shelf life remaining days index Supplier response time metrics ; This represents a preset threshold for the coefficient of variation, such as setting a threshold for the coefficient of variation of the inventory level indicator. Daily average consumption index coefficient of variation threshold The threshold for the coefficient of variation of the remaining shelf life index Supplier response time coefficient of variation threshold ; Represents the standardized first The average daily consumption of this type of drug; This represents the logarithmic function, with a default base of 10. This represents the maximum daily average consumption of all medications. The dynamic weighting of average daily consumption is expressed as follows: ; Indicates function substitution ; This indicates an indicator function that takes the value 1 when the condition is met and 0 otherwise. Represents the standardized first The remaining shelf life of this type of drug; Represents the natural exponential function; The dynamic weight representing the remaining days of shelf life is calculated as follows: ; Indicates function substitution ; For the standardized first Supplier response time for this type of drug; This represents the maximum response time across all suppliers. The dynamic weighting of supplier response time is expressed as follows: ; This parameter represents the sensitivity to controlling shelf life.

[0049] Example 2 In this embodiment, the differences in solution space distribution between the dynamic weight adjustment mechanism and the traditional fixed weight strategy are compared using a three-dimensional surface plot, such as... Figure 2 and Figure 3 As shown, the horizontal and vertical axes represent the weighting of inventory costs and stockout risk, respectively, while the vertical axis reflects the distribution density of Pareto front solutions. The dynamic weight surface exhibits a multi-peaked undulating structure, indicating that the algorithm can discover multiple high-quality solution sets under different weight combinations, while the fixed weight surface presents a single sloping plane, showing its limited ability to explore the solution space. The difference in the complexity of the surface morphology intuitively reveals the ability of this technology to automatically focus on key contradictory objectives by dynamically sensing the changing trend of the objective function. The dynamic weight mechanism enables the algorithm to enhance risk control weights when inventory costs surge and increase loss suppression weights when the shelf life is approaching.

[0050] Example 3 In this embodiment, the response process of the storage capacity constraint strategy to demand fluctuations is dynamically displayed using vector arrows, such as... Figure 4 and Figure 5 As shown, the horizontal and vertical axes represent the rate of change in demand and the capacity slack, respectively. The direction of the arrows indicates the direction of strategy adjustment, and the color intensity reflects the strength of strategy adjustment. The streamlines of the adaptive constraint strategy exhibit a multi-directional radial pattern, especially in regions of surging demand (…). Figure 5 This creates an outward-expanding flow field, particularly in areas where demand drops sharply. Figure 4 This generates inward-contracting clusters of streamlines, demonstrating the system's ability to flexibly adjust storage space according to changing demand. In contrast, the streamlines of a fixed-constraint strategy are parallel straight lines, allowing mechanical adjustment only in a single direction. Experimental results prove that this technique avoids inventory stagnation caused by traditional hard constraints while preventing resource waste due to excessive relaxation.

[0051] Example 4 In this embodiment, as Figure 6As shown, the convergence process of the algorithm is compared by the change trajectory of the hypervolume index. The horizontal axis represents the number of iterations, and the vertical axis represents the comprehensive evaluation index of the solution set quality. The curve fluctuations reflect the stability of the algorithm. The convergence curve of this technique shows a steep drop followed by rapid stabilization, with a narrow shaded bandwidth, indicating that the algorithm quickly locks into the optimal region like a precision-guided missile. The curve of the traditional method drops gently and oscillates significantly in the later stages, indicating that it is prone to getting trapped in local optima. Experimental results show that the initial population generated based on historical Pareto solutions is like implanting prior knowledge into the algorithm, allowing it to start close to the high-quality solution region. Furthermore, the directional mutation operation combined with inventory trends guides the search direction to avoid invalid regions, making the optimization process both efficient and robust, and significantly shortening the computation cycle.

[0052] like Figure 7 As shown, a Pareto front comparison plot is used to verify the optimization performance of the improved non-dominated sorting genetic algorithm. The horizontal and vertical axes represent normalized inventory cost and stockout risk, respectively. The black dashed line represents the theoretical optimal front, the cyan scatter dots represent the solution set of the traditional algorithm, and the red diamond dots represent the solution set of this technique. It can be observed that the solution set of this technique exhibits two significant characteristics: first, the distribution is closer to the theoretical front, proving that the improved neighborhood crossover strategy and trend-oriented mutation mechanism effectively improve the quality of the solution; second, the uniformity of the solution set distribution is significantly improved, reflecting the success of the dynamic weight adjustment and diversity preservation strategies. Experimental results show that this technique achieves a dual breakthrough in solution set convergence and distribution, providing a higher-quality Pareto solution selection space for practical applications.

[0053] Example 5 This embodiment provides a hospital drug inventory management system based on multi-objective collaborative optimization, including: The hospital drug inventory management definition module is used to define the optimization parameters, objective function, and inventory constraints for hospital drug inventory management. The optimization parameters include inventory level, average daily consumption, remaining shelf life in days, and supplier response time. The objective function includes inventory cost, stockout risk, shelf-life loss, and storage space utilization. Spatiotemporal encoding vector construction module: used to construct spatiotemporal encoding vectors based on optimization parameters, including inventory quantity, average change rate of inventory quantity in the past 3 days, correlation coefficient of drug inventory quantity in the same storage area, coefficient of variation of consumption in the past 7 days, and cumulative shortage probability. Multi-objective optimization module: used to input the spatiotemporal encoded vector into the improved NSGA-II non-dominated sorting genetic algorithm for multi-objective optimization; Trigger dynamic environment re-optimization operation module: Used to monitor the comprehensive environmental change index of the standard deviation of the 3-day consumption change rate of all drugs and the change rate of storage unit volume. When the comprehensive environmental change index exceeds the threshold, local optimization iteration is performed starting from the current optimal solution. The results generation module is used to generate Pareto front solution sets using the improved NSGA-II non-dominated sorting genetic algorithm. Fuzzy C-means clustering is used to group the solution sets, and the comprehensive utility value of each cluster is calculated. The solution center of the cluster with the highest utility value is selected as the final output result, thus achieving four-dimensional objective equilibrium optimization.

[0054] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A hospital drug inventory management method based on multi-objective collaborative optimization, characterized in that, Includes the following steps: S1. Define the optimization parameters, objective function, and inventory constraints for hospital drug inventory management; The optimization parameters include inventory level, average daily consumption, remaining shelf life days, and supplier response time; the objective function includes inventory cost, stockout risk, shelf-life loss, and storage space utilization. S2. Construct a spatiotemporal encoding vector based on optimized parameters, including inventory quantity, average change rate of inventory quantity in the past 3 days, correlation coefficient of drug inventory quantity in the same storage area, coefficient of variation of consumption in the past 7 days, and cumulative shortage probability. S3. Input the spatiotemporal encoding vector into the improved NSGA-II non-dominated sorting genetic algorithm for multi-objective optimization; S4. Trigger dynamic environment re-optimization operation: Monitor the comprehensive environmental change index of the standard deviation of the 3-day consumption change rate of all drugs and the change rate of storage unit volume. When the comprehensive environmental change index exceeds the threshold, execute local optimization iteration starting from the current optimal solution. S5. A Pareto front solution set is generated by the improved non-dominated sorting genetic algorithm NSGA-II. Fuzzy C-means clustering is used to group the solution set and calculate the comprehensive utility value of each cluster. The cluster center solution with the highest utility value is selected as the final output result to achieve four-dimensional objective equilibrium optimization.

2. The hospital drug inventory management method based on multi-objective collaborative optimization according to claim 1, characterized in that, Step S1 defines the optimization parameters for hospital drug inventory management, specifically including: Definition of the first The inventory of this type of drug is , No. The average daily consumption of this type of drug is , No. The remaining shelf life of this type of drug is [number] days. , No. The supplier response time for this type of drug is ; , This indicates the total number of drug categories managed by the hospital pharmacy.

3. The hospital drug inventory management method based on multi-objective collaborative optimization according to claim 2, characterized in that, A multi-objective collaborative optimization algorithm is used for hospital drug inventory management, which dynamically balances multiple objectives such as inventory cost, stockout risk, shelf-life loss, and storage space utilization. Inventory cost objective function The stockout risk objective function is the sum of holding costs and ordering costs, where the number of orders is calculated by rounding up the ratio of the maximum safety stock difference to the purchase quantity; A logical function mapping is used to represent the relationship between inventory and consumption; the objective function for shelf-life loss is... Modeling the inverse relationship between inventory level and remaining shelf life sensitivity; space utilization objective function. Calculated by the average of the volume deviations of each storage cell; By dynamically adjusting the inventory of each type of drug, a multi-objective optimization is performed. The Pareto optimal solution is found among the four objective functions, yielding a weighted summation of the overall objective function, expressed as follows: ， in, Represent the overall objective function; Indicates the first Dynamic weights of each objective function; Indicates the first An adaptive normalization factor for each objective function; Indicates the first The mean of the objective functions; Represents a constant; Indicates the constraint penalty coefficient; Indicates the total number of storage units; Indicates the first Non-negative truncation for overcapacity of a storage cell, when Punishment should be imposed at the appropriate time; Indicates the first Capacity constraints for each storage unit; No. The formula for the dynamic weights of the objective function is expressed as follows: ， in, This represents the weighted smoothing factor; Indicates the first The rate of change of the objective function; Indicates the first The rate of change of the objective function; This represents the natural exponential function.

4. The hospital drug inventory management method based on multi-objective collaborative optimization according to claim 3, characterized in that, No. The specific capacity constraints for each storage cell are as follows: An adaptive relaxation strategy is adopted, which combines storage capacity constraints with dynamic relaxation variables. When the rate of change of storage unit demand exceeds the upper threshold, the capacity limit is relaxed proportionally in the positive direction. The relaxation amount is determined by the product of a preset coefficient and the rate of change of demand. When the rate of change of storage unit demand is lower than the lower threshold, the capacity limit is contracted in the reverse direction based on the volume ratio. The relaxation amount is determined by the product of a preset coefficient and the volume benchmark value.

5. A hospital drug inventory management method based on multi-objective collaborative optimization according to claim 4, characterized in that, In step S2, the correlation coefficient of drug inventory within the same storage area is calculated by the ratio of the covariance to the standard deviation of inventory within the sliding time window. The cumulative stockout probability is derived from the joint probability of stockout probabilities at each time point. Spatiotemporal encoding vector of drug class The formula is expressed as follows: ， in, Indicates the first Inventory levels of this type of drug; This indicates the average rate of change in inventory over the past three days. Indicates the first Class of drugs and the first in the same storage area Correlation coefficient of inventory levels for this type of drug; Indicates the first Coefficient of variation of consumption of this type of drug over the past 7 days; Indicates the first The cumulative out-of-stock probability of this type of drug.

6. A hospital drug inventory management method based on multi-objective collaborative optimization according to claim 5, characterized in that, Step S3 specifically includes: S31. Population initialization based on Pareto front: The average inventory of historical Pareto optimal solutions is used as the initial population center point. An initial solution is generated by superimposing a Gaussian distribution perturbation that follows the variance of historical solutions. The population distribution is guided to the high-quality solution region by the statistical properties of the reference solution set. S32. Neighborhood Crossover and Mutation Strategy: Design a neighborhood crossover and mutation strategy based on inventory correlation. Parent selection adopts an exponential probability distribution based on solution ranking, prioritizing the selection of high-quality individuals. Meanwhile, the crossover operation is limited to gene exchange at positions where the spatial correlation coefficient exceeds a preset threshold. The mutation operation applies perturbation according to the direction of recent inventory change trends, and the perturbation amplitude is proportional to the consumption mutation coefficient. S33. Dynamic Weight Adjustment of the Objective Function: Based on the rate of change of the objective function in the current iteration, the dynamic weights of the objective function are adjusted and updated. The calculation method is expressed as follows: ， in, Indicates the first The first generation of the population The rate of change of the objective function; Indicates the first The first generation of the population The mean of the objective functions; Indicates the first The first generation of the population The mean of the objective functions; The first The first generation of the population The rate of change of the first objective function is substituted into the formula for calculating the dynamic weight of the objective function, replacing the first... The rate of change of each objective function is used to update the dynamic weights of the objective function, and the updated dynamic weights of the objective function are used as the dynamic weights of the objective function in the next iteration. S34. Iteration Stopping Criteria Judgment: The iteration stopping criteria adopt a multi-index joint criterion. If any of the following stopping criteria are met, the iteration will stop. The stopping criteria include Pareto front hypervolume stagnation criteria, population diversity threshold criteria, and maximum number of iterations criteria. S341. Criterion for hypervolume stagnation at the Pareto front: Calculate the nearest... The relative improvement rate of the Pareto front hypervolume is considered to be converged if the relative improvement rate is less than a preset threshold, and the iteration is stopped. The algebraic window threshold for determining supervolume stagnation; S342. Population diversity threshold judgment condition: Calculate the crowding distance mean of the current generation population. ,like If the population diversity is less than a preset threshold, the iteration is considered insufficient and the iteration stops. S343. Maximum iteration count determination condition: Set a hard upper limit and stop iteration.

7. A hospital drug inventory management method based on multi-objective collaborative optimization according to claim 6, characterized in that, Comprehensive environmental change indicators in step S4: The formula for the comprehensive environmental change index is as follows: ， in, Indicates a comprehensive index of environmental change; Indicates the first The demand change rate of each storage unit over the past 3 days; The standard deviation of all drug consumption; Indicates the first The change rate of storage unit volume demand over the past 3 days; Indicates the first The maximum volume of each storage unit; if If the value exceeds a preset threshold, a local search is performed starting from the current optimal solution for a preset number of iterations.

8. A hospital drug inventory management method based on multi-objective collaborative optimization according to claim 7, characterized in that, In step S5 The formula for calculating the overall utility value of a cluster is as follows: ， in, For the first The overall utility value of a cluster; For the first Cluster in the The weights on the objective are obtained by learning the cluster centers of the objective function weights from historical decision data, and the calculation method is expressed as follows: ; For the first The first historical solution Each target weight; For the historical Pareto solution set, it belongs to the first The set of indices of the solutions to the cluster; For the first The cluster in the th The average function value over each objective; Indicates the first The maximum value of the objective function at the current frontier; Indicates the first The minimum value of the objective function at the current frontier.

9. A hospital drug inventory management method based on multi-objective collaborative optimization according to claim 8, characterized in that, The inventory level, average daily consumption, remaining shelf life, and supplier response time in step S1 are dynamically weighted to obtain standardized optimization parameters. These standardized optimization parameters can replace the optimization parameters involved in subsequent steps S2 to S5. The inventory level is standardized based on the mean and standard deviation, and a dynamic weight reflecting its historical fluctuation frequency is added. The average daily consumption is logarithmically scaled and then adjusted with dynamic weights. The remaining shelf life is mapped to sensitivity using a sigmoid function and then multiplied by the dynamic weights. Supplier response time is weighted after being scaled linearly; dynamic weights are calculated based on the percentage of times each indicator's coefficient of variation exceeds a preset threshold over the past 7 days.

10. A hospital drug inventory management system based on multi-objective collaborative optimization, executing the hospital drug inventory management method based on multi-objective collaborative optimization as described in claim 1, characterized in that, include: Hospital Drug Inventory Management Definition Module: Used to define the optimization parameters, objective function, and inventory constraints for hospital drug inventory management; The optimization parameters include inventory level, average daily consumption, remaining shelf life in days, and supplier response time. The objective function includes inventory cost, stockout risk, shelf-life loss, and storage space utilization. Spatiotemporal encoding vector construction module: used to construct spatiotemporal encoding vectors based on optimization parameters, including inventory quantity, average change rate of inventory quantity in the past 3 days, correlation coefficient of drug inventory quantity in the same storage area, coefficient of variation of consumption in the past 7 days, and cumulative shortage probability. Multi-objective optimization module: used to input the spatiotemporal encoded vector into the improved NSGA-II non-dominated sorting genetic algorithm for multi-objective optimization; Trigger dynamic environment re-optimization operation module: Used to monitor the comprehensive environmental change index of the standard deviation of the 3-day consumption change rate of all drugs and the change rate of storage unit volume. When the comprehensive environmental change index exceeds the threshold, local optimization iteration is performed starting from the current optimal solution. The results generation module is used to generate Pareto front solution sets using the improved NSGA-II non-dominated sorting genetic algorithm. Fuzzy C-means clustering is used to group the solution sets, and the comprehensive utility value of each cluster is calculated. The solution center of the cluster with the highest utility value is selected as the final output result, thus achieving four-dimensional objective equilibrium optimization.

Images