A laboratory consumable management system and consumable management method based on multi-dimensional dynamic programming

Abstract

The application discloses a kind of laboratory consumable management system and method based on multidimensional dynamic programming, belong to laboratory resource management and intelligent decision-making technical field, by collecting consumable historical consumption data, improved association rule mining algorithm is used to construct consumable association network;Demand prediction is carried out using HoltWinters exponential smoothing model combined with association correction;Establish the total cost minimization model including stockout cost, inventory holding cost and expiration scrap cost, dynamically calculate optimal purchase quantity;Introduce expiration risk coefficient and realize joint early warning and FEFO delivery strategy using association network;Online update model parameters by adaptive feedback regulator.The application can accurately identify the real consumption coupling relationship between consumables, simultaneously optimize inventory level, purchase quantity and expiration risk, significantly reduce the stockout rate, overstock rate and expiration scrap rate of laboratory consumables, realize intelligent, whole-process closed-loop management, and improve laboratory operation efficiency.

Description

Technical Field

[0001] This invention relates to the field of laboratory resource management and intelligent decision-making technology, and more specifically to a laboratory consumables management system and consumables management method based on multi-dimensional dynamic programming. Background Technology

[0002] Currently, in laboratories in fields such as biomedicine, chemical analysis, and scientific research and education, the management of consumables (such as pipette tips, centrifuge tubes, chemical reagents, culture media, etc.) is a key aspect of daily operations.

[0003] Current laboratory consumables management relies primarily on manual registration or simple spreadsheets, leading to issues such as inaccurate inventory, coexistence of stockouts and overstocking, and a lack of expiration date management. Published patents, such as CN111915261B, use a weighted moving average method to predict monthly consumption, but their safety stock is fixed and does not consider expiration dates; CN106780890A only addresses physical access control and lacks an intelligent replenishment algorithm. None of these solutions resolve the issues of coupled consumption among consumables (e.g., experimental combinations consuming multiple consumables simultaneously), balancing multi-objective costs (stockouts, holding, expiration), and the adaptive updating of predictive models.

[0004] Therefore, the pain points of existing technologies are: Low inventory accuracy: Traditional manual bookkeeping or simple Excel management cannot reflect the actual consumption and remaining amount of consumables in real time, resulting in "discrepancy between accounts and actual inventory".

[0005] Rigid safety stock: Existing ERP systems typically use fixed "maximum-minimum" inventory warnings, ignoring the seasonal and cyclical fluctuations (such as concentrated experiments at the end of the month) and randomness of experimental consumable consumption, resulting in both "stockouts when urgently needed" and "large amounts of near-expiry waste".

[0006] Lack of expiration date management: For consumables with strict expiration dates, such as biochemical reagents, the existing system lacks dynamic early warning based on consumption rate, often resulting in the scrapping of entire batches of reagents due to expiration, causing huge waste.

[0007] Disconnect between requisition and approval: Unable to automatically verify the logical relationship between requisition applications and actual project budgets and remaining inventory.

[0008] Therefore, how to provide a laboratory consumables management system and consumables management method based on multi-dimensional dynamic programming is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0009] In view of this, the present invention provides a laboratory consumables management system and consumables management method based on multi-dimensional dynamic programming to solve the problems in the background art.

[0010] To achieve the above objectives, the present invention adopts the following technical solution: On one hand, this invention discloses a laboratory consumables management method based on multi-dimensional dynamic programming, comprising: Historical consumption data for each type of consumable is collected to create a dynamic profile of the consumables. An improved association rule mining algorithm is used to identify the consumption relationships between different consumables, forming a consumable association network. Based on the historical consumption data and the consumable association network, the HoltWinters exponential smoothing model is used to predict the daily consumption demand of each consumable in the future within a preset time period. The predicted value of the current consumable is then corrected based on the prediction results of the associated consumables to obtain the corrected predicted demand. Based on the standard deviation of the revised forecast demand and historical daily consumption, safety stock and replenishment point are dynamically calculated. Establish an optimization model with the goal of minimizing total cost, and calculate the recommended purchase quantity for each consumable. The total cost includes stockout cost, inventory holding cost, and expired / scrap cost. For different batches of the same consumable, calculate the expiration date risk coefficient for each batch, and use the consumable association network to automatically push a joint use suggestion to the administrator of the associated consumables when the risk coefficient of a certain batch of a certain consumable exceeds the threshold. When a requisition request is received, the system recommends batches for shipment in descending order of the expiration date risk coefficient, and issues an alert when the risk coefficient exceeds a preset threshold. The system compares the actual consumption data with the predicted value each time, and dynamically adjusts the smoothing parameters in the HoltWinters model and the cost coefficient in the optimization model using the exponentially weighted moving average method.

[0011] The above technical solutions can simultaneously address multiple complex factors such as the cyclical fluctuations in laboratory consumable consumption, the coupled consumption of multiple consumables, random demand fluctuations, and expiration dates. They significantly reduce stockout rates, inventory backlog rates, and expiration rates, achieving a leap from passive replenishment to proactive optimization decision-making, thereby improving the efficiency of laboratory resource utilization and economic benefits.

[0012] Preferably, in the above-mentioned laboratory consumables management method based on multi-dimensional dynamic programming, the specific steps of the improved association rule mining algorithm to identify the consumption association relationships between different consumables are as follows: Historical consumable requisition records are collected, and a transaction database is constructed using a single experimental operation as the transaction boundary. Each transaction includes a timestamp of the transaction occurrence, the identifier of the consumables involved, and their consumption quantity. A time decay weight is assigned to each transaction, which decreases exponentially as the interval between the transaction occurrence time and the current time increases. For each type of consumable in each transaction, its consumption quantity is discretized into quantity levels based on the ratio of its consumption quantity to the historical daily average consumption quantity of that consumable. Frequent itemsets and association rules are mined from the transaction database using weighted support counting and weighted confidence calculation methods, where both weighted support counting and weighted confidence calculation use the time decay weight as a multiplier. For the mined association rules, a quantity level correction factor is introduced to enhance or weaken the confidence. The quantity level correction factor is the ratio of the average quantity level of the consequent consumable under the condition that the antecedent of the rule occurs to the average quantity level of that consumable in all transactions. Rules with confidence exceeding a preset threshold in the association rules after quantity level correction are stored as positive associations, forming a consumable association network.

[0013] The above technical solution eliminates the spurious associations caused by the fixed time window in the traditional Apriori algorithm, making recent consumption behavior have a greater impact on association rules. At the same time, it incorporates the quantity of consumption (rather than just whether it occurs) into the association strength assessment, thereby more accurately identifying real and meaningful consumable associations in the laboratory, and providing a reliable data foundation for subsequent joint prediction and joint early warning.

[0014] Preferably, in the above-mentioned laboratory consumables management method based on multi-dimensional dynamic programming, the calculation method for the corrected predicted demand is as follows: ; in In order to consumables A collection of consumables that are related. For the association strength weight, For related consumables The predicted value, For its historical average, This represents the daily consumption demand.

[0015] The above technical solution can transmit information on changes in demand for related consumables to the current consumables, thereby avoiding simultaneous stockouts caused by changes in demand for supporting consumables and improving the accuracy and robustness of forecasts.

[0016] Preferably, in the above-mentioned laboratory consumables management method based on multi-dimensional dynamic programming, the formula for calculating the safety stock is: ; in For the service level coefficient, To allow for lead time in procurement, Corrected redundancy introduced for associated consumables.

[0017] The above technical solution enables the safety stock to not only reflect the random fluctuations in its own consumption, but also the cascading consumption impact that fluctuations in related consumables may bring. This provides an additional buffer stock when multiple related consumables face the risk of shortage at the same time, effectively reducing the risk of experimental interruption due to the lack of supporting materials.

[0018] Preferably, in the above-mentioned laboratory consumables management method based on multi-dimensional dynamic programming, the expression of the optimization model is: ; in, , , These are the unit cost of stockouts, the unit inventory holding cost rate, and the unit cost of obsolescence / disposal. In order to purchase volume The expected shortage volume In order to purchase volume The probability of being scrapped due to expiration date issues. This refers to the procurement cycle.

[0019] The above technical solution breaks through the limitations of traditional economic order quantity which only considers holding and ordering costs. It can automatically balance the trade-off between "buying more to prevent stockouts" and "buying less to prevent expiration", making the purchase quantity economically optimal each time and significantly reducing the overall operating cost of laboratory consumables management.

[0020] Preferably, in the above-mentioned laboratory consumables management method based on multi-dimensional dynamic programming, the shelf-life risk coefficient for each batch is calculated: ; in, To revise the forecasted demand, This represents the current inventory quantity of this batch of consumables. The expiration date of this batch of consumables. This is the current system date.

[0021] The above technical solution can quickly and intuitively determine whether a batch can be used up within its expiration date based on the current predicted consumption rate. When the risk coefficient exceeds the threshold, a timely warning is issued, providing a clear quantitative basis for priority outbound delivery and internal allocation, and effectively reducing the loss of expired reagents.

[0022] Preferably, in the above-mentioned laboratory consumables management method based on multi-dimensional dynamic programming, the formula for calculating the corrected redundancy introduced by associated consumables is as follows: ; in, To indicate consumables Inventory uncertainty affects consumables The extent of the impact of safety stock For consumables for which safety stock is currently being calculated, In order to consumables A positive correlation exists between consumables that are frequently consumed simultaneously. Belonging to One of the related consumables, This serves as a safety stock for related consumables.

[0023] The above technical solution addresses the uncertainty of associated consumables (reflected by their safety stock) by adjusting the degree of association. By passing this information to the current consumables, a "risk contagion" model of safety stock is achieved, making inventory decisions more refined and avoiding the problem of ignoring coupled risks when calculating the safety stock of each consumable separately.

[0024] Preferably, in the above-mentioned laboratory consumables management method based on multi-dimensional dynamic programming, the expected stockout quantity is... The prediction error distribution is calculated based on the prediction error distribution function, which is obtained by kernel density estimation of historical prediction residuals.

[0025] The above technical solution eliminates the need to assume that the error follows a specific parameter distribution (such as a normal distribution), and can more realistically fit any shape of the actual prediction error, thereby more accurately estimating the probability of stockout and the expected stockout amount, and improving the reliability and practicality of the optimization model.

[0026] On the other hand, this invention discloses a laboratory consumables management system based on multi-dimensional dynamic programming, used to execute the above method, including: a data acquisition and association mining module, used to acquire consumables inbound and outbound information in real time and update real-time inventory, while periodically mining association rules between consumables to generate and update a consumables association network; a multi-level prediction engine, including: an individual demand prediction unit, used to calculate the basic demand for each consumable using the HoltWinters model; an association correction unit, used to correct the current demand for consumables based on the prediction results of other consumables in the consumables association network, and output the corrected predicted demand; and a dynamic inventory optimizer, which includes: An optimization model aimed at minimizing total cost is used to calculate the dynamic safety stock, replenishment point, and optimal purchase quantity for each consumable based on revised forecast demand and historical fluctuation data. An expiration risk manager calculates the risk coefficient for each batch of consumables and generates joint use warnings by combining the consumable association network. An adaptive feedback regulator updates the smoothing parameters of the HoltWinters model and the cost coefficients in the optimization model online based on the deviation between actual consumption data and forecast values. A decision execution module triggers orders to be placed at the optimal purchase quantity when real-time inventory falls below the replenishment point, and outputs priority outbound batches and joint use suggestions during requisition applications.

[0027] By organically combining the above technical solutions with functional modules, a complete automated closed loop from data collection, correlation mining, prediction, optimization to decision execution is realized. This enables real-time response to inventory changes and demand fluctuations, reduces manual intervention, and improves the intelligence level and response speed of laboratory consumables management.

[0028] Preferably, in the above-mentioned laboratory consumables management system based on multi-dimensional dynamic programming, the adaptive feedback regulator uses an exponentially weighted moving average method to update the smoothing parameter in real time, and its update formula is: ; in, This represents the current prediction error. For learning rate, The updated smoothing parameter value; The smoothing parameter values ​​before the update. For predicted values For smoothing parameters The partial derivatives of represent the sensitivity of parameter changes to the predicted output.

[0029] The above technical solutions enable the prediction model to automatically adapt to slow changes in laboratory consumption habits (such as adjustments to experimental projects and changes in seasonal fluctuation patterns) without the need for manual parameter readjustment, maintaining high prediction accuracy at all times. At the same time, they avoid instability caused by parameter mutations, enhancing the robustness and adaptability of the system.

[0030] As can be seen from the above technical solution, compared with the prior art, this invention discloses a laboratory consumables management method and system based on multi-dimensional dynamic programming. By constructing a consumables association network and adopting an improved association rule mining algorithm, it achieves accurate identification of the actual consumption coupling relationship between laboratory consumables. On this basis, it uses the HoltWinters exponential smoothing model and association correction mechanism for joint demand forecasting, combined with dynamic safety stock (introducing the standard deviation of demand fluctuation and association redundancy) and an optimal procurement model with the goal of minimizing total cost (simultaneously optimizing the costs of stockouts, holding, and expired disposal), significantly reducing the risks of stockouts and backlogs. At the same time, through the calculation of expiration risk coefficients, joint early warning, and multi-batch priority outbound strategies, it effectively reduces reagent expiration and disposal losses. In addition, the adaptive feedback regulator uses the exponential weighted moving average method to update model parameters in real time, enabling the system to automatically adapt to changes in consumption habits. Overall, this invention realizes a fully intelligent closed-loop management system from data collection, association mining, forecasting, optimization to decision execution, significantly improving the inventory turnover rate, capital utilization efficiency, and operational economy of laboratory consumables. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0032] Figure 1 The overall method flowchart provided by the present invention; Figure 2 Flowchart of the improved association rule mining algorithm provided by this invention; Figure 3 This is a system block diagram provided for the present invention. Detailed Implementation

[0033] 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.

[0034] This invention discloses a laboratory consumables management method based on multi-dimensional dynamic programming, such as... Figure 1 As shown, it includes: S1 collects historical consumption data for each type of consumable, establishes a dynamic profile of the consumables, and identifies the consumption relationships between different consumables through an improved association rule mining algorithm, forming a consumable association network. S2 uses the HoltWinters exponential smoothing model to predict the daily consumption demand of each consumable within a preset time period based on the historical consumption data and the consumable association network, and corrects the predicted value of the current consumable based on the prediction results of the associated consumables to obtain the corrected predicted demand. S3 dynamically calculates safety stock and replenishment points based on the standard deviation of the corrected forecast demand and historical daily consumption. S4 establishes an optimization model with the goal of minimizing total cost, and calculates the recommended purchase quantity for each consumable. The total cost includes stockout cost, inventory holding cost, and expired / scrap cost. S5 calculates the expiration date risk coefficient for each batch of the same consumable and uses the consumable association network to automatically push a joint use suggestion to the administrator of the associated consumables when the risk coefficient of a certain batch of a certain consumable exceeds the threshold. S6 When a requisition request is received, the system recommends batches for shipment in descending order of the expiration date risk coefficient, and issues an early warning when the risk coefficient exceeds a preset threshold; S7 The system compares the actual consumption data with the predicted value each time, and dynamically adjusts the smoothing parameters in the HoltWinters model and the cost coefficient in the optimization model using the exponentially weighted moving average method.

[0035] To further optimize the above technical solutions, such as Figure 2 As shown, the improved association rule mining algorithm identifies the consumption relationships between different consumables through the following specific steps: S101 Collect historical consumable requisition records, construct a transaction database with each experimental operation as the transaction boundary, and each transaction includes a transaction timestamp, the identifier of the consumables involved, and their consumption quantity; S102 Assign a time decay weight to each transaction, the time decay weight decreasing exponentially as the interval between the transaction occurrence time and the current time increases; S103 For each type of consumable in each transaction, discretize it into quantity levels according to the ratio of its consumption quantity to the historical daily average consumption quantity of that consumable; S104 Use weighted support counting and weighted confidence calculation methods to analyze the data from the transaction... Frequent itemsets and association rules are mined from the database, wherein the weighted support count and weighted confidence calculation are both multiplied by the time decay weight; S105 For the mined association rules, a quantity level correction factor is introduced to enhance or decay the confidence, wherein the quantity level correction factor is: the ratio of the average quantity level of the consequent consumable under the condition that the antecedent of the rule appears to the average quantity level of the consumable in all transactions; S106 Rules with confidence exceeding a preset threshold in the association rules after quantity level correction are stored as positive associations, forming a consumable association network.

[0036] Specifically, the method described in S101, which uses one experimental operation as the boundary of a transaction, is as follows: when the system can obtain the start and end times of the experiment, the set of all consumables used in one experiment is taken as a transaction; when the start and end times of the experiment cannot be obtained, a density clustering algorithm is used to cluster the timestamps of the usage records, and the set of consumables continuously used by the same experimenter within a preset time window is taken as a transaction.

[0037] Furthermore, the time decay weight adopts an exponential decay function. ,in For the current date, The date the transaction occurred. This is the preset attenuation rate parameter.

[0038] Furthermore, the discretization method for the quantity level described in S103 is as follows: calculate the consumption amount of consumable i in transaction T. Compared with the historical average daily consumption of this consumable ratio ;like Then the level is 1 (low cost), if Then the level is 2 (medium consumption), if The level is 3 (high consumption); unused consumables are level 0.

[0039] Furthermore, the formula for calculating the weighted support count described in S104 is as follows: in The time decay weight for transaction T. For a transactional database; the formula for calculating the weighted confidence level is: When mining frequent itemsets, a dynamic support threshold is used, with the minimum support threshold for each consumable i. Set as 0.1), where occ The weighted number of occurrences of consumable i. For the weighted total number of transactions, This is the preset minimum absolute support.

[0040] Furthermore, the quantity order correction factor described in S104 The calculation formula is: in Let Y be the quantity level of consumable Y in transaction T (1, 2, 3). The average rank of Y when the antecedent X occurs. Y represents the average rank of Y across all transactions; final confidence level. .

[0041] Furthermore, it also includes the step of mining negative correlation rules: calculating the negative correlation coefficient between any two consumables i and j. : in This is a binary flag indicating whether consumable i appears in transaction T. Its frequency of occurrence; when When the support of both consumables i and j is greater than the preset threshold, consumable i is marked as having a negative association relationship with j and stored in the association network as alternative information.

[0042] To further optimize the above technical solutions, the daily consumption of laboratory consumables often exhibits the following characteristics: Trend: As the size of the laboratory expands or the number of experimental projects increases or decreases, the consumption may show a slow growth or decline trend.

[0043] Seasonality: Consumption may vary regularly on different workdays within a week (e.g., Monday and Friday); consumption increases significantly in certain months (e.g., final exam season, project completion month).

[0044] Random fluctuations: In addition to trends and seasonality, there are still unpredictable random components.

[0045] This invention employs an additive model (suitable for scenarios where seasonal fluctuations do not increase with horizontal growth), for consumables Its daily consumption sequence is denoted as ,in This represents the current date (in days). The model contains three smoothing equations and one prediction equation.

[0046] Smoothing equation: Level: ; Trend: ; Seasonal: ; in: : No. The horizontal value after smoothing (the baseline value after removing the season); : No. Trend slope over the day (unit: quantity / day / day); : No. Seasonal component of a day (additive form, representing the offset of that day relative to a baseline value); Seasonal cycle length. For consumption patterns with a weekly cycle, ; Smoothing parameter, the range of values ​​is... These control the update speed of the level, trend, and seasonal terms, respectively; the closer to 1, the more sensitive the model is to recent changes.

[0047] Prediction equation: based on the first The smoothed value of the day predicts the future number of days. step( ,in To predict the step size, the consumption of 7 is usually taken: To facilitate daily replenishment decisions, the system typically outputs daily forecasts for the next 7 days and calculates their arithmetic mean as the average daily forecast demand. .

[0048] For consumables Find all sets of consumables that are related to it from the association network. For each associated consumable The system also calculates its predicted demand. The revised demand is: ; in, This refers to the association strength (i.e., the rule confidence or the normalized weight). For consumables The average daily actual consumption over the past 30 days. The significance of this correction is: when related consumables... When projected demand is higher than its historical average, it indicates that there may be more [demand / demand] in the future. and The experiment was conducted simultaneously, therefore the adjustment was increased. The forecast demand is adjusted accordingly; otherwise, it is lowered.

[0049] Further fine-tuning: If the directionality of the association rule is clear (e.g., A→B means that the use of A is usually accompanied by B, but B is not necessarily accompanied by A), then the correction is only performed along the direction of the rule. The system stores the directions of directed edges in the association network.

[0050] Example: Consumable "Taq enzyme" ( The individual predicted daily demand is 50 units, and its associated consumable "primers" ( The predicted daily demand is 120 units, while the historical average primer demand is 80 units. (Association strength) =0.85. Therefore, the correction value is... The system then prepares higher inventory levels to avoid simultaneous stockouts.

[0051] Traditional methods use fixed safety stock, while this invention combines safety stock with demand fluctuations and related redundancy correction.

[0052] First, calculate the consumables. Historical daily consumption standard deviation (Using data from the past 90 days). Procurement lead time (Days) is calculated from supplier data or historical procurement records. Service Level Coefficient Based on the criticality of consumables: critical consumables (such as those essential for the experiment and with no substitutes) are selected. =1.96 (corresponding to a 97.5% probability of no stockout); ordinary consumables take =1.28 (corresponding to a 90% probability).

[0053] The basic safety stock is: ; Considering the additional uncertainties introduced by correlation correction, define the correction redundancy. ;in The correlation factor can be taken as the correlation strength. Half of (or calibrated by historical joint out-of-stock rate). For related consumables Safety stock. Final safety stock: ; Replenishment point ,in The average daily demand at the current stage can be calculated as a weighted average of the average daily consumption over the past 30 days and the revised forecast.

[0054] Example: A certain consumable =10 units / day =3 days =1.96→ If the sum of the associated consumable safety stock is 50 units, and λ=0.3, then =15 units, =48.96 units. If =40 units / day, then =40*3+49=169 units.

[0055] This invention overcomes the shortcomings of traditional Economic Order Quantity (EOQ) which only considers holding costs and ordering costs, and specifically introduces stockout costs and expired disposal costs for laboratory consumables, establishing the following optimization model: Decision variable: Purchase quantity Q (recommended to be an integer multiple of the minimum order quantity) Objective function: ; The meanings and calculation methods for each item are as follows: Out-of-stock cost item : Unit stockout cost. This refers to the loss incurred for each unit of consumable that is missing, including experimental interruptions, emergency procurement premiums, and wasted manpower. It is usually set at 2 to 5 times the unit price of the consumable (depending on the criticality of the experiment).

[0056] Given a purchase quantity The expected stockout quantity under the existing inventory strategy (units consistent with consumables). Stockouts occur during the procurement lead time. When actual demand exceeds available inventory at the time.

[0057] calculate Method: Assume the replenishment trigger point is the reorder point. Demand during the procurement lead time Follows probability distribution When inventory levels fall to Purchase orders are issued at this time, and the inventory is [missing information]. In the lead time If demand The out-of-stock quantity is However, due to the quantity of this purchase... This will affect the inventory level after replenishment, and stockouts usually occur before orders arrive, therefore the expression for the expected stockout quantity is: in ,and This is for safety stock. Note that here... In fact with There is no direct relationship, because The peak inventory level after replenishment is determined, but stockouts only occur during the lead time, and... Size is irrelevant (premise) (Enough to cover one cycle).

[0058] In order to reflect Indirect impacts of stockouts (e.g.) (Excessive shortages may lead to increased purchasing frequency, thereby increasing the chance of stockouts). This invention employs a more accurate model: expressing the expected stockout quantity as the average stockout quantity within a period, which is compared with... The relationship is indirectly reflected through service levels. For simplicity and ease of engineering implementation, the system adopts the following approximation: in , The standard normal loss function (i.e. ), , The formula means: when the purchase quantity is At that time, it is equivalent to the inventory being at the beginning of the replenishment cycle. The portion of demand exceeding the inventory during the lead time period constitutes the expected stockout. This process will... It is directly linked to the risk of stockouts.

[0059] Nonparametric estimation of prediction error distribution: The prediction error distribution is obtained using kernel density estimation; therefore, this invention does not use the normality assumption, but instead relies on historical prediction residuals. The probability density function is obtained through kernel density estimation. Then, the distribution of total demand during the lead time is obtained through convolution, and the expected construction volume is calculated by numerical integration. The distribution is updated once a week.

[0060] Inventory holding costs : Daily inventory holding cost per unit of consumables. Typically converted to daily cost using an annual holding cost rate (e.g., 25%). unit price Annual cost of ownership .

[0061] Average inventory level. Assuming consumables are consumed at a constant rate during the procurement cycle, inventory levels decrease from... The linear decrease to 0 results in an average inventory of [value missing]. .

[0062] Procurement cycle (days). ,in This represents the average daily demand. Therefore, the cost of holding items can be expanded as follows: This form indicates that holding costs and This is proportional to the size of the purchase and encourages smaller purchase batches.

[0063] Expired and scrapped cost items : The loss from expired and scrapped consumables is usually equal to the unit price of the consumables (if they cannot be returned or reused).

[0064] Purchase quantity is At that time, the probability that the batch cannot be completely consumed within the expiration date and will be scrapped.

[0065] Calculation method: Assume the remaining shelf life of this batch (calculated from the planned arrival date) is... The daily average consumption forecast is [value]. The standard deviation of daily consumption is The total consumption over the entire validity period. Follows the distribution: The condition for this batch to expire and be scrapped is: the actual consumption is less than the purchase quantity. .therefore: in This is the standard normal cumulative distribution function. If a nonparametric method is used, an empirical distribution is used instead. Note: If... ,but A value less than 0.5 indicates that the product can be used up within its expiration date at the average consumption rate; conversely, a value greater than 0.5 indicates a significantly increased risk of expiration.

[0066] The solution method involves using the golden section search or enumeration method to find the minimum cost Q within the feasible region, denoted as the optimal purchase quantity Q*, since the objective function is a non-convex discrete function.

[0067] Example: A reagent costs 100 yuan. =200 yuan, =0.0685 yuan / day (25% per year) =100 yuan. =10 units / day =3, =5 days =30 days. Calculated. The total cost for 200 units is approximately 120 yuan. The total cost is approximately 145 yuan when producing 300 units. When the cost is 250 units, the minimum cost is 110 yuan, so Q* = 250 units.

[0068] To further optimize the above technical solution, the system assesses the risk factor of each batch of consumables (with a unique batch number and expiration date) in real time: ; in Take the moving average of the actual consumption and the revised forecast over the most recent 7 days; The remaining valid days; This represents the current inventory level for this batch.

[0069] Risk Factor Interpretation: Low risk; can be used up within the expiration date with normal consumption.

[0070] Medium risk; priority should be given to outbound shipments.

[0071] High risk. The system issues a red alert and performs the following actions: This batch will be prioritized when recommending shipments.

[0072] The administrator will be notified via push notification, suggesting that promotions or internal adjustments be made.

[0073] Triggering a correlation alert: Searching the correlation network for other consumables with a confidence level greater than 0.6 related to this consumable, and generating a "Combined Use Recommendation": "Batch A (XX days remaining) is about to expire, it is recommended to use it in combination with batch B (corresponding consumable) for the experiment." At the same time, it automatically calculates whether the inventory of consumable B required for the combined experiment is sufficient, and if not, reminds the user to replenish the stock.

[0074] In another embodiment, Using a fixed daily average consumption, random fluctuations in consumption are ignored.

[0075] To overcome the limitations of point estimation and to quantify the uncertainty of consumption, a forecast interval is used instead of point estimation to calculate the probability that the inventory cannot be consumed within its validity period.

[0076] This also includes dynamic assessment and management of the shelf-life risk of different batches of the same consumable. The specific steps are as follows: For each batch, based on its remaining expiration date Current inventory Predicted distribution of daily consumption (mean) and standard deviation ), calculate the probability that this batch cannot be completely consumed within its expiration date. ; Obtain the real-time inventory of other consumables that are related to this consumable, and calculate the effective daily consumption rate for correlation correction. And recalculate the corrected expiration probability. ; when Exceeding the dynamic threshold associated with the consumable When this happens, an alert is triggered; When a requisition request is made, the order of issuance is determined according to a multi-batch priority model, where the priority model includes batches. priority and Positive correlation.

[0077] Specifically, the remaining validity period is defined as: (days); Current inventory of this batch ;future Total consumption within a day Follows the distribution: ,in: ; ( If the standard deviation of daily consumption is taken as the standard deviation of daily consumption, and the residuals are derived from historical data or predictive models, then the probability of this batch expiring is: ; in It is the standard normal cumulative distribution function.

[0078] Risk level classification: Low risk (over 90% chance of running out of money); Medium risk; High risk (significant probability of expiration); Furthermore, in the laboratory, the consumption of certain consumables often occurs simultaneously with related consumables (such as reagent kits and specific pipette tips). When the inventory of related consumables is insufficient, the actual consumption rate of the current consumable will decrease, thereby increasing the risk of expiration. Let the consumables... The associated consumables set is Define the effective daily consumption rate correction factor: in For related consumables Inventory adequacy function: When related consumables Inventory Below the threshold When the factor is less than 1, the actual consumption rate will decrease. Substituting into the risk probability formula, we obtain the associated correction expiration risk: This correction can provide early warnings: even if the current inventory of consumables is not high, if the necessary supporting consumables are about to run out of stock, it will cause the current consumables to be consumed more slowly and face the risk of expiration.

[0079] Furthermore, when multiple batches of the same consumable exist, each batch has a different expiration date and inventory level. Traditional FIFO (First-In, First-Out) simply sorts them by remaining time, without considering minimizing overall obsolescence.

[0080] This invention proposes a multi-batch dynamic allocation model.

[0081] Objective: To predict a given future demand sequence In this case, we decide which batch to release from at each time point to minimize the total number of expired items. Let the batch size be... The initial inventory is The expiry date is .

[0082] Define indicator variables For the first Tiancong batch The quantity shipped out satisfies: (Meeting daily needs); (Cumulative outbound shipments shall not exceed inventory); for , (Cannot be shipped after expiration); This problem is a linear programming problem, but solving it in real time is complex.

[0083] This invention employs a heuristic greedy algorithm: daily outbound batches are selected according to the following priority: The denominator is the remaining number of days available multiplied by the standard deviation of the daily consumption of the batch (batches with large fluctuations are given priority for release because of their high uncertainty). The associated penalty term is used to avoid prematurely consuming batches that are strongly associated with other consumables (reserved to match future experiments).

[0084] This model allows the system to dynamically adjust the outbound order, rather than using a fixed FEFO (Feed Outbound Order).

[0085] Traditional early warning systems use fixed thresholds (e.g., 80%), but the consumption volatility and value vary among different consumables. This invention employs dynamic thresholds, automatically adjusted based on the consumable's historical expiration losses and stockout costs.

[0086] Define consumables Optimal risk warning threshold By minimizing the expected total cost, we obtain: Using historical data or simulation optimization, the system updates the personalized threshold for each consumable weekly.

[0087] To enable the model to evolve automatically as experimental habits change, an online feedback adjustment mechanism was designed for the system.

[0088] This invention discloses a laboratory consumables management system based on multi-dimensional dynamic programming, such as... Figure 3 As shown, it includes: Data Acquisition and Association Mining Module: This module collects real-time inbound and outbound records for each type of consumable through intelligent shelves (including gravity sensors, RFID readers, and QR code scanners), including consumable ID, batch number, quantity, operation time, operator, and associated experimental project. This module periodically (e.g., monthly) runs an association rule mining algorithm to discover consumption relationships between consumables and generate a consumable association network.

[0089] The multi-level forecasting engine consists of an individual demand forecasting unit and a correlation correction unit. The individual demand forecasting unit independently uses the HoltWinters exponential smoothing model to forecast the demand for the next 7 days for each consumable. The correlation correction unit corrects the forecast result of the current consumable based on the forecast deviation or forecast value of other consumables in the correlation network, and outputs the corrected forecast demand.

[0090] Dynamic Inventory Optimizer: Built-in mathematical optimization model with the goal of minimizing total cost (out-of-stock cost + inventory holding cost + expired and scrap cost). Inputs include corrected forecast demand, historical consumption standard deviation, procurement lead time, service level coefficient, etc., and outputs dynamic safety stock, replenishment point and optimal purchase quantity for each consumable.

[0091] Expiry Date Risk Manager: Calculates the expiry date risk coefficient for each batch and generates a joint use alert based on the associated network. When the risk coefficient of a consumable batch exceeds a threshold (e.g., 80%), it automatically identifies other consumables strongly associated with that batch and sends a "Recommended Joint Use" notification to the administrator.

[0092] Adaptive feedback regulator: Compares actual consumption data with predicted values ​​each time to calculate prediction error. It uses exponentially weighted moving average or gradient descent to update the smoothing parameters (α, β, γ) in the HoltWinters model online and optimizes cost coefficients (shortage unit cost, holding cost rate, etc.) in the model.

[0093] The decision execution module includes a replenishment trigger (which periodically checks whether the real-time inventory is below the replenishment point; if so, it generates a purchase order with the purchase quantity given as the optimal batch size) and an outbound recommender (which sorts inventory by expiration risk coefficient from high to low and displays related usage suggestions on the requisition interface).

[0094] To further optimize the above technical solution, the adaptive feedback regulator uses an exponentially weighted moving average method to update the smoothing parameters in real time. The update formula is as follows: ;in, This represents the current prediction error. For learning rate, The updated smoothing parameter value; The smoothing parameter values ​​before the update. For predicted values For smoothing parameters The partial derivatives of represent the sensitivity of parameter changes to the predicted output.

[0095] The system provides an optional interface that allows importing laboratory project schedules for a future period (e.g., one week), including experiment names, required consumables and quantities, and planned dates. A multi-level prediction engine uses this information as a priori constraint: if an experiment plan explicitly requires a certain consumable, the predicted value is directly the larger of the planned quantity and the statistical prediction; if an experiment plan is cancelled, the system automatically marks the corresponding ordered purchase requests as cancelable (if the goods have not yet been shipped) and adjusts the replenishment plan for the associated consumables.

[0096] Example: The experimental plan shows that an experiment requiring 50 special centrifuge tubes will be conducted next Tuesday, while the statistical forecast is only 30. The system will use 50 as the forecast value. If the experiment is temporarily canceled, the system will send a "suggest order cancellation" notification to the purchasing agent and recalculate the safety stock and replenishment point.

[0097] This system can be deployed in a cloud-based SaaS model or a local server model. The data acquisition layer utilizes intelligent consumable cabinets (each compartment equipped with a weight sensor and RFID antenna) or ordinary shelves combined with handheld barcode scanners. The server-side is developed using Python, with a PostgreSQL database and NumPy, SciPy, and Scikit-learn algorithm libraries. The front-end provides a web interface and a WeChat mini-program. The system automatically performs forecasting and replenishment point calculations every morning at midnight, processes inbound and outbound events in real time, and asynchronously runs association rule mining (once a week).

[0098] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0099] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A laboratory consumables management method based on multi-dimensional dynamic programming, characterized in that, include: Historical consumption data for each type of consumable is collected to create a dynamic profile of the consumables. An improved association rule mining algorithm is used to identify the consumption relationships between different consumables, forming a consumable association network. Based on the historical consumption data and the consumable association network, the HoltWinters exponential smoothing model is used to predict the daily consumption demand of each consumable in the future within a preset time period. The predicted value of the current consumable is then corrected based on the prediction results of the associated consumables to obtain the corrected predicted demand. Based on the standard deviation of the revised forecast demand and historical daily consumption, safety stock and replenishment point are dynamically calculated. Establish an optimization model with the goal of minimizing total cost, and calculate the recommended purchase quantity for each consumable. The total cost includes stockout cost, inventory holding cost, and expired / scrap cost. For different batches of the same consumable, calculate the expiration date risk coefficient for each batch, and use the consumable association network to automatically push a joint use suggestion to the administrator of the associated consumables when the risk coefficient of a certain batch of a certain consumable exceeds the threshold. When a requisition request is received, the system recommends batches for release in descending order of the expiration date risk coefficient, and issues an alert when the risk coefficient exceeds a preset threshold. The actual consumption data is compared with the predicted value each time, and the smoothing parameters in the HoltWinters model and the cost coefficients in the optimization model are dynamically adjusted using the exponentially weighted moving average method.

2. The laboratory consumables management method based on multi-dimensional dynamic programming according to claim 1, characterized in that, The improved association rule mining algorithm identifies the consumption relationships between different consumables through the following specific steps: Historical consumable requisition records are collected, and a transaction database is constructed using a single experimental operation as the transaction boundary. Each transaction includes a timestamp of the transaction occurrence, the identifier of the consumables involved, and their consumption quantity. A time decay weight is assigned to each transaction, which decreases exponentially as the interval between the transaction occurrence time and the current time increases. For each type of consumable in each transaction, its consumption quantity is discretized into quantity levels based on the ratio of its consumption quantity to the historical daily average consumption quantity of that consumable. Frequent itemsets and association rules are mined from the transaction database using weighted support counting and weighted confidence calculation methods, where both weighted support counting and weighted confidence calculation use the time decay weight as a multiplier. For the mined association rules, a quantity level correction factor is introduced to enhance or weaken the confidence. The quantity level correction factor is the ratio of the average quantity level of the consumable in the consequent, given the occurrence of the antecedent, to the average quantity level of that consumable in all transactions. Rules with confidence levels exceeding a preset threshold in the association rules after quantity level correction are stored as positive associations, forming a consumable association network.

3. The laboratory consumables management method based on multi-dimensional dynamic programming according to claim 1, characterized in that, The method for calculating the revised forecast demand is as follows: ； in In order to consumables A collection of consumables that are related. For the association strength weight, For related consumables The predicted value, For its historical average, This represents the daily consumption demand.

4. The laboratory consumables management method based on multi-dimensional dynamic programming according to claim 1, characterized in that, The formula for calculating the safety stock is: ； in For the service level coefficient, To allow for lead time in procurement, Corrected redundancy introduced for associated consumables.

5. The laboratory consumables management method based on multi-dimensional dynamic programming according to claim 1, characterized in that, The expression for the optimization model is: ； in, , , These are the unit cost of stockouts, the unit inventory holding cost rate, and the unit cost of obsolescence / disposal. In order to purchase volume The expected shortage volume In order to purchase volume The probability of being scrapped due to expiration date issues. This refers to the procurement cycle.

6. The laboratory consumables management method based on multi-dimensional dynamic programming according to claim 1, characterized in that, Calculate the shelf-life risk factor for each batch: ； in, To revise the forecasted demand, This represents the current inventory quantity of this batch of consumables. The expiration date of this batch of consumables. This is the current system date.

7. The laboratory consumables management method based on multi-dimensional dynamic programming according to claim 4, characterized in that, The formula for calculating the corrected redundancy introduced by associated consumables is as follows: ； in, To indicate consumables Inventory uncertainty affects consumables The extent of the impact of safety stock For consumables for which safety stock is currently being calculated, In order to consumables A positive correlation exists between consumables that are frequently consumed simultaneously. Belonging to One of the related consumables, This serves as a safety stock for related consumables.

8. The laboratory consumables management method based on multi-dimensional dynamic programming according to claim 5, characterized in that, Expected shortage The prediction error distribution is calculated based on the prediction error distribution function, which is obtained by kernel density estimation of historical prediction residuals.

9. A laboratory consumables management system based on multi-dimensional dynamic programming, used to execute the method according to any one of claims 1 to 8, characterized in that, include: The data acquisition and association mining module is used to acquire consumables inbound and outbound information in real time and update real-time inventory. At the same time, it periodically mines the association rules between consumables to generate and update the consumables association network. The multi-level forecasting engine includes: an individual demand forecasting unit, used to calculate the basic demand for each consumable using the HoltWinters model; an association correction unit, used to correct the current demand for consumables based on the forecast results of other consumables in the consumable association network, and output the corrected forecast demand; a dynamic inventory optimizer, which has a built-in optimization model with the goal of minimizing total cost, and calculates the dynamic safety stock, replenishment point, and optimal purchase batch for each consumable based on the corrected forecast demand and historical fluctuation data; an expiration date risk manager, used to calculate the risk coefficient for each batch of consumables and generate a joint use warning in combination with the consumable association network; an adaptive feedback regulator, used to update the smoothing parameters of the HoltWinters model and the cost coefficient in the optimization model online based on the deviation between actual consumption data and forecast values; and a decision execution module, used to trigger an order to be placed at the optimal purchase batch when the real-time inventory is lower than the replenishment point, and to output priority outbound batches and joint use suggestions when a requisition application is submitted.

10. A laboratory consumables management system based on multi-dimensional dynamic programming according to claim 9, characterized in that, The adaptive feedback regulator uses an exponentially weighted moving average method to update the smoothing parameters in real time. The update formula is as follows: ;in, This represents the current prediction error. For learning rate, The updated smoothing parameter value; The smoothing parameter values ​​before the update. For predicted values For smoothing parameters The partial derivatives of represent the sensitivity of parameter changes to the predicted output.

Images