Drug sales inventory management system based on artificial intelligence

Abstract

The invention relates to the technical field of medicine sales inventory management, and discloses a medicine sales inventory management system based on artificial intelligence, which comprises a multi-modal anomaly detection module, a multi-modal anomaly detection framework is constructed, and medicine sales anomaly is detected by using an isolated forest algorithm, a depth auto-encoder and a sequence anomaly detection algorithm; the causal structure learning and attribution analysis module is used for constructing a drug sales causal structure learning and attribution analysis system and analyzing abnormal reasons by adopting a NOTEARS algorithm and an SHAP attribution algorithm; the intervention strategy generation and evaluation module is used for constructing an intervention strategy generation and evaluation system and simulating different strategy effects through a causal intervention formula; and the inventory strategy optimization and execution module is used for implementing inventory strategy optimization and execution and dynamically adjusting inventory management. According to the invention, automatic detection, cause attribution and intelligent correction of sales abnormity are realized, and a closed-loop management system is formed.

Description

Technical Field

[0001] The present invention relates to the field of drug sales inventory management, and more specifically, to a drug sales inventory management system based on artificial intelligence. Background Art

[0002] Technical issues include the lack of automatic attribution capabilities for anomaly detection in traditional pharmaceutical sales inventory management systems, the lack of deep-level intervention capabilities in anomaly attribution systems, and the difficulty for traditional monitoring systems to cope with complex and changeable abnormal sales patterns due to their reliance on manually preset rules.

[0003] Therefore, an artificial intelligence-based drug sales inventory management system is proposed. By constructing a structured causal model and a deep learning framework, the automatic detection, cause attribution and intelligent correction of drug sales anomalies are realized, forming a closed-loop system for anomaly management. Summary of the invention

[0004] The present invention provides a medicine sales inventory management system based on artificial intelligence to solve the technical problems in related technologies.

[0005] The present invention proposes an artificial intelligence-based drug sales inventory management system, which performs the following steps:

[0006] Step 1: Build a multimodal anomaly detection framework, use the isolation forest algorithm, deep autoencoder and sequence anomaly detection algorithm to detect drug sales anomalies and generate a comprehensive anomaly score;

[0007] Step 2: Build a drug sales causal structure learning and attribution analysis system, use the NOTEARS algorithm to learn the causal relationship between variables to form a causal structure diagram, and use the SHAP attribution algorithm to calculate the contribution of each factor to the anomaly;

[0008] Step 3: Construct an intervention strategy generation and evaluation system, apply the causal intervention formula to simulate the impact of different intervention strategies on the sales system, and generate and evaluate intervention strategy combinations;

[0009] Step 4: Implement inventory strategy optimization and execution, generate inventory adjustment plans based on intervention strategy evaluation results, monitor execution results and dynamically adjust intervention strategies;

[0010] The multimodal anomaly detection framework generates a comprehensive anomaly score by weighted fusion of the results of multiple anomaly detection algorithms.

[0011] Furthermore, the step of constructing a multimodal anomaly detection framework includes:

[0012] Build a detection framework that integrates multiple anomaly detection algorithms, including a data preprocessing module, an algorithm integration module, and a result fusion module;

[0013] Train anomaly pattern classifier based on historical anomaly case data and expert knowledge base;

[0014] Build a multi-level risk early warning system to map anomaly scores to predefined risk levels and generate corresponding early warning signals and notifications.

[0015] Furthermore, the abnormal pattern classifier adopts a gradient boosting decision tree algorithm, including a feature extraction layer, a tree model training layer and a model integration layer, the input is a sales data feature vector, and the output is an abnormal pattern category and its probability distribution.

[0016] Furthermore, in the step of constructing the drug sales causal structure learning and attribution analysis system, the NOTEARS algorithm learns the causal relationship between variables by solving the optimization problem:

[0017] ;

[0018] in represents the weight matrix to be learned, Represents drug sales-related characteristic data, is the loss function, is acyclic constraint.

[0019] Furthermore, the SHAP attribution algorithm calculates feature importance using the following formula:

[0020] ;

[0021] in Representation characteristics Contribution to the anomaly, represents the set of all features, Indicates that it does not contain features A subset of features, Represents the output of the prediction model on the feature set.

[0022] Furthermore, in the step of constructing the intervention strategy generation and evaluation system, the causal intervention formula is:

[0023] ;

[0024] in Indicates that the intervention variable The value is Time target variable The distribution of Represents the intervention variable With the target variable The confounding factors between To make a budget, represents the variable being intervened, It indicates the specific value that a variable takes after being intervened. Represents confounding variables A value of , It affects simultaneously and variables.

[0025] Furthermore, the step of constructing an intervention strategy generation and evaluation system also includes: constructing a multi-objective adaptive intervention strategy generation module, selecting and combining the optimal intervention operation set from the intervention operation library based on a genetic algorithm; establishing an intervention effect evaluation system, comprehensively considering the intervention cost, implementation difficulty and expected effect, and conducting quantitative evaluation of each intervention strategy.

[0026] Furthermore, the steps of implementing inventory strategy optimization and execution include:

[0027] Build an intelligent inventory adjustment plan generator, using reinforcement learning algorithms to generate detailed adjustment plans including price adjustment, promotion strategy, inventory level adjustment, and delivery method optimization;

[0028] Build an inventory strategy execution monitoring system to continuously collect key indicator data during strategy execution;

[0029] Build a strategy feedback optimization module to dynamically adjust and optimize the intervention strategy based on the online learning algorithm.

[0030] Furthermore, the drug sales data includes fields such as product ID, sales quantity, sales amount, sales time, sales store, payment method, etc.

[0031] The inventory data includes fields such as product ID, inventory quantity, inventory cost, inventory location, batch information, and expiration date;

[0032] The external factor data includes weather information, holiday information, competitor promotion activities, supplier delivery status and other data.

[0033] The present invention also proposes a drug sales inventory management system based on artificial intelligence, comprising:

[0034] Multimodal anomaly detection module, which uses the isolation forest algorithm, deep autoencoder and sequence anomaly detection algorithm to detect drug sales anomalies and generate a comprehensive anomaly score;

[0035] The causal structure learning and attribution analysis module is used to learn the causal relationship between variables using the NOTEARS algorithm to form a causal structure diagram, and use the SHAP attribution algorithm to calculate the contribution of each factor to the anomaly;

[0036] The intervention strategy generation and evaluation module is used to apply the causal intervention formula to simulate the impact of different intervention strategies on the sales system, and to generate and evaluate intervention strategy combinations;

[0037] Inventory strategy optimization and execution module, which is used to generate inventory adjustment plans based on intervention strategy evaluation results, monitor execution effects and dynamically adjust intervention strategies;

[0038] The system is used to execute the aforementioned artificial intelligence-based drug sales inventory management system.

[0039] The beneficial effects of the present invention are:

[0040] The present invention integrates multiple anomaly detection algorithms to construct a multimodal anomaly detection framework. Compared with traditional rule-based methods, it can effectively cope with complex and changeable abnormal sales patterns, greatly improve the detection rate of abnormal sales behaviors, reduce the false alarm rate, and discover abnormal sales trends in advance, thus gaining valuable time for management decision-making.

[0041] The present invention realizes automatic attribution of abnormal causes by constructing a causal structure learning model based on the NOTEARS algorithm and a factor attribution algorithm of SHAP, which greatly shortens the time of locating the abnormal causes, improves the attribution accuracy, and enables managers to quickly determine the real causes behind the abnormalities.

[0042] The intervention strategy generation and evaluation system based on causal reasoning of the present invention can generate the optimal intervention strategy combination for different abnormality types and causes, greatly improving the success rate of abnormality correction, shortening the correction time, and enhancing the system's deep-level intervention capability on the abnormality generation mechanism.

[0043] The present invention realizes closed-loop optimization of inventory management through an intelligent inventory adjustment scheme generator and a strategy feedback optimization module, improves inventory turnover, reduces inventory costs and out-of-stock rates, and comprehensively improves the efficiency of drug sales inventory management.

[0044] The present invention realizes the effect evaluation and dynamic adjustment of different intervention strategies through the counterfactual scenario simulation engine and the strategy feedback optimization module, so that the system can continuously learn and optimize, and constantly improve its adaptability to complex and changing market environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a flowchart of module execution in a drug sales inventory management system based on artificial intelligence proposed by the present invention;

[0046] Figure 2 The present invention Figure 1 A flowchart of the sub-steps of step 1;

[0047] Figure 3The present invention Figure 1 A flowchart of the sub-steps of step 2;

[0048] Figure 4 The present invention Figure 1 A flowchart of the sub-steps of step 3;

[0049] Figure 5 The present invention Figure 1 A flowchart of the sub-steps of step 4;

[0050] Figure 6 It is a structural block diagram of an artificial intelligence-based drug sales inventory management system proposed by the present invention.

[0051] In the figure: 101, multimodal anomaly detection module; 102, causal structure learning and attribution analysis module; 103, intervention strategy generation and evaluation module; 104, inventory strategy optimization and execution module. DETAILED DESCRIPTION

[0052] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0053] like Figure 1-Figure 6 As shown in the figure, an artificial intelligence-based drug sales inventory management system includes the following four modules:

[0054] The multimodal anomaly detection module 101 constructs a multimodal anomaly detection framework, uses an isolation forest algorithm, a deep autoencoder, and a sequence anomaly detection algorithm to detect drug sales anomalies and generate a comprehensive anomaly score;

[0055] The causal structure learning and attribution analysis module 102 builds a drug sales causal structure learning and attribution analysis system, uses the NOTEARS algorithm to learn the causal relationship between variables to form a causal structure diagram, and uses the SHAP attribution algorithm to calculate the contribution of each factor to the anomaly;

[0056] An intervention strategy generation and evaluation module 103 constructs an intervention strategy generation and evaluation system, applies a causal intervention formula to simulate the impact of different intervention strategies on the sales system, and generates and evaluates intervention strategy combinations;

[0057] The inventory strategy optimization and execution module 104 implements inventory strategy optimization and execution, generates inventory adjustment plans based on intervention strategy evaluation results, monitors execution effects, and dynamically adjusts intervention strategies.

[0058] The four modules execute steps 1 to 4 respectively, and the specific steps are as follows:

[0059] Step 1: Build a multimodal anomaly detection framework to detect anomalies and provide early warnings for drug sales data. This step includes the following sub-steps:

[0060] 1.1: Construct a detection framework that integrates multiple anomaly detection algorithms. The framework includes a data preprocessing module, an algorithm integration module, and a result fusion module. The data preprocessing module cleans and extracts features from sales data. The algorithm integration module runs multiple detection algorithms in parallel. The result fusion module fuses the results of each algorithm by weighted voting. The framework uses the isolation forest algorithm to analyze the spatial distribution data of drug sales and identify spatial anomalies. A deep autoencoder is used to process sales pattern feature data and detect pattern anomalies. The deep autoencoder consists of an encoder and a decoder. The encoder maps high-dimensional input data to a low-dimensional latent space, and the decoder reconstructs the low-dimensional representation into the original input. Anomalies are identified by comparing reconstruction errors. A sequence anomaly detection algorithm is used to analyze sales time series data and find time series anomalies. The integrated framework weightedly fuses the results of multiple anomaly detection algorithms to generate a comprehensive anomaly score.

[0061] 1.2: Based on historical abnormal case data and expert knowledge base, train an abnormal pattern classifier. The classifier uses a gradient boosting decision tree algorithm and includes a feature extraction layer, a tree model training layer, and a model integration layer. The feature extraction layer extracts time series features, spatial features, and product features from sales data. The tree model training layer builds multiple decision tree models. The model integration layer generates the final prediction by weighted combination of the prediction results of multiple tree models. The classifier inputs the sales data feature vector and outputs the abnormal pattern category and its probability distribution, realizing automatic discovery of unknown abnormal patterns and real-time monitoring of known abnormal patterns.

[0062] 1.3: According to the severity and urgency of the anomaly, a multi-level risk warning system is constructed. The system includes a risk level classification module, a warning signal generation module and a notification distribution module. The risk level classification module maps the anomaly score to a predefined risk level, the warning signal generation module generates visual and sound warning signals according to the risk level, and the notification distribution module sends warning information to relevant personnel according to the preset notification rules; the system generates corresponding warning signals and notifications based on the risk level.

[0063] Step 2: Build a drug sales causal structure learning and attribution analysis system to determine the abnormal causes. This step includes the following sub-steps:

[0064] 2.1: Construct a drug sales causal structure learning model based on the NOTEARS algorithm. The model learns the causal relationship between variables by solving the following optimization problem:

[0065] ;

[0066] in represents the weight matrix to be learned, Represents drug sales-related characteristic data, is the loss function, Acyclic constraints ensure that the learned causal graph is acyclic;

[0067] The model can automatically discover the causal links from sales data, such as "delayed supplier delivery" leading to "inadequate inventory" and then causing "abnormal sales", and output the causal structure diagram of the drug sales system;

[0068] 2.2: Construct a factor attribution algorithm based on SHAP to calculate the contribution of each factor to the anomaly. The algorithm calculates the feature importance through the following formula:

[0069] ;

[0070] in Representation characteristics Contribution to the anomaly, represents the set of all features, Indicates that it does not contain features A subset of features, represents the output of the prediction model on the feature set, represents the predictive model to be analyzed, whose output depends on a subset of the input, Represents the sampled dataset instance, used to calculate the contribution value, Representation Subset The size of Representation characteristics Join Subset The difference in model output before and after, i.e., the feature The marginal contribution to the outcome, Represents the combined weight, indicating that among all possible subset arrangements, the current subset The weight of .

[0071] The algorithm can determine the contribution of each factor in a specific abnormal event. For example, it can identify that in a certain sales anomaly, the contribution of "promotional delay" is 45%, the contribution of "competing product price changes" is 30%, and the contribution of "seasonal factors" is 25%. The output is a ranked list of the contribution of each factor to the anomaly.

[0072] 2.3: Construct a counterfactual scenario simulation engine, which includes a variable intervention module, a causal propagation module, and an effect comparison module. By modifying the values ​​of specific nodes in the causal model and propagating the changes along the causal graph, it simulates the counterfactual scenario of "what would happen if this factor was not present", such as simulating the scenario of "whether sales anomalies would still occur if there were no delivery delays from suppliers", calculates the degree of influence of different factors on anomalies in counterfactual scenarios, verifies the validity of causal relationships, and outputs a causal verification report.

[0073] Step 3: Build an intervention strategy generation and evaluation system and formulate anomaly correction plans. This step includes the following sub-steps:

[0074] 3.1: Construct an intervention effect reasoning engine based on causal structure, which includes intervention operation definition module, reasoning calculation module and effect prediction module, and applies causal intervention formula:

[0075] ;

[0076] Simulate the impact of different intervention strategies on the sales system, including Indicates that the intervention variable The value is Time target variable The distribution of Represents the intervention variable With the target variable The confounding factors between To intervene, you force the variable to a specific value instead of observing its value. represents the variable being intervened, It indicates the specific value that a variable takes after being intervened. Represents confounding variables A value of , It affects simultaneously and variables; for example, the engine can predict the impact of the intervention of "increasing the inventory of a specific drug by 15%" on "out-of-stock rate" and "inventory cost", and the output is the expected effect of each intervention strategy;

[0077] 3.2: Construct a multi-objective adaptive intervention strategy generation module, which includes a strategy library management unit, a combination optimization unit and an adaptive adjustment unit. Based on the genetic algorithm, the module selects and combines the optimal intervention operation set from the intervention operation library for different sales anomaly types and severities. For example, for the "inadequate inventory caused by seasonal demand fluctuations" anomaly, the module can generate a combination strategy including "increase the purchase volume by 20% 2 weeks in advance", "adjust the safety inventory level" and "start emergency supply from alternative suppliers". The input is the anomaly type, severity and current system status, and the output is a feasible combination of intervention strategies.

[0078] 3.3: Establish an intervention effect evaluation system, which includes a multi-dimensional evaluation index module, a strategy simulation test module and a comprehensive score calculation module. It comprehensively considers the intervention cost, implementation difficulty and expected effect, and conducts quantitative evaluation of each intervention strategy through a multi-dimensional scoring matrix. For example, the "price adjustment" strategy is scored in terms of cost, timeliness, customer satisfaction and long-term impact, and a comprehensive score is generated, from which the optimal intervention plan is selected.

[0079] Step 4: Implement inventory strategy optimization and execution to complete the anomaly correction process. This step includes the following sub-steps:

[0080] 4.1: Based on the evaluation results of the intervention strategy, an intelligent inventory adjustment plan generator is constructed. The generator includes a plan template library, a parameter optimization module, and an execution instruction conversion module. The generator uses a reinforcement learning algorithm to generate detailed adjustment plans including price adjustment, promotion strategy, inventory level adjustment, and distribution method optimization for specific abnormality types and causes. For example, for the abnormality of "sudden increase in demand for specific drugs", a comprehensive adjustment plan including "temporarily increase the safety inventory level of the drug by 30%", "increase the distribution frequency", and "adjust the order point" is generated, and executable inventory optimization instructions are output;

[0081] 4.2: Build an inventory strategy execution monitoring system, which includes a data collection module, a real-time analysis module, and a feedback generation module. It uses a real-time data processing framework to continuously collect key indicator data during the strategy execution process, including inventory levels, sales volume, order completion rate, etc., and monitor the execution effect. For example, it tracks the changes in inventory turnover rate and out-of-stock rate after the implementation of the "increase safety stock" strategy, calculates abnormal indicator changes after the implementation of the strategy, and generates an execution effect report;

[0082] 4.3: Construct a strategy feedback optimization module, which includes an effect evaluation unit, a strategy adjustment unit, and a learning update unit. Based on the online learning algorithm, the intervention strategy is dynamically adjusted and optimized according to the strategy execution effect data. For example, according to the actual execution effect of the "price adjustment" strategy, the price change range or duration is automatically adjusted to form a closed-loop optimization system to achieve continuous improvement in exception management.

[0083] The technical solution of the present invention is described in detail below in conjunction with specific embodiments, but the protection scope of the present invention is not limited to the following embodiments.

[0084] Application scenarios:

[0085] The present invention is applied to the drug sales inventory management system of a large chain drugstore enterprise. The enterprise has 500 stores, sells more than 20,000 kinds of drugs, and generates about 1 million sales records every day. Before implementing the present invention, the enterprise faced the following problems:

[0086] (1) Sales anomaly detection mainly relies on manually preset rules, with a false alarm rate as high as 40%;

[0087] (2) After the abnormality is discovered, professional personnel are required to analyze the cause, which takes an average of 4-5 hours;

[0088] (3) Anomaly correction strategies are mainly developed based on manual experience, with a success rate of approximately 60%;

[0089] (4) Inventory turnover efficiency is low, with average annual inventory costs accounting for 18% of sales.

[0090] System architecture:

[0091] The drug sales inventory management system implemented by the present invention adopts a layered architecture, as shown in Table 1:

[0092] Table 1 System architecture level and function allocation

[0093]

[0094] Data preparation:

[0095] The data used in this embodiment include:

[0096] Sales data: includes fields such as product ID, sales quantity, sales amount, sales time, sales store, payment method, etc. The data spans 3 years and contains approximately 1 billion records in total.

[0097] Inventory data: Contains fields such as product ID, inventory quantity, inventory cost, inventory location, batch information, expiration date, etc. 20,000 kinds of drugs are updated daily, totaling about 22 million records.

[0098] External factor data: includes weather information, holiday information, competitor promotions, supplier delivery conditions, etc., as external variables that affect sales and inventory.

[0099] Historical abnormal case data: Contains 2,500 annotated abnormal cases, each of which contains information such as abnormality type, cause, severity, handling method and handling result.

[0100] Data preprocessing includes: (1) Missing value filling: the median is used to fill in the numerical features, and the mode is used to fill in the categorical features; (2) Outlier processing: the 3σ rule is used to identify and replace outliers; (3) Feature engineering: constructing more than 200 dimensional features such as time series features, sales trend features, and product association features.

[0101] Multimodal anomaly detection framework implementation details:

[0102] In this embodiment, the specific implementation process of the multimodal anomaly detection framework is as follows:

[0103] Detection algorithm configuration: Configure the parameters of each anomaly detection algorithm, as shown in Table 2.

[0104] Table 2 Anomaly detection algorithm parameter configuration

[0105]

[0106] Real-time data processing flow: The system collects sales data from the previous day every day, and after preprocessing, it is input into three anomaly detection algorithms respectively.

[0107] Fusion strategy implementation: The fusion weights of the output results of each algorithm are shown in Table 3.

[0108] Table 3 Anomaly detection algorithm fusion weight configuration

[0109]

[0110] Actual detection effect: In the anomaly detection results of a Class C cold medicine sales data, the red marked points are anomalies identified by the system. Further analysis found that these anomalies are mainly related to seasonal influenza outbreaks and competitive product promotions.

[0111] Application case: During the Spring Festival in 2022, the system detected an abnormal sales peak of a certain prescription drug in three stores, with a spatial anomaly score of 0.92, a pattern anomaly score of 0.85, a temporal anomaly score of 0.78, and a comprehensive anomaly score of 0.86 (full score 1.0). The system automatically issued a high-level warning and discovered possible abnormal sales trends three days in advance.

[0112] Implementation details of causal structure learning and attribution analysis:

[0113] In this embodiment, the specific implementation process of causal structure learning and attribution analysis is as follows:

[0114] NOTEARS algorithm implementation: Causal structure learning is performed on the variables in the drug sales system. The key parameter configuration is shown in Table 4.

[0115] Table 4: NOTEARS algorithm parameter configuration

[0116]

[0117] Learning results: The system's key causal paths for abnormal drug sales events include: supplier delay → insufficient inventory → abnormal sales (weight 0.82), promotion activities → competitor sales → own sales (weight 0.65).

[0118] SHAP attribution implementation: Based on the above causal structure, the contribution of each factor in the abnormal event is calculated. The SHAP algorithm configuration is shown in Table 5.

[0119] Table 5: SHAP attribution algorithm configuration

[0120]

[0121] Attribution results: Regarding the abnormal sales volume of a certain cold medicine in February 2022, the attribution analysis results generated by the system are shown in Table 6.

[0122] Table 6: Results of abnormal sales attribution analysis

[0123]

[0124] Counterfactual verification: To verify whether the "flu outbreak" was the main cause, the system constructed a counterfactual scenario "whether sales would still be abnormal if the flu had not broken out". The simulation results showed that in the absence of flu, the probability of abnormal sales dropped to 12%, verifying that the flu outbreak was indeed the main cause.

[0125] Application effect: After adopting the causal attribution method of the present invention, the average time for abnormal cause analysis was reduced from 4.3 hours to 18 minutes, and the analysis accuracy was increased from 65% to 87%, providing a reliable basis for the formulation of subsequent intervention strategies.

[0126] Intervention strategy generation and inventory optimization implementation details:

[0127] In this embodiment, the specific implementation process of intervention strategy generation and inventory optimization is as follows:

[0128] Construction of intervention strategy library: The system's preset intervention strategy library contains 100 basic strategies, which are divided into five categories: price adjustment, promotion management, inventory control, supply chain optimization and channel management. These include: temporary price reduction of 10%, member-only price (price adjustment category); buy two get one free, limited-time rush sale (promotion management category); increase safety inventory by 20%, adjust replenishment frequency (inventory control category); enable alternative suppliers (supply chain optimization category); inter-store inventory transfer (channel management category), etc. Different strategies are suitable for different types of anomalies, and have different implementation costs and expected impact durations.

[0129] Intervention effect reasoning implementation: In response to the anomaly of "surge in sales of cold medicines caused by influenza", the system uses causal intervention reasoning to simulate the effects of multiple intervention strategies. The main results include: without intervention, the out-of-stock rate increased by 25%, sales decreased by 15%, and customer satisfaction decreased by 20%; the strategy of increasing safety stock by 30% can reduce the out-of-stock rate by 20% and increase sales by 8%; the strategy of enabling alternative suppliers can reduce the out-of-stock rate by 22% and increase sales by 10%; the best combination strategy (alternative suppliers + inventory transfer) can reduce the out-of-stock rate by 28%, increase sales by 15%, improve customer satisfaction by 22%, and the overall score reaches +11.75%.

[0130] Inventory adjustment plan implementation: The system selects "Combination Strategy 2" as the optimal solution for this case and generates specific execution instructions:

[0131] Immediately purchase cold medicine X from alternative supplier A in an amount that is 150% of the normal monthly purchase volume;

[0132] Transfer inventory from stores in the southern region with sufficient inventory to stores in the northern region. The quantity transferred from a single store shall not exceed 40% of the safety stock.

[0133] Increase the safety stock level of cold medicine X in all stores to 130% of the original level, which will last until the peak of the flu season;

[0134] Increase the frequency of delivery from the central warehouse to stores from twice a week to three times a week;

[0135] Strategy execution monitoring: The system monitors the changes in key indicators after the implementation of the above strategies. After the intervention measures were taken, the out-of-stock rate dropped from 25% to 6%, customer satisfaction increased by 20 percentage points, and sales were 18% higher than those without the intervention plan.

[0136] Strategy feedback optimization: Based on the execution monitoring data, the system learns the actual effect of combination strategy 2 and optimizes the strategy parameters: the proportion of alternative supplier procurement is adjusted from 150% to 135%, and the safety stock increase ratio is adjusted from 130% to 120% to reduce inventory costs while maintaining service levels.

[0137] Comprehensive effect verification: In the management of cold medicines throughout 2022, this system successfully predicted and handled 35 abnormal sales events, discovered abnormal trends 2.3 days in advance on average, the average execution time of the intervention strategy was 3.5 hours, the success rate of abnormal correction reached 88%, the inventory turnover rate increased by 28%, and the overall inventory cost was reduced by 17%.

[0138] System implementation case summary:

[0139] After one year of implementing the intelligent inventory management system of the present invention in a large chain drugstore, significant economic and management benefits have been achieved:

[0140] Anomaly detection accuracy: The system detection rate reaches 92%, which is 46% higher than the traditional method; the false alarm rate is reduced to 15%, which is 25% lower than the traditional method; the average warning is 2.5 days in advance, which buys sufficient time for management decision-making.

[0141] Attribution analysis efficiency: The average time for anomaly attribution was reduced from 4.3 hours to 15 minutes, and the accuracy rate was increased from 65% to 87%, significantly improving the decision-making efficiency of managers.

[0142] Improved intervention effectiveness: The average time required to formulate an intervention strategy was reduced from 3.2 hours to 42 minutes, the strategy execution success rate was increased from 60% to 85%, and the anomaly correction time was shortened by 58%.

[0143] Economic benefits: Inventory turnover rate increased by 26%, inventory cost as a percentage of sales decreased from 18% to 14.5%, out-of-stock rate decreased from 8.5% to 5.2%, and inventory costs were saved by approximately RMB 12.5 million throughout the year, while sales increased by approximately RMB 18.5 million.

[0144] The present invention realizes the intelligent, automated and precise management of drug sales inventory, breaks the information barriers between anomaly detection, attribution analysis and intervention strategies in traditional systems, forms a closed-loop anomaly management system, and provides efficient inventory management solutions for large-scale drug sales companies.

[0145] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. A drug sales inventory management system based on artificial intelligence, characterized in that: include: Multimodal anomaly detection module, which builds a multimodal anomaly detection framework, uses the isolation forest algorithm, deep autoencoder and sequence anomaly detection algorithm to detect drug sales anomalies and generate a comprehensive anomaly score; The causal structure learning and attribution analysis module builds a drug sales causal structure learning and attribution analysis system, uses the NOTEARS algorithm to learn the causal relationship between variables to form a causal structure diagram, and uses the SHAP attribution algorithm to calculate the contribution of each factor to the anomaly; Intervention strategy generation and evaluation module: builds an intervention strategy generation and evaluation system, applies causal intervention formulas to simulate the impact of different intervention strategies on the sales system, and generates and evaluates intervention strategy combinations; The inventory strategy optimization and execution module implements inventory strategy optimization and execution, generates inventory adjustment plans based on intervention strategy evaluation results, monitors execution effects and dynamically adjusts intervention strategies.

2. According to claim 1, the drug sales inventory management system based on artificial intelligence is characterized in that: The multimodal anomaly detection framework in the multimodal anomaly detection module generates a comprehensive anomaly score by weighted fusion of the results of multiple anomaly detection algorithms.

3. According to claim 2, the drug sales inventory management system based on artificial intelligence is characterized in that: The step of constructing a multimodal anomaly detection framework in the multimodal anomaly detection module includes: Build a detection framework that integrates multiple anomaly detection algorithms, including a data preprocessing module, an algorithm integration module, and a result fusion module; Train anomaly pattern classifier based on historical anomaly case data and expert knowledge base; Build a multi-level risk early warning system to map anomaly scores to predefined risk levels and generate corresponding early warning signals and notifications.

4. According to claim 3, the drug sales inventory management system based on artificial intelligence is characterized in that: The abnormal pattern classifier adopts the gradient boosting decision tree algorithm, includes a feature extraction layer, a tree model training layer and a model integration layer, the input is the sales data feature vector, and the output is the abnormal pattern category and its probability distribution.

5. According to claim 4, the drug sales inventory management system based on artificial intelligence is characterized in that: The NOTEARS algorithm in the causal structure learning and attribution analysis module forms a causal structure diagram by solving the optimization problem to learn the causal relationship between variables. The calculation formula is as follows: ； in represents the weight matrix to be learned, Represents drug sales-related characteristic data, is the loss function, is acyclic constraint.

6. The drug sales inventory management system based on artificial intelligence according to claim 5 is characterized in that: The SHAP attribution algorithm in the causal structure learning and attribution analysis module calculates feature importance using the following formula: ； in Representation characteristics Contribution to the anomaly, represents the set of all features, Indicates that it does not contain features A subset of features, represents the output of the prediction model on the feature set, Represents a model to be analyzed whose output depends on a subset of its input , Represents the sampled dataset instance, used to calculate the contribution value, Representation Subset The size of Representation characteristics Join Subset The difference in model output before and after, i.e., the feature The marginal contribution to the outcome, Represents the combined weight, indicating that among all possible subset arrangements, the current subset The weight of .

7. The drug sales inventory management system based on artificial intelligence according to claim 6 is characterized in that: In the step of constructing the intervention strategy generation and evaluation system in the intervention strategy generation and evaluation module, the causal intervention formula is: ； in Indicates that the intervention variable The value is Time target variable The distribution of Represents the intervention variable With the target variable The confounding factors between To make a budget, represents the variable being intervened, It indicates the specific value that a variable takes after being intervened. Represents confounding variables A value of , It affects simultaneously and variables.

8. The drug sales inventory management system based on artificial intelligence according to claim 7 is characterized in that: The step of constructing the intervention strategy generation and evaluation system also includes: Construct a multi-objective adaptive intervention strategy generation module to select and combine the optimal intervention operation set from the intervention operation library based on genetic algorithm; Establish an intervention effect evaluation system, comprehensively consider the intervention cost, implementation difficulty and expected effect, and conduct quantitative evaluation of each intervention strategy.

9. The drug sales inventory management system based on artificial intelligence according to claim 8 is characterized in that: The steps of implementing inventory strategy optimization and execution in the inventory strategy optimization and execution module include: Build an intelligent inventory adjustment plan generator, using reinforcement learning algorithms to generate detailed adjustment plans including price adjustment, promotion strategy, inventory level adjustment, and delivery method optimization; Build an inventory strategy execution monitoring system to continuously collect key indicator data during strategy execution; Build a strategy feedback optimization module to dynamically adjust and optimize the intervention strategy based on the online learning algorithm.

10. An artificial intelligence-based drug sales inventory management system according to any one of claims 1 to 9, characterized in that: The data used in the drug sales inventory management system include drug sales data, inventory data, external factor data, and historical abnormal case data; The drug sales data includes fields such as product ID, sales quantity, sales amount, sales time, sales store, and payment method; The inventory data includes fields for product ID, inventory quantity, inventory cost, inventory location, batch information, and expiration date; The external factor data includes weather information, holiday information, competitor promotion activities, and supplier delivery data; The historical abnormal case data includes a number of annotated abnormal cases, each of which includes information on the abnormality type, cause, severity, handling method and handling result.

Images