Drug sales management method and system based on information monitoring

Abstract

The invention discloses a medicine sales management method and system based on information monitoring, and relates to the technical field of medicine sales management, and the method comprises the steps: collecting medicine sales time sequence data, carrying out the cleaning and structural processing, and generating standardized sales data; based on the standardized sales data, constructing a sales behavior time sequence causal network, embedding virtual sales intervention nodes, and generating a time sequence causal graph; the intervention strategy is coded to the virtual intervention node in the time sequence causal atlas, the graph neural network is utilized to simulate time sequence propagation of the intervention strategy, and a sales prediction result is obtained; and combining the sales prediction result with the real-time inventory, constructing a multi-level distributed sales dependency network, and forming the sales dependency network. According to the method, the preset intervention strategy is embedded into the causal structure in the node form, so that dynamic modeling of the multivariable causal relationship in the sales behavior is realized, and the prediction result not only reflects the historical trend, but also can respond to the possible influence of different intervention strategies.

Description

Technical Field

[0001] This invention relates to the field of pharmaceutical sales management technology, and in particular to a method and system for pharmaceutical sales management based on information monitoring. Background Technology

[0002] With the rapid development of information technology and big data technology, pharmaceutical sales management is gradually transforming towards digitalization and intelligence. Existing pharmaceutical sales management systems typically rely on the collection and analysis of sales data, using historical sales volume and inventory information to achieve basic inventory replenishment management and sales forecasting. Some methods combine time series analysis or machine learning models to predict sales trends and guide replenishment and allocation decisions accordingly.

[0003] However, existing technologies still have certain limitations in optimizing sales forecasting and replenishment allocation schemes. Most methods lack a sufficient understanding of the temporal causal relationships in sales data, making it difficult to accurately capture the complex dynamic influencing factors and interactions in sales behavior, resulting in a lack of specificity and adaptability in forecasting results and replenishment decisions. In addition, existing solutions often ignore the multi-dimensional impact of sales interventions (such as promotions and price adjustments) on sales trends, lack in-depth modeling of the temporal propagation effects of intervention strategies, and are difficult to achieve precise dynamic control of the sales process. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for drug sales management based on information monitoring to solve the problems of accurate prediction of sales intervention strategies and optimization of dynamic replenishment and allocation.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for drug sales management based on information monitoring, comprising: collecting drug sales time-series data, cleaning and structuring the data to generate standardized sales data; constructing a time-series causal network of sales behavior based on the standardized sales data, embedding virtual sales intervention nodes, and generating a time-series causal graph; encoding intervention strategies into the virtual intervention nodes in the time-series causal graph, simulating the time-series propagation of the intervention strategies using a graph neural network, and obtaining sales forecast results; combining the sales forecast results with real-time inventory to construct a multi-level distributed sales dependency network; automatically adjusting store replenishment and warehouse allocation plans using a multi-objective optimization algorithm based on the sales dependency network and sales forecast results, and generating replenishment and allocation execution instructions; and dynamically correcting and updating the standardized sales data according to the feedback information of the replenishment and allocation execution instructions and actual sales data to generate a feedback fusion sales dataset.

[0008] As a preferred embodiment of the information monitoring-based drug sales management method of the present invention, the specific steps for generating standardized sales data are as follows:

[0009] Collect sales time, store number, drug code, sales quantity, price, inventory snapshot and promotional labels to form actual sales data;

[0010] The actual sales data is formatted, fields are standardized, and redundant fields are removed to output preliminary cleaned data. Missing values ​​are filled in, outliers are identified and corrected, and valid sales data is output.

[0011] Extract time-series feature fields from valid sales data, normalize them, output structured feature data, and align them with sales scenario labels according to a unified dimension to generate standardized sales data.

[0012] As a preferred embodiment of the drug sales management method based on information monitoring according to the present invention, the specific steps for generating the time-series causal graph are as follows:

[0013] Extract causal feature variables from standardized sales data and construct an initial causal relationship graph;

[0014] The initial causal relationship graph is subjected to structural learning and parameter estimation to enhance the credibility and temporal dependence of edge weights. The temporal causal network of sales behavior is output, and different types of preset virtual sales intervention nodes are inserted to generate a causal structure graph.

[0015] Intervention propagation modeling is performed on the causal structure graph to simulate the propagation process of different types of virtual sales intervention nodes in the causal chain, and the dynamic impact on downstream sales variables is analyzed to generate a time-series causal graph.

[0016] In a preferred embodiment of the information-monitoring-based drug sales management method of the present invention, the specific steps for obtaining sales forecast results are as follows:

[0017] The virtual sales intervention nodes in the time-series causal graph are encoded with intervention strategy features to generate intervention node embedding vectors.

[0018] Graph neural networks are used to model the temporal propagation of intervention graph embedding vectors, learn the temporal influence relationship between virtual sales intervention nodes, and output a dynamic representation of propagation.

[0019] Based on the dynamic representation of propagation, the multidimensional impact of sales intervention strategies on future sales trends is predicted, and sales forecast results are generated.

[0020] In a preferred embodiment of the information monitoring-based drug sales management method of the present invention, the specific steps for forming a sales dependency network are as follows:

[0021] Clean and standardize the real-time inventory data, and combine it with sales forecast results to generate standardized sales-inventory matching data;

[0022] Standardized sales-inventory matching data is used to analyze the coupling relationship between sales demand and inventory flow, identify the dependencies between flow nodes, inventory nodes and demand nodes, and construct a preliminary sales dependency graph.

[0023] A hierarchical structure is constructed based on the initial sales dependency graph to map the dependencies between various links in the supply chain and generate a sales dependency network.

[0024] As a preferred embodiment of the information monitoring-based drug sales management method of the present invention, the specific steps for generating replenishment and allocation execution instructions are as follows:

[0025] Based on the sales dependency network and sales forecast results, relevant information from circulation nodes, inventory nodes and demand nodes is integrated to construct a multi-objective optimization problem;

[0026] A multi-objective optimization algorithm is used to weigh the various objectives of the optimization problem, optimize the store replenishment and warehouse transfer scheme, obtain the optimization results, and generate replenishment and transfer execution instructions.

[0027] As a preferred embodiment of the drug sales management method based on information monitoring described in this invention, the specific steps for generating the feedback fusion sales dataset are as follows:

[0028] Collect feedback information on replenishment and transfer execution instructions and actual sales data to generate raw feedback data;

[0029] The raw feedback data is cleaned and formatted to generate standardized feedback data, which is then merged with standardized sales data to generate a fused dataset.

[0030] The fused dataset is dynamically corrected to adjust sales data deviations, generate corrected sales data, and update the sales database to generate a feedback fused sales dataset.

[0031] Secondly, this invention provides a system for drug sales management based on information monitoring, comprising: a data acquisition module for collecting drug sales time-series data, cleaning and structuring it to generate standardized sales data; a network construction module for constructing a time-series causal network of sales behavior based on the standardized sales data, embedding virtual sales intervention nodes, and generating a time-series causal graph; a propagation simulation module for encoding intervention strategies into the virtual intervention nodes in the time-series causal graph, simulating the time-series propagation of the intervention strategies using a graph neural network, and obtaining sales forecast results; a network construction module for combining sales forecast results and real-time inventory to construct a multi-level distributed sales dependency network; an allocation optimization module for automatically adjusting store replenishment and warehouse allocation schemes based on the sales dependency network and sales forecast results using a multi-objective optimization algorithm, generating replenishment and allocation execution instructions; and a data update module for dynamically correcting and updating the standardized sales data based on feedback information from replenishment and allocation execution instructions and actual sales data, generating a feedback fusion sales dataset.

[0032] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the method for drug sales management based on information monitoring as described in the first aspect of the present invention.

[0033] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the method for drug sales management based on information monitoring as described in the first aspect of the present invention.

[0034] The beneficial effects of this invention are as follows: By embedding preset intervention strategies into the causal structure as nodes, dynamic modeling of multivariate causal relationships in sales behavior is achieved, enabling prediction results to not only reflect historical trends but also respond to the potential impact of different intervention strategies. Furthermore, by modeling temporal relationships using graph neural networks, the transmission effect and temporal dependence of intervention strategies in the causal chain are revealed, enhancing the model's adaptability to complex sales scenarios and the interpretability of prediction logic, thereby significantly improving the accuracy of sales forecasting and the foresight of strategic decisions. Attached Figure Description

[0035] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1This is a flowchart of a method for drug sales management based on information monitoring.

[0037] Figure 2 This is a schematic diagram of a drug sales management system based on information monitoring.

[0038] Figure 3 The flowchart for generating a time-series causal graph.

[0039] Figure 4 A flowchart generated from the sales forecast results. Detailed Implementation

[0040] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0041] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0042] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0043] Reference Figures 1-4 This is one embodiment of the present invention, which provides a method for drug sales management based on information monitoring, including the following steps:

[0044] S1. Collect time-series data on drug sales, and perform cleaning and structuring processing to generate standardized sales data.

[0045] S1.1 Collect sales time, store number, drug code, sales quantity, price, inventory snapshot and promotional labels to form actual sales data.

[0046] Specifically, each transaction record is retrieved from the sales terminal. The sales time field in the record is read and converted to a standard time format. The corresponding store number field in the transaction record is read, the store number information is extracted and converted to a standard format. The product barcode or coding information is read as the drug code and matched and verified with the drug code field in the drug database. The sales quantity field in the transaction details is extracted, the integer part is retained and converted to a standard unit. The transaction amount field is read as the price field and two decimal places are retained for standardized processing. Real-time inventory snapshot data in the inventory management system is called synchronously, and the inventory quantity of the corresponding product is extracted according to the sales time point as the inventory snapshot field. The sales strategy record table is called to extract whether there is a promotional activity when the transaction occurs, and the corresponding promotional label information is extracted and summarized into a structured record in a standard format. Finally, the actual sales data containing sales time, store number, drug code, sales quantity, price, inventory snapshot and promotional label are generated.

[0047] S1.2. Standardize the format, fields, and remove redundant fields from the actual sales data to output preliminary cleaned data. Then, fill in missing values, identify and correct outliers, and output valid sales data.

[0048] Specifically, the sales time field in the actual sales data is uniformly converted to the format "YYYY-MM-DDHH:MM:SS", the store number is uniformly padded to 6 digits, and the drug code is uniformly converted to all uppercase. The sales quantity and price fields are uniformly retained to two decimal places, and the inventory snapshot field is uniformly converted to an integer. The promotional label fields are uniformly converted to standard labels in a fixed label set, for example, "100 minus 10" and "100-10" are uniformly converted to "full reduction". Unused intermediate processing fields in the actual sales data are removed, such as temporary cache fields or duplicate fields. For the sales quantity, price, and inventory snapshot fields in the actual sales data with missing values, the missing values ​​are padded using the median of the same store number and drug code within adjacent dates. Records in the actual sales data where the sales quantity field exceeds the historical daily sales average of similar drugs by more than 5 times are marked as outliers and corrected using the average of adjacent non-outlier values. Finally, the output is valid sales data with standardized format, complete fields, and no redundant information.

[0049] S1.3 Extract time-series feature fields from valid sales data, normalize them, output structured feature data, and align them with sales scenario labels according to unified dimension encoding to generate standardized sales data.

[0050] Specifically, based on valid sales data, time-series feature fields such as "hour," "weekday," and "month" are extracted from the sales time field. The sales quantity field, price field, and inventory snapshot field are used as numerical feature fields. All numerical feature fields are normalized using the min-max normalization method, with a normalization interval of [0,1]. After normalization, structured feature data is formed. The store number field, drug code field, and promotional label field are uniformly encoded using one-hot encoding, and then concatenated with the normalized structured feature data along the field dimensions. By aligning the field positions, the concatenated structured feature data is associated with the sales scenario labels, and numerical mapping is performed on the sales scenario labels. For example, "holidays" are mapped to 1 and "weekdays" are mapped to 0. After unified encoding, these are incorporated into the structured feature data, ultimately generating standardized sales data.

[0051] S2. Based on standardized sales data, construct a temporal causal network of sales behavior, embed virtual sales intervention nodes, and generate a temporal causal graph.

[0052] S2.1 Extract causal characteristic variables from standardized sales data and construct an initial causal relationship diagram.

[0053] Specifically, based on standardized sales data, the sales quantity field is selected as the dependent variable, and the "hour," "weekday," and "month" fields from the sales time field, the store number field, the drug code field, the price field, the inventory snapshot field, and the promotional label field are selected as candidate causal feature variables. The Granger causality test is used to conduct bivariate regression analysis on the causal relationship between the candidate causal feature variables and the sales quantity field. Under the condition of setting the significance level, for example, 0.05, causal feature variables with a probability value of less than 0.05 are selected as effective causal variables. According to the significant causal relationship between the selected effective causal variables and the sales quantity field, the effective causal variables are used as nodes, and the sales quantity field is used as the target node. The nodes are connected according to the direction of the causal relationship to form an initial causal relationship graph representing the path of influence of each causal variable on the sales quantity.

[0054] S2.2 Perform structural learning and parameter estimation on the initial causal relationship graph to enhance the credibility and time dependence of edge weights, output the temporal causal network of sales behavior, and insert different types of preset virtual sales intervention nodes to generate a causal structure graph.

[0055] Specifically, based on the temporal relationship between causal characteristic variables and the sales quantity field in standardized sales data, edge connections between nodes are identified using methods such as Granger causality tests or constraint-based structural learning, and causal directions are formed. Maximum likelihood estimation is used to numerically estimate the edge weight parameters, quantifying the influence of different causal characteristic variables on the sales quantity field. Furthermore, the timeliness of the influence relationship between nodes is modeled by setting time windows, enhancing the time dependence in the edge weights. After adjusting the edge weight parameters, a temporal causal network of sales behavior is output. Based on predefined sales intervention strategies such as promotional activities and pricing adjustments, corresponding types of virtual sales intervention nodes are added to the temporal causal network of sales behavior, and directed connections are established between the virtual sales intervention nodes and relevant causal characteristic variables, ultimately generating a causal structure graph containing intervention information.

[0056] It should also be noted that the predefined sales intervention strategies, such as promotional activities and pricing adjustments, are determined based on historical sales data and marketing plans. The specific steps are as follows: collect and organize information on the types, timing, duration, and corresponding price adjustment ranges of past promotional activities; combine the current and future promotional plans provided by the marketing department to clarify the specific parameters of various promotional methods and pricing strategies; and based on the time nodes and scope of impact of promotional activities and pricing adjustments, summarize the corresponding strategy content into a standardized set of intervention strategies as the basis for constructing virtual sales intervention nodes.

[0057] Virtual sales intervention nodes originate from predefined sales intervention strategies. Based on actual sales intervention records and business rules in standardized sales data, intervention events such as promotional activities, pricing adjustments, and changes in marketing methods are extracted. Intervention types are categorized through field labeling, and virtual nodes corresponding to actual causal characteristic variables are constructed in the causal network of sales behavior according to the intervention type. These virtual nodes serve as carriers of intervention information and are used to simulate the propagation effect of specific sales intervention strategies in the causal structure diagram.

[0058] S2.3. Model the intervention propagation of the causal structure diagram, simulate the propagation process of different types of virtual sales intervention nodes in the causal link, analyze the dynamic impact on downstream sales variables, and generate a time-series causal graph.

[0059] Specifically, based on the types of virtual sales intervention nodes in the causal structure graph, these nodes are classified according to different intervention strategy characteristics, and the intervention trigger time is used as the starting propagation time point. Using a graph traversal approach, such as depth-first search, the propagation path is traced sequentially along the outgoing edges of the virtual sales intervention nodes in the causal structure graph, recording the levels and lag times the intervention signal passes through to each downstream sales variable node. Then, along each propagation path, based on the edge weights and time decay function, the intervention signal is propagated downstream along the temporal causal network of sales behavior, starting from the virtual sales intervention node. Upon reaching each sales variable node, the influence value of the upstream node is weighted and accumulated based on the edge weights and time decay function of the path. For nodes with multiple upstream paths, the results of all paths are accumulated separately to obtain the dynamic influence value of the intervention signal at each sales variable node. In this example, the edge weight can be set to 0.8, the time decay coefficient to 0.9, and the influence value at the third level to be 0.8 × 0.9. 3 The impact results of all propagation paths are aggregated at the node level to obtain the impact distribution of each sales variable node under the current intervention strategy; finally, the node state information of the causal structure graph is updated based on the impact value of each sales variable node to generate a time-series causal graph containing dynamic information of intervention propagation.

[0060] S3. Encode the intervention strategy into virtual intervention nodes in the temporal causal graph, and use a graph neural network to simulate the temporal propagation of the intervention strategy to obtain sales forecast results.

[0061] S3.1 Encode the intervention strategy features of the virtual sales intervention nodes in the time-series causal graph to generate the intervention node embedding vector.

[0062] Specifically, the intervention strategy feature fields associated with virtual sales intervention nodes are extracted, including intervention strategy type, intervention start time, intervention duration, intervention object code, intervention product category, intervention region number, and intervention intensity level. One-hot encoding is performed on categorical fields such as intervention strategy type, intervention object code, intervention product category, and intervention region number. Intervention start time and duration are normalized, and the intervention intensity level is numerically mapped, for example, mapping "high," "medium," and "low" to 1.0, 0.5, and 0.2, respectively. The processed field encoding results are then concatenated into a one-dimensional vector in a fixed order to form the initial feature vector for each virtual sales intervention node. Finally, methods such as principal component analysis are used to compress the dimensionality of the initial feature vector, generating a unified-dimensional intervention node embedding vector.

[0063] S3.2 Utilize graph neural networks to model the temporal propagation of the intervention graph embedding vectors, learn the temporal influence relationship between virtual sales intervention nodes, and output the dynamic representation of propagation.

[0064] Specifically, the embedding vectors of each virtual sales intervention node in the temporal causal graph are used as input features to construct the feature matrix and adjacency matrix of the virtual sales intervention node. Then, a graph convolution operation is employed, based on graph convolutional network layers, to perform layer-by-layer aggregation and nonlinear transformation on the feature vectors of the virtual sales intervention node and its neighboring nodes with the adjacency matrix of the temporal causal network of sales behavior. In each graph convolutional network layer, the neighbor set of the virtual sales intervention node is determined according to the adjacency matrix, and the feature vectors of the neighboring nodes are linearly weighted and combined with the feature vector of the virtual sales intervention node itself, then input into an activation function for nonlinear mapping. This process is repeated multiple times, ultimately outputting the hidden state representation of the virtual sales intervention node at each time step. Next, a recurrent neural network structure is used to sequentially model the hidden states of the virtual sales intervention node at each time step, capturing the temporal influence relationship between virtual sales intervention nodes. During training, the backpropagation algorithm is used to optimize the graph neural network parameters, learning the temporal dynamic changes of the virtual sales intervention node through a combination of multi-layer graph convolution and recurrent units. Finally, a propagation dynamic representation integrating temporal and graph structure information is output.

[0065] S3.3 Based on the dynamic representation of propagation, predict the multidimensional impact of sales intervention strategies on future sales trends and generate sales forecast results.

[0066] Specifically, based on the propagation dynamic representation H t ∈R d Sales forecasts for future points in time The calculation is performed, and the expression is:

[0067]

[0068] in, H represents the sales indicator vector predicted at a future time point t+τ, W represents the regression weight matrix, and H represents the regression weight matrix. t Let b represent the propagation dynamic representation vector at the current time point t, b represent the bias vector, t represent the current time point, τ represent the prediction time step, R represent the sales prediction result matrix, d represent the feature dimension of the propagation dynamic representation, and m represent the number of dimensions of the sales indicator.

[0069] For the predicted sales indicator vector Perform smoothing and outlier correction to output the final sales forecast.

[0070] S4. Combine sales forecast results with real-time inventory to construct a multi-level distributed sales dependency network, thus forming a sales dependency network.

[0071] S4.1 Clean and standardize the real-time inventory data, and generate standardized sales-inventory matching data by combining the sales forecast results.

[0072] Specifically, the inventory quantity, drug code, store number, and time fields in the real-time inventory data are standardized in terms of field names and data types. For example, the time field format is standardized to "YYYY-MM-DDHH:MM:SS". Secondly, null values ​​in the real-time inventory data are filled in, for example, by using adjacent time point interpolation to fill in missing inventory quantities, and outliers are identified and removed using a quantile-based method. Thirdly, the cleaned real-time inventory data is aligned with the sales forecast time, drug code, and store number fields in the sales forecast results. Primary key matching is performed using the drug code and store number fields, and time synchronization is performed using the sales forecast time field. Finally, the inventory quantity field and the sales forecast quantity field are concatenated using the drug code, store number, and sales forecast time fields as a composite primary key, outputting standardized sales-inventory matching data that includes both the sales forecast quantity and inventory quantity fields.

[0073] S4.2 Analyze the coupling relationship between sales demand and inventory flow using standardized sales-inventory matching data, identify the dependencies between flow nodes, inventory nodes and demand nodes, and construct a preliminary sales dependency graph.

[0074] Specifically, the drug code field, store number field, sales forecast quantity field, inventory quantity field, and sales forecast time field are extracted from standardized sales-inventory matching data. Using the drug code field and store number field as a joint primary key, a time window sequence with the sales forecast time field as the sequence is constructed. Difference calculations are performed on the changes in the inventory quantity field of the same drug code across different stores to identify store nodes exhibiting inventory outflow and inflow behavior, defining them as circulation nodes. Store nodes with a sales forecast quantity field greater than zero and an insufficient inventory quantity field are marked as demand nodes, while store nodes with a sales forecast quantity field of zero but an inventory quantity field greater than zero are marked as inventory nodes. Then, based on the directionality of inventory quantity field changes and the sequential order of the time window, directional edges are established between inventory nodes and circulation nodes, and between circulation nodes and demand nodes. Finally, based on the drug code field as the classification condition, a preliminary sales dependency graph containing a set of nodes including inventory nodes, circulation nodes, and demand nodes, as well as their dependencies, is generated.

[0075] S4.3 Construct a hierarchical structure based on the preliminary sales dependency graph to map the dependencies between various links in the supply chain and generate a sales dependency network.

[0076] Specifically, based on the node type field in the preliminary sales dependency graph, nodes are divided into three levels: inventory nodes, circulation nodes, and demand nodes. Based on the established directional edges between inventory nodes and circulation nodes, and between circulation nodes and demand nodes, a top-down hierarchical dependency path is constructed. Then, using the drug code field as the classification condition, node hierarchical labels are assigned to each type of drug in the preliminary sales dependency graph, ensuring that all inventory nodes belong to the upper layer, circulation nodes to the middle layer, and demand nodes to the lower layer. Next, a set of dependency links is formed by connecting the directional edges between nodes of different levels, and all link sets are deduplicated and merged. Finally, using the node type field as the hierarchical classification condition, a hierarchical structure containing inventory nodes, circulation nodes, demand nodes, and dependency links is output, generating a sales dependency network.

[0077] S5. Based on the sales dependency network and sales forecast results, a multi-objective optimization algorithm is used to automatically adjust the store replenishment and warehouse allocation plans and generate replenishment and allocation execution instructions.

[0078] S5.1 Based on the sales dependency network and sales forecast results, integrate relevant information from circulation nodes, inventory nodes and demand nodes to construct a multi-objective optimization problem.

[0079] Specifically, the dependency path from each inventory node to the demand node is extracted from the sales dependency network. Combining the future sales quantity field of each demand node in the sales forecast results, the expected demand for each path is calculated using the path summation method. The current inventory quantity field and replenishment capacity field of the corresponding inventory node, the transfer cost field and maximum processing capacity field of the circulation node, and the service period field and demand priority field of the demand node are extracted and uniformly transformed into a constraint parameter matrix. Then, a multi-objective optimization problem is constructed with the inventory cost function, transfer cost function, and stockout risk function as objective functions, including the variables of inventory replenishment quantity and transfer circulation quantity.

[0080] S5.2. Employ a multi-objective optimization algorithm to weigh the various objectives of the optimization problem, optimize the store replenishment and warehouse allocation schemes, obtain the optimization results, and generate replenishment and allocation execution instructions.

[0081] Specifically, the Non-Dominated Sorting Genetic Algorithm II (NSGA-II) is selected as the solution method to iteratively optimize the inventory cost function, transfer cost function, and stockout risk function in the constructed multi-objective optimization problem simultaneously. Each chromosome in the initial population is encoded as a combination of an inventory replenishment vector and a transfer flow matrix. Individual fitness is evaluated, and the non-dominated level and crowding distance of each individual in the objective space are calculated based on the values ​​of the three objective functions. A new generation of population is generated through selection, crossover, and mutation operations. The above iterative process is repeated until the maximum number of generations or convergence condition is met. The Pareto front solution set is extracted from the final population, and the optimal inventory replenishment vector and transfer flow matrix are selected based on business weights or scheduling rules. The optimal inventory replenishment vector is converted into store replenishment execution instructions, and the optimal transfer flow matrix is ​​converted into warehouse transfer execution instructions. The execution data format is output, including fields such as "transfer origin number," "transfer destination number," "transfer quantity," "replenishment store number," and "replenishment quantity," forming the final replenishment and transfer execution instructions.

[0082] S6. Based on the feedback information from replenishment and allocation execution instructions and actual sales data, dynamically correct and update the standardized sales data to generate a feedback-integrated sales dataset.

[0083] S6.1 Collect feedback information from replenishment and transfer execution instructions and actual sales data to generate raw feedback data.

[0084] Specifically, after replenishment and transfer execution instructions are issued, execution feedback information is collected from each node of the store and warehouse, including replenishment completion time, replenishment quantity, transfer path, transfer completion time, and transfer quantity, as recorded in actual operation. Secondly, actual sales data corresponding to the replenishment and transfer cycle is collected simultaneously, including sales occurrence time, sales category, sales quantity, and sales price. The collected data is matched according to instruction number, product code, and timestamp to construct a joint structured data table containing three types of data fields: "instruction type - actual execution result - sales result". Finally, data consistency verification and missing value filling are performed on each field. For example, the missing sales quantity field is handled using the previous value filling method to ensure completeness and time alignment, generating original feedback data.

[0085] S6.2 Clean and standardize the format of the original feedback data to generate standardized feedback data, and merge it with standardized sales data to generate a fused dataset.

[0086] Specifically, anomaly detection and missing value handling are performed on the field values ​​in the original feedback data. For example, data records with negative replenishment quantities are deleted, and missing sales quantity fields are filled with the average sales of adjacent time periods for the same store. Secondly, the time format in the original feedback data is standardized to the standard format "YYYY-MM-DDHH:MM:SS", the product code format is standardized to an eight-digit code, and the unit field is standardized as "piece" as the unit of measurement for inventory and sales quantities. Key fields are extracted from the replenishment feedback data and transfer feedback data, including replenishment quantity, transfer quantity, execution time, and actual arrival time, and these fields are uniformly named "Replenishment Quantity," "Transfer Quantity," "Execution Time," and "Arrival Time," respectively. The cleaned data is sorted by timestamp, product code, and store code to output standardized feedback data. Using product code, store code, and timestamp as the primary key, the standardized feedback data and standardized sales data are cascaded and horizontally concatenated to ensure that the sales and replenishment / transfer activities of the same product in the same store correspond within the same record, generating a fused dataset.

[0087] S6.3 Dynamically correct the fused dataset, adjust the sales data deviation, generate corrected sales data, update the sales database, and generate a feedback fused sales dataset.

[0088] Specifically, the daily difference method is used to process each record in the fused dataset, calculating the daily difference between the actual sales quantity and the sales forecast result to obtain the sales error sequence. Based on the feedback execution time of replenishment and allocation instructions and the actual arrival time, it is determined whether there are sales gaps caused by delayed arrivals. The average sales cycle is traced back from the actual arrival time to adjust the sales data for delay compensation. The historical sales error sequence is smoothed using a moving average method. For example, a sliding window size of 7 days is set, and the average error of the previous 7 days is calculated as the correction offset for each time point. This correction offset is then added to the original sales data to generate the corrected sales quantity, forming the corrected sales data. Subsequently, the corrected sales data is updated to the original sales data records in the sales database according to the product code, store code, and timestamp. Finally, the fused standardized feedback data and the corrected sales data are concatenated to output a feedback fused sales dataset containing feedback information and corrected sales data.

[0089] It should also be noted that the specific method for determining whether there is a sales gap due to delayed delivery is as follows: A dynamic inventory comparison method is used. The daily inventory change in the fused dataset is compared with the sales forecast results. If the sales forecast result is significantly higher than the actual sales quantity within a certain period, and the inventory for that period does not decrease significantly or is in a zero-inventory state, then it is determined that there is a sales gap due to delayed delivery for that period. Example: If the sales forecast result for day X is 80, the actual sales quantity is 20, and the inventory record for day X is 0 or continuously 0, then it is determined that there is a sales gap due to delayed delivery on day X.

[0090] This embodiment also provides a drug sales management system based on information monitoring, including: a data acquisition module for collecting drug sales time-series data, cleaning and structuring it to generate standardized sales data; a network construction module for constructing a sales behavior time-series causal network based on the standardized sales data, embedding virtual sales intervention nodes, and generating a time-series causal graph; a propagation simulation module for encoding intervention strategies into the virtual intervention nodes in the time-series causal graph, simulating the time-series propagation of the intervention strategies using a graph neural network, and obtaining sales forecast results; a network construction module for combining sales forecast results and real-time inventory to construct a multi-level distributed sales dependency network; an allocation optimization module for automatically adjusting store replenishment and warehouse allocation schemes based on the sales dependency network and sales forecast results using a multi-objective optimization algorithm, generating replenishment and allocation execution instructions; and a data update module for dynamically correcting and updating the standardized sales data based on feedback information from replenishment and allocation execution instructions and actual sales data, generating a feedback fusion sales dataset.

[0091] This embodiment also provides a computer device applicable to the method of drug sales management based on information monitoring, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the drug sales management method based on information monitoring as proposed in the above embodiment.

[0092] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0093] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the method for drug sales management based on information monitoring as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0094] In summary, this invention achieves dynamic modeling of multivariate causal relationships in sales behavior by embedding pre-defined intervention strategies as nodes within a causal structure. This allows the prediction results to not only reflect historical trends but also respond to the potential impacts of different intervention strategies. Furthermore, by modeling temporal relationships using graph neural networks, the transmission effect and temporal dependence of intervention strategies in the causal chain are revealed, enhancing the model's adaptability to complex sales scenarios and the interpretability of the prediction logic. This significantly improves the accuracy of sales forecasting and the foresight of strategic decisions.

[0095] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for drug sales management based on information monitoring, characterized in that: include, Collect time-series data on drug sales, and perform cleaning and structuring processing to generate standardized sales data; Based on standardized sales data, a temporal causal network of sales behavior is constructed, virtual sales intervention nodes are embedded, and a temporal causal graph is generated. Intervention strategies are encoded into virtual intervention nodes in a temporal causal graph, and the temporal propagation of the intervention strategies is simulated using a graph neural network to obtain sales forecast results. By combining sales forecasts with real-time inventory, a multi-level distributed sales dependency network is constructed, forming a sales dependency network. Based on the sales dependency network and sales forecast results, a multi-objective optimization algorithm is used to automatically adjust store replenishment and warehouse allocation plans and generate replenishment and allocation execution instructions. Based on the feedback information from replenishment and allocation execution instructions and actual sales data, standardized sales data is dynamically corrected and updated to generate a feedback-integrated sales dataset.

2. The method for drug sales management based on information monitoring as described in claim 1, characterized in that: The specific steps for generating standardized sales data are as follows: Collect sales time, store number, drug code, sales quantity, price, inventory snapshot and promotional labels to form actual sales data; The actual sales data is formatted, fields are standardized, and redundant fields are removed to output preliminary cleaned data. Missing values ​​are filled in, outliers are identified and corrected, and valid sales data is output. Extract time-series feature fields from valid sales data, normalize them, output structured feature data, and align them with sales scenario labels according to a unified dimension to generate standardized sales data.

3. The method for drug sales management based on information monitoring as described in claim 1, characterized in that: The specific steps for generating the time-series causal graph are as follows. Extract causal feature variables from standardized sales data and construct an initial causal relationship graph; The initial causal relationship graph is subjected to structural learning and parameter estimation to enhance the credibility and temporal dependence of edge weights. The temporal causal network of sales behavior is output, and different types of preset virtual sales intervention nodes are inserted to generate a causal structure graph. Intervention propagation modeling is performed on the causal structure graph to simulate the propagation process of different types of virtual sales intervention nodes in the causal chain, and the dynamic impact on downstream sales variables is analyzed to generate a time-series causal graph.

4. The method for drug sales management based on information monitoring as described in claim 1, characterized in that: The specific steps to obtain the sales forecast results are as follows: The virtual sales intervention nodes in the time-series causal graph are encoded with intervention strategy features to generate intervention node embedding vectors. Graph neural networks are used to model the temporal propagation of intervention graph embedding vectors, learn the temporal influence relationship between virtual sales intervention nodes, and output a dynamic representation of propagation. Based on the dynamic representation of propagation, the multidimensional impact of sales intervention strategies on future sales trends is predicted, and sales forecast results are generated.

5. The method for drug sales management based on information monitoring as described in claim 1, characterized in that: The specific steps for forming a sales dependency network are as follows. Clean and standardize the real-time inventory data, and combine it with sales forecast results to generate standardized sales-inventory matching data; Standardized sales-inventory matching data is used to analyze the coupling relationship between sales demand and inventory flow, identify the dependencies between flow nodes, inventory nodes and demand nodes, and construct a preliminary sales dependency graph. A hierarchical structure is constructed based on the initial sales dependency graph to map the dependencies between various links in the supply chain and generate a sales dependency network.

6. The method for drug sales management based on information monitoring as described in claim 1, characterized in that: The specific steps for generating replenishment and transfer execution instructions are as follows: Based on the sales dependency network and sales forecast results, relevant information from circulation nodes, inventory nodes and demand nodes is integrated to construct a multi-objective optimization problem; A multi-objective optimization algorithm is used to weigh the various objectives of the optimization problem, optimize the store replenishment and warehouse transfer scheme, obtain the optimization results, and generate replenishment and transfer execution instructions.

7. The method for drug sales management based on information monitoring as described in claim 1, characterized in that: The specific steps for generating the feedback fusion sales dataset are as follows: Collect feedback information on replenishment and allocation execution instructions and actual sales data to generate raw feedback data; The raw feedback data is cleaned and formatted to generate standardized feedback data, which is then merged with standardized sales data to generate a fused dataset. The fused dataset is dynamically corrected to adjust sales data deviations, generate corrected sales data, and update the sales database to generate a feedback fused sales dataset.

8. A system for drug sales management based on information monitoring, based on the drug sales management method based on information monitoring according to any one of claims 1 to 7, characterized in that: include, The data acquisition module is used to collect time-series data on drug sales, and to clean and structure the data to generate standardized sales data. The network construction module is used to build a time-series causal network of sales behavior based on standardized sales data, embed virtual sales intervention nodes, and generate a time-series causal graph. The propagation simulation module is used to encode intervention strategies into virtual intervention nodes in a temporal causal graph, and to use a graph neural network to simulate the temporal propagation of the intervention strategies to obtain sales forecast results. The network building module is used to combine sales forecast results and real-time inventory to build a multi-level distributed sales dependency network, thus forming a sales dependency network. The allocation optimization module is used to automatically adjust store replenishment and warehouse allocation plans based on sales dependency networks and sales forecast results, and generate replenishment and allocation execution instructions. The data update module is used to dynamically correct and update standardized sales data based on feedback information from replenishment and transfer execution instructions and actual sales data, generating a feedback-integrated sales dataset.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the method for drug sales management based on information monitoring as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the method for drug sales management based on information monitoring as described in any one of claims 1 to 7.

Images