A digital enterprise operation refined management system and method based on big data

Abstract

The present invention discloses a digital enterprise operation refined management system and method based on big data, which relates to the field of digital system technology. The system includes: a feature engineering unit, a model building unit and an optimization solution unit; wherein the feature engineering unit is used to collect enterprise business data based on business needs, extract operational features, and build a decision feature set; the model building unit is used to establish a refined model based on the decision feature set, and the refined model includes decision variables, objective functions and constraints; the optimization solution unit is used to solve the refined model using a cutting plane method, obtain the optimal decision variables, and apply them to actual business. The present invention realizes the refined management of the entire process from data collection to decision execution, so that enterprises can make more accurate decisions in a complex and changing market environment, achieve the optimal allocation of resources, and significantly improve the accuracy of enterprise operation decisions.

Description

Technical Field

[0001] The present invention relates to the technical field of digital systems, and in particular to a digital enterprise operation refined management system and method based on big data. Background Art

[0002] With the rapid development of the digital economy, enterprise operations management is undergoing unprecedented changes. The widespread application of emerging technologies such as big data, artificial intelligence, and cloud computing has provided enterprises with massive data resources and powerful analytical tools. The integration of these technologies has not only changed the traditional business model, but also provided new possibilities for refined management of enterprises. In this context, how to effectively use digital technologies to improve enterprise operational efficiency and achieve refined management has become the core issue of current enterprise management; digital transformation is not only a technological upgrade, but also an innovation in enterprise management concepts and methods. It requires enterprises to be able to fully integrate internal and external data resources and establish a data-driven decision-making mechanism, so as to maintain competitive advantages in a complex and changing market environment.

[0003] Under this trend, many enterprises have begun to try to introduce various digital management tools and systems. Existing digital enterprise operation management systems usually include modules such as data collection, data analysis, and decision support. These systems can collect enterprise operation data from multiple channels and provide decision-making suggestions to managers through data mining and analysis technology. Some advanced systems have also introduced machine learning algorithms to try to predict future business trends through learning from historical data. These systems have improved the data processing capabilities and decision-making efficiency of enterprises to a certain extent, and provided technical support for the refinement and scientific management of enterprises. However, with the expansion of enterprise scale and the increase of business complexity, these systems have also exposed some limitations in actual applications.

[0004] Although the existing digital management systems have made some progress in data processing and analysis, they still have obvious deficiencies in achieving truly refined management. First, most systems lack depth in feature engineering and are unable to effectively extract and screen the most representative operational features, resulting in a lack of accuracy in subsequent analysis and decision-making; second, in terms of model construction, existing systems often adopt a fixed model structure, lack flexibility, and are difficult to adapt to the specific needs of different companies and the rapidly changing market environment; finally, in terms of optimization and practical application, many systems cannot effectively handle complex integer programming problems, and lack the ability to transform theoretical models into operational management solutions; these problems seriously restrict the refinement and efficiency improvement of enterprise operations management.

[0005] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0006] In response to the problems in the related technologies, the present invention proposes a digital enterprise operation refined management system and method based on big data, which has the advantages of realizing refined management of the entire process from data collection to decision execution, enabling enterprises to make more accurate decisions in a complex and changing market environment and achieve optimal allocation of resources, thereby solving the problems in the existing technology that the most representative operational characteristics cannot be effectively extracted and screened, the analysis and decision-making accuracy is not high, and there is a lack of the ability to convert theoretical models into operational management solutions.

[0007] To this end, the specific technical solution adopted by the present invention is as follows:

[0008] According to one aspect of the present invention, a digital enterprise operation refined management system based on big data is provided, the digital enterprise operation refined management system based on big data comprises a feature engineering unit, a model building unit and an optimization solution unit, the optimization solution unit comprises an initial solution acquisition module, a cutting plane method solution module and a management and application module;

[0009] Among them, the feature engineering unit is used to collect enterprise business data based on business needs, extract operational features, and build a decision feature set;

[0010] A model building unit, used to build a refined model based on a decision feature set, wherein the refined model includes decision variables, objective functions, and constraint conditions;

[0011] The optimization solution unit is used to solve the refined model using the cutting plane method, obtain the optimal decision variables, and apply them to actual business;

[0012] An initial solution acquisition module is used to obtain an initial feasible solution using the simplex method according to the objective function and constraints;

[0013] The cutting plane method solution module is used to obtain an integer feasible solution based on the initial feasible solution using the cutting plane method;

[0014] The management and application module is used to obtain the final optimization solution based on integer feasible solutions and apply it to the business through sensitivity analysis.

[0015] Furthermore, the feature engineering unit includes a data acquisition module, a feature extraction module and a feature screening module;

[0016] The data collection module is used to collect enterprise business data from multiple sources and channels using data collection tools to generate a business demand set;

[0017] The feature extraction module is used to extract operational features based on the business requirement set using feature engineering technology. The operational features include numerical features, classification features, time series features, and text features.

[0018] The feature screening module is used to screen out the most representative features based on operational characteristics and generate a decision feature set using linear discriminant analysis.

[0019] Furthermore, data collection tools are used to collect enterprise business data from multiple sources and generate business requirements including:

[0020] Analyze the specific needs of each business segment of the enterprise and determine the internal and external data sources;

[0021] Design a data collection plan based on specific needs and data sources, including collection frequency, time points and data formats;

[0022] Based on the data collection plan, use the selected tools to collect and pre-process data from various data sources to establish a set of business requirements.

[0023] Furthermore, based on the operational characteristics, linear discriminant analysis is used to screen out the most representative features and generate a decision feature set including:

[0024] Preprocess operational features and establish a standard measurement feature set;

[0025] Based on the standard metric feature set, a first-level operation matrix is ​​constructed, and the mean vector of each operation feature is calculated;

[0026] According to the mean vector, a conversion matrix is ​​constructed, and the first-level operation matrix is ​​converted using the conversion matrix to obtain a second-level operation matrix;

[0027] Based on the secondary operation matrix, the variance of each operation feature is calculated, and the operation feature with the highest variance is selected to construct the decision feature set.

[0028] Furthermore, the expression for calculating the mean vector of each operation feature is:

[0029] ;

[0030] In the formula, represents the mean vector of the operating feature c; Nc represents the number of samples in the operating feature c, The feature vector representing the i-th sample of the operational feature c;

[0031] The expression for calculating the variance of each operating characteristic is:

[0032] ;

[0033] In the formula, represents the variance of the jth feature dimension in the operational feature c; represents the observed value of the i-th sample in the j-th feature dimension in the operational feature c; Represents the average value of all samples in the operation feature c on the jth feature dimension.

[0034] Furthermore, the model building unit includes a feature analysis and variable definition module, an objective function building module and a constraint setting module;

[0035] Among them, the feature analysis and variable definition module is used to analyze the statistical characteristics of each high variance feature based on the constructed decision feature set and define the decision variables;

[0036] An objective function construction module is used to construct a linear objective function based on a decision feature set and a decision variable;

[0037] The constraint setting module is used to establish linear constraints based on statistical characteristics and decision variables.

[0038] Furthermore, based on the constructed decision feature set, the statistical characteristics of each high variance feature are analyzed, and the decision variables are defined as follows:

[0039] Based on the decision feature set, the descriptive statistics method is used to obtain the data type and distribution characteristics of each high variance feature;

[0040] Based on the data type and distribution characteristics of each high variance feature, the variable type and value range of each feature are determined by using data type judgment and threshold statistics. The variable types include continuous decision variables and integer decision variables.

[0041] Using standardization and encoding techniques, continuous decision variables are standardized and integer decision variables are encoded.

[0042] Furthermore, based on the decision feature set and decision variables, the linear objective function is constructed including:

[0043] Based on the decision feature set, the weights of decision variables are determined using fuzzy hierarchical analysis;

[0044] Based on the decision variable weights, the initial linear objective function is obtained by weighted summation;

[0045] Using sensitivity analysis, the initial linear objective function is optimized to obtain the final linear objective function.

[0046] Furthermore, based on the statistical characteristics and decision variables, the linear constraints are established including:

[0047] Based on statistical characteristics, confidence intervals are calculated and constraints on variable value ranges are constructed;

[0048] Based on business rules, business logic constraints are obtained by using linear inequality transformation;

[0049] According to the relationship between decision variables, strongly correlated variable pairs are determined and variable association constraints are established.

[0050] According to another aspect of the present invention, a digital enterprise operation refined management method based on big data is also provided, and the digital enterprise operation refined management method based on big data comprises the following steps:

[0051] S1. Based on business needs, collect enterprise business data, extract operational features, and build a decision feature set;

[0052] S2. Based on the decision feature set, a refined model is established, wherein the refined model includes decision variables, objective functions and constraints;

[0053] S3. Use the cutting plane method to solve the refined model, obtain the optimal decision variables, and apply them to actual business.

[0054] The beneficial effects of the present invention are:

[0055] (1) Accurate feature engineering and decision support: The present invention realizes the collection and efficient processing of multi-source channel data through the feature engineering unit. With the help of feature extraction module and feature screening module, the system can extract multi-dimensional operational features including numerical values, classifications, time series and texts, and use linear discriminant analysis to screen out the most representative features. This refined feature engineering provides a high-quality data foundation for subsequent decision support, significantly improving the accuracy of enterprise operational decisions.

[0056] (2) Flexible optimization model construction: The present invention realizes the transformation from features to decision variables and the flexible construction of objective functions and constraints through the feature analysis and variable definition module, the objective function construction module and the constraint setting module. At the same time, the fuzzy hierarchical analysis is used to determine the weights of decision variables, and the method of optimizing the objective function through sensitivity analysis is used. This enables the system to construct the most suitable optimization model according to the specific situation of the enterprise, providing strong theoretical support for the refined management of enterprise operations.

[0057] (3) Efficient solution and management application: The optimization solution unit adopts a combination of the initial solution acquisition module, the cutting plane method solution module and the management and application module to achieve efficient transformation from theoretical model to practical application. In particular, it uses the cutting plane method to solve integer programming problems and obtains the final optimization solution through sensitivity analysis. This enables the system to quickly obtain feasible optimization solutions and directly apply them to actual operation management business. This seamless connection from theory to practice has greatly improved the efficiency and accuracy of enterprise operation management. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0059] Figure 1 It is a principle block diagram of a digital enterprise operation refined management system based on big data according to an embodiment of the present invention;

[0060] Figure 2 It is a flowchart of a method for refined management of digital enterprise operations based on big data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0061] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention and are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, ordinary technicians in the field should be able to understand other possible implementations and advantages of the present invention. The components in the figures are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0062] According to an embodiment of the present invention, a digital enterprise operation refined management system and method based on big data are provided.

[0063] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to an embodiment of the present invention, a digital enterprise operation fine management system based on big data is provided, and the digital enterprise operation fine management system based on big data includes a feature engineering unit 1, a model building unit 2 and an optimization solution unit 3, and the optimization solution unit 3 includes an initial solution acquisition module 301, a cutting plane method solution module 302 and a management and application module 303;

[0064] Among them, the feature engineering unit 1 is used to collect enterprise business data based on business needs, extract operational features, and build a decision feature set;

[0065] A model building unit 2 is used to build a refined model based on the decision feature set, where the refined model includes decision variables, objective functions and constraints;

[0066] The optimization solving unit 3 is used to solve the refined model by using the cutting plane method, obtain the optimal decision variables, and apply them to actual business;

[0067] An initial solution acquisition module 301 is used to obtain an initial feasible solution using the simplex method according to the objective function and the constraints;

[0068] A cutting plane method solution module 302 is used to obtain an integer feasible solution based on the initial feasible solution using the cutting plane method;

[0069] The management and application module 303 is used to obtain the final optimization solution based on the integer feasible solution and apply it to the business by using sensitivity analysis.

[0070] Specifically, based on the linear objective function constructed by the objective function construction module 202 and the linear constraints established by the constraint setting module 203, the objective function and the constraints are first converted into a standard form. In this embodiment, the converted standard form is to maximize Z=0.270x1+0.165x2+0.400x3+0.110x4+0.055x5; the constraints are:

[0071] 6.228≤x1≤18.772;

[0072] 0.052≤x2≤0.248;

[0073] 0.093≤x3≤0.407;

[0074] 20600≤x4≤79400;

[0075] 4120≤x5≤15880;

[0076] x2+0.5x3≤0.3;

[0077] x1-0.5(x4 / 10000)≥5;

[0078] x2-0.002(x5 / 1000)≥0.05;

[0079] 0.8x3+0.0000005x4≤0.5;

[0080] 0.01x1+0.1x2≤0.4;

[0081] x4-5x5≤10000;

[0082] Then use the simplex method to solve this linear programming problem. After several iterations, the initial feasible solution is obtained: x1=15.5, x2=0.22, x3=0.35, x4=70000, x5=12000.

[0083] Specifically, based on the initial feasible solution, the integer feasible solution is obtained by using the cutting plane method. First, check whether the integer decision variable (x5 in this embodiment) in the initial feasible solution is an integer. If it is not an integer, add a cutting plane constraint. In this embodiment, x5=12000 is not an integer solution, so the following cutting plane constraint is added: x5≤floor(12000) or x5≥ceil(12000); where floor(x) represents the floor rounding function, that is, the maximum integer not greater than x; ceil(x) represents the ceiling rounding function, that is, the minimum integer not less than x. Then, solve the new linear programming problem again and repeat this process until all integer variables take integer values; in the above embodiment, the integer feasible solution finally obtained is: x1=15.3, x2=0.21, x3=0.36, x4=69500, x5=12000.

[0084] Specifically, based on the integer feasible solution, sensitivity analysis is used to obtain the final optimization solution and apply it to the business. In the above embodiment, a sensitivity analysis is first performed on the integer feasible solution to check the stability of the solution and the sensitivity to changes in constraint conditions. The analysis results show that x3 (product profit margin) and x4 (average daily sales) have the greatest impact on the objective function value, and changes within a certain range will not lead to significant changes in the solution; therefore, based on the sensitivity analysis results, the final optimization plan is formulated as follows: increase the inventory turnover rate (x1) to 15.3 times / year; increase the market share (x2) to 21%; increase the product profit margin (x3) to 36%; increase the average daily sales The amount (x4) increased to 69,500 yuan; the average daily website visits (x5) were maintained at 12,000 times / day; and this optimization plan was applied to actual business: a. In the inventory management section, optimize the replenishment cycle and improve inventory turnover; b. In the marketing section, increase marketing investment, especially online marketing, to increase market share; c. In the product pricing section, appropriately increase product prices while controlling costs to increase profit margins; d. In the sales strategy section, optimize sales channels and improve sales team efficiency to increase average daily sales; e. In the website optimization section, maintain the current website traffic while optimizing user experience and improving conversion rate.

[0085] Specifically, through the above optimization and solution process, not only the theoretical optimal solution is obtained, but also it is transformed into an operational business strategy. This method combines mathematical optimization and business practice, provides strong support for the refined management of enterprises, and helps enterprises make better decisions in a complex market environment.

[0086] In one embodiment, the feature engineering unit 1 includes a data acquisition module 101, a feature extraction module 102 and a feature screening module 103;

[0087] The data collection module 101 is used to collect enterprise business data from multiple sources and channels using data collection tools to generate a business demand set;

[0088] A feature extraction module 102 is used to extract operational features based on a business requirement set using feature engineering technology, where the operational features include numerical features, classification features, time series features, and text features;

[0089] Specifically, based on the business requirement set, the feature types to be extracted are determined. In this embodiment, four main feature types are focused on: numerical features, classification features, time series features, and text features.

[0090] Specifically, numerical features mainly focus on continuous variables such as sales volume, inventory level and price. The extraction methods include: a. Calculating basic statistics such as mean, median, standard deviation, maximum and minimum values; b. Performing normalization, using Min-Max scaling or Z-score standardization; c. Creating interactive features, such as the ratio of sales volume to price. In this embodiment, for each product, the average sales volume, standard deviation of sales volume, and price elasticity (sales volume change rate divided by price change rate) over the past 30 days are calculated.

[0091] Specifically, classification features mainly focus on discrete variables such as product categories, customer types, and promotional activities, and the extraction method includes: a. performing one-hot encoding to convert classification variables into binary features; b. calculating frequency encoding to replace the original categories with the frequency of the categories appearing in the data set; c. creating grouped statistical features, such as the average sales volume of each product category. In the above embodiment, by creating one-hot encoding features for each product category, the proportion of each category in the total sales is calculated.

[0092] Specifically, time series features mainly focus on sales trends, seasonal patterns and cyclical changes, and the extraction method includes: a. Creating time-related features, such as year, month, day, day of the week, and whether it is a holiday; b. Calculating rolling statistics, 7-day moving average sales volume, and 30-day sales volume standard deviation; c. Decomposing the time series to extract trends, seasonality and residual components. In the above embodiment, the sales growth rate of each product at different time scales (day, week, month) is calculated, and the weekend effect and holiday effect are identified.

[0093] Specifically, the text features mainly focus on product reviews and social media mentions, and the extraction method includes: a. Using TF-IDF (term frequency-inverse document frequency) to calculate the importance of keywords; b. Performing sentiment analysis to evaluate the positive, negative or neutral tendency of the text; c. Extracting the topic model and using LDA (latent Dirichlet allocation) to identify the main discussion topics. In the above embodiment, by analyzing the most frequently occurring keywords in product reviews, the average sentiment score of each product is calculated, and the main discussion topics related to product quality, price or service are identified.

[0094] Specifically, based on the extracted features, feature selection and dimensionality reduction are performed. The specific methods include: a. Using correlation analysis to remove highly correlated redundant features; b. Applying principal component analysis (PCA) to reduce the dimension of the feature space; c. Using feature importance scoring, in this embodiment, the most relevant features are selected based on the feature importance of random forests. Finally, the selected features are integrated into a unified feature matrix as input for subsequent modeling. This feature matrix contains information from different data sources and different feature types, providing a comprehensive data foundation for the refined model.

[0095] Specifically, through the above feature extraction process, the original business data can be converted into a set of structured and information-rich feature sets. These features not only capture key information such as sales, inventory, customer behavior, etc., but also contain time patterns and text data. This provides a solid foundation for the subsequent establishment of refined models, enabling the subsequent refined models to better understand and predict complex patterns in business operations.

[0096] The feature screening module 103 is used to screen out the most representative features based on the operational features and generate a decision feature set by using linear discriminant analysis.

[0097] In one embodiment, using a data collection tool to collect enterprise business data from multiple sources, generating a business requirement set includes:

[0098] Analyze the specific needs of each business segment of the enterprise and determine the internal and external data sources;

[0099] Specifically, taking the optimization of inventory management and sales strategies in enterprise operations as an example, first, analyze the specific needs of each business segment of the enterprise, and determine the internal data source and the external data source. In this embodiment, the internal data source includes a. sales system (including daily sales records, product information), b. inventory management system (including real-time inventory levels, replenishment records), c. customer relationship management system (including customer purchase history, member information); the external data source includes a. meteorological data API (including daily weather forecast), b. social media platform (including product reviews, brand mentions), c. competitor website (including competing product price information).

[0100] Design a data collection plan based on specific needs and data sources, including collection frequency, time points and data formats;

[0101] Specifically, a data collection plan is designed. In the above embodiment, it includes a. inventory data, which is updated in real time and continuously monitored in JSON format, and includes fields such as product ID, current inventory and last updated time; b. customer data, which is updated daily and updated at 2:00 a.m. every day, and is in the form of a SQL database table, and includes fields such as customer ID, purchase history and membership level; c. competitor price data, which is updated daily and updated at 12:00 noon every day, and is in the form of an Excel file, and includes fields such as date, competitor ID and price.

[0102] Based on the data collection plan, use the selected tools to collect and pre-process data from various data sources to establish a set of business requirements.

[0103] Specifically, based on the data collection plan, use the selected tools to collect and pre-process data from various data sources, and establish a set of business requirements, including a. Internal data collection, first use ETL tools to extract data from sales systems, inventory systems, and CRM systems, then set up automated tasks to perform data extraction according to the planned frequency and time points, and finally perform preliminary cleaning on the extracted data, such as removing duplicate records and processing missing values; b. External data collection, first use scripts to obtain weather data through APIs, then use web crawlers to obtain data from social media platforms and competitor websites, and finally set up scheduled tasks to ensure that data collection is performed as planned; c. Data preprocessing, first perform data cleaning and format unification, then integrate data from different sources into a unified data warehouse, and finally create a data quality inspection process to ensure the integrity and accuracy of the data; d. Establish a set of business requirements, first create a unified view in the data warehouse that contains all relevant data; then design a data model to ensure that the relationships between data sources are correctly mapped, and finally create a data dictionary to record the meaning, source, and update frequency of each field in detail.

[0104] Specifically, through the above-mentioned embodiments, enterprise data from multiple internal and external sources can be systematically collected and integrated, providing a comprehensive data foundation for subsequent feature extraction and decision support, ensuring the timeliness, completeness and relevance of the data, and providing solid data support for the refined management of the enterprise.

[0105] In one embodiment, based on the operational characteristics, linear discriminant analysis is used to screen out the most representative features, and the decision feature set generated includes:

[0106] Preprocess operational features and establish a standard measurement feature set;

[0107] Specifically, first, preprocess the operational features and establish a standard measurement feature set. In this embodiment, there are 10 operational features, including average daily sales, customer satisfaction, inventory turnover, promotion effect, product profit margin, customer retention rate, website traffic, social media interaction rate, market share and employee productivity. These features are standardized and the Z-score method is used to ensure that all features are on the same scale.

[0108] Specifically, then, for each feature x, calculate its standardized value z, expressed as: z=(x-μ) / σ; where μ is the mean value of the feature and σ is the standard deviation; in the above embodiment, for the average daily sales feature, the mean value of the original data is 10,000 yuan and the standard deviation is 2,000 yuan, then a sales volume of 12,000 yuan will be standardized to (12,000-10,000) / 2,000=1.

[0109] Based on the standard metric feature set, a first-level operation matrix is ​​constructed, and the mean vector of each operation feature is calculated;

[0110] Specifically, based on the standard metric feature set, a first-level operation matrix is ​​constructed, and the mean vector of each operation feature is calculated. In the above embodiment, there are three main operation categories: sales performance, customer relationship, and operation efficiency. The 10 standardized features are assigned to these three categories, and the mean vector of each category is calculated.

[0111] Specifically, the sales performance category includes three features: average daily sales, promotion effect, and market share. In the above embodiment, there are 100 samples, and the average values ​​of these 100 samples on these three features are calculated to obtain the mean vector of the sales performance category.

[0112] According to the mean vector, a conversion matrix is ​​constructed, and the first-level operation matrix is ​​converted using the conversion matrix to obtain a second-level operation matrix;

[0113] Specifically, according to the mean vector, a transformation matrix is ​​constructed, and the transformation matrix is ​​used to transform the primary operation matrix to obtain the secondary operation matrix. The construction of the transformation matrix is ​​based on the generalized eigenvalue solution of the inter-class scatter matrix and the intra-class scatter matrix.

[0114] Specifically, first calculate the inter-class scatter matrix and the intra-class scatter matrix , the expression is:

[0115] ;

[0116] ;

[0117] In the formula, C represents the total number of operation characteristics c; Nc represents the number of samples in operation characteristics c, represents the mean vector of the operational feature c; Represents the population mean vector of all samples.

[0118] Specifically, then calculate , the obtained eigenvector w constitutes the transformation matrix W, and finally the transformation matrix W is used to transform the first-level operation matrix to obtain the second-level operation matrix, which is expressed as X2=WX1; where X1 represents the first-level operation matrix; X2 represents the second-level operation matrix.

[0119] Based on the secondary operation matrix, the variance of each operation feature is calculated, and the operation feature with the highest variance is selected to construct the decision feature set.

[0120] Specifically, according to the secondary operation matrix, the features with the highest variance are selected. The features with the highest variance indicate that these features provide more information in distinguishing different categories. Through the above steps, the most representative operation features can be selected according to the size of the variance to construct a decision feature set, which will be used for subsequent data analysis and model training.

[0121] In one embodiment, the expression for calculating the mean vector of each operation feature is:

[0122] ;

[0123] In the formula, represents the mean vector of the operating feature c; Nc represents the number of samples in the operating feature c, The feature vector representing the i-th sample of the operational feature c;

[0124] The expression for calculating the variance of each operating characteristic is:

[0125] ;

[0126] In the formula, represents the variance of the jth feature dimension in the operational feature c; represents the observed value of the i-th sample in the j-th feature dimension in the operational feature c; Represents the average value of all samples in the operation feature c on the jth feature dimension.

[0127] In one embodiment, the model building unit 2 includes a feature analysis and variable definition module 201, an objective function building module 202, and a constraint condition setting module 203;

[0128] The feature analysis and variable definition module 201 is used to analyze the statistical characteristics of each high variance feature based on the constructed decision feature set and define the decision variables;

[0129] An objective function construction module 202 is used to construct a linear objective function based on a decision feature set and a decision variable;

[0130] The constraint condition setting module 203 is used to establish linear constraint conditions according to statistical characteristics and decision variables.

[0131] In one embodiment, based on the constructed decision feature set, the statistical characteristics of each high variance feature are analyzed, and the decision variables are defined to include:

[0132] Based on the decision feature set, the descriptive statistics method is used to obtain the data type and distribution characteristics of each high variance feature;

[0133] Based on the data type and distribution characteristics of each high variance feature, the variable type and value range of each feature are determined by using data type judgment and threshold statistics. The variable types include continuous decision variables and integer decision variables.

[0134] Using standardization and encoding techniques, continuous decision variables are standardized and integer decision variables are encoded.

[0135] Specifically, firstly, based on the decision feature set, the descriptive statistics method is used to obtain the data type and distribution characteristics of each high variance feature. In this embodiment, the five high variance features selected above are analyzed, and the results are: inventory turnover rate (continuous, mean 12.5, standard deviation 3.2, range 5.0-25.0, approximate normal distribution slightly right-skewed); market share (continuous, mean 0.15, standard deviation 0.05, range 0.05-0.30, approximate normal distribution); product profit margin (continuous, mean 0.25, standard deviation 0.08, range 0.10-0.50, approximate normal distribution slightly left-skewed); average daily sales (continuous, mean 50000, standard deviation 15000, range 20000-100000, lognormal distribution); website traffic (discrete, mean 10000, standard deviation 3000, range 5000-20000, Poisson distribution).

[0136] Specifically, based on the data type and distribution characteristics of each high variance feature, the variable type and value range of each feature are determined by using data type judgment and threshold statistics. In the above embodiment, the results of defining the decision variables are as follows: x1 is the inventory turnover rate (continuous decision variable, value range [5.0, 25.0]); x2 is the market share (continuous decision variable, value range [0.05, 0.30]); x3 is the product profit margin (continuous decision variable, value range [0.10, 0.50]); x4 is the average daily sales (continuous decision variable, value range [20000, 100000]); x5 is the average daily website visits (integer decision variable, value range [5000, 20000]).

[0137] Specifically, finally, the standardization and encoding technology is used to standardize the continuous decision variables and encode the integer decision variables. In the above embodiment, the continuous decision variables are subjected to maximum and minimum standardization to obtain: x1_standardization=(x1-5.0) / (25.0-5.0), x2_standardization=(x2-0.05) / (0.30-0.05), x3_standardization=(x3-0.10) / (0.50-0.10), x4_standardization=(x4-20000) / (100000-20000); at the same time, for the integer decision variable x5 (average daily website visits), the interval encoding technology is used to obtain 5000≤x5<10000 encoded as 0, 10000≤x5<15000 encoded as 1, and 15000≤x5≤20000 encoded as 2.

[0138] Specifically, through the above process, the original high-variance features are converted into standardized decision variables. This conversion unifies the scales of different features, retains the distribution characteristics of the original data, and limits the value range to [0, 1] (for continuous variables). For integer variables, interval coding simplifies the problem while retaining the ordinal characteristics of the data. These standardized and encoded decision variables will serve as the basis for subsequent objective function construction and constraint setting, which will help to build a more accurate and effective optimization model, and also facilitate the interpretation and practical application of model results.

[0139] In one embodiment, based on the decision feature set and the decision variables, constructing a linear objective function includes:

[0140] Based on the decision feature set, the weights of decision variables are determined using fuzzy hierarchical analysis;

[0141] Based on the decision variable weights, the initial linear objective function is obtained by weighted summation;

[0142] Using sensitivity analysis, the initial linear objective function is optimized to obtain the final linear objective function.

[0143] Specifically, firstly, based on the decision feature set, the weights of the decision variables are determined by using fuzzy hierarchical analysis. In the above embodiment, the five decision variables are compared in pairs to construct a fuzzy judgment matrix, as shown in Table 1:

[0144] Table 1 Fuzzy judgment matrix

[0145]

[0146] Specifically, the eigenvector is then calculated to obtain the weights of each decision variable: w1 (inventory turnover rate) = 0.263; w2 (market share) = 0.160; w3 (product profit margin) = 0.419; w4 (average daily sales) = 0.096; w5 (average daily website visits) = 0.062. Then, based on the decision variable weights, the initial linear objective function is obtained by weighted summation. In the above embodiment, the following initial objective function is constructed: MaxZ = 0.263x1 + 0.160x2 + 0.419x3 + 0.096x4 + 0.062x5; where x1, x2, x3, x4 are the continuous variables corresponding to the above decision variables after standardization, and x5 is the encoded discrete variable.

[0147] Specifically, the sensitivity analysis is finally used to optimize the initial linear objective function to obtain the final linear objective function. In the above embodiment, the following sensitivity analysis is performed: a. Each weight is changed by ±10% to observe the change in the objective function value; b. The impact of different decision variable combinations on the objective function is analyzed; c. The performance of the objective function in extreme cases (such as a variable taking the maximum or minimum value) is considered; d. Based on the sensitivity analysis results, the initial objective function is fine-tuned. Assuming that the analysis shows that the impact of product profit margins may be slightly overestimated, while the impact of average daily sales is underestimated, the weights are appropriately adjusted to obtain the final linear objective function:

[0148] MaxZ=0.270x1+0.165x2+0.400x3+0.110x4+0.055x5;

[0149] This final linear objective function reflects the combined impact of each decision variable on the company's operations, among which product profit margin is still the most important factor, but the importance of inventory turnover rate and average daily sales has increased slightly. This objective function will be used in the subsequent optimization process to find the best combination of decision variables, so as to achieve refined management of corporate operations.

[0150] Specifically, through the above process, a linear objective function is constructed that takes into account the relative importance of each decision variable and is adjusted through sensitivity analysis. This objective function can better reflect the actual situation of enterprise operations and provide a reliable mathematical basis for subsequent optimization decisions. At the same time, the flexibility of this method also allows the objective function to be adjusted and optimized at any time according to changes in the actual business environment.

[0151] In one embodiment, establishing linear constraints based on statistical characteristics and decision variables includes:

[0152] Based on statistical characteristics, confidence intervals are calculated and constraints on variable value ranges are constructed;

[0153] Based on business rules, business logic constraints are obtained by using linear inequality transformation;

[0154] According to the relationship between decision variables, strongly correlated variable pairs are determined and variable association constraints are established.

[0155] Specifically, firstly, based on the statistical characteristics, the confidence interval is calculated and the variable value range constraint is constructed. In this embodiment, the 95% confidence interval is calculated for each decision variable and used as the variable value range constraint, including a. Inventory turnover rate (x1), mean 12.5, standard deviation 3.2, 95% confidence interval [6.228, 18.772], constraint 6.228 ≤ x1 ≤ 18.772; b. Market share (x2), mean 0.15, standard deviation 0.05, 95% confidence interval [0.052, 0.248], constraint 0.052 ≤ x2 ≤ 0.248; c. Product profit margin (x3), mean 0.25, standard deviation 0.08, 95% confidence interval [0.093, 0.407], constraint 0.093≤x3≤0.407; d. Average daily sales (x4), mean 50,000, standard deviation 15,000, 95% confidence interval [20,600, 79,400], constraint 20,600≤x4≤79,400; e. Average daily website visits (x5), mean 10,000, standard deviation 3,000, 95% confidence interval [4,120, 15,880], constraint 4,120≤x5≤15,880.

[0156] Specifically, based on the business rules, linear inequality conversion is used to obtain business logic constraints. In this embodiment, the following business rules are considered and converted into linear constraints: a. The trade-off between product profit margin and market share. It is set that for every 0.01 increase in profit margin, the market share will decrease by 0.005, and the constraint condition is: x2+0.5x3≤0.3; b. The relationship between inventory turnover rate and average daily sales. It is set that for every increase in average daily sales of 10,000, the inventory turnover rate will increase by at least 0.5, and the constraint condition is: x1≥0.5(x4 / 10000)+5; c. The relationship between website traffic and market share. It is set that for every increase in website traffic of 1,000, the market share will increase by at least 0.002, and the constraint condition is: x2≥0.002(x5 / 1000)+0.05.

[0157] Specifically, according to the relationship between the decision variables, strongly correlated variable pairs are determined, and variable association constraints are established. In this embodiment, the following strongly correlated variable pairs are identified and constraints are established: a. Average daily sales (x4) and product profit margin (x3), it is assumed that the two are positively correlated, and when the average daily sales increase, the profit margin increases less, and the constraint condition is obtained: 0.8x3+0.0000005x4≤0.5; b. Inventory turnover rate (x1) and market share (x2), it is assumed that the two are positively correlated, but the growth rates are different, and the constraint condition is obtained: 0.01x1+0.1x2≤0.4; c. Average daily website visits (x5) and average daily sales (x4), it is assumed that an increase in website traffic drives an increase in sales, but there is an upper limit, and the constraint condition is obtained: x4≤5x5+10000.

[0158] Specifically, through the above process, a series of linear constraints are established, including variable value range constraints, business logic constraints, and variable association constraints. These constraints reflect the actual limitations of business operations and the complex relationships between variables. In the above embodiment, the trade-off constraint between product profit margin and market share reflects the reality that increasing profit margin may lead to a decrease in market share; the relationship constraint between inventory turnover rate and average daily sales ensures that inventory management efficiency is improved accordingly when sales increase; the relationship constraint between website traffic and market share reflects the impact of online marketing on market share.

[0159] Specifically, these constraints, together with the objective function constructed above, form a complete linear programming model. This model not only takes into account the goals pursued by the enterprise (through the objective function), but also reflects various restrictions and balances in reality (through the constraints). By solving this model, we can find the decision variable values ​​that can maximize the objective function under given constraints, thereby providing specific and actionable decision suggestions for the refined management of the enterprise.

[0160] like Figure 2According to another embodiment of the present invention, a digital enterprise operation refined management method based on big data is provided. The digital enterprise operation refined management method based on big data includes the following steps:

[0161] S1. Based on business needs, collect enterprise business data, extract operational features, and build a decision feature set;

[0162] S2. Based on the decision feature set, a refined model is established, wherein the refined model includes decision variables, objective functions and constraints;

[0163] S3. Use the cutting plane method to solve the refined model, obtain the optimal decision variables, and apply them to actual business.

[0164] In summary, with the help of the above technical solutions of the present invention, the present invention realizes the collection and efficient processing of multi-source channel data through the feature engineering unit. With the help of the feature extraction module and the feature screening module, the system can extract multi-dimensional operational features including numerical values, classifications, time series and texts, and use linear discriminant analysis to screen out the most representative features. This refined feature engineering provides a high-quality data foundation for subsequent decision support and significantly improves the accuracy of enterprise operation decisions. At the same time, the present invention realizes the transformation from features to decision variables and the flexible construction of objective functions and constraints through the feature analysis and variable definition module, the objective function construction module and the constraint setting module. At the same time, using the model The fuzzy hierarchy analysis determines the weights of decision variables, and the method of optimizing the objective function through sensitivity analysis enables the system to construct the most suitable optimization model according to the specific situation of the enterprise, providing a strong theoretical support for the refined management of enterprise operations; in addition, the present invention realizes the efficient transformation from theoretical model to practical application through the combination of initial solution acquisition module, cutting plane method solution module and management and application module, especially the method of solving integer programming problems by cutting plane method and obtaining the final optimization solution through sensitivity analysis, so that the system can quickly obtain an implementable optimization solution and directly apply it to actual business. This seamless connection from theory to practice greatly improves the efficiency and accuracy of enterprise operation management.

[0165] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A digital enterprise operation refined management system based on big data, characterized in that: The digital enterprise operation refined management system based on big data includes a feature engineering unit (1), a model building unit (2) and an optimization solution unit (3); The feature engineering unit (1) is used to collect enterprise business data based on business needs, extract operational features, and construct a decision feature set; The feature engineering unit (1) comprises a data acquisition module (101), a feature extraction module (102) and a feature screening module (103); The data collection module (101) is used to collect enterprise business data from multiple source channels using data collection tools to generate a business demand set; The data collection tool is used to collect enterprise business data from multiple sources and channels, and generate a business requirement set including: Analyze the specific needs of each business segment of the enterprise and determine the internal and external data sources; Design a data collection plan based on specific needs and data sources, including collection frequency, time points, and data formats; Based on the data collection plan, use the selected tools to collect and pre-process data from various data sources to establish a set of business requirements; The feature extraction module (102) is used to extract operational features based on the business requirement set using feature engineering technology, wherein the operational features include numerical features, classification features, time series features, and text features; The numerical features include sales volume, inventory level and price, and the extraction method thereof includes: a. calculating basic statistics, including mean, median, standard deviation, maximum value and minimum value; b. performing normalization processing, using Min-Max scaling or Z-score standardization; c. creating interactive features, including the ratio of sales volume to price; for each product, calculating the average sales volume, standard deviation of sales volume and price elasticity in the past 30 days; The classification features include product categories, customer types, and promotional activities, and the extraction method thereof includes: a. performing one-hot encoding to convert the classification variables into binary features; b. calculating frequency encoding to replace the original categories with the frequencies of the categories appearing in the data set; c. creating grouped statistical features, including the average sales volume of each product category; by creating one-hot encoding features for each product category, calculating the proportion of each category in the total sales volume; The time series features include sales trends, seasonal patterns and cyclical changes, and the extraction method includes: a. creating time-related features, such as year, month, day, day of the week, and whether it is a holiday; b. calculating rolling statistics, such as 7-day moving average sales volume and 30-day sales volume standard deviation; c. performing time series decomposition to extract trend, seasonality and residual components; calculating the sales growth rate of each product at different time scales, and identifying weekend effects and holiday effects; The text features include product reviews and social media mentions, and the extraction method thereof includes: a. using TF-IDF to calculate the importance of keywords; b. performing sentiment analysis to evaluate the positive, negative or neutral tendency of the text; c. extracting a topic model and using LDA to identify the main discussion topics; by analyzing the most frequently occurring keywords in product reviews, calculating the average sentiment score for each product, and identifying discussion topics related to product quality, price or service; The feature screening module (103) is used to screen out the most representative features based on the operational features by using linear discriminant analysis to generate a decision feature set; Based on the operational characteristics, the most representative features are screened out by using linear discriminant analysis to generate a decision feature set including: Preprocess operational features and establish a standard measurement feature set; Based on the standard metric feature set, a primary operation matrix is ​​constructed, and the mean vector of each operation feature is calculated; the expression for calculating the mean vector of each operation feature is: ； In the formula, μ c Indicates operational characteristics c The mean vector of ; N c Indicates operational characteristics c The number of samples in x i (c) Indicates operational characteristics c No. i The feature vector of samples; According to the mean vector, a conversion matrix is ​​constructed, and the first-level operation matrix is ​​converted using the conversion matrix to obtain a second-level operation matrix; Based on the secondary operation matrix, the variance of each operation feature is calculated, and the operation feature with the highest variance is selected to construct a decision feature set; the expression for calculating the variance of each operation feature is: ； In the formula, Indicates operational characteristics c Middle j The variance of the feature dimension; Indicates operational characteristics c Middle i The sample in j Observations on feature dimensions; Indicates operational characteristics c All samples in j The average value of the feature dimensions; The model building unit (2) is used to establish a refined model based on the decision feature set, wherein the refined model includes decision variables, objective functions and constraint conditions; The model building unit (2) comprises a feature analysis and variable definition module (201), an objective function building module (202) and a constraint condition setting module (203); The feature analysis and variable definition module (201) is used to analyze the statistical characteristics of each high variance feature based on the constructed decision feature set, and define decision variables; The decision feature set constructed based on the above method analyzes the statistical characteristics of each high variance feature and defines the decision variables including: Based on the decision feature set, the descriptive statistics method is used to obtain the data type and distribution characteristics of each high variance feature; Based on the data type and distribution characteristics of each high variance feature, the variable type and value range of each feature are determined by using data type judgment and threshold statistics, wherein the variable type includes continuous decision variables and integer decision variables; Using standardization and encoding techniques, continuous decision variables are standardized and integer decision variables are encoded; The objective function construction module (202) is used to construct a linear objective function based on a decision feature set and a decision variable; The construction of a linear objective function based on the decision feature set and the decision variables includes: Based on the decision feature set, the weights of decision variables are determined using fuzzy hierarchical analysis; Based on the decision variable weights, the initial linear objective function is obtained by weighted summation; Using sensitivity analysis, the initial linear objective function is optimized to obtain the final linear objective function; The constraint condition setting module (203) is used to establish linear constraint conditions according to statistical characteristics and decision variables; The establishment of linear constraints based on statistical characteristics and decision variables includes: Based on statistical characteristics, confidence intervals are calculated and constraints on variable value ranges are constructed; Based on business rules, business logic constraints are obtained by using linear inequality transformation; According to the relationship between decision variables, strongly correlated variable pairs are determined and variable association constraints are established; The optimization solving unit (3) is used to solve the refined model using the cutting plane method to obtain the optimal decision variables and apply them to actual business; The optimization solution unit (3) comprises an initial solution acquisition module (301), a cutting plane method solution module (302) and a management and application module (303); The initial solution acquisition module (301) is used to obtain an initial feasible solution using the simplex method according to the objective function and the constraint conditions; The cutting plane method solution module (302) is used to obtain an integer feasible solution based on the initial feasible solution using the cutting plane method; The management and application module (303) is used to obtain the final optimization solution based on the integer feasible solution and apply it to the business by using sensitivity analysis.

2. A digital enterprise operation refined management method based on big data, used to implement the digital enterprise operation refined management system based on big data as claimed in claim 1, characterized in that: The digital enterprise operation refined management method based on big data includes the following steps: S1. Based on business needs, collect enterprise business data, extract operational features, and build a decision feature set; S2. Based on the decision feature set, a refined model is established, wherein the refined model includes decision variables, objective functions and constraints; S3. Use the cutting plane method to solve the refined model, obtain the optimal decision variables, and apply them to actual business.

Images