Enterprise flow monitoring and early warning method and system

Abstract

The invention provides an enterprise flow monitoring and early warning method and system, and belongs to the field of big data analysis, and the method comprises the steps: collecting the historical flow data of an enterprise, such as the information of relocation, merging and purchasing, closing or transformation, and the like, and collecting the economic data related to the enterprise. After the data are preprocessed, the data are combined to construct a feature matrix so as to construct an early warning model, and the model is trained. After model training, enterprise flow risks can be predicted regularly or in real time. According to the system, enterprise risk management and prevention are facilitated, and the accuracy and efficiency of enterprise decision making are improved. Meanwhile, a homogeneity multi-task learning algorithm is adopted for model establishment, so that the accuracy and stability of model prediction are guaranteed.

Description

Technical Field

[0001] The present invention belongs to the field of big data analysis, and more specifically relates to an enterprise flow monitoring and early warning method and system. Background Art

[0002] In today's globalized economy, businesses face a constantly changing environment. As they grow, they often need to engage in a series of mobility activities, such as relocation, mergers and acquisitions, bankruptcy, or restructuring. These activities often carry significant economic risks. Therefore, effectively predicting a company's mobility behavior can help them respond to market changes and mitigate risks.

[0003] However, current enterprise mobility monitoring relies primarily on human judgment and empirical predictions, which often result in significant errors and lack real-time monitoring. Furthermore, existing monitoring and early warning systems often rely on specialized knowledge of specific industries or markets, requiring high data acquisition and processing capabilities and resulting in high costs.

[0004] At the same time, a company's mobility behavior is influenced by numerous factors, such as its industry environment, market demand, company operating conditions, and the policy environment. These factors interact with each other, forming a nonlinear, high-dimensional, complex system. This places higher technical requirements on the prediction and analysis of corporate mobility behavior.

[0005] Therefore, how to use existing data and advanced technologies to establish a method that can comprehensively, accurately and in real time predict corporate liquidity risks, and develop corresponding monitoring and early warning systems based on this, has become an urgent problem to be solved in the industry. Summary of the Invention

[0006] The present invention aims to solve the problems of low accuracy and efficiency in enterprise liquidity risk warning in the existing technology, and provides a new enterprise liquidity monitoring and early warning method and system. By collecting and processing the historical liquidity data and economic data of the enterprise, training the early warning model, and then using the trained early warning model to predict the enterprise liquidity risk in real time or periodically, the purpose of real-time warning, high accuracy and low cost is achieved.

[0007] In order to achieve the above object, the present invention is implemented by adopting the following technical solutions: the early warning method includes: Collect historical enterprise mobility data, including historical information on enterprise relocation, mergers and acquisitions, bankruptcy or transformation; Collect economic data related to the enterprise, including macroeconomic data related to the industry in which the enterprise is located, and microeconomic data related to the enterprise's daily business activities; Data preprocessing: preprocessing the collected historical enterprise flow data and economic data; Construct a feature matrix to combine the historical flow data of the enterprise with the economic data related to the enterprise; Build and train early warning models; Use the trained early warning model to predict enterprise liquidity risks regularly or in real time.

[0008] In one embodiment, the collection of historical enterprise flow data includes: Data collection: Define the types of corporate movements of interest, such as relocations, mergers and acquisitions, closures, or transformations; and scrape information about these companies from public data sources, databases, or social media. Data annotation: Use relocation, merger and acquisition, bankruptcy or transformation as labels to mark the relocation, merger and acquisition, bankruptcy or transformation events that have occurred in these companies in history.

[0009] In one embodiment, the collection of enterprise-related economic data includes: Use web crawlers to regularly crawl enterprise-related economic data from public data sources. Macroeconomic data includes GDP growth rate, industrial production index, and consumer confidence index; microeconomic data includes product prices, inventory, and sales.

[0010] In one embodiment, constructing a feature matrix includes: Data integration: Integrate the company's historical flow data, arrange each company's flow events in chronological order, and record the specific time and type of each flow; Normalization: normalize macroeconomic data and microeconomic data; Constructing a feature matrix: For each flow event, extract the macroeconomic and microeconomic data before it occurs. These data together constitute the feature vector of this event. The feature vectors of all events are combined into a feature matrix. There are N flow events in total, and each event has M features. The feature matrix is N*M dimensional, denoted as X=(x ij ), where x ij represents the jth feature of the i-th event; Construct a label vector: Label each mobility event by its type, with relocation as 1, merger and acquisition as 2, and bankruptcy or transformation as 3. These labels form the label vector; There are N flow events in total, so the label vector is N-dimensional, denoted as Y=(y i ), where y i Indicates the label of the i-th event; Get the feature matrix X and label vector Y, and input them into the early warning model for training.

[0011] In one solution, the early warning model is constructed using a homogeneous multi-task learning algorithm HMTL, and the constructed feature matrix and label vector are input into the HMTL model; There are M tasks and N samples, each sample has P features, then the feature matrix X is a matrix of NP, and the label vector Y is a matrix of NM; The objective function is: ; Among them, Θ is an M*P matrix, each row represents the model parameter θj of each task j; is the feature vector corresponding to sample i; L is the loss function, is the Frobenius norm of the parameter matrix Θ, which is used to regularize the model to prevent overfitting; λ is the specified regularization strength; The training process includes: The loss function is gradually minimized through repeated iterations. In each iteration, the gradient of the objective function with respect to the parameters is first calculated, and then the parameters are updated along the negative direction of the gradient. For the above objective function, the parameter update rule is expressed as: ; Among them, α is the learning rate, which controls the step size of parameter update, and ∇L is the gradient of the objective function; For each task j, the gradient of the corresponding model parameter θj is calculated and updated, that is, ; The above method is continuously iterated until the loss function converges, and a trained early warning model is obtained.

[0012] In one embodiment, the learning rate is 0.001.

[0013] In another aspect, an enterprise mobility monitoring and early warning system is provided, wherein the system is applicable to the method described above, and the monitoring and early warning system comprises: Data collection module: used to collect historical flow data of enterprises, macroeconomic data and microeconomic data related to the industry in which the enterprises are located; Data preprocessing module: used to preprocess the collected historical flow data and economic data of enterprises; Feature matrix construction module: used to combine the enterprise's historical flow data with the enterprise's related economic data to construct a feature matrix; Early warning model training module: responsible for training the early warning model, including setting the given learning rate and regularization strength parameters, as well as initializing and updating the parameters; Risk prediction module: uses the trained early warning model to determine the company's potential liquidity risks in the future; Early warning result output module: outputs the predicted enterprise liquidity risk results for reference by decision makers.

[0014] Beneficial effects of the present invention: By combining historical enterprise liquidity data with relevant economic data and training an early warning model, enterprise liquidity risk can be predicted in real time or periodically, improving the accuracy and real-time nature of early warnings. This can help enterprises better manage risks, enhance decision-making efficiency, and improve economic benefits. Furthermore, compared to traditional manual early warning methods, the method and system of the present invention reduce the cost and complexity of early warnings, offering high practicality and broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 Flow chart of the method of the present invention; Figure 2 Construct a feature matrix flow chart for the method of the present invention; Figure 3 This is a block diagram of the early warning system of the present invention. DETAILED DESCRIPTION

[0016] To facilitate understanding of the present invention, the present invention will be described more fully below with reference to the accompanying drawings. The drawings illustrate exemplary embodiments of the present invention. However, the present invention may be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and comprehensive understanding of the present invention.

[0017] Unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0018] Example 1: Figure 1 As shown, the enterprise flow monitoring and early warning method includes: S1. Collect historical enterprise mobility data, including but not limited to historical information on enterprise relocations, mergers and acquisitions, bankruptcies, or restructuring. Typically, this data can be obtained from public company records, news reports, social media, economic databases, and other channels.

[0019] The collection of enterprise historical flow data includes: Data collection: First, you need to define the types of corporate movements you are interested in, such as relocations, mergers and acquisitions, closures, or transformations. Then, you can scrape information about these companies from various public data sources, databases, or social media. You can also use internal company records or documents.

[0020] Data Annotation: Using CAT (Relocation, Mergers and Acquisitions, Bankruptcy, or Transformation) as the label, identify historical relocation, mergers and acquisitions, bankruptcies, or transformation events that occurred at these companies. S2. Collect relevant economic data for these companies, such as GDP growth rate, industrial production index, consumer confidence index, etc., as well as microeconomic data related to the company's daily operations, such as product prices, inventory levels, and sales. These data sources may include, but are not limited to, public reports or data published by the National Bureau of Statistics, financial institutions, and economic research institutions.

[0021] Use web crawlers to regularly crawl enterprise-related economic data from public data sources. Macroeconomic data includes GDP growth rate, industrial production index, and consumer confidence index; microeconomic data includes product prices, inventory, and sales.

[0022] S3. Data preprocessing: preprocessing the collected historical enterprise flow data and economic data; Missing Value Handling: First, use descriptive statistics to identify columns with missing values. Use the isnull() function in the pandas library to find columns with missing values, and then choose to fill, delete, or ignore them based on the actual situation.

[0023] Outlier handling: Based on empirical rules, values exceeding (mean + 3 × standard deviation) are generally considered outliers and can be deleted or replaced. Alternatively, quartiles and interquartile ranges (IQR) can be used to identify outliers. Values exceeding or falling below (Q1 - 1.5 × IQR) or (Q3 + 1.5 × IQR) are considered outliers. The formula is as follows: ; Q1 and Q3 represent the first quartile and the third quartile respectively, and IQR=Q3-Q1.

[0024] Data normalization: Normalize the data to a common numerical range to eliminate the impact caused by scale differences between data. Z-score normalization formula: ; Where x represents a single data point, μ and σ represent the mean and standard deviation of the dataset, respectively.

[0025] like Figure 2 As shown, S4, construct a feature matrix, combining the enterprise's historical flow data with the enterprise's related economic data to facilitate the subsequent early warning model to identify and serve as the input of the early warning model. The construction of the feature matrix includes: S401. Data integration: Integrate the historical flow data of the enterprise, arrange the flow events of each enterprise in chronological order, and record the specific time and type of each flow; S402. Normalization: Normalize the macroeconomic and microeconomic data. Use the maximum-minimum normalization method to linearly transform the original data to the range [0, 1]. The formula is as follows: ; Where X is the original data, Xmin and Xmax are the minimum and maximum values of the data respectively, and Xnorm is the normalized data.

[0026] S403, constructing a feature matrix: for each flow event, extracting the macroeconomic data and microeconomic data before the event, which together constitute the feature vector of the event; combining the feature vectors of all events into a feature matrix; There are N flow events in total, and each event has M features. The feature matrix is N*M dimensional, denoted as X=(x ij ), where x ij represents the jth feature of the i-th event.

[0027] S404. Construct a label vector: Label each mobility event by its type, with relocation as 1, merger and acquisition as 2, and bankruptcy or transformation as 3. These labels constitute the label vector. There are N flow events in total, so the label vector is N-dimensional, denoted as Y=(y i ), where y i Indicates the label of the i-th event.

[0028] Get the feature matrix X and label vector Y, and input them into the early warning model for training.

[0029] S5. Build an early warning model and train it. Different enterprise mobility events (such as relocation, mergers and acquisitions, bankruptcy or transformation, etc.) can be considered as different tasks. There is a certain degree of correlation between these tasks. Homogeneous multi-task learning can be used to share some features or rules between these tasks to improve the accuracy of early warning.

[0030] The aforementioned early warning model is constructed using the homogeneous multi-task learning algorithm HMTL. The constructed feature matrix and label vector are input into the HMTL model. There are M tasks and N samples, and each sample has P features. Then the feature matrix X is a matrix of NP, and the label vector Y is a matrix of NM. The objective function is: ; Among them, Θ is an M*P matrix, each row represents the model parameter θj of each task j; is the feature vector corresponding to sample i; L is the loss function, is the Frobenius norm of the parameter matrix Θ, which is used to regularize the model to prevent overfitting; λ is the specified regularization strength; The training process includes: S501. Minimize the loss function gradually through repeated iterations. In each iteration, first calculate the gradient of the objective function with respect to the parameters, and then update the parameters along the negative direction of the gradient. For the above objective function, the parameter update rule is expressed as: ; Where α is the learning rate, which is 0.001. It controls the step size of parameter updates, and ∇L is the gradient of the objective function. S502: For each task j, calculate the gradient of the corresponding model parameter θj and update it, that is, ; The above method is continuously iterated until the loss function converges, and a trained early warning model is obtained.

[0031] The advantages of using the HMTL algorithm in the field of enterprise liquidity risk warning compared with other algorithms are mainly reflected in the following aspects: Corporate mobility activities include relocation, mergers and acquisitions, bankruptcy, and transformation, among others. These activities may be inherently interconnected. HMTL is a multi-task algorithm that can simultaneously handle risk warning tasks for multiple mobility activities and share information between tasks. This helps comprehensively assess a company's mobility risk from multiple perspectives.

[0032] When dealing with homogeneous tasks, multi-task learning can improve the model's predictive accuracy by sharing information between tasks. Because various enterprise mobility behaviors are often driven by similar economic factors, they can be considered homogeneous tasks, making HMTL suitable for use. In this case, HMTL can effectively improve the accuracy of early warnings by capturing the correlations between tasks.

[0033] HMTL boasts high computational efficiency, particularly when processing large amounts of data. In the field of enterprise liquidity risk early warning, the volume of data required is often enormous, including historical liquidity data from various companies and related economic data. The HMTL algorithm can rapidly train on this large-scale data and provide real-time early warning information, which is crucial for timely identifying and addressing enterprise liquidity risks.

[0034] S6. Use the trained early warning model to predict enterprise liquidity risks regularly or in real time.

[0035] The latest enterprise-related economic data is collected and preprocessed. For any monitored enterprise, if a new liquidity event occurs, its information is marked as a new sample, and the macroeconomic and microeconomic data preceding the event are extracted as features. The features of this new sample are input into the trained early warning model to obtain a prediction result, which includes the probability of relocation, merger and acquisition, bankruptcy, or transformation. If the predicted probability exceeds a set threshold, a liquidity warning is issued. This method can monitor and predict enterprise liquidity risks in real time and issue early warning signals in a timely manner, which is of great significance to enterprise decision-making and economic forecasting.

[0036] Example 2: Figure 3 As shown, an enterprise flow monitoring and early warning system is applicable to the method described above, and the monitoring and early warning system includes: Data Collection Module: This module is primarily responsible for collecting historical enterprise flow data, as well as macroeconomic and microeconomic data related to the enterprise's industry. Its role is to obtain the required raw data and provide the necessary data support for subsequent data preprocessing, feature matrix construction, and early warning model training.

[0037] Data Preprocessing: This module is responsible for preprocessing the collected historical enterprise flow and economic data, including screening, cleaning, filling missing values, and handling outliers. This processing step is crucial for improving the accuracy of subsequent model predictions. Preprocessed data helps enhance the accuracy and robustness of the model.

[0038] Feature Matrix Construction Module: This module primarily constructs a feature matrix by integrating and normalizing historical liquidity and economic data. This matrix forms the foundation for training early warning models and provides key data features for predicting enterprise liquidity risk. An effective feature matrix can enhance the model's predictive capabilities.

[0039] Early Warning Model Training Module: This module is responsible for training the early warning model based on the feature matrix, including setting the learning rate, regularization parameters, and initializing and updating the model parameters. A successfully trained early warning model can accurately predict the risks a company may face. The accuracy of the early warning model directly affects the effectiveness of the early warning system.

[0040] Risk Prediction Module: Utilizes trained early warning models to predict the company's future liquidity risks. This allows for early detection of potential risks and provides a certain level of risk prevention and control warning.

[0041] Early warning result output module: The predicted enterprise liquidity risk results are sorted, counted and output so that enterprise decision makers can clearly and intuitively understand the risks that the enterprise may face, and provide decision support for the enterprise's future strategic adjustments, risk prevention, and emergency plan formulation.

[0042] Example 3: Assume that the liquidity of three companies (Company 1, Company 2, and Company 3) needs to be monitored. The following is their key historical information: Company 1: 1 relocation in the past five years, macroeconomic data (GDP growth rate: 5%, industrial production index: 105, consumer confidence index: 75), microeconomic data (product price: $100, inventory: 500, sales: $1,000,000) Company 2: Has undergone two mergers in the past five years. Macroeconomic data (GDP growth rate: 3%, industrial production index: 110, consumer confidence index: 80), microeconomic data (product price: $120, inventory: 800, sales: $2,000,000) Company 3: 1 bankruptcy in the past five years, macroeconomic data (GDP growth rate: 2%, industrial production index: 95, consumer confidence index: 70), microeconomic data (product price: $80, inventory: 1000, sales: $500,000) First, the historical flow data and microeconomic data of the three companies are collected, and then preprocessing is performed, such as data cleaning, filling missing values, and outlier processing.

[0043] Next, integrate these data to construct a feature matrix as shown in Table 1: Table 1

[0044] The above feature matrix is then normalized to eliminate the dimensional influence between different indicators.

[0045] Afterwards, a warning model algorithm, such as a homogeneous multi-task learning algorithm, was used for model training. During the training process, based on the feature weights of GDP growth rate, industrial production index, consumer confidence index, product prices, inventory level, and sales volume, it was found that GDP growth rate and sales volume had a greater impact on enterprise liquidity.

[0046] The trained early warning model is used to predict the liquidity risks of the three companies. For example, the model might predict that Company 1 may face relocation risk in the next year, Company 2 may face merger risk, and Company 3 may initiate a new transformation plan.

[0047] Such early warning results are for reference. Enterprise decision makers can formulate corresponding strategic deployments and take corresponding risk control measures based on these forecast results to deal with possible adverse situations.

[0048] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0049] It should be understood that the detailed description of the technical solutions of the present invention using the preferred embodiments above is illustrative and not restrictive. A person skilled in the art, after reading the present specification, may modify the technical solutions described in the embodiments or replace some of the technical features therein with equivalents; such modifications or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for monitoring and early warning of enterprise mobility, characterized by: The early warning method includes: Collect historical enterprise flow data, including historical information on enterprise relocation, mergers and acquisitions, bankruptcy or transformation; Collect economic data related to the enterprise, including macroeconomic data related to the industry in which the enterprise is located, and microeconomic data related to the daily business activities of the enterprise; Data preprocessing: preprocess the collected historical flow data and economic data of enterprises; Construct a feature matrix to combine the historical flow data of the enterprise with the economic data related to the enterprise; Build an early warning model and train the early warning model; Use the trained early warning model to predict enterprise liquidity risks regularly or in real time.

2. The enterprise flow monitoring and early warning method according to claim 1 is characterized by: The collection of enterprise historical flow data includes: Data collection: define the types of corporate movements of interest, relocations, mergers and acquisitions, closures or transformations; crawl information about these companies from public data sources, databases or social media; Data annotation: Use relocation, mergers and acquisitions, bankruptcy or transformation as labels to mark the relocation, mergers and acquisitions, bankruptcy or transformation events that have occurred in these companies in history.

3. The enterprise flow monitoring and early warning method according to claim 1 is characterized by: The economic data collected about enterprises include: Use web crawlers to regularly crawl enterprise-related economic data from public data sources. Macroeconomic data includes GDP growth rate, industrial production index, and consumer confidence index; microeconomic data includes product prices, inventory, and sales.

4. The enterprise flow monitoring and early warning method according to claim 1 is characterized by: The construction of the feature matrix includes: Data integration: Integrate the historical flow data of enterprises, arrange the flow events of each enterprise in chronological order, and record the specific time and type of each flow; Normalization: normalize macroeconomic data and microeconomic data; Constructing a feature matrix: For each flow event, extract the macroeconomic and microeconomic data before it occurs, which together constitute the feature vector of this event; the feature vectors of all events are combined into a feature matrix; There are N flow events in total, and each event has M features. The feature matrix is ​​N*M dimensional, denoted by X=(x ij ), where x ij represents the jth feature of the i-th event; Construct a label vector: label each flow event with its type, relocation as 1, merger and acquisition as 2, and bankruptcy or transformation as 3. These labels constitute the label vector; There are N flow events in total, so the label vector is N-dimensional, denoted by Y=(y i ), where y i Represents the label of the i-th event; The feature matrix X and label vector Y are obtained and input into the early warning model for training.

5. The enterprise flow monitoring and early warning method according to claim 1 is characterized by: The described construction of the early warning model adopts the homogeneous multi-task learning algorithm HMTL, and the constructed feature matrix and label vector are input into the HMTL model; There are M tasks and N samples, each sample has P features, then the feature matrix X is a matrix of NP, and the label vector Y is a matrix of NM; The objective function is: ； Among them, Θ is an M*P matrix, each row represents the model parameter θj of each task j; is the feature vector corresponding to sample i; L is the loss function, is the Frobenius norm of the parameter matrix Θ, which is used for regularization of the model to prevent overfitting; λ is the specified regularization strength; The training process includes: The loss function is gradually minimized through repeated iterations. In each iteration, the gradient of the objective function with respect to the parameters is first calculated, and then the parameters are updated along the negative direction of the gradient. For the above objective function, the parameter update rule is expressed as: ； Among them, α is the learning rate, which controls the step size of parameter update, and ∇L is the gradient of the objective function; For each task j, the gradient of the corresponding model parameter θj is calculated and updated, that is, ； The above method is continuously iterated until the loss function converges and a trained early warning model is obtained.

6. The enterprise flow monitoring and early warning method according to claim 5 is characterized by: The learning rate is 0.

001.

7. An enterprise mobility monitoring and early warning system, the system being applicable to the method according to any one of claims 1 to 6, characterized in that: The monitoring and early warning system comprises: Data collection module: used to collect the company's historical flow data, macroeconomic data and microeconomic data related to the industry in which the company is located; Data preprocessing module: used to preprocess the collected historical flow data and economic data of enterprises; Feature matrix building module: used to combine the enterprise's historical flow data with the enterprise's related economic data to build a feature matrix; Early warning model training module: responsible for training the early warning model, including setting the given learning rate and regularization strength parameters, as well as initializing and updating the parameters; Risk prediction module: Use the trained early warning model to derive the potential liquidity risks of the enterprise in the future; Early warning result output module: outputs the predicted enterprise liquidity risk results for reference by decision makers.

Images