SOM-GAN-based enterprise asset flow risk identification and early warning method

Abstract

The invention relates to the technical field of enterprise asset risk management, and provides an SOM-GAN-based enterprise asset flow risk identification and early warning method, which comprises a data acquisition module, a data preprocessing module, an SOM module, a GAN module, a risk identification and evaluation module and an early warning and feedback module. Through comprehensive data acquisition, various data inside and outside an enterprise are covered, and the asset flow condition of the enterprise can be reflected more comprehensively. A cleaning and feature extraction method in data preprocessing ensures data quality and simplifies a data structure. The improved algorithm of the SOM module improves the clustering accuracy and efficiency, and facilitates the accurate analysis of a risk mode. The confrontation training of the GAN module optimizes the data, and the risk identification accuracy is enhanced. And the risk identification and evaluation module performs evaluation based on a multi-module result and outputs comprehensive risk indexes. And the early warning and feedback module timely warns the risk and optimizes the performance of each module of the system, so that the management capability of the enterprise on the asset flow risk is integrally improved.

Description

Technical Field

[0001] The present invention relates to the technical field of enterprise asset risk management, and in particular to a method for identifying and early warning enterprise asset liquidity risks based on SOM-GAN. Background Art

[0002] In today's complex and ever-changing business environment, enterprises are faced with many uncertainties. The effective identification and early warning of enterprise asset liquidity risks have become a key link in enterprise risk management.

[0003] Traditional methods for identifying and warning of enterprise asset flow risks have numerous limitations. For one thing, they often rely solely on a subset of internal enterprise data, such as purely financial data, while neglecting the comprehensive consideration of multiple aspects of information, including business and operational data, as well as external market, industry, and macroeconomic data. This one-sided data source fails to fully reflect the complex factors influencing enterprise asset flows, leading to inaccurate risk identification.

[0004] On the other hand, traditional methods have technical deficiencies in data analysis. For example, a lack of effective data cleaning and feature extraction methods can result in low data quality, with significant noise and redundant information, impacting subsequent analysis results. Furthermore, traditional cluster analysis methods suffer from low clustering accuracy and efficiency when processing enterprise asset liquidity risk data, making it difficult to accurately capture complex patterns within the data. Furthermore, there is a lack of effective optimization mechanisms to improve risk identification accuracy, making them inadequately adaptable to the dynamic nature of enterprise asset liquidity risk.

[0005] Therefore, this paper proposes a SOM-GAN-based enterprise asset liquidity risk identification and early warning method to solve the above problems. Summary of the Invention

[0006] In order to overcome the defects of the prior art, the purpose of the present invention is to provide an enterprise asset liquidity risk identification and early warning method based on SOM-GAN.

[0007] To achieve the purpose, the technical solution of the present invention is implemented as follows: a method for identifying and early warning enterprise asset liquidity risks based on SOM-GAN, characterized by comprising: Data collection module, used to collect relevant data from inside and outside the enterprise; The data preprocessing module is connected to the data acquisition module and is used to clean and extract features from the collected data; The SOM module is connected to the data preprocessing module and is used to perform cluster analysis on the preprocessed data to achieve data dimensionality reduction and pattern recognition. This module uses an improved self-organizing map (SOM) algorithm to improve the accuracy and efficiency of clustering by introducing a dynamic learning rate and adaptive adjustment of the neighborhood function; The GAN module is connected to the data preprocessing module and optimizes the data to improve the accuracy of risk identification through adversarial training of the generator and the discriminator; The risk identification and assessment module is connected to the SOM module and the GAN module respectively, and is used to assess the enterprise asset flow risk level based on the processing results of the SOM module and the GAN module, and output the corresponding risk assessment indicators; The early warning and feedback module is connected to the risk identification and assessment module. When the risk assessment indicator exceeds the preset risk threshold, it issues a risk early warning signal and continuously optimizes the system performance based on the early warning situation.

[0008] Preferably, the internal enterprise data collected by the data acquisition module includes financial data, business data, and operational data, and the external enterprise data includes market data, industry data, and macroeconomic data.

[0009] 3. A method for identifying and warning enterprise asset liquidity risks based on SOM-GAN according to claim 1, characterized in that the data preprocessing module adopts data cleaning methods including removing duplicate data, processing missing values, and correcting erroneous data, and the feature extraction methods include principal component analysis and factor analysis.

[0010] Preferably, the generator of the GAN module is used to generate simulated data, and the discriminator is used to determine whether the data is real data or simulated data generated by the generator. Through adversarial training of the two, the performance of the generator and the discriminator is continuously optimized to improve the accuracy of risk identification.

[0011] Preferably, the risk identification and assessment module assesses the enterprise asset liquidity risk according to a preset risk assessment model and indicator system, determines the risk level, and outputs risk assessment indicators including risk probability, risk loss, etc.

[0012] Preferably, the early warning and feedback module feeds back the warning information to the data acquisition module, data preprocessing module, SOM module, GAN module and risk identification and assessment module while issuing the risk warning signal, so as to optimize the performance and parameters of each module and improve the risk management capability of the system.

[0013] Preferably, the improved self-organizing map (SOM) algorithm in the SOM module specifically includes: Initialization phase: The number and topology of the output layer neurons are determined based on the dimensionality of the enterprise asset liquidity risk data and the expected clustering pattern. The output layer neurons are arranged in a rectangular or hexagonal pattern on a two-dimensional plane. The topology of the neurons is set to adapt to the distribution pattern of the enterprise asset liquidity risk data in different dimensions. The weight vectors of the output layer neurons are randomly initialized. The dimension of the weight vector is the same as the characteristic dimension of the input enterprise asset liquidity risk data. The initial weights are set based on the value range and distribution characteristics of the enterprise asset liquidity risk data. Competition stage: For each input enterprise asset liquidity risk data sample, the distance between it and the weight vector of each neuron in the output layer is calculated. The distance metric uses the Euclidean distance formula. When calculating the distance, normalization is performed on the different types of features in the enterprise asset liquidity risk data to ensure that the weights of different features in the distance calculation are reasonable. In this way, the neuron that best matches the risk data sample is found and is selected as the winning neuron. The winning neuron represents a local optimal pattern of the enterprise asset liquidity risk data in the current cluster analysis. Its potential relationship with the enterprise asset liquidity risk will be further explored in the subsequent learning phase. Learning stage: A dynamic learning rate is adopted, and the initial value of the learning rate is (\eta_0). As the number of iterations (t) increases, it is adjusted according to the formula (\eta(t)=\eta_0\times(1 - t / T)), where (T) is the total number of iterations. This dynamic learning rate adjustment method adapts to the complexity and diversity of enterprise asset liquidity risk data. As the enterprise asset liquidity risk data pattern is continuously learned, the learning rate gradually decreases to stabilize the clustering results and avoid over-adjustment. For the winning neuron and the neurons in its neighborhood, the weight vector is updated according to the formula (w_j(t + 1)=w_j(t)+\eta(t)\times h_{ij}(t)\times(x - w_j(t))), where (w_j(t)) is the weight vector of the (j)th neuron at the (t)th iteration, (x) is the input enterprise asset liquidity risk data sample, and (h_{ij}(t)) is the neighborhood function. The temporal sequence and correlation of enterprise asset liquidity risk data are considered when updating the weight vector. Adaptive adjustment of neighborhood function: The neighborhood function uses a Gaussian function (h_{ij}(t)=\exp(-\frac{d_{ij}^2}{2\sigma^2(t)})), where (d_{ij}) is the distance between the winning neuron (i) and the neuron (j), (\sigma(t)) is the neighborhood radius, and the initial value is (\sigma_0). As the number of iterations (t) increases, it is adjusted according to the formula (\sigma(t)=\sigma_0\times(1 - t / T)). In this way, adaptive adjustment of the neighborhood function is achieved, so that at the beginning of the iteration, a larger range of neurons participate in the weight update to capture the global pattern in the enterprise asset flow risk data. As the iteration proceeds, the range of neurons participating in the update gradually shrinks, thereby focusing on more local and detailed risk data patterns. The GAN module is connected to the data preprocessing module and optimizes the data to improve the accuracy of risk identification through adversarial training of the generator and the discriminator; The risk identification and assessment module is connected to the SOM module and the GAN module respectively, and is used to assess the enterprise asset flow risk level based on the processing results of the SOM module and the GAN module, and output the corresponding risk assessment indicators; The early warning and feedback module is connected to the risk identification and assessment module. When the risk assessment indicator exceeds the preset risk threshold, it issues a risk early warning signal and continuously optimizes the system performance based on the early warning situation.

[0014] Preferably, the Euclidean distance formula is (d=\sqrt{\sum_{k = 1}^{n}(x_k - w_{jk})^2}), where (x_k) is the (k)th eigenvalue of the input enterprise asset liquidity risk data sample, (w_{jk}) is the (k)th component of the (j)th neuron weight vector, and (n) is the characteristic dimension of the input enterprise asset liquidity risk data.

[0015] The beneficial effects of the present invention are embodied in: The present invention covers a variety of data inside and outside the enterprise through comprehensive data collection, which can more comprehensively reflect the asset flow status of the enterprise. The cleaning and feature extraction methods in data preprocessing ensure data quality and simplify the data structure. The improved algorithm of the SOM module improves clustering accuracy and efficiency, which helps to accurately analyze risk patterns. The adversarial training of the GAN module optimizes data and enhances the accuracy of risk identification. The risk identification and assessment module evaluates based on multi-module results and outputs comprehensive risk indicators. The early warning and feedback module timely warns of risks and optimizes the performance of each module of the system, thereby improving the overall management ability of the enterprise for asset flow risks, effectively responding to potential risks, and ensuring the stable flow of enterprise assets and the healthy development of the enterprise. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In the attached figure: Figure 1 Schematic diagram of the connection relationship of each module of the present invention. DETAILED DESCRIPTION

[0017] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Obviously, the embodiments described are only some embodiments of the invention, not all embodiments. In the absence of conflict, the embodiments and features in the embodiments of this application can be combined with each other. Based on the embodiments of the invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the invention.

[0018] Please refer to the instruction manual Figure 1 The present invention provides a specific implementation method of an enterprise asset flow risk identification and early warning method based on SOM-GAN: 1. Data Collection Internal data collection Obtain financial data from the company's financial department, such as the data in the balance sheet, income statement, and cash flow statement, including total assets, total liabilities, operating income, net profit, cash flow from operating activities, etc.

[0019] Collect business data from business departments, such as order quantity, order amount, number of customers, product sales volume, etc.

[0020] The operations department provides operational data, such as the operating time of production equipment, employee working hours, inventory turnover rate, etc.

[0021] External data collection Market data collection: Obtain market data such as market share, competitor data, product price fluctuations, etc. through market research institutions, financial data providers, etc.

[0022] Industry data collection: Pay attention to industry reports released by industry associations to obtain data such as the industry's average profit margin, industry growth rate, and the average debt-to-asset ratio of companies in the industry.

[0023] Macroeconomic data collection: Obtain macroeconomic data from government economic departments and economic research institutions, such as gross domestic product (GDP) growth rate, inflation rate, interest rate, etc.

[0024] 2. Data Preprocessing Data cleaning De-duplicate data: Check the collected data and identify and delete duplicate data records through the data's unique identifier (such as order number, transaction time, etc.).

[0025] Handling missing values: For missing values ​​in financial data, if they are small and have a certain pattern (such as missing data in a certain time period), you can use mean filling, median filling, or interpolation filling based on the trend of the previous and next data; for missing values ​​in business data, fill them according to business logic. For example, missing order amounts can be calculated based on product unit prices and order quantities.

[0026] Correct erroneous data: Check for outliers in the data, such as if the sales volume in a certain month in the financial data is suddenly abnormally high or low, and make corrections by verifying with the business department or based on the distribution of historical data.

[0027] Feature extraction Principal component analysis (PCA) is performed on collected multidimensional data (such as numerous financial indicators). By calculating the data's covariance matrix, the principal components of the data are identified, reducing the data's dimensionality while retaining the data's key information. For example, PCA can be used to transform multiple related financial ratio indicators into a few comprehensive indicators.

[0028] Factor analysis: For some business data and operational data with potential structures, factor analysis methods are used to extract common factors to simplify the data structure and highlight the key features of the data.

[0029] 3. SOM module operation Initialization phase The number and topology of output layer neurons are determined based on the dimensionality of the enterprise asset liquidity risk data (e.g., the dimensions of the data after feature extraction) and the expected clustering patterns. If the data distribution is relatively regular and compact clustering is desired, a rectangular arrangement of output layer neurons can be chosen. If the data exhibits more directional or local correlations, a hexagonal arrangement can be chosen. Assuming that the enterprise asset liquidity risk data has 10 feature dimensions after feature extraction, and based on experience and preliminary analysis, five major clustering patterns are expected, the number of output layer neurons can be determined to be 25 (a 5×5 rectangular arrangement).

[0030] The topology of the neurons is set to adapt to the distribution of asset flow risk data in different dimensions. For example, for asset flow risk data related to time series, the topology can be set to associate them in chronological order.

[0031] The weight vectors of the output layer neurons are randomly initialized. The dimension of the weight vector is the same as the characteristic dimension of the input enterprise asset liquidity risk data, that is, 10 dimensions. The initial weights are set based on the value range and distribution characteristics of the enterprise asset liquidity risk data. If the data range is between [0, 100], and the majority of the data is concentrated between [20, 80], then the initial weights can be randomly generated within this range.

[0032] Competition stage For each input enterprise asset liquidity risk data sample, the distance between it and the weight vector of each neuron in the output layer is calculated. The Euclidean distance formula (d = \sqrt{\sum_{k = 1}^{n}(x_k - w_{jk})^2}) is used, where (x_k) is the kth eigenvalue of the input enterprise asset liquidity risk data sample, (w_{jk}) is the kth component of the weight vector of the jth neuron, and (n) is the characteristic dimension of the input enterprise asset liquidity risk data. When calculating the distance, normalization is performed on the different types of features in the enterprise asset liquidity risk data. For example, the amount data and ratio data in the financial data are normalized separately to ensure that the different features are weighted appropriately in the distance calculation. This allows the neuron that best matches the risk data sample to be identified and designated as the winning neuron.

[0033] Learning stage A dynamic learning rate is used, with an initial value of eta0 = 0.5. Assuming a total number of iterations (T = 100), the learning rate is adjusted as the number of iterations (t) increases according to the formula (eta(t) = eta0 times (1 - t / T)). For the input enterprise asset liquidity risk data sample (x), the weight vector of the winning neuron and its neighboring neurons is updated according to the formula (wj(t + 1) = wj(t) + eta(t) times hij(t) times (x - wj(t))), where wj(t) is the weight vector of the jth neuron at iteration t, and hij(t) is the neighborhood function. The weight vector update considers the temporal nature and correlation of enterprise asset liquidity risk data. For example, for asset liquidity data for consecutive time periods, the weight adjustment of the previous time period is used to adjust the impact on the current time period.

[0034] Adaptive adjustment of neighborhood function The neighborhood function uses a Gaussian function (h_{ij}(t)=\exp(-\frac{d_{ij}^2}{2\sigma^2(t)})), where (d_{ij}) is the distance between the winning neuron (i) and neuron (j), and (\sigma(t)) is the neighborhood radius, with an initial value of (\sigma_0 = 5). As the number of iterations (t) increases, the neighborhood radius is adjusted according to the formula (\sigma(t)=\sigma_0\times(1 -t / T)). For example, in the early stages of the iteration, a relatively large number of neurons participate in weight updates to capture global patterns in the enterprise's asset liquidity risk data. As the iteration progresses (t = 50), the range of neurons participating in the update gradually decreases, focusing on more localized and detailed risk data patterns.

[0035] 4. GAN module operation Generator and Discriminator Initialization Initialize the network structure and parameters of the generator and discriminator of the GAN module. The generator generates simulated data, while the discriminator determines whether the data is real or simulated. For example, the generator can use a multi-layer perceptron (MLP) structure, with input being a random noise vector and output being simulated enterprise asset liquidity risk data. The discriminator also uses a similar MLP structure, with input being real data or simulated data output by the generator, and outputting the probability of whether the data is real or simulated.

[0036] Adversarial training process During training, the discriminator's parameters are first fixed, and the generator is trained. The generator generates simulated enterprise asset liquidity risk data, and the discriminator distinguishes between this simulated data and real data. Based on the discrimination results, the backpropagation algorithm is used to adjust the generator's parameters, making the simulated data closer to real data and thus deceiving the discriminator.

[0037] The parameters of the generator are then fixed and the discriminator is trained. The discriminator distinguishes between real data and simulated data generated by the generator, and adjusts its own parameters to improve the accuracy of the discrimination.

[0038] By repeating the above process, the performance of the generator and discriminator is continuously optimized, and the accuracy of risk identification is improved.

[0039] V. Risk Identification and Assessment Establishment of risk assessment model and indicator system Establish a risk assessment model and indicator system based on the characteristics of corporate asset flows and relevant industry standards. For example, risk assessment indicators might include risk probability and risk loss. Risk probability can be estimated based on the frequency of risks associated with similar asset flows in historical data, while risk loss can be determined based on factors such as the asset's value and the potential extent of impairment.

[0040] Risk level assessment Based on the processing results of the SOM and GAN modules, combined with the risk assessment model and indicator system, the company's asset liquidity risk is assessed. For example, if the SOM cluster analysis shows that the company's asset liquidity data is in a cluster similar to a high-risk pattern, and the GAN module's optimization of the data shows large abnormal fluctuations, then the risk assessment indicator system calculates that the risk probability is high and the risk loss is likely to be large. The company's asset liquidity risk level is determined to be high risk, and risk assessment indicators such as risk probability and risk loss are output.

[0041] VI. Early Warning and Feedback Risk Warning When risk assessment indicators exceed preset risk thresholds—for example, when the risk probability exceeds 50% or the risk loss is expected to exceed the company's tolerance—the early warning and feedback module issues a risk warning signal. This warning signal can be sent to relevant management personnel, such as financial managers and senior executives, via the company's internal information systems (e.g., email systems, instant messaging tools, etc.).

[0042] System performance optimization feedback The early warning information is fed back to the data acquisition module, data preprocessing module, SOM module, GAN module, and risk identification and assessment module. For example, feedback to the data acquisition module can prompt it to increase the frequency of data collection or expand the scope of data collection; feedback to the data preprocessing module can adjust the data cleaning and feature extraction strategies; feedback to the SOM module can adjust the clustering parameters; feedback to the GAN module can optimize the structure and parameters of the generator and discriminator; feedback to the risk identification and assessment module can adjust the risk assessment model and indicator system to optimize the performance and parameters of each module and improve the system's risk management capabilities.

[0043] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for identifying and warning enterprise asset liquidity risks based on SOM-GAN, characterized by: include: Data collection module, used to collect relevant data from inside and outside the enterprise; A data preprocessing module, connected to the data acquisition module, for cleaning and feature extraction of the collected data; The SOM module is connected to the data preprocessing module and is used to perform cluster analysis on the preprocessed data to achieve data dimensionality reduction and pattern recognition. This module adopts an improved self-organizing map (SOM) algorithm and improves the accuracy and efficiency of clustering by introducing a dynamic learning rate and adaptive adjustment of the neighborhood function; A GAN module, connected to the data preprocessing module, optimizes the data through adversarial training of the generator and the discriminator to improve the accuracy of risk identification; A risk identification and assessment module, connected to the SOM module and the GAN module respectively, for assessing the enterprise asset flow risk level based on the processing results of the SOM module and the GAN module, and outputting corresponding risk assessment indicators; The early warning and feedback module is connected to the risk identification and assessment module. When the risk assessment index exceeds the preset risk threshold, it issues a risk early warning signal and continuously optimizes the system performance according to the early warning situation.

2. The enterprise asset liquidity risk identification and early warning method based on SOM-GAN according to claim 1 is characterized in that: The internal enterprise data collected by the data acquisition module includes financial data, business data, and operational data, and the external enterprise data includes market data, industry data, and macroeconomic data.

3. The enterprise asset liquidity risk identification and early warning method based on SOM-GAN according to claim 1 is characterized in that: The data cleaning methods adopted by the data preprocessing module include removing duplicate data, processing missing values, and correcting erroneous data, and the feature extraction methods include principal component analysis and factor analysis.

4. The enterprise asset liquidity risk identification and early warning method based on SOM-GAN according to claim 1 is characterized in that: The generator of the GAN module is used to generate simulated data, and the discriminator is used to determine whether the data is real data or simulated data generated by the generator. Through adversarial training between the two, the performance of the generator and the discriminator is continuously optimized, thereby improving the accuracy of risk identification.

5. The enterprise asset liquidity risk identification and early warning method based on SOM-GAN according to claim 1 is characterized in that: The risk identification and assessment module assesses the enterprise asset liquidity risk based on a preset risk assessment model and indicator system, determines the risk level, and outputs risk assessment indicators including risk probability, risk loss, etc.

6. The enterprise asset liquidity risk identification and early warning method based on SOM-GAN according to claim 1 is characterized in that: When issuing a risk warning signal, the warning and feedback module feeds back the warning information to the data acquisition module, data preprocessing module, SOM module, GAN module and risk identification and assessment module to optimize the performance and parameters of each module and improve the risk management capability of the system.

7. The enterprise asset liquidity risk identification and early warning method based on SOM-GAN according to claim 1 is characterized in that: The improved self-organizing map (SOM) algorithm in the SOM module specifically includes: Initialization phase: The number and topology of the output layer neurons are determined based on the dimensionality of the enterprise asset liquidity risk data and the expected clustering pattern. The output layer neurons are arranged in a rectangular or hexagonal pattern on a two-dimensional plane. The topology of the neurons is set to adapt to the distribution pattern of the enterprise asset liquidity risk data in different dimensions. The weight vectors of the output layer neurons are randomly initialized. The dimension of the weight vector is the same as the characteristic dimension of the input enterprise asset liquidity risk data. The initial weights are set based on the value range and distribution characteristics of the enterprise asset liquidity risk data. Competition stage: For each input enterprise asset liquidity risk data sample, the distance between it and the weight vector of each neuron in the output layer is calculated. The distance metric uses the Euclidean distance formula. When calculating the distance, normalization is performed on the different types of features in the enterprise asset liquidity risk data to ensure that the weights of different features in the distance calculation are reasonable. In this way, the neuron that best matches the risk data sample is found and is selected as the winning neuron. The winning neuron represents a local optimal pattern of the enterprise asset liquidity risk data in the current cluster analysis. Its potential relationship with the enterprise asset liquidity risk will be further explored in the subsequent learning phase. Learning stage: A dynamic learning rate is adopted, and the initial value of the learning rate is (\eta_0). As the number of iterations (t) increases, it is adjusted according to the formula (\eta(t)=\eta_0\times(1 - t / T)), where (T) is the total number of iterations. This dynamic learning rate adjustment method adapts to the complexity and diversity of enterprise asset liquidity risk data. As the enterprise asset liquidity risk data pattern is continuously learned, the learning rate gradually decreases to stabilize the clustering results and avoid over-adjustment. For the winning neuron and the neurons in its neighborhood, the weight vector is updated according to the formula (w_j(t + 1)=w_j(t)+\eta(t)\times h_{ij}(t)\times(x - w_j(t))), where (w_j(t)) is the weight vector of the (j)th neuron at the (t)th iteration, (x) is the input enterprise asset liquidity risk data sample, and (h_{ij}(t)) is the neighborhood function. The temporal sequence and correlation of enterprise asset liquidity risk data are considered when updating the weight vector. Adaptive adjustment of neighborhood function: The neighborhood function uses a Gaussian function (h_{ij}(t)=\exp(-\frac{d_{ij}^2}{2\sigma^2(t)})), where (d_{ij}) is the distance between the winning neuron (i) and the neuron (j), (\sigma(t)) is the neighborhood radius, and the initial value is (\sigma_0). As the number of iterations (t) increases, it is adjusted according to the formula (\sigma(t)=\sigma_0\times(1 - t / T)). In this way, adaptive adjustment of the neighborhood function is achieved, so that at the beginning of the iteration, a larger range of neurons participate in the weight update to capture the global pattern in the enterprise asset flow risk data. As the iteration proceeds, the range of neurons participating in the update gradually shrinks, thereby focusing on more local and detailed risk data patterns. A GAN module, connected to the data preprocessing module, optimizes the data through adversarial training of the generator and the discriminator to improve the accuracy of risk identification; A risk identification and assessment module, connected to the SOM module and the GAN module respectively, for assessing the enterprise asset flow risk level based on the processing results of the SOM module and the GAN module, and outputting corresponding risk assessment indicators; The early warning and feedback module is connected to the risk identification and assessment module. When the risk assessment index exceeds the preset risk threshold, it issues a risk early warning signal and continuously optimizes the system performance according to the early warning situation.

8. The enterprise asset liquidity risk identification and early warning method based on SOM-GAN according to claim 1 is characterized in that: The Euclidean distance formula is (d=\sqrt{\sum_{k = 1}^{n}(x_k - w_{jk})^2}), where (x_k) is the (k)th eigenvalue of the input enterprise asset liquidity risk data sample, (w_{jk}) is the (k)th component of the (j)th neuron weight vector, and (n) is the characteristic dimension of the input enterprise asset liquidity risk data.

Images