Multi-dimensional data analysis-based cross-institution unhealthy asset risk assessment system

Abstract

The invention discloses a multi-dimensional data analysis-based cross-institution risk assessment system for unhealthy assets, which relates to the technical field of unhealthy asset management and comprises a data management module, a trend analysis module, a factor identification module and a risk assessment module. Through cooperative operation of four modules of data management, trend analysis, factor identification and risk assessment, comprehensive perception and dynamic assessment of cross-institution non-performing assets are realized, so that multi-source heterogeneous data can be effectively integrated, key risk features can be extracted, and the value change trend of the non-performing assets can be identified, thereby accurately locking core risk factors, and improving the risk assessment efficiency. And the identification result is compared with the known risk factor, so that the quantitative evaluation of the risk level is realized, the financial institution is assisted in dynamically adjusting the risk prevention and control strategy, and the cross-institution risk linkage response capability and the asset management efficiency are further improved.

Description

Technical Field

[0001] This invention relates to the field of non-performing asset management technology, and more specifically, to a cross-institutional non-performing asset risk assessment system based on multi-dimensional data analysis. Background Technology

[0002] Non-performing assets (NPAs) refer to assets that cannot be recovered on schedule, cannot generate expected returns, or are of poor quality. Specifically, it includes the sum of assets on a financial institution's balance sheet that cannot be recovered on schedule and their expected returns. Besides banks, non-bank financial institutions such as securities companies, funds, trusts, and leasing companies also generate NPAs. These NPAs typically stem from borrower defaults, changes in the market environment, or liquidity problems, affecting the asset quality and capital adequacy of financial institutions. NPAs are diverse, not limited to non-performing loans from banks, but also including NPAs in asset products managed by other financial institutions. With the diversification of financial markets, more and more assets from non-bank financial institutions are gradually becoming sources of NPAs. Therefore, risk assessment for NPAs is not limited to internal bank operations but requires cross-institutional and cross-sectoral analysis and monitoring.

[0003] Existing technologies are not conducive to accurately capturing complex trends in the value of non-performing assets, making it impossible to accurately predict future risk trends. Furthermore, they are not conducive to identifying key factors affecting non-performing asset risks, leading to risk assessments often relying on traditional, static models that lack the ability to respond quickly to market changes. At the same time, they are not conducive to providing timely and accurate risk warnings and effective adjustments to prevention and control strategies, failing to meet the risk management needs of financial institutions in a rapidly changing environment, and increasing potential non-performing asset losses and the instability of the financial system.

[0004] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention proposes a cross-institutional non-performing asset risk assessment system based on multi-dimensional data analysis. This system solves the problems mentioned in the background, such as the inability to accurately capture complex trends in non-performing asset value changes, leading to an inability to accurately predict future risk trends. Furthermore, it struggles to identify key factors influencing non-performing asset risk, resulting in risk assessments often relying on traditional, static models that lack rapid response to market changes. Simultaneously, it fails to provide timely and accurate risk warnings and effective adjustments to control strategies, thus failing to meet the risk management needs of financial institutions in a rapidly changing environment and increasing potential non-performing asset losses and financial system instability.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A cross-institutional non-performing asset risk assessment system based on multidimensional data analysis includes:

[0008] The data management module is used to acquire non-performing asset data across institutions and extract non-performing asset characteristic data;

[0009] The trend analysis module is used to analyze the characteristic data of non-performing assets using value change analysis algorithms to identify the value change trend of non-performing assets.

[0010] The factor identification module is used to identify the value change trend of non-performing assets using risk factor identification algorithms, and to obtain the key factors affecting the risk of non-performing assets.

[0011] The risk assessment module is used to evaluate key factors against known risk factors and, based on the assessment results, adjust the risk control strategy for non-performing assets across institutions.

[0012] Furthermore, the data management module includes:

[0013] The data acquisition module is used to acquire non-performing asset data across institutions, and to perform noise reduction, filtering and smoothing on duplicate data, missing values ​​and outliers in the non-performing asset data to obtain complete non-performing asset data.

[0014] The feature extraction module is used to extract features from complete non-performing asset data to obtain non-performing asset feature data.

[0015] Furthermore, feature extraction is performed on the complete non-performing asset data to obtain non-performing asset feature data, including:

[0016] Standardize the complete dataset of non-performing assets, extract key time-series indicators, and establish a basic analytical framework.

[0017] Based on the industry risk transmission model and asset quality curve characteristics, calculate the benchmark parameters required for feature extraction;

[0018] Hilbert-Huang transform is used to extract the envelope signal reflecting the nature of risk from the original non-performing asset data, thus separating the real risk fluctuations from data noise;

[0019] An iterative generalized demodulation algorithm is used to iteratively demodulate the envelope signal and decompose risk feature components of different dimensions;

[0020] The decomposed risk feature components are subjected to a fast Fourier transform to generate a quantifiable risk feature spectrum.

[0021] Based on the matching relationship between the peak distribution of the spectrum and the preset risk nodes, the characteristic data of non-performing assets are extracted.

[0022] Furthermore, by using value change analysis algorithms to analyze the characteristic data of non-performing assets, the value change trends of non-performing assets are identified, including:

[0023] Correlation analysis was conducted on various influencing factors of non-performing asset characteristic data to screen out factors related to changes in asset value, construct an optimized non-performing asset characteristic dataset, and divide it into training set and test set;

[0024] The value change sequence of non-performing assets is decomposed into several sub-sequences using decomposition techniques; and a trend identification model is established for each sub-sequence to assess the value change trend of non-performing assets in each sub-sequence.

[0025] The parameters of the trend recognition model for each subsequence are optimized using a search algorithm to obtain the optimized trend recognition model.

[0026] The optimized trend identification model is used to predict the value change trend of non-performing assets in each subsequence, and the value change trends of non-performing assets in each subsequence are integrated and reconstructed to form the final value change trend of non-performing assets.

[0027] Furthermore, a correlation analysis was conducted on various influencing factors of non-performing asset characteristic data, and factors related to changes in asset value were identified, including:

[0028] Extract the various influencing factors to be analyzed, and set the initial parameters and maximum number of iterations for the correlation analysis algorithm;

[0029] Candidate combinations of influencing factors are generated based on the characteristics of factor changes, and perturbation factors and inverse factors are introduced to expand the dimensions of analysis;

[0030] Calculate the correlation between each candidate combination of influencing factors and changes in asset value to form a multi-objective optimal set of influencing factors and preliminarily screen key combinations of influencing factors;

[0031] If there are the most candidate combinations of influencing factors, then redundant factors and combinations of influencing factors with the lowest correlation are eliminated, and the key factor combinations that have the greatest impact on changes in asset value are retained to optimize the set of influencing factors.

[0032] Determine if the maximum number of iterations has been reached. If so, output the combination of key factors that have the greatest impact on changes in asset value as factors related to changes in asset value. Otherwise, continue to expand the analysis dimensions.

[0033] Furthermore, the parameters of the trend recognition model for each subsequence are optimized using a search algorithm, resulting in the optimized trend recognition model, which includes:

[0034] Set the maximum number of iterations and channel time for the search algorithm, and initialize parameter configurations for each subsequence model;

[0035] Evaluate the performance of the trend recognition model in the trend recognition task based on the current parameter configuration of the trend recognition model for each subsequence;

[0036] If the parameter configuration of the trend recognition model for a certain subsequence performs optimally, then update the parameters of the trend recognition model for that subsequence and expand its search range.

[0037] When the channel time reaches the threshold, the search step size and direction are reset, the search strategy is adjusted, and the global search continues.

[0038] Compare the performance of the parameter configurations of the trend recognition model for the current subsequence with those of the historical subsequence trend recognition models, retain the best-performing parameter configuration, and discard the worst-performing parameter configuration.

[0039] When the maximum number of iterations is reached, the optimal parameter configuration of the trend recognition model for all subsequences is output, resulting in the optimized trend recognition model.

[0040] Furthermore, by using risk factor identification algorithms to identify the value change trends of non-performing assets, the key factors affecting the risk of non-performing assets include:

[0041] Input historical data on non-performing assets and related risk factors, and set the parameters and maximum number of iterations for the risk factor identification algorithm;

[0042] Chaotic mapping is used to initialize the risk factor combination, and the initial risk factor combination is non-dominated and ranked to evaluate the quality of each risk factor combination.

[0043] According to the rules of the risk factor identification algorithm, selection, crossover and mutation operations are performed to generate offspring risk factor combinations, and the parent risk factor combinations and offspring risk factor combinations are merged into a risk factor combination set.

[0044] The metaheuristic algorithm is used to optimize the risk factor combination set, generate new offspring risk factor combinations, and merge the parent risk factor combination with the new offspring risk factor combination to form a new risk factor combination set.

[0045] Apply a fast non-dominated sorting, crowding calculation, and elite selection strategy to the new set of risk factor combinations to generate the optimal set of risk factor combinations.

[0046] Determine if the maximum number of iterations has been reached. If not, continue with selection, crossover, and mutation operations. Otherwise, output the final optimal set of risk factors as the key factors affecting the risk of non-performing assets.

[0047] Furthermore, a non-dominated ranking is performed on the initialized risk factor combinations, and the merits of each risk factor combination are evaluated, including:

[0048] Configure a list of the number of times each risk factor is dominated and the combinations thereof;

[0049] The set of risk factors that are not controlled by any portfolio is classified as the first-level non-controlled portfolio risk factor portfolio;

[0050] Iterate through the first-level non-dominated risk factor combinations and reduce the number of times the risk factor combinations it controls are dominated.

[0051] The risk factor combination that has zero dominance in this round is classified as the second-level non-dominant risk factor combination;

[0052] Continue the segmentation process until all risk factor combinations are fully segmented into multi-level non-dominated risk factor combination sets and classified into different Pareto frontier levels, thus completing the classification of all risk factor combinations.

[0053] After all risk factor combinations are categorized, the merits of each risk factor combination are assessed based on its performance in the multi-objective evaluation.

[0054] Furthermore, the risk factor combination set is optimized using a metaheuristic algorithm to generate new offspring risk factor combinations, including:

[0055] Set the parameters and maximum number of iterations for the metaheuristic algorithm;

[0056] Obtain the initial set of risk factor combinations and map the initial set of risk factor combinations to the original search space;

[0057] Each combination of risk factors is evaluated, and invalid combinations of risk factors that do not meet the constraints are eliminated;

[0058] The optimization phase is determined based on the escape energy intensity of the current combination of risk factors. When the energy is at its highest, the exploration phase is entered; otherwise, the partial development phase is entered.

[0059] Update the position of the risk factor combination at the corresponding stage, perform Cauchy-Gaussian perturbation on the best performing risk factor combination, and retain the new best risk factor combination if the effect is the best after mutation.

[0060] If the maximum number of iterations is reached, output the new optimal combination of risk factors; otherwise, continue iterative optimization.

[0061] Furthermore, the risk assessment module includes:

[0062] The risk comparison module is used to compare and analyze the key factors affecting the risk of non-performing assets with the risk factors in the known database to identify risk patterns.

[0063] The strategy adjustment module is used to adjust the corresponding risk control strategies for non-performing assets across institutions based on the identified risk patterns.

[0064] The beneficial effects of this invention are as follows:

[0065] 1. This invention achieves comprehensive perception and dynamic assessment of non-performing assets across institutions through the coordinated operation of four modules: data management, trend analysis, factor identification, and risk assessment. It enables the effective integration of multi-source heterogeneous data, extraction of key risk characteristics, identification of trends in the value of non-performing assets, and precise identification of core risk factors. Furthermore, by comparing the identification results with known risk factors, it achieves quantitative assessment of risk levels and assists financial institutions in dynamically adjusting risk prevention and control strategies, thereby improving cross-institutional risk response capabilities and asset management efficiency.

[0066] 2. This invention achieves accurate prediction of non-performing asset value and in-depth mining of risk factors through value change analysis algorithms. It uses correlation analysis to screen key influencing factors, combines empirical mode decomposition technology to decompose complex value changes into multiple predictable subsequences, and then optimizes the prediction model parameters of each subsequence through search algorithms. Finally, it integrates these subsequences to form an accurate prediction of value change trends, thereby improving the accuracy and timeliness of risk assessment. This enables financial institutions to identify risk signals earlier and locate the source of risk more accurately, providing reliable data support for formulating precise risk prevention and control strategies.

[0067] 3. This invention, through a risk factor identification algorithm, combined with non-dominated sorting and metaheuristic optimization, can accurately identify key factors affecting non-performing asset risk from complex value change trends. It uses chaotic mapping to initialize risk factor combinations and systematically judges the merits of each factor through rapid non-dominated sorting and multi-objective evaluation. At the same time, it combines Cauchy-Gaussian perturbation mechanism to improve global search and local optimization capabilities, making the finally identified risk factors more scientific and comprehensive, and effectively enhancing the accuracy and adaptability of cross-institutional non-performing asset risk identification. Attached Figure Description

[0068] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0069] Figure 1This is a schematic diagram of a cross-institutional non-performing asset risk assessment system based on multidimensional data analysis according to an embodiment of the present invention.

[0070] In the picture:

[0071] 1. Data Management Module; 2. Trend Analysis Module; 3. Factor Identification Module; 4. Risk Assessment Module. Detailed Implementation

[0072] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0073] In the description of this invention, unless otherwise stated, "a plurality of" means two or more. Furthermore, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0074] According to an embodiment of the present invention, a cross-institutional non-performing asset risk assessment system based on multidimensional data analysis is provided.

[0075] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, the cross-institutional non-performing asset risk assessment system based on multidimensional data analysis according to an embodiment of the present invention includes:

[0076] Data management module 1 is used to acquire non-performing asset data across institutions and extract non-performing asset characteristic data;

[0077] Specifically, obtaining non-performing asset data across institutions includes:

[0078] Financial institutions (such as banks, trust companies, and securities companies) regularly publish their balance sheets, which include information on non-performing loans and non-performing assets. Third-party financial data platforms, such as Wind and Choice, provide financial statements and risk assessment reports from various financial institutions, including data related to non-performing assets. Regulatory agencies regularly publish statistical data on non-performing assets for the banking, securities, and other financial sectors, covering the overall non-performing asset situation across institutions.

[0079] Specifically, non-performing asset data includes: non-performing loans, non-performing debts, bad debts and impairment provisions, default event data, etc.

[0080] Specifically, the characteristics of non-performing assets include: asset type, debtor default information, credit rating, market environment factors, and number of overdue days.

[0081] Trend Analysis Module 2 is used to analyze the characteristic data of non-performing assets using a value change analysis algorithm to identify the value change trend of non-performing assets.

[0082] Factor identification module 3 is used to identify the value change trend of non-performing assets using risk factor identification algorithms, and to obtain the key factors affecting the risk of non-performing assets.

[0083] Specifically, key factors affecting the risk of non-performing assets include: asset-level factors, debtor or asset-related parties factors, market and economic environment factors, internal factors of financial institutions, and asset transaction and circulation factors.

[0084] Risk assessment module 4 is used to evaluate key factors against known risk factors and adjust risk control strategies for non-performing assets across institutions based on the assessment results.

[0085] In this optional embodiment, the data management module 1 includes:

[0086] The data acquisition module is used to acquire non-performing asset data across institutions, and to perform noise reduction, filtering and smoothing on duplicate data, missing values ​​and outliers in the non-performing asset data to obtain complete non-performing asset data.

[0087] The feature extraction module is used to extract features from complete non-performing asset data to obtain non-performing asset feature data.

[0088] In this optional embodiment, feature extraction is performed on the complete non-performing asset data to obtain non-performing asset feature data, including:

[0089] Standardize the complete dataset of non-performing assets, extract key time-series indicators, and establish a basic analytical framework.

[0090] Based on the industry risk transmission model and asset quality curve characteristics, calculate the benchmark parameters required for feature extraction;

[0091] Hilbert-Huang transform is used to extract the envelope signal reflecting the nature of risk from the original non-performing asset data, thus separating the real risk fluctuations from data noise;

[0092] An iterative generalized demodulation algorithm is used to iteratively demodulate the envelope signal and decompose risk feature components of different dimensions;

[0093] The decomposed risk feature components are subjected to a fast Fourier transform to generate a quantifiable risk feature spectrum.

[0094] Based on the matching relationship between the peak distribution of the spectrum and the preset risk nodes, the characteristic data of non-performing assets are extracted.

[0095] It should be explained that, firstly, the original non-performing asset data from across institutions is cleaned and standardized, and key time-series indicators are extracted. Based on the industry risk transmission model and asset quality curves, the baseline parameters required for feature extraction are calculated. The Hilbert-Huang transform is used to separate the envelope signal reflecting the essence of risk, and then an iterative generalized demodulation algorithm is used for deep decomposition to achieve multi-dimensional analysis of risk characteristics. Subsequently, a fast Fourier transform is performed on the risk characteristics of each dimension, converting them into a frequency domain representation to generate a risk characteristic spectrum, which intuitively reflects the risk intensity of each frequency range. Finally, by combining the spectrum peak values ​​with preset node matching, non-performing asset characteristic data is extracted.

[0096] In this optional embodiment, the value change analysis algorithm is used to analyze the characteristic data of non-performing assets and identify the value change trend of non-performing assets, including:

[0097] Correlation analysis was conducted on various influencing factors of non-performing asset characteristic data to screen out factors related to changes in asset value, construct an optimized non-performing asset characteristic dataset, and divide it into training set and test set;

[0098] The value change sequence of non-performing assets is decomposed into several sub-sequences using decomposition techniques (i.e., empirical mode decomposition techniques); and a trend identification model is established for each sub-sequence to evaluate the value change trend of non-performing assets in each sub-sequence.

[0099] The parameters of the trend recognition model for each subsequence are optimized using a search algorithm to obtain the optimized trend recognition model.

[0100] The optimized trend identification model is used to predict the value change trend of non-performing assets in each subsequence, and the value change trends of non-performing assets in each subsequence are integrated and reconstructed to form the final value change trend of non-performing assets.

[0101] Specifically, the value change analysis algorithm is the kernel extreme learning machine, which introduces the kernel function concept on the basis of extreme learning machine (ELM), thereby effectively overcoming the problem of low output stability caused by ELM due to random generation of initial weights and thresholds, and improving output stability.

[0102] To facilitate understanding of the above technical solutions of the present invention, the following provides a detailed description of how the present invention uses a value change analysis algorithm to analyze the characteristic data of non-performing assets and identify the value change trend of non-performing assets in actual practice.

[0103] Step 1: Collect cross-institutional non-performing asset data (1000 records in total). Each record includes: historical asset value (monthly value, unit: yuan), such as [3100, 3200, 3150, ..., 3300]. Debtor's financial condition (debt ratio %), such as 85%. Macroeconomic indicators (such as GDP growth rate %), such as 4.5%. Industry development stage (scoring system 0-1), such as 0.6. Standardize all features to the range [0, 1]. For example, the minimum asset value is 1000, the maximum is 5000, and the current value is 3000. Standardized value = (3000-1000) / (5000-1000) = 0.5. Debt ratio from 0% to 100%, currently 85% → standardized value is 0.85. Use the Pearson correlation coefficient to calculate the relationship with changes in asset value. For example, asset value vs. debt ratio: r = 0.85. Asset value vs. GDP growth rate: r = -0.60. Asset value vs. industry development stage: r = 0.70. Features with high correlation coefficients are selected for modeling. Using empirical mode decomposition, the standardized value time series (past 12 months) of a certain non-performing asset is assumed to be:

[0104] x(m)=[0.45, 0.47, 0.49, 0.48, 0.50, 0.51, ..., 0.55].

[0105] The sequence was decomposed into four subsequences (IMF) using EMD:

[0106] x(m)=IMF1(m)+IMF2(m)+IMF3(m)+IMF4(m).

[0107] For example, suppose that EMD decomposition yields four subsequences (IMF1, IMF2, IMF3, IMF4).

[0108] in:

[0109] IMF1: Long-term trend, value range such as [0.40~0.53].

[0110] IMF2: Medium-term cyclical fluctuations (quarterly fluctuations).

[0111] IMF3: Short-term fluctuations (e.g., brief declines or rebounds).

[0112] IMF4: High-frequency noise (perturbation values ​​[-0.01, 0.01]).

[0113] Step 2: Establish a trend forecasting model using the ARIMA model. The initial model settings and forecasting results are shown in Table 1.

[0114] Table 1: Initial Model Setup and Prediction Results

[0115] IMF subsequence Initial model Current forecast value <![CDATA[IMF1]]> ARIMA(2, 1, 2) 0.51 <![CDATA[IMF2]]> ARIMA(1, 0, 1) 0.03 <![CDATA[IMF3]]> ARIMA(1, 0, 0) -0.01 <![CDATA[IMF4]]> ARIMA(0, 0, 0) 0.05

[0116] Step 3: Optimize the parameters of the trend identification model for each subsequence using a search algorithm. Taking IMF1 as an example, the parameters are updated from (2, 1, 2) to (4, 1, 2) through the search algorithm. The optimization result is that the original mean squared error was 0.05, and the optimized mean squared error is 0.02.

[0117] The updated predicted values ​​are shown in Table 2:

[0118] Table 2: Updated Predictions

[0119]

[0120]

[0121] Step 4: Use the optimized model to predict the trend value of each subsequence, including: IMF1 = 0.53, IMF2 = 0.04, IMF3 = -0.015, and IMF4 = 0.003. The influence weights of each IMF are shown in Table 3.

[0122] Table 3: Impact Weights of Each IMF

[0123] IMF subsequence Weight <![CDATA[IMF1]]> 0.4 <![CDATA[IMF2]]> 0.3 <![CDATA[IMF3]]> 0.2 <![CDATA[IMF4]]> 0.1

[0124] The final predicted trend value is calculated as follows:

[0125]

[0126] Denormalize the standardized values ​​to the actual asset value (minimum 1000, maximum 5000):

[0127] Asset value forecast = 0.2213 × (5000 - 1000) + 1000 = 1885.2.

[0128] The final prediction of the value change trend of the non-performing asset in the future cycle is 1885.2, which provides quantifiable support for risk assessment and asset management.

[0129] In this optional embodiment, correlation analysis is performed on various influencing factors of non-performing asset characteristic data to screen out factors related to changes in asset value, including:

[0130] Extract the various influencing factors to be analyzed, and set the initial parameters and maximum number of iterations for the correlation analysis algorithm;

[0131] Candidate combinations of influencing factors are generated based on the characteristics of factor changes, and perturbation factors and inverse factors are introduced to expand the dimensions of analysis;

[0132] Calculate the correlation between each candidate combination of influencing factors and changes in asset value to form a multi-objective optimal set of influencing factors and preliminarily screen key combinations of influencing factors;

[0133] If there are the most candidate combinations of influencing factors, then redundant factors and combinations of influencing factors with the lowest correlation are eliminated, and the key factor combinations that have the greatest impact on changes in asset value are retained to optimize the set of influencing factors.

[0134] Determine if the maximum number of iterations has been reached. If so, output the combination of key factors that have the greatest impact on changes in asset value as factors related to changes in asset value. Otherwise, continue to expand the analysis dimensions.

[0135] Specifically, the correlation analysis algorithm is a quantitative association rule mining algorithm based on multi-objective fireworks algorithm and reverse learning. This algorithm comprehensively searches for association rules through multi-objective fireworks algorithm, introduces reverse learning to improve the convergence speed of the algorithm and reduce the probability of the algorithm getting stuck in local optima, and uses a similarity-based redundancy elimination mechanism to maintain the diversity of association rules in the library. After multiple iterations, the final set of association rules is obtained.

[0136] It's important to explain that, firstly, various factors that may influence changes in asset value need to be extracted from historical data on non-performing assets. Based on these extracted factors, multiple candidate combinations of influencing factors are generated. To increase the dimensionality of the analysis, perturbation factors and inverse factors are introduced. The inclusion of these factors helps detect potential, non-linear correlations. A correlation analysis algorithm is then used to analyze each candidate combination of influencing factors and calculate its correlation with changes in the value of non-performing assets. Once the maximum number of iterations is reached or convergence occurs, the combination of key factors that has the greatest impact on changes in asset value is output. This combination will serve as the final set of influencing factors related to changes in asset value, used for subsequent asset valuation, risk management, and decision support.

[0137] In this optional embodiment, the parameters of the trend recognition model for each subsequence are optimized using a search algorithm to obtain the optimized trend recognition model, which includes:

[0138] Set the maximum number of iterations and channel time for the search algorithm, and initialize parameter configurations for each subsequence model;

[0139] Evaluate the performance of the trend recognition model in the trend recognition task based on the current parameter configuration of the trend recognition model for each subsequence;

[0140] If the parameter configuration of the trend recognition model for a certain subsequence performs optimally, then update the parameters of the trend recognition model for that subsequence and expand its search range.

[0141] When the channel time reaches the threshold, the search step size and direction are reset, the search strategy is adjusted, and the global search continues.

[0142] Compare the performance of the parameter configurations of the trend recognition model for the current subsequence with those of the historical subsequence trend recognition models, retain the best-performing parameter configuration, and discard the worst-performing parameter configuration.

[0143] When the maximum number of iterations is reached, the optimal parameter configuration of the trend recognition model for all subsequences is output, resulting in the optimized trend recognition model.

[0144] Specifically, the search algorithm is the lightning search algorithm, which is an optimization algorithm inspired by the natural phenomenon of lightning. It achieves global optimization by mimicking the characteristics of lightning propagation.

[0145] It's important to explain the process: First, a search algorithm is selected, and the maximum number of iterations and channel time are set. The maximum number of iterations is the upper limit of the executable search, and channel time refers to the search time for each subsequence model. The parameter configuration for each subsequence trend recognition model is initialized, and the task performance of the current configuration is evaluated in each search. If it's better than historical configurations, the new configuration is updated, and the search scope is expanded. If the channel time reaches a threshold, the search step size and direction are reset to avoid getting trapped in local optima and enhance global search capabilities. The best-performing configuration is retained, and the worst-performing configuration is discarded. Finally, the optimal parameters for each subsequence are output for subsequent trend prediction tasks.

[0146] In this optional embodiment, a risk factor identification algorithm is used to identify the value change trend of non-performing assets, and the key factors affecting the risk of non-performing assets include:

[0147] Input historical data on non-performing assets and related risk factors, and set the parameters and maximum number of iterations for the risk factor identification algorithm;

[0148] Chaotic mapping is used to initialize the risk factor combination, and the initial risk factor combination is non-dominated and ranked to evaluate the quality of each risk factor combination.

[0149] According to the rules of the risk factor identification algorithm, selection, crossover and mutation operations are performed to generate offspring risk factor combinations, and the parent risk factor combinations and offspring risk factor combinations are merged into a risk factor combination set.

[0150] The metaheuristic algorithm is used to optimize the risk factor combination set, generate new offspring risk factor combinations, and merge the parent risk factor combination with the new offspring risk factor combination to form a new risk factor combination set.

[0151] Apply a fast non-dominated sorting, crowding calculation, and elite selection strategy to the new set of risk factor combinations to generate the optimal set of risk factor combinations.

[0152] Determine if the maximum number of iterations has been reached. If not, continue with selection, crossover, and mutation operations. Otherwise, output the final optimal set of risk factors as the key factors affecting the risk of non-performing assets.

[0153] Specifically, the risk factor identification algorithm, NSGA-II (Non-dominated Sorting Genetic Algorithm II), is a classic multi-objective optimization algorithm. By improving the selection mechanism of genetic algorithms and combining non-dominated sorting and crowding distance strategies, it effectively solves multi-objective optimization problems (MOPs), simultaneously optimizing multiple conflicting objectives and generating a set of uniformly distributed Pareto optimal solutions.

[0154] Specifically, the metaheuristic algorithm is the Harris Eagle Optimization Algorithm, a novel metaheuristic intelligent optimization algorithm. Its idea comes from the hunting mechanism of the Harris Eagle. The optimization process is divided into an exploration phase and a development phase. The process transition is controlled by the escape energy of the prey. As the prey escapes, the escape energy gradually decreases, thus transforming from the exploration phase of global exploration to the development phase of local search.

[0155] It is important to explain that the process begins with collecting historical data on non-performing assets and the relevant risk factors affecting their value. To enhance global search capabilities, chaotic mapping is used to initialize the risk factor combinations, ensuring diversity in the search space. Subsequently, the initialized combinations are non-dominated and ranked to identify combinations that perform well across multiple objectives, and the priority of each combination is evaluated accordingly. The ranking results reflect the superior performance of different combinations in multi-objective optimization. Next, according to the rules of the risk factor identification algorithm, selection, crossover, and mutation operations are performed on the combinations to further optimize the combination structure and improve the quality of the solution. This generates the final set of optimal risk factor combinations. Finally, the generated set of optimal risk factor combinations is output. These combinations, as key factors influencing the risk of non-performing assets, can help identify and predict asset risks.

[0156] In this optional embodiment, performing a non-dominated ranking of the initialized risk factor combinations and evaluating the merits of each risk factor combination includes:

[0157] Configure a list of the number of times each risk factor is dominated and the combinations thereof;

[0158] The set of risk factors that are not controlled by any portfolio is classified as the first-level non-controlled portfolio risk factor portfolio;

[0159] Iterate through the first-level non-dominated risk factor combinations and reduce the number of times the risk factor combinations it controls are dominated.

[0160] The risk factor combination that has zero dominance in this round is classified as the second-level non-dominant risk factor combination;

[0161] Continue the segmentation process until all risk factor combinations are fully segmented into multi-level non-dominated risk factor combination sets and classified into different Pareto frontier levels, thus completing the classification of all risk factor combinations.

[0162] After all risk factor combinations are categorized, the merits of each risk factor combination are assessed based on its performance in the multi-objective evaluation.

[0163] It's important to explain that before starting the non-dominated ranking, each risk factor combination needs to be assigned a dominance count and a list of combinations it dominates. By comparing each combination pairwise, combinations not dominated by other combinations are identified and classified as the first-level non-dominated combinations, i.e., the optimal solution set. The first-level combinations are then iterated through, reducing the dominance count of their dominated combinations; if it reaches zero, they are moved to the second level. This process continues until all combinations are assigned to their respective non-dominated levels. Each level represents solutions of the same priority. After ranking, a further evaluation of the merits of each combination is conducted based on multi-objective assessment.

[0164] In this optional embodiment, optimizing the risk factor combination set using a metaheuristic algorithm to generate new offspring risk factor combinations includes:

[0165] Set the parameters and maximum number of iterations for the metaheuristic algorithm;

[0166] Obtain the initial set of risk factor combinations and map the initial set of risk factor combinations to the original search space;

[0167] Each combination of risk factors is evaluated, and invalid combinations of risk factors that do not meet the constraints are eliminated;

[0168] The optimization phase is determined based on the escape energy intensity of the current combination of risk factors. When the energy is at its highest, the exploration phase is entered; otherwise, the partial development phase is entered.

[0169] Update the position of the risk factor combination at the corresponding stage, perform Cauchy-Gaussian perturbation on the best performing risk factor combination, and retain the new best risk factor combination if the effect is the best after mutation.

[0170] Specifically, the formula for applying the Cauchy-Gaussian perturbation to the optimal combination of risk factors is:

[0171] W(t)=Ha(t)[1+μ1Cauchy(0,1)+μ2Gauss(0,1)];

[0172] In the formula, W(t) represents the position of the risk factor combination generated after the perturbation after the t-th iteration; H a (t) represents the position of the optimal combination of risk factors after the t-th iteration; μ1 represents the adjustment factor controlling the amplitude of the Cauchy perturbation; μ2 represents the adjustment factor controlling the amplitude of the Gaussian perturbation; Cauchy(0,1) represents the Cauchy distribution with a median of 0 and a scale parameter of 1; Gauss(0,1) represents the Gaussian distribution with a mean of 0 and a standard deviation of 1.

[0173] If the maximum number of iterations is reached, output the new optimal combination of risk factors; otherwise, continue iterative optimization.

[0174] It should be explained that, firstly, a metaheuristic algorithm is selected and parameters are set. Initial solutions are generated using historical data, expert experience, or random methods, and then mapped back to the original search space to ensure that the combinations do not deviate from the actual problem domain. The performance of each risk factor combination is evaluated; if the constraints are not met, it is considered invalid and discarded. The search stage is determined based on the "escape energy intensity," which reflects the algorithm's ability to escape local optima. Depending on the current stage, the combination position is updated using Cauchy-Gaussian perturbations. If the maximum number of iterations is reached, optimization stops, and the current optimal combination is output as the final result.

[0175] In this optional embodiment, the risk assessment module 4 includes:

[0176] The risk comparison module is used to compare and analyze the key factors affecting the risk of non-performing assets with the risk factors in the known database to identify risk patterns.

[0177] The strategy adjustment module is used to adjust the corresponding risk control strategies for non-performing assets across institutions based on the identified risk patterns.

[0178] Specifically, risk prevention and control strategies include risk mitigation and diversification strategies, risk control and management strategies, non-performing asset disposal strategies, and capital adequacy and reserve strategies.

[0179] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A cross-institutional non-performing asset risk assessment system based on multi-dimensional data analysis, characterized in that: include: The data management module is used to acquire non-performing asset data across institutions and extract non-performing asset characteristic data; The trend analysis module is used to analyze the characteristic data of non-performing assets using value change analysis algorithms to identify the value change trend of non-performing assets. The factor identification module is used to identify the value change trend of non-performing assets using risk factor identification algorithms, and to obtain the key factors affecting the risk of non-performing assets. The risk assessment module is used to evaluate key factors against known risk factors and, based on the assessment results, adjust the risk control strategy for non-performing assets across institutions.

2. The cross-institutional non-performing asset risk assessment system based on multi-dimensional data analysis according to claim 1, characterized in that, The data management module includes: The data acquisition module is used to acquire non-performing asset data across institutions, and to perform noise reduction, filtering and smoothing on duplicate data, missing values ​​and outliers in the non-performing asset data to obtain complete non-performing asset data. The feature extraction module is used to extract features from complete non-performing asset data to obtain non-performing asset feature data.

3. The cross-institutional non-performing asset risk assessment system based on multi-dimensional data analysis according to claim 2, characterized in that, The feature extraction from the complete non-performing asset data to obtain non-performing asset feature data includes: Standardize the complete dataset of non-performing assets, extract key time-series indicators, and establish a basic analytical framework. Based on the industry risk transmission model and asset quality curve characteristics, calculate the benchmark parameters required for feature extraction; Hilbert-Huang transform is used to extract the envelope signal reflecting the nature of risk from the original non-performing asset data, thus separating the real risk fluctuations from data noise; An iterative generalized demodulation algorithm is used to iteratively demodulate the envelope signal and decompose risk feature components of different dimensions; The decomposed risk feature components are subjected to a fast Fourier transform to generate a quantifiable risk feature spectrum. Based on the matching relationship between the peak distribution of the spectrum and the preset risk nodes, the characteristic data of non-performing assets are extracted.

4. The cross-institutional non-performing asset risk assessment system based on multi-dimensional data analysis according to claim 1, characterized in that, The analysis of non-performing asset characteristic data using the value change analysis algorithm to identify the value change trend of non-performing assets includes: Correlation analysis was conducted on various influencing factors of non-performing asset characteristic data to screen out factors related to changes in asset value, construct an optimized non-performing asset characteristic dataset, and divide it into training set and test set; The value change sequence of non-performing assets is decomposed into several sub-sequences using decomposition techniques; and a trend identification model is established for each sub-sequence to assess the value change trend of non-performing assets in each sub-sequence. The parameters of the trend recognition model for each subsequence are optimized using a search algorithm to obtain the optimized trend recognition model. The optimized trend identification model is used to predict the value change trend of non-performing assets in each subsequence, and the value change trends of non-performing assets in each subsequence are integrated and reconstructed to form the final value change trend of non-performing assets.

5. The cross-institutional non-performing asset risk assessment system based on multi-dimensional data analysis according to claim 4, characterized in that, The correlation analysis of various influencing factors on the characteristic data of non-performing assets identified factors related to changes in asset value as follows: Extract the various influencing factors to be analyzed, and set the initial parameters and maximum number of iterations for the correlation analysis algorithm; Candidate combinations of influencing factors are generated based on the characteristics of factor changes, and perturbation factors and inverse factors are introduced to expand the dimensions of analysis; Calculate the correlation between each candidate combination of influencing factors and changes in asset value to form a multi-objective optimal set of influencing factors and preliminarily screen key combinations of influencing factors; If there are the most candidate combinations of influencing factors, then redundant factors and combinations of influencing factors with the lowest correlation are eliminated, and the key factor combinations that have the greatest impact on changes in asset value are retained to optimize the set of influencing factors. Determine if the maximum number of iterations has been reached. If so, output the combination of key factors that have the greatest impact on changes in asset value as factors related to changes in asset value. Otherwise, continue to expand the analysis dimensions.

6. The cross-institutional non-performing asset risk assessment system based on multi-dimensional data analysis according to claim 4, characterized in that, The process of optimizing the parameters of the trend recognition model for each sub-sequence using a search algorithm to obtain the optimized trend recognition model includes: Set the maximum number of iterations and channel time for the search algorithm, and initialize parameter configurations for each subsequence model; Evaluate the performance of the trend recognition model in the trend recognition task based on the current parameter configuration of the trend recognition model for each subsequence; If the parameter configuration of the trend recognition model for a certain subsequence performs optimally, then update the parameters of the trend recognition model for that subsequence and expand its search range. When the channel time reaches the threshold, the search step size and direction are reset, the search strategy is adjusted, and the global search continues. Compare the performance of the parameter configuration of the trend recognition model of the current subsequence with the parameter configuration of the trend recognition model of the historical subsequence, retain the parameter configuration with the best performance, and discard the parameter configuration with the worst performance. When the maximum number of iterations is reached, the optimal parameter configuration of the trend recognition model for all subsequences is output, resulting in the optimized trend recognition model.

7. The cross-institutional non-performing asset risk assessment system based on multi-dimensional data analysis according to claim 1, characterized in that, The risk factor identification algorithm is used to identify the value change trend of non-performing assets, and the key factors affecting the risk of non-performing assets include: Input historical data on non-performing assets and related risk factors, and set the parameters and maximum number of iterations for the risk factor identification algorithm; Chaotic mapping is used to initialize the risk factor combination, and the initial risk factor combination is non-dominated and ranked to evaluate the quality of each risk factor combination. According to the rules of the risk factor identification algorithm, selection, crossover and mutation operations are performed to generate offspring risk factor combinations, and the parent risk factor combinations and offspring risk factor combinations are merged into a risk factor combination set. The metaheuristic algorithm is used to optimize the risk factor combination set, generate new offspring risk factor combinations, and merge the parent risk factor combination with the new offspring risk factor combination to form a new risk factor combination set. Apply a fast non-dominated sorting, crowding calculation, and elite selection strategy to the new set of risk factor combinations to generate the optimal set of risk factor combinations. Determine if the maximum number of iterations has been reached. If not, continue with selection, crossover, and mutation operations. Otherwise, output the final optimal set of risk factors as the key factors affecting the risk of non-performing assets.

8. The cross-institutional non-performing asset risk assessment system based on multidimensional data analysis according to claim 7, characterized in that, The process of performing a non-dominated ranking of the initialized risk factor combinations and evaluating the merits of each risk factor combination includes: Configure a list of the number of times each risk factor is dominated and the combinations thereof; The set of risk factors that are not controlled by any portfolio is classified as the first-level non-controlled portfolio risk factor portfolio; Iterate through the first-level non-dominated risk factor combinations and reduce the number of times the risk factor combinations it controls are dominated. The risk factor combination that has zero dominance in this round is classified as the second-level non-dominant risk factor combination; Continue the segmentation process until all risk factor combinations are fully segmented into multi-level non-dominated risk factor combination sets and classified into different Pareto frontier levels, thus completing the classification of all risk factor combinations. After all risk factor combinations are categorized, the merits of each risk factor combination are assessed based on its performance in the multi-objective evaluation.

9. The cross-institutional non-performing asset risk assessment system based on multidimensional data analysis according to claim 7, characterized in that, The optimization of the risk factor combination set using a metaheuristic algorithm to generate new offspring risk factor combinations includes: Set the parameters and maximum number of iterations for the metaheuristic algorithm; Obtain the initial set of risk factor combinations and map the initial set of risk factor combinations to the original search space; Each combination of risk factors is evaluated, and invalid combinations of risk factors that do not meet the constraints are eliminated; The optimization phase is determined based on the escape energy intensity of the current combination of risk factors. When the energy is at its highest, the exploration phase is entered; otherwise, the partial development phase is entered. Update the position of the risk factor combination at the corresponding stage, perform Cauchy-Gaussian perturbation on the best performing risk factor combination, and retain the new best risk factor combination if the effect is the best after mutation. If the maximum number of iterations is reached, output the new optimal combination of risk factors; otherwise, continue iterative optimization.

10. The cross-institutional non-performing asset risk assessment system based on multi-dimensional data analysis according to claim 1, characterized in that, The risk assessment module includes: The risk comparison module is used to compare and analyze the key factors affecting the risk of non-performing assets with the risk factors in the known database to identify risk patterns. The strategy adjustment module is used to adjust the corresponding risk control strategies for non-performing assets across institutions based on the identified risk patterns.

Images