Enterprise health monitoring and evaluation method and system based on multi-dimensional analysis

Abstract

The invention provides an enterprise health monitoring and evaluation method and system based on multi-dimensional analysis, and the method comprises the steps: obtaining enterprise data in a statistical period, the enterprise data comprising production data, employment data, order data, fund data and research and development data; preprocessing the enterprise data; performing feature extraction on the preprocessed production data, order data, fund data and research and development data to determine production features, order features, fund features and research and development features; and on the basis of the production features, the order features, the fund features and the research and development features, carrying out cross-dimensional feature conjoint analysis, and determining a multi-dimensional health score of the enterprise so as to realize multi-dimensional evaluation of the enterprise and provide decision assistance for the enterprise.

Description

Technical Field

[0001] This application relates to the field of enterprise status assessment, and more specifically, to an enterprise health monitoring and assessment method and system based on multi-dimensional analysis. Background Technology

[0002] As a service platform serving enterprise clients, providing status assessments and decision-making support is a powerful way for the platform to enhance its competitiveness. However, traditional enterprise status assessments primarily analyze only financial performance, making it difficult to conduct comprehensive, multi-dimensional analyses of the enterprise.

[0003] Factors influencing a company's status include production status, order data, cash flow, R&D direction and results, and many others. These factors reflect the company's development prospects and current state, involving numerous indicators. Furthermore, different indicators across different dimensions often exhibit correlations; for example, order data affects production status, and vice versa. Therefore, achieving a comprehensive, multi-dimensional assessment of a company is a challenging problem. Summary of the Invention

[0004] The purpose of this application is to provide a method and system for enterprise health monitoring and assessment based on multi-dimensional analysis, so as to realize multi-dimensional assessment of enterprises and provide decision-making assistance for enterprises.

[0005] To achieve the above objectives, the embodiments of this application are implemented in the following manner: This application provides a method for enterprise health monitoring and assessment based on multi-dimensional analysis, including: acquiring enterprise data within a statistical period, wherein the enterprise data includes production data, employment data, order data, financial data, and R&D data; preprocessing the enterprise data; extracting features from the preprocessed production data, order data, financial data, and R&D data to determine production features, order features, financial features, and R&D features; and performing cross-dimensional feature joint analysis based on the production features, order features, financial features, and R&D features to determine the enterprise's multi-dimensional health score.

[0006] Furthermore, feature extraction is performed on the preprocessed production data, order data, financial data, and R&D data to determine production features, order features, financial features, and R&D features. This includes: feature extraction of preprocessed production data to determine production features; feature extraction of preprocessed order data to determine order features; feature extraction of preprocessed financial data to determine financial features; and feature extraction of preprocessed R&D data to determine R&D features.

[0007] Furthermore, the production data includes raw material consumption, energy consumption, planned production time, actual production time, daily number of workers, total daily working hours, output of each product, pass rate, scrap rate, and rework rate, the conversion relationship between each product and the standard product, and the production time of the standard product. Feature extraction is performed on the preprocessed production data to determine production characteristics, including: determining the output, pass rate, scrap rate, and rework rate of the standard product based on the output, pass rate, scrap rate, and rework rate of each product, and the conversion relationship between each product and the standard product; determining the time utilization rate based on the planned production time and actual production time; determining performance efficiency based on the actual production time, the output of the standard product, and the production time of the standard product; and determining performance efficiency based on the time utilization rate, performance efficiency, and standard product output. The overall equipment efficiency is determined by the product qualification rate; the output per unit of labor hour is determined based on the product output of standard products and the total daily working hours; the output per person per day is determined based on the product output of standard products and the number of workers employed daily; the raw material output rate of standard products is determined based on the product output, qualification rate, and raw material consumption of standard products; and the energy consumption per unit of output is determined based on energy consumption and product output of standard products. The numerical characteristics of raw material consumption, energy consumption, planned production time, actual production time, number of workers employed daily, total daily working hours, product output of standard products, qualification rate, scrap rate, and rework rate, as well as time utilization rate, performance efficiency, overall equipment efficiency, output per unit of labor hour, output per person per day, raw material output rate, and energy consumption per unit of output of standard products are extracted and combined to form production characteristics.

[0008] Furthermore, each order data point includes the order creation date, order deadline, order completion date, customer information, product type, quantity, amount, and payment amount. Feature extraction is performed on the preprocessed order data to determine order characteristics, including: determining the order delay duration based on the order deadline and completion date; determining the order processing cycle based on the order creation and completion dates; and determining the on-time delivery rate, delayed delivery rate, average processing cycle, and average delayed delivery rate based on the order delay duration and total order volume. Duration; Based on the order customer information for each order, determine the total number of customers and the number of repeat customers; Based on the order product type and order product quantity for each order, determine the average quantity of products per order and the total product sales; Based on the order product amount for each order, determine the total product sales; Based on the order product amount and order payment amount for each order, determine the order payment ratio; Extract numerical features of order quantity, on-time delivery rate, late delivery rate, average order processing cycle, average late delivery duration, total number of customers, number of repeat customers, average quantity of products per order, total product sales, total product sales, and order payment ratio, and concatenate them to form order features.

[0009] Furthermore, the financial data is used for feature extraction. Feature extraction is performed on the preprocessed financial data to determine its characteristics, including: identifying the financial data and determining the values ​​of various indicators; determining net profit margin, return on equity, accounts receivable turnover, inventory turnover, cash flow ratio, cash ratio, revenue growth rate, profit cash ratio, and revenue cash ratio based on these indicators; and extracting the values ​​of multiple financial indicators and the numerical characteristics of net profit margin, return on equity, accounts receivable turnover, inventory turnover, cash flow ratio, cash ratio, revenue growth rate, profit cash ratio, and revenue cash ratio, and then combining them to form the financial characteristics.

[0010] Furthermore, the R&D data includes the number of R&D personnel, R&D working hours, R&D investment, and the number of R&D achievements. Feature extraction is performed on the preprocessed R&D data to determine R&D characteristics, including: determining R&D intensity based on R&D investment and operating revenue; determining the R&D personnel ratio based on the number of R&D personnel and the total number of employees; determining per capita R&D investment based on the number of R&D personnel and R&D investment; determining per capita achievement output based on the number of R&D achievements and the number of R&D personnel; and determining the average working hours per unit of achievement based on the number of R&D achievements and R&D working hours. Numerical features of the number of R&D personnel, R&D working hours, R&D investment, number of R&D achievements, R&D intensity, R&D personnel ratio, per capita R&D investment, per capita achievement output, and average working hours per unit of achievement are extracted and concatenated to form the R&D characteristics.

[0011] Furthermore, based on production characteristics, order characteristics, capital characteristics, and R&D characteristics, cross-dimensional feature joint analysis is conducted to determine the enterprise's multi-dimensional health score. This includes: assigning values ​​to each parameter in the production, order, capital, and R&D characteristics to obtain normalized production feature vectors, order feature vectors, capital feature vectors, and R&D feature vectors; and using an attention mechanism to conduct cross-dimensional feature joint analysis on the normalized production feature vectors, order feature vectors, capital feature vectors, and R&D feature vectors to determine the enterprise's multi-dimensional health score.

[0012] Furthermore, an attention mechanism is used to perform cross-dimensional joint feature analysis on the normalized production feature vector, order feature vector, capital feature vector, and R&D feature vector to determine the enterprise's multi-dimensional health score. This includes: determining the production dimension score based on the normalized production feature vector, order feature vector, capital feature vector, and R&D feature vector, using multiple sets of production weight matrices in the trained attention mechanism; determining the order dimension score based on the normalized production feature vector, order feature vector, capital feature vector, and R&D feature vector, using multiple sets of order weight matrices in the trained attention mechanism; determining the capital dimension score based on the normalized production feature vector, order feature vector, capital feature vector, and R&D feature vector, using multiple sets of R&D weight matrices in the trained attention mechanism; and determining the R&D dimension score based on the normalized production feature vector, order feature vector, capital feature vector, and R&D feature vector, using multiple sets of R&D weight matrices in the trained attention mechanism. Finally, a weighted sum of the production dimension score, order dimension score, capital dimension score, and R&D dimension score is performed to obtain the enterprise's multi-dimensional health score.

[0013] Furthermore, based on normalized production feature vectors, order feature vectors, funding feature vectors, and R&D feature vectors, and utilizing multiple sets of production weight matrices in the trained attention mechanism, the production dimension score is determined, including: Based on the normalized production feature vector, order feature vector, and the production-order weight matrix in the production weight matrix, calculate the production-order enhanced features: , , in, To enhance the production-order feature, For the production-order attention weight vector, To generate feature vectors, For order feature vectors, This indicates element-wise addition. This indicates element-wise multiplication. This represents the query matrix between production feature vectors and order feature vectors. This represents the key matrix between the production feature vector and the order feature vector. This represents the value matrix between the production feature vector and the order feature vector; Based on the normalized production feature vector, capital feature vector, and the production-capital weight matrix in the production weight matrix, calculate the production-capital enhancement feature: , , in, For production-capital enhancement features, For the production-capital attention weight vector, To generate feature vectors, For the feature vector of funds, This indicates element-wise addition. This indicates element-wise multiplication. This represents the query matrix between the production feature vector and the capital feature vector. This represents the key matrix between the production feature vector and the capital feature vector. A value matrix representing the relationship between the production feature vector and the capital feature vector; Based on the normalized production feature vector, R&D feature vector, and the production-R&D weight matrix in the production weight matrix, calculate the production-R&D enhancement features: , , in, Enhanced features for production and R&D For the production-R&D attention weight vector, To generate feature vectors, To develop feature vectors, This indicates element-wise addition. This indicates element-wise multiplication. This represents the query matrix between production feature vectors and R&D feature vectors. This represents the key matrix between the production feature vector and the R&D feature vector. A value matrix representing the relationship between production feature vectors and R&D feature vectors; The production dimension score is obtained by using a trained production dimension regressor to perform nonlinear mapping and scoring on the production feature vector, production-order enhancement feature, production-fund enhancement feature and production-R&D enhancement feature. The production dimension regressor is a fully connected neural network with no more than 3 layers.

[0014] Furthermore, this application embodiment also provides an enterprise health monitoring and assessment system based on multi-dimensional analysis, including: a data acquisition unit for acquiring enterprise data within a statistical period, wherein the enterprise data includes production data, employment data, order data, financial data, and R&D data; a preprocessing unit for preprocessing the enterprise data; a feature extraction unit for extracting features from the preprocessed production data, order data, financial data, and R&D data to determine production features, order features, financial features, and R&D features; and a multi-dimensional scoring unit for performing cross-dimensional feature joint analysis based on production features, order features, financial features, and R&D features to determine the enterprise's multi-dimensional health score.

[0015] Beneficial effects: This solution acquires and preprocesses enterprise data (including production, employment, order, financial, and R&D data) over a statistical period (e.g., one month, three months, six months, one year). Then, it extracts features from the preprocessed production, order, financial, and R&D data (the feature extraction process mainly involves acquiring various data and calculating multiple indicators) to determine production, order, financial, and R&D characteristics. Based on these characteristics, a cross-dimensional feature joint analysis is performed to determine the enterprise's multi-dimensional health score. This approach enables multi-dimensional enterprise evaluation, considers the correlations between several indicators within each dimension, and achieves cross-dimensional feature joint analysis to calculate the enterprise's multi-dimensional health score, thus providing decision-making support for the enterprise.

[0016] By assigning values ​​to each parameter in the production, order, capital, and R&D features (considering the impact of each indicator on the enterprise's status, with values ​​between 0 and 1; the larger the value, the healthier the enterprise's status in that indicator), normalized production feature vectors, order feature vectors, capital feature vectors, and R&D feature vectors are obtained. Then, an attention mechanism is used to perform cross-dimensional feature joint analysis on the normalized production feature vectors, order feature vectors, capital feature vectors, and R&D feature vectors: by introducing an end-to-end jointly trained attention mechanism + regressors for each dimension, cross-dimensional feature joint analysis is performed. Specifically, using the normalized production feature vectors, order feature vectors, capital feature vectors, and R&D feature vectors, and using multiple sets of weight matrices in the trained attention mechanism, the order dimension score, capital dimension score, and R&D dimension score are determined. Specifically, when analyzing scores for a particular dimension, the weight matrices obtained after training between each pair of dimensions are used to calculate an enhancement matrix (which takes into account the influence of other dimensions on the features of this dimension, thus achieving feature enhancement). Simultaneously, the feature vectors of that dimension are concatenated and input into the corresponding dimension's regressor (a small, fully connected neural network with no more than three layers) for score determination. Finally, the determined scores for production, orders, funding, and R&D are weighted (the weights are determined by the weights allocated to each dimension and the number of indicators contained in each dimension) and summed to obtain the enterprise's multi-dimensional health score. This approach enables cross-dimensional joint analysis of feature vectors from different dimensions. The creative application of the attention mechanism to this scenario (originally, attention mechanisms were mainly used in matrix analysis, while in this scenario, feature vectors) considers the potential positive or negative interactions between indicators, deriving the enhancement matrix for the corresponding dimension. However, matrices are difficult to quantify scores simply by using matrices; therefore, a regressor combined with the attention mechanism is designed for joint training to achieve accurate quantification of scores for each dimension. This approach enables cross-dimensional joint analysis and status scoring for each dimension. Compared to traditional single-dimensional analysis methods, this solution can achieve multi-dimensional assessment, more accurately evaluate the health status of enterprises, and provide assistance for enterprise decision-making.

[0017] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described in detail below. Detailed Implementation

[0018] The technical solutions in the embodiments of this application will be described below.

[0019] This embodiment provides a multi-dimensional analysis-based enterprise health monitoring and assessment method, which is applied to electronic devices and is mainly aimed at enterprises in the unit's service platform. It includes steps S10, S20, S30 and S40.

[0020] In this embodiment, in order to achieve multi-dimensional health monitoring and evaluation of enterprises (this embodiment mainly focuses on product manufacturing enterprises; for service enterprises, the evaluation method in this embodiment is not very applicable and requires another evaluation mechanism, which will not be elaborated in this application), the electronic device can first run step S10.

[0021] Step S10: Obtain enterprise data within the statistical period, including production data, employment data, order data, financial data, and R&D data.

[0022] In this embodiment, the electronic device can acquire enterprise data of the target enterprise within a statistical period (e.g., one month, three months, six months, one year, etc.). The enterprise data includes production data, employment data, order data, financial data, and R&D data. The enterprise needs to collect this enterprise data and upload it to the system.

[0023] Production data mainly includes raw material consumption, energy consumption, planned production time, actual production time, daily number of workers, total daily working hours, output of each product, pass rate, scrap rate, and rework rate.

[0024] Raw material consumption represents the types of raw materials and their corresponding consumption amounts (in kilograms or tons) within the statistical period. Energy consumption represents electricity consumption (in kilowatt-hours). Planned production time is the planned production time for standard products (in days or hours). Actual production time is the actual production time for standard products (in days or hours). Daily number of workers, total daily working hours (the sum of working hours for each worker), output of each product (the quantity of each product), pass rate of each product (first-pass rate, i.e., the ratio of the number of products of this type that pass the inspection on the first attempt to the total quantity of products of this type), scrap rate (the ratio of the number of scrapped products of this type to the total quantity of products of this type), and rework rate (the ratio of the number of products of this type that fail the inspection but are not scrapped to the total quantity of products of this type).

[0025] Order data includes the following for each order: order creation date (the date the order is generated, down to the hour), order deadline (the latest delivery date for the order, down to the hour; unless otherwise specified, the deadline is the last hour of that date), order completion date (the date the order is completed, down to the hour), order customer information (the customer unit for this order, but only used to distinguish between new and returning customers; specific customer information must not be disclosed), order product type (the type of products ordered in this order), order product quantity (the quantity of each product type in this order), order product amount (the total amount of this order), and order payment amount (the amount already paid for this order).

[0026] Financial data mainly consists of three sections: data related to the cash flow statement, data related to the income statement, and data related to the balance sheet.

[0027] The cash flow statement includes data on cash flows from operating activities, investing activities, financing activities, and the beginning and ending amounts of cash. Each type of cash flow data contains many sub-items, such as cash received from sales of goods and services, tax refunds, and other cash related to operating activities; cash paid for purchases of goods and services; cash paid to and on behalf of employees; taxes paid; and other cash related to operating activities. The list is extensive and can be found in the standard cash flow statement.

[0028] The income statement data includes operating revenue, operating costs, taxes and surcharges, selling expenses, administrative expenses, R&D expenses, financial expenses, credit impairment losses, asset impairment losses, fair value change gains, investment income, operating profit, non-operating income, non-operating expenses, total profit, income tax expense, and net profit.

[0029] The balance sheet data includes changes in asset items (such as increases or decreases in accounts receivable and notes receivable, increases or decreases in inventory, and increases or decreases in prepayments), changes in liability items (such as increases or decreases in accounts payable and notes payable, and increases or decreases in advances from customers), and changes in equity items (such as changes in share capital and capital reserves).

[0030] Research and development data includes the number of R&D personnel, R&D working hours (total working hours of all R&D personnel within the statistical period), R&D investment (total investment amount converted within the statistical period), and the number of R&D achievements (the number of intellectual property achievements produced within the statistical period, such as trademarks, patents, copyrights, technical documents, etc.).

[0031] After obtaining the enterprise data, the electronic device can proceed to step S20.

[0032] Step S20: Preprocess the enterprise data.

[0033] In this embodiment, the electronic device can perform data cleaning (including missing value handling and outlier handling) and data transformation (e.g., unifying units and date formats) on enterprise data. This embodiment does not introduce data standardization (or normalization) processing, but standardization of some data can be introduced, which is not limited here.

[0034] After the preprocessing is completed, the electronic device can proceed to step S30.

[0035] Step S30: Extract features from the preprocessed production data, order data, financial data, and R&D data to determine the production features, order features, financial features, and R&D features.

[0036] In this embodiment, the electronic device can extract features from the preprocessed production data to determine the production characteristics.

[0037] For example, electronic devices can determine the output, pass rate, scrap rate, and rework rate of standard products based on the output, pass rate, scrap rate, and rework rate of each product, as well as the conversion relationship between each product and the standard product.

[0038] For example, to facilitate analysis, electronic devices can utilize the conversion relationships between each product and a standard product (e.g., product A = 1.0 standard product, product B = 0.8 standard products, product C = 1.5 standard products, etc.). Using this conversion relationship, the output, pass rate, scrap rate, and rework rate of each product can be converted into the output, pass rate, scrap rate, and rework rate of the standard product. After summing them up, the output, pass rate, scrap rate, and rework rate of the standard product can be obtained.

[0039] For example, electronic devices can determine their time-to-work rate based on planned production time and actual production time. For instance, the time-to-work rate can be obtained as the ratio of actual production time to planned production time.

[0040] For example, electronic devices can determine their performance efficiency based on actual production time, the output of standard products, and the production time of standard products (i.e., standard cycle time, representing the production time of a single standard product). For instance, the performance efficiency of an electronic device can be calculated as (total output × standard cycle time) / actual production time.

[0041] For example, the overall efficiency of an electronic device can be determined based on time utilization rate, performance efficiency, and the yield rate of standard products. For instance, the overall efficiency of an electronic device can be calculated as: time utilization rate × performance efficiency × yield rate.

[0042] For example, electronic devices can determine the output per unit of labor hour based on the production output of standard products and the total daily labor hours. For instance, the output per unit of labor hour can be calculated by the ratio of the total production output of standard products to the total daily labor hours. For example, electronic devices can determine the daily output per employee based on the production volume of standard products and the number of employees employed daily. For instance, the daily output per employee can be calculated as the ratio of the total production volume of standard products to the number of employees employed daily.

[0043] For example, electronic devices can determine the raw material yield of a standard product based on its output, pass rate, and raw material consumption. For instance, electronic devices can calculate the yield rate of several different raw materials by using the formula: (output of the standard product × pass rate of the standard product / raw material consumption) × 100%.

[0044] For example, electronic devices can determine the energy consumption per unit of output based on energy consumption and the output of standard products. For instance, an electronic device can calculate the energy consumption per unit of output by dividing the energy consumption by the total output of standard products.

[0045] Subsequently, the electronic equipment can extract the following numerical characteristics: raw material consumption (e.g., converted into monetary value based on raw material costs, in ten thousand yuan), energy consumption, planned production time, actual production time, daily number of workers, total daily working hours, production time of standard products, product output of standard products, pass rate of standard products, scrap rate of standard products and rework rate of standard products, time utilization rate, performance efficiency, overall equipment efficiency, output per unit working hour, output per person per day, raw material output rate, and energy consumption per unit output of standard products. These values ​​are then pieced together to form production characteristics.

[0046] In this embodiment, the electronic device can extract features from the preprocessed order data to determine the order features.

[0047] For example, an electronic device can determine the order delay duration (in days or hours) for each order based on the order deadline and order completion date by subtracting the order deadline from the order completion date. If the difference between the order completion date and the order deadline is negative, it is taken as zero, which is considered as no delay.

[0048] For example, an electronic device can determine the order processing cycle (in days or hours) for each order based on the order creation date and the order completion date of each order data by subtracting the order creation date from the order completion date.

[0049] For example, electronic devices can determine the on-time delivery rate (number of orders with zero delay divided by total order quantity), the delayed delivery rate (number of orders with non-zero delay divided by total order quantity), the average order processing cycle (sum of order processing cycles for each order data point divided by total order quantity) and the average delayed delivery time (sum of delay times for orders with non-zero delay divided by the number of orders with non-zero delay) based on the order delay duration and total order quantity for each order data point.

[0050] For example, electronic devices can determine the total number of customers and the number of repeat customers based on the order customer information of each order. The customer information is mainly used to distinguish between new customers and repeat customers, and does not involve the leakage of customer information (or, the customer information in the enterprise data is directly marked with an identifier, 1 represents old customers, 2 represents new customers, and there is no risk of customer information leakage), so the total number of customers and the number of repeat customers can be counted.

[0051] For example, electronic devices can determine the total product sales (the sum of the product quantities converted to standard products for each order) and the average number of products per order (which requires dividing the total product sales by the total order volume) based on the order product type and order product quantity for each order.

[0052] For example, electronic devices can determine the total product sales amount (summing up the product sales amount of each order) based on the order product amount of each order data. Of course, they can also determine the average sales amount per order (total product sales amount divided by the total number of orders).

[0053] For example, electronic devices can determine the order return ratio based on the order product amount and order payment amount for each order. For instance, the order return ratio can be obtained by dividing the sum of all order product amounts (i.e., total product sales) by the sum of all order payment amounts.

[0054] Based on this, electronic devices can extract numerical features such as order quantity, on-time delivery rate, delayed delivery rate, average order processing cycle, average delayed delivery duration, total number of customers, number of repeat customers, average number of products per order, total product sales, total product sales amount, and order payment collection ratio, and then combine them to form order features.

[0055] In this embodiment, the electronic device can extract features from the preprocessed order data to determine the order features.

[0056] For example, electronic devices can identify financial data (e.g., using natural language processing technology; if it is a scanned document, it is also necessary to first use optical character recognition technology to recognize characters in the scanned document, convert the scanned document into text or tabular data, and then use natural language processing technology for recognition), and identify the financial data as structured data, thereby determining the values ​​of various indicators of the financial data.

[0057] After obtaining the various indicators and corresponding values ​​of the financial data, the electronic device can determine the net profit margin, return on net assets, accounts receivable turnover, inventory turnover, cash flow ratio, cash ratio, operating revenue growth rate, profit cash ratio, and revenue cash ratio based on the values ​​of the various indicators of the financial data. The calculation method is shown in Table 1.

[0058] Table 1. Calculation methods for each feature

[0059] The data for the above indicators comes from financial statements. Some data can be directly obtained from the financial data, while others require intermediate calculations. For example, the net profit and operating revenue needed to calculate the net profit margin can be directly obtained from the income statement; the net profit needed to calculate the return on equity can also be directly obtained from the income statement. However, the average shareholders' equity needs to be calculated by dividing the sum of the beginning and ending shareholders' equity by 2, and the data comes from the balance sheet. The calculation of these indicators is very common in the calculation of existing financial data indicators, and these indicators are relatively effective in measuring the financial and operational health of a company. Therefore, they will not be elaborated on here.

[0060] Accordingly, electronic devices can extract multiple indicators from financial data (including at least some important data such as operating revenue, operating costs, net profit, accounts receivable, inventory, net operating cash flow, etc.) and the numerical characteristics of net profit margin, return on net assets, accounts receivable turnover, inventory turnover, cash flow ratio, cash ratio, operating revenue growth rate, profit cash ratio, and revenue cash ratio, and then combine them to form financial characteristics.

[0061] In this embodiment, the electronic device can extract features from the preprocessed R&D data to determine the R&D features.

[0062] For example, for electronic devices, the R&D intensity can be determined based on R&D investment and operating revenue (obtained by dividing R&D investment by operating revenue). The R&D personnel ratio can be determined based on the number of R&D personnel and the total number of employees (obtained by dividing the number of R&D personnel by the total number of employees). The R&D investment per person can be determined based on the number of R&D personnel and R&D investment (obtained by dividing R&D investment by the number of R&D personnel). The output per person can be determined based on the number of R&D achievements and the number of R&D personnel (obtained by dividing the number of R&D achievements by the number of R&D personnel); and the average working hours per unit of output can be determined based on the number of R&D achievements and R&D working hours (obtained by dividing R&D working hours by the number of R&D achievements).

[0063] Based on this, electronic devices can extract numerical features such as the number of R&D personnel, R&D working hours, R&D investment, number of R&D results, R&D intensity, R&D personnel ratio, R&D investment per person, output per person, and average working hours per unit of results, and splice them together to form R&D features.

[0064] In this way, feature extraction can be achieved from preprocessed production data, order data, financial data, and R&D data. The resulting production features, order features, financial features, and R&D features are all vectors, with vector lengths of [missing information]. , , , .

[0065] Afterwards, the electronic device can run step S40.

[0066] Step S40: Based on production characteristics, order characteristics, capital characteristics, and R&D characteristics, conduct cross-dimensional characteristic joint analysis to determine the enterprise's multi-dimensional health score.

[0067] In this embodiment, the electronic device can assign values ​​to each parameter in the production feature, order feature, funding feature, and R&D feature to obtain normalized production feature vector, order feature vector, funding feature vector, and R&D feature vector.

[0068] In this embodiment, the production characteristics are in the form of: [raw material consumption, energy consumption, planned production time, actual production time, daily number of workers, total daily working hours, production time of standard products, product output of standard products, pass rate of standard products, scrap rate of standard products, rework rate of standard products, time utilization rate, performance efficiency, overall equipment efficiency, output per unit working hour, output per person per day, raw material output rate, energy consumption per unit output of standard products]. For each indicator value of the production characteristics, the range to which the indicator value belongs is determined, thereby realizing the assignment (the corresponding assignment is determined according to the range to which the indicator value belongs, and the assignment is between 0 and 1).

[0069] For example, for a company in a certain commodity manufacturing industry, the assignment standards for each indicator in the production characteristics are shown in Table 2 below: Table 2. Assignment Standards for Each Indicator in Production Characteristics

[0070] This allows us to obtain the standardized production feature vector.

[0071] In this embodiment, the order characteristics are in the form of: [order quantity, on-time delivery rate, delayed delivery rate, average order processing cycle, average delayed delivery duration, total number of customers, number of repeat customers, average number of products per order, total product sales, total product sales amount, and order payment collection ratio]. For each indicator value of the order characteristics, the range to which the indicator value belongs can be determined, thereby realizing the assignment (the corresponding assignment is determined according to the range to which the indicator value belongs, and the assignment is between 0 and 1).

[0072] For example, for a company in a certain commodity manufacturing industry, the assignment standards for each indicator in the order characteristics are shown in Table 3 below: Table 3. Assignment Criteria for Each Indicator in Order Characteristics

[0073] This allows us to obtain the standardized order feature vector.

[0074] In this embodiment, the capital characteristics are in the form of: [operating revenue, operating cost, net profit, accounts receivable, inventory, net operating cash flow, net profit margin, return on net assets, accounts receivable turnover, inventory turnover, cash flow ratio, cash ratio, operating revenue growth rate, profit cash ratio, revenue cash ratio]. For each indicator value of the capital characteristics, the range to which the indicator value belongs can be determined, thereby realizing the assignment (the corresponding assignment is determined according to the range to which the indicator value belongs, and the assignment is between 0 and 1).

[0075] For example, for a company in a certain commodity manufacturing industry, the assignment standards for each indicator in the capital characteristics are shown in Table 4 below: Table 4. Assignment Standards for Each Indicator in the Funding Characteristics

[0076] This allows us to obtain the standardized capital feature vector.

[0077] In this embodiment, the R&D characteristics are in the form of: [number of R&D personnel, R&D working hours, R&D investment, number of R&D results, R&D intensity, R&D personnel ratio, R&D investment per person, output per person, and average working hours per unit of result]. For each indicator value of the R&D characteristics, the range to which the indicator value belongs can be determined, thereby realizing the assignment (the corresponding assignment is determined according to the range to which the indicator value belongs, and the assignment is between 0 and 1).

[0078] For example, for a company in a certain commodity manufacturing industry, the assignment standards for each indicator in the R&D characteristics are shown in Table 5 below: Table 5. Assignment Criteria for Each Indicator in R&D Characteristics

[0079] This allows us to obtain the standardized R&D feature vector.

[0080] Therefore, electronic devices can use attention mechanisms to perform cross-dimensional joint analysis of normalized production feature vectors, order feature vectors, capital feature vectors, and R&D feature vectors to determine the enterprise's multi-dimensional health score.

[0081] For example, electronic devices can determine the production dimension score based on normalized production feature vectors, order feature vectors, funding feature vectors, and R&D feature vectors, using multiple sets of production weight matrices in a trained attention mechanism.

[0082] Specifically, electronic devices can calculate production-order enhancement features based on normalized production feature vectors, order feature vectors, and the production-order weight matrix in the production weight matrix: , (1) , (2) in, Enhanced features for production-order ( (vector) For the production-order attention weight vector ( (vector) To generate feature vectors ( (vector) For order feature vectors ( (vector) This indicates element-wise addition. This indicates element-wise multiplication. The query matrix representing the relationship between production feature vectors and order feature vectors (is a...) (matrix) The key matrix (which is a key matrix between production feature vectors and order feature vectors) represents the relationship between production feature vectors and order feature vectors. (matrix) The value matrix representing the relationship between the production feature vector and the order feature vector (is a...) (matrix).

[0083] Similarly, electronic devices can calculate production-capital enhancement features based on normalized production feature vectors, capital feature vectors, and the production-capital weight matrix in the production weight matrix: , (3) , (4) in, For production-capital enhancement features ( (vector) For the production-capital attention weight vector ( (vector) To generate feature vectors ( (vector) For capital feature vector ( (vector) This indicates element-wise addition. This indicates element-wise multiplication. The query matrix representing the relationship between the production feature vector and the capital feature vector (is a...) (matrix) The key matrix (which represents the relationship between the production feature vector and the capital feature vector) is a key matrix. (matrix) The value matrix representing the relationship between the production feature vector and the capital feature vector (is a...) (matrix).

[0084] Furthermore, electronic devices can calculate production-R&D enhancement features based on normalized production feature vectors, R&D feature vectors, and the production-R&D weight matrix in the production weight matrix: , (5) , (6) in, Enhanced features for production and R&D ( (vector) For the production-R&D attention weight vector ( (vector) To generate feature vectors ( (vector) For the research and development of feature vectors ( (vector) This indicates element-wise addition. This indicates element-wise multiplication. The query matrix representing the relationship between production feature vectors and R&D feature vectors (is a...) (matrix) The key matrix representing the relationship between production feature vectors and R&D feature vectors (is a...) (matrix) The value matrix representing the relationship between the production feature vector and the R&D feature vector (is a...) (matrix).

[0085] In this embodiment, It can be In order to facilitate subsequent feature fusion processing (e.g., Therefore, we can use a weighted summation method, which adds the main elements of the vectors, to achieve fusion, instead of being limited to vector concatenation schemes.

[0086] Therefore, electronic devices perform feature fusion on production feature vectors, production-order enhancement features, production-capital enhancement features, and production-R&D enhancement features (if set). If so, feature fusion can be achieved by weighted addition of elements. Then, the trained production dimension regressor is used to perform non-linear mapping and scoring on the production feature vector after feature fusion to obtain the production dimension score. The production dimension regressor is a fully connected neural network with no more than 3 layers (in practice, 2 layers are sufficient, which is a small and lightweight fully connected neural network. During training, it needs to be trained together with the production weight matrix of the attention mechanism mentioned above).

[0087] For example, electronic devices can determine the order dimension score based on normalized production feature vectors, order feature vectors, funding feature vectors, and R&D feature vectors, using multiple sets of order weight matrices in a trained attention mechanism.

[0088] Specifically, electronic devices can calculate order-production enhancement features based on normalized order feature vectors, production feature vectors, and the order-production weight matrix in the order weight matrix: , (7) , (8) in, Enhanced features for order-to-production ( (vector) For the order-production attention weight vector ( (vector) For order feature vectors ( (vector) To generate feature vectors ( (vector) This indicates element-wise addition. This indicates element-wise multiplication. The query matrix representing the relationship between order feature vectors and production feature vectors (is a...) (matrix) The key matrix (which represents the relationship between the order feature vector and the production feature vector) is a key matrix. (matrix) The value matrix representing the relationship between the order feature vector and the production feature vector (is a...) (matrix).

[0089] Similarly, electronic devices can calculate order-fund enhancement features based on normalized order feature vectors, fund feature vectors, and the order-fund weight matrix in the order weight matrix: , (9) , (10) in, Enhanced features for order-funding ( (vector) For the order-fund attention weight vector ( (vector) For order feature vectors ( (vector) For capital feature vector ( (vector) This indicates element-wise addition. This indicates element-wise multiplication. The query matrix representing the relationship between the order feature vector and the capital feature vector (is a...) (matrix) The key matrix (which represents the relationship between the order feature vector and the capital feature vector) is a key matrix. (matrix) The value matrix representing the relationship between the order feature vector and the capital feature vector (is a...) (matrix).

[0090] Furthermore, electronic devices can calculate order-R&D enhanced features based on normalized order feature vectors, R&D feature vectors, and the order-R&D weight matrix in the order weight matrix: , (11) , (12) in, Enhanced features for order-R&D ( (vector) For the order-R&D attention weight vector ( (vector) For order feature vectors ( (vector) For the research and development of feature vectors ( (vector) This indicates element-wise addition. This indicates element-wise multiplication. This represents the query matrix between the order feature vector and the R&D feature vector (which is a...). (matrix) The key matrix representing the relationship between the order feature vector and the R&D feature vector (is a...) (matrix) This represents the value matrix between the order feature vector and the R&D feature vector (which is a...). (matrix).

[0091] In this embodiment, It can be In order to facilitate subsequent feature fusion processing (e.g., Therefore, we can use a weighted summation method, which adds the main elements of the vectors, to achieve fusion, instead of being limited to vector concatenation schemes.

[0092] Then, electronic devices perform feature fusion on order feature vectors, order-production enhancement features, order-funding enhancement features, and order-R&D enhancement features (if set). If so, feature fusion can be achieved by weighted addition of elements. Then, the trained order dimension regressor is used to perform non-linear mapping and scoring on the fused order feature vector to obtain the order dimension score. The order dimension regressor is a fully connected neural network with no more than 3 layers (in practice, 2 layers are sufficient, which is a small and lightweight fully connected neural network. During training, it needs to be trained together with the order weight matrix of the attention mechanism mentioned above).

[0093] For example, electronic devices can determine a funding dimension score based on normalized production feature vectors, order feature vectors, funding feature vectors, and R&D feature vectors, using multiple sets of funding weight matrices in a trained attention mechanism.

[0094] Specifically, electronic devices can calculate capital-production enhancement features based on normalized capital feature vectors, production feature vectors, and the capital-production weight matrix in the capital weight matrix: , (13) , (14) in, Enhanced capital-production characteristics ( (vector) For the capital-production attention weight vector ( (vector) For capital feature vector ( (vector) To generate feature vectors ( (vector) This indicates element-wise addition. This indicates element-wise multiplication. The query matrix representing the relationship between the capital feature vector and the production feature vector (is a...) (matrix) The key matrix (which represents the relationship between the capital eigenvector and the production eigenvector) is a... (matrix) The value matrix representing the relationship between the capital eigenvector and the production eigenvector is a... (matrix).

[0095] Similarly, electronic devices can calculate enhanced fund-order features based on normalized fund feature vectors, order feature vectors, and the fund-order weight matrix in the fund weight matrix: , (15) , (16) in, Enhanced features for funds-orders ( (vector) For the fund-order attention weight vector ( (vector) For capital feature vector ( (vector) For order feature vectors ( (vector) This indicates element-wise addition. This indicates element-wise multiplication. The query matrix representing the relationship between the capital feature vector and the order feature vector (is a...) (matrix) The key matrix (which represents the relationship between the capital feature vector and the order feature vector) is a key matrix. (matrix) The value matrix representing the relationship between the capital feature vector and the order feature vector (is a...) (matrix).

[0096] Furthermore, electronic devices can calculate funding-R&D enhancement features based on normalized funding feature vectors, R&D feature vectors, and the funding-R&D weight matrix in the funding weight matrix: , (17) , (18) in, Enhanced features for funding and R&D ( (vector) For the funding-R&D attention weight vector ( (vector) For capital feature vector ( (vector) For the research and development of feature vectors ( (vector) This indicates element-wise addition. This indicates element-wise multiplication. This represents the query matrix between the funding feature vector and the R&D feature vector (which is a...). (matrix) The key matrix representing the relationship between the capital feature vector and the R&D feature vector (is a...) (matrix) The value matrix representing the relationship between the capital feature vector and the R&D feature vector (is a...) (matrix).

[0097] In this embodiment, It can be In order to facilitate subsequent feature fusion processing (e.g., Therefore, we can use a weighted summation method, which adds the main elements of the vectors, to achieve fusion, instead of being limited to vector concatenation schemes.

[0098] Then, electronic devices perform feature fusion on the capital feature vector, capital-production enhancement features, capital-order enhancement features, and capital-R&D enhancement features (if set). If so, feature fusion can be achieved by weighted addition of elements. Then, the trained capital dimension regressor is used to perform non-linear mapping and scoring on the capital feature vector after feature fusion to obtain the capital dimension score. The capital dimension regressor is a fully connected neural network with no more than 3 layers (in practice, 2 layers are sufficient, which is a small and lightweight fully connected neural network. During training, it needs to be trained together with the capital weight matrix of the attention mechanism mentioned above).

[0099] For example, electronic devices can determine the R&D dimension score based on normalized production feature vectors, order feature vectors, funding feature vectors, and R&D feature vectors, using multiple sets of R&D weight matrices in a trained attention mechanism.

[0100] Specifically, electronic devices can calculate R&D-production enhancement features based on normalized R&D feature vectors, production feature vectors, and the R&D-production weight matrix in the R&D weight matrix: , (19) , (20) in, Enhanced features for R&D and production ( (vector) For R&D-production attention weight vectors ( (vector) For the research and development of feature vectors ( (vector) To generate feature vectors ( (vector) This indicates element-wise addition. This indicates element-wise multiplication. This represents the query matrix between R&D feature vectors and production feature vectors (it is a...) (matrix) The key matrix representing the relationship between R&D feature vectors and production feature vectors (is a...) (matrix) The value matrix representing the relationship between the R&D feature vector and the production feature vector (is a...) (matrix).

[0101] Similarly, electronic devices can calculate R&D-order enhanced features based on normalized R&D feature vectors, order feature vectors, and the R&D-order weight matrix in the R&D weight matrix: , (twenty one) , (twenty two) in, Enhanced features for R&D-orders ( (vector) For R&D-order attention weight vector ( (vector) For the research and development of feature vectors ( (vector) For order feature vectors ( (vector) This indicates element-wise addition. This indicates element-wise multiplication. This represents the query matrix between the R&D feature vector and the order feature vector (which is a...). (matrix) The key matrix representing the relationship between the R&D feature vector and the order feature vector (is a...) (matrix) This represents the value matrix between the R&D feature vector and the order feature vector (which is a...). (matrix).

[0102] Furthermore, electronic devices can calculate R&D-funding enhanced features based on normalized R&D feature vectors, funding feature vectors, and the R&D-funding weight matrix in the R&D weight matrix: , (twenty three) , (twenty four) in, Enhanced features for R&D-funding ( (vector) For the R&D-fund attention weight vector ( (vector) For the research and development of feature vectors ( (vector) For capital feature vector ( (vector) This indicates element-wise addition. This indicates element-wise multiplication. This represents the query matrix between the R&D feature vector and the funding feature vector (which is a...). (matrix) The key matrix representing the relationship between the R&D feature vector and the funding feature vector (is a...) (matrix) The value matrix representing the relationship between the R&D feature vector and the funding feature vector (is a...) (matrix).

[0103] In this embodiment, It can be In order to facilitate subsequent feature fusion processing (e.g., Therefore, we can use a weighted summation method, which adds the main elements of the vectors, to achieve fusion, instead of being limited to vector concatenation schemes.

[0104] Therefore, electronic devices perform feature fusion on R&D feature vectors, R&D-production enhancement features, R&D-order enhancement features, and R&D-funding enhancement features (if set). If so, feature fusion can be achieved by weighted addition of elements. Then, the trained R&D dimension regressor is used to perform non-linear mapping and scoring on the R&D feature vector after feature fusion to obtain the R&D dimension score. The R&D dimension regressor is a fully connected neural network with no more than 3 layers (in practice, 2 layers are sufficient, which is a small and lightweight fully connected neural network. During training, it needs to be trained together with the R&D weight matrix of the attention mechanism mentioned above).

[0105] In the process of determining enhanced features, unlike the traditional attention fusion mechanism, element-wise addition and element-wise multiplication operations are adopted, which can more effectively introduce the influence of other dimensions on the target dimension, and facilitate cross-dimensional analysis.

[0106] Finally, electronic devices can be weighted and summed based on production dimension scores, order dimension scores, funding dimension scores, and R&D dimension scores to obtain a multi-dimensional health score for the enterprise.

[0107] Furthermore, embodiments of this application also provide an enterprise health monitoring and assessment system based on multi-dimensional analysis, including: The data acquisition unit is used to acquire enterprise data within the statistical period, including production data, employment data, order data, financial data, and R&D data.

[0108] The preprocessing unit is used to preprocess enterprise data.

[0109] The feature extraction unit is used to extract features from preprocessed production data, order data, financial data, and R&D data to determine production features, order features, financial features, and R&D features.

[0110] The multi-dimensional scoring unit is used to conduct cross-dimensional feature joint analysis based on production characteristics, order characteristics, capital characteristics, and R&D characteristics to determine the enterprise's multi-dimensional health score.

[0111] The units of the enterprise health monitoring and assessment system based on multi-dimensional analysis correspond to the steps in the method described above, and will not be repeated in this embodiment.

[0112] In summary, this application provides a method and system for enterprise health monitoring and assessment based on multi-dimensional analysis. It acquires and preprocesses enterprise data (including production data, employment data, order data, financial data, and R&D data) within a statistical period (e.g., one month, three months, six months, one year), then extracts features from the preprocessed production, order, financial, and R&D data (the feature extraction process mainly involves acquiring various data and calculating multiple indicators) to determine production characteristics, order characteristics, financial characteristics, and R&D characteristics. Based on these characteristics, cross-dimensional feature joint analysis is then performed to determine the enterprise's multi-dimensional health score. This approach enables multi-dimensional assessment of enterprises, considers the correlation between several indicators within each dimension, achieves cross-dimensional feature joint analysis, calculates the enterprise's multi-dimensional health score, and thus provides decision-making support for enterprises.

[0113] By assigning values ​​to each parameter in the production, order, capital, and R&D features (considering the impact of each indicator on the enterprise's status, with values ​​between 0 and 1; the larger the value, the healthier the enterprise's status in that indicator), normalized production feature vectors, order feature vectors, capital feature vectors, and R&D feature vectors are obtained. Then, an attention mechanism is used to perform cross-dimensional feature joint analysis on the normalized production feature vectors, order feature vectors, capital feature vectors, and R&D feature vectors: by introducing an end-to-end jointly trained attention mechanism + regressors for each dimension, cross-dimensional feature joint analysis is performed. Specifically, using the normalized production feature vectors, order feature vectors, capital feature vectors, and R&D feature vectors, and using multiple sets of weight matrices in the trained attention mechanism, the order dimension score, capital dimension score, and R&D dimension score are determined. Specifically, when analyzing scores for a particular dimension, the weight matrices obtained after training between each pair of dimensions are used to calculate an enhancement matrix (which takes into account the influence of other dimensions on the features of this dimension, thus achieving feature enhancement). Simultaneously, the feature vectors of that dimension are concatenated and input into the corresponding dimension's regressor (a small, fully connected neural network with no more than three layers) for score determination. Finally, the determined scores for production, orders, funding, and R&D are weighted (the weights are determined by the weights allocated to each dimension and the number of indicators contained in each dimension) and summed to obtain the enterprise's multi-dimensional health score. This approach enables cross-dimensional joint analysis of feature vectors from different dimensions. The creative application of the attention mechanism to this scenario (originally, attention mechanisms were mainly used in matrix analysis, while in this scenario, feature vectors) considers the potential positive or negative interactions between indicators, deriving the enhancement matrix for the corresponding dimension. However, matrices are difficult to quantify scores simply by using matrices; therefore, a regressor combined with the attention mechanism is designed for joint training to achieve accurate quantification of scores for each dimension. This approach enables cross-dimensional joint analysis and status scoring for each dimension. Compared to traditional single-dimensional analysis methods, this solution can achieve multi-dimensional assessment, more accurately evaluate the health status of enterprises, and provide assistance for enterprise decision-making.

[0114] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for enterprise health monitoring and assessment based on multi-dimensional analysis, characterized in that, include: Obtain enterprise data within the statistical period, including production data, employment data, order data, financial data, and R&D data; Preprocess enterprise data; Feature extraction is performed on the preprocessed production data, order data, financial data, and R&D data to determine production features, order features, financial features, and R&D features; Based on production characteristics, order characteristics, capital characteristics, and R&D characteristics, a cross-dimensional characteristic joint analysis is conducted to determine the enterprise's multi-dimensional health score.

2. The enterprise health monitoring and assessment method based on multi-dimensional analysis according to claim 1, characterized in that, Feature extraction was performed on the preprocessed production data, order data, financial data, and R&D data to determine production features, order features, financial features, and R&D features, including: Feature extraction is performed on the preprocessed production data to determine the production characteristics; Feature extraction is performed on the preprocessed order data to determine the order characteristics; Feature extraction is performed on the preprocessed financial data to determine the financial characteristics; Feature extraction is performed on the preprocessed R&D data to determine the R&D characteristics.

3. The enterprise health monitoring and assessment method based on multi-dimensional analysis according to claim 2, characterized in that, Production data includes raw material consumption, energy consumption, planned production time, actual production time, daily number of workers, total daily working hours, output of each product, pass rate, scrap rate, and rework rate, the conversion relationship between each product and the standard product, and the production time of the standard product. Feature extraction is performed on the preprocessed production data to determine production characteristics, including: Based on the output, pass rate, scrap rate and rework rate of each product, as well as the conversion relationship between each product and the standard product, the output, pass rate, scrap rate and rework rate of the standard product are determined. The time utilization rate is determined based on the planned production time and the actual production time. Performance efficiency is determined based on actual production time, standard product output, and standard product production time. The overall equipment efficiency is determined based on time uptime, performance efficiency, and the pass rate of standard products. Based on the product output and total daily working hours of standard products, the output per unit working hour is determined. Based on the product output of standard products and the number of workers employed daily, the average daily output per person is determined. Based on the product output, qualification rate and raw material consumption of standard products, the raw material output rate of standard products is determined. Based on energy consumption and the output of standard products, the energy consumption per unit output is determined. Extract the numerical characteristics of raw material consumption, energy consumption, planned production time, actual production time, daily number of workers, total daily working hours, production time of standard products, product output, pass rate, scrap rate and rework rate of standard products, time utilization rate, performance efficiency, overall equipment efficiency, output per unit working hour, output per person per day, raw material output rate, and energy consumption per unit output of standard products, and piece them together to form production characteristics.

4. The enterprise health monitoring and assessment method based on multi-dimensional analysis according to claim 2, characterized in that, Each order data item includes the order creation date, order deadline, order completion date, order customer information, order product type, order product quantity, order product amount, and order payment amount. Feature extraction is performed on the preprocessed order data to determine the order characteristics, including: Based on the order deadline and order completion date of each order data, the order delay duration of each order data is determined, and based on the order generation date and order completion date of each order data, the order processing cycle of each order data is determined; Based on the order delay duration and total number of orders for each order, the on-time delivery rate, delayed delivery rate, average order processing cycle, and average delayed delivery duration are determined. Based on the order customer information for each order, the total number of customers and the number of repeat customers are determined. Based on the order product type and order product quantity for each order, the average order product quantity and total product sales volume are determined. Based on the order product amount of each order, the total product sales amount is determined. Based on the order product amount and order payment amount of each order, the order payment ratio is determined; Extract numerical features such as order quantity, on-time delivery rate, late delivery rate, average order processing cycle, average late delivery duration, total number of customers, number of repeat customers, average number of products per order, total product sales, total product sales amount, and order payment collection ratio, and combine them to form order features.

5. The enterprise health monitoring and assessment method based on multi-dimensional analysis according to claim 2, characterized in that, The financial data is used for preprocessing. Feature extraction is performed on the preprocessed financial data to determine its characteristics, including: Identify financial data and determine the values ​​of various financial indicators; Based on the various indicators of financial data, the following are determined: net profit margin, return on net assets, accounts receivable turnover, inventory turnover, cash flow ratio, cash ratio, operating revenue growth rate, profit cash ratio, and revenue cash ratio. Extract multiple indicators from financial data, including net profit margin, return on net assets, accounts receivable turnover, inventory turnover, cash flow ratio, cash ratio, operating revenue growth rate, profit cash ratio, and revenue cash ratio, and combine them to form financial characteristics.

6. The enterprise health monitoring and assessment method based on multi-dimensional analysis according to claim 5, characterized in that, R&D data includes the number of R&D personnel, R&D man-hours, R&D investment, and the number of R&D achievements. Feature extraction is performed on the preprocessed R&D data to determine R&D characteristics, including: The intensity of research and development is determined based on research and development investment and operating revenue; The proportion of R&D personnel is determined based on the number of R&D personnel and the total number of employees in the company; The per capita R&D investment was determined based on the number of R&D personnel and R&D investment. The per capita output is determined based on the number of R&D achievements and the number of R&D personnel. The average working hours per unit of R&D output were determined based on the number of R&D results and R&D working hours. Extract numerical features such as the number of R&D personnel, R&D working hours, R&D investment, number of R&D results, R&D intensity, R&D personnel ratio, R&D investment per person, R&D output per person, and average working hours per unit of results, and splice them together to form R&D features.

7. The enterprise health monitoring and assessment method based on multi-dimensional analysis according to claim 1, characterized in that, Based on production characteristics, order characteristics, capital characteristics, and R&D characteristics, a cross-dimensional joint analysis is conducted to determine the enterprise's multi-dimensional health score, including: Assign values ​​to each parameter in the production feature, order feature, capital feature, and R&D feature to obtain normalized production feature vector, order feature vector, capital feature vector, and R&D feature vector; By utilizing the attention mechanism, cross-dimensional joint analysis of normalized production feature vectors, order feature vectors, capital feature vectors, and R&D feature vectors is conducted to determine the enterprise's multi-dimensional health score.

8. The enterprise health monitoring and assessment method based on multi-dimensional analysis according to claim 7, characterized in that, By utilizing an attention mechanism to perform cross-dimensional joint feature analysis on normalized production feature vectors, order feature vectors, capital feature vectors, and R&D feature vectors, a multi-dimensional health score for the enterprise is determined, including: Based on normalized production feature vectors, order feature vectors, funding feature vectors, and R&D feature vectors, and using multiple sets of production weight matrices in a trained attention mechanism, the production dimension score is determined. Based on normalized production feature vectors, order feature vectors, funding feature vectors, and R&D feature vectors, and using multiple sets of order weight matrices in a trained attention mechanism, the order dimension score is determined. Based on normalized production feature vectors, order feature vectors, capital feature vectors, and R&D feature vectors, and using multiple sets of capital weight matrices in a trained attention mechanism, the capital dimension score is determined. Based on normalized production feature vectors, order feature vectors, funding feature vectors, and R&D feature vectors, and using multiple sets of R&D weight matrices in a trained attention mechanism, the R&D dimension score is determined. A multi-dimensional health score for a company is obtained by weighting and summing the scores from production, orders, funding, and R&D dimensions.

9. The enterprise health monitoring and assessment method based on multi-dimensional analysis according to claim 8, characterized in that, Based on normalized production feature vectors, order feature vectors, funding feature vectors, and R&D feature vectors, and utilizing multiple sets of production weight matrices in a trained attention mechanism, the production dimension score is determined, including: Based on the normalized production feature vector, order feature vector, and the production-order weight matrix in the production weight matrix, calculate the production-order enhanced features: ， ， in, To enhance the production-order feature, For the production-order attention weight vector, To generate feature vectors, For order feature vectors, This indicates element-wise addition. This indicates element-wise multiplication. This represents the query matrix between production feature vectors and order feature vectors. This represents the key matrix between the production feature vector and the order feature vector. This represents the value matrix between the production feature vector and the order feature vector; Based on the normalized production feature vector, capital feature vector, and the production-capital weight matrix in the production weight matrix, calculate the production-capital enhancement feature: ， ， in, For production-capital enhancement features, For the production-capital attention weight vector, To generate feature vectors, For the feature vector of funds, This indicates element-wise addition. This indicates element-wise multiplication. This represents the query matrix between the production feature vector and the capital feature vector. This represents the key matrix between the production feature vector and the capital feature vector. A value matrix representing the relationship between the production feature vector and the capital feature vector; Based on the normalized production feature vector, R&D feature vector, and the production-R&D weight matrix in the production weight matrix, calculate the production-R&D enhancement features: ， ， in, Enhanced features for production and R&D For the production-R&D attention weight vector, To generate feature vectors, To develop feature vectors, This indicates element-wise addition. This indicates element-wise multiplication. This represents the query matrix between production feature vectors and R&D feature vectors. This represents the key matrix between the production feature vector and the R&D feature vector. A value matrix representing the relationship between production feature vectors and R&D feature vectors; The production dimension score is obtained by using a trained production dimension regressor to perform nonlinear mapping and scoring on the production feature vector, production-order enhancement feature, production-fund enhancement feature and production-R&D enhancement feature. The production dimension regressor is a fully connected neural network with no more than 3 layers.

10. A corporate health monitoring and assessment system based on multi-dimensional analysis, characterized in that, include: The data acquisition unit is used to acquire enterprise data within the statistical period, including production data, employment data, order data, financial data, and R&D data. The preprocessing unit is used to preprocess enterprise data; The feature extraction unit is used to extract features from preprocessed production data, order data, financial data, and R&D data to determine production features, order features, financial features, and R&D features. The multi-dimensional scoring unit is used to conduct cross-dimensional feature joint analysis based on production characteristics, order characteristics, capital characteristics, and R&D characteristics to determine the enterprise's multi-dimensional health score.

Images