Enterprise credit risk prevention and control method and system fusing electric power data and financial data

Abstract

The invention discloses an enterprise credit risk prevention and control method and system fusing electric power data and financial data, and the method comprises the following steps: obtaining the electric power data and financial data of an enterprise, and fusing the electric power data and financial data to generate fused data of the enterprise; performing feature zooming according to the fused data, and creating new feature items to generate interactive features; and constructing a credit risk assessment model, inputting the fusion data and the interaction features after feature zooming into the credit risk assessment model, and obtaining a credit risk assessment value of the customer. By combining the power data and the financial data, the financial institution can assess the operation condition and financial health degree of the enterprise more comprehensively and accurately, and the effectiveness and stability of risk control are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of financial risk management, and in particular relates to an enterprise credit risk prevention and control method and system integrating power data and financial data. Background Art

[0002] Credit risk management refers to a loan management activity in which banks use scientific methods to effectively predict, analyze, prevent, control and deal with various subjective factors that may lead to loan losses, so as to reduce loan risks, reduce loan losses, improve loan quality, and thus enhance the bank's risk control ability and loss compensation ability. Commercial banks must not only rely on the continuous and regular return of assets to obtain income and conduct re-lending business to meet the demand for funds for social and economic development, but also ensure the liquidity of liability business through the return of loans. If loans are in arrears and difficult to recover, it will have a direct catastrophic impact on the normal operation of commercial banks. Therefore, it is very necessary to manage loan risks. In the current credit risk management, there is a problem of excessive reliance on single financial data and inability to accurately capture the daily operations of enterprises, which leads to the inaccuracy of risk assessment models and difficulty in discovering and preventing potential credit risks in advance.

[0003] The Chinese invention patent application with publication number CN116823433A discloses a method and system for credit risk prevention and control of distributed photovoltaic projects. The method, as a data sharing mechanism, adopts a user two-way authorization mechanism. The data ownership belongs to the enterprise. The enterprise authorizes the power company to provide data and authorizes the bank to use the power company's data. The enterprise obtains loan project related information through the above-mentioned sharing mechanism, integrates artificial intelligence algorithms and balanced scorecard principles, calculates some key indicator values ​​in the balanced scorecard through artificial intelligence algorithms, and integrates the key indicators formed by the data of all parties by weight assignment to obtain risk quantification points, thereby expanding the risk assessment dimension.

[0004] The model design of the above scheme is originally intended for specific types of photovoltaic projects, while other types of projects may require different modeling strategies, which may limit the applicability in different scenarios. In addition, during the weight optimization process, if the selected historical data samples are not representative, the weight distribution of key indicators may be unreasonable. Historical data may be affected by market fluctuations and policy changes, resulting in insufficient data stability. Summary of the invention

[0005] The present invention provides an enterprise credit risk prevention and control method and system integrating power data and financial data, aiming to solve the problems of insufficient applicability and stability of the prior art in enterprise credit risk prevention and control.

[0006] In order to solve the above technical problems, the enterprise credit risk prevention and control method proposed in the present invention comprises the following steps:

[0007] Acquire power data and financial data of the enterprise, and fuse the power data and financial data to generate fused data of the enterprise;

[0008] Perform feature scaling according to the fused data, and create new feature items to generate interactive features;

[0009] A credit risk assessment model is constructed, and the fused data and interactive features after feature scaling are input into the credit risk assessment model to obtain the credit risk assessment value of the customer.

[0010] Preferably, the method for generating the fusion data is specifically as follows:

[0011] Set the target frequency and resample the power data according to the target frequency;

[0012] Setting a financial data time window, and generating an average financial indicator during the period according to the financial data time window and the target frequency;

[0013] Align the resampled power data and average financial indicators according to the timestamps corresponding to the target frequency;

[0014] The aligned data are combined together to form a time series dataset containing power indicators and financial indicators at each timestamp.

[0015] Preferably, the method for generating the interaction feature is specifically:

[0016] Select two features in the fused data, and multiply the two features to obtain an initial interaction feature;

[0017] The initial interaction features are standardized to obtain the interaction features. The standardization method is as follows:

[0018]

[0019] In the formula, F i is the calculated interaction feature, F o is the initial interaction feature, is the mean of all initial interaction features, and is the standard deviation of all initial interaction features.

[0020] Preferably, the credit risk assessment model is a decision tree model, which uses entropy to assess the importance of features:

[0021]

[0022] In the formula, Entropy(S) is entropy, c is the number of categories, and p is i is the probability of category i, S is the sample set;

[0023] Then calculate the information gain of each feature to select the best split feature:

[0024]

[0025] Where Gain(S, A) is the information gain of feature A in sample set S, S v It is the subset when feature A takes the value of v, and the feature with the largest information gain is selected as the best split feature.

[0026] Preferably, the decision tree model calculates the Gini index of each feature, and selects the feature with the smallest weighted Gini index as the best split feature. The calculation method of the Gini index is:

[0027] Calculate all possible values ​​of a feature and divide them;

[0028] Calculate the Gini index of the subset for each value:

[0029]

[0030] In the formula, S v is the subset when feature A takes value v, Gini(S v ) is the corresponding Gini index, c is the number of categories, p i|v It is a subset S v The proportion of category i in ;

[0031] Calculate the weighted average of the Gini index:

[0032]

[0033] Where Gain w (S, A) is the weighted Gini index of feature A, and S is the sample set.

[0034] Preferably, the method for the credit risk assessment model to obtain the credit risk assessment value of the customer is specifically:

[0035] Divide the dataset into training and testing sets;

[0036] Constructing a credit risk assessment model:

[0037]

[0038] Where P(Y=1|X) is the probability of default of a customer given feature X, β 0 The intercept term, β 1 , β 2 , …, β n is the corresponding feature X 1 , X 2 , …, X nThe regression coefficient of

[0039] Estimate the parameter β using the maximum likelihood estimation method using the training set 0 , β 1 , β 2 , …, β n ;

[0040] Use the testing machine to test the constructed credit risk assessment model and select the best model;

[0041] Use the best model to obtain the customer's default probability as the customer's credit risk assessment value.

[0042] Preferably, the credit risk assessment model calculation method is as follows:

[0043] Y t =μ+φ 1 Y t-1 +φ 2 Y t-2 +…+φ p Y t-p +θ 1 ε t-1 +θ 2 ε t-2 +…+θ q ε t-q +ε t

[0044] Where Y t is the predicted value of the time series at time t, μ is the intercept term, φ i is the autoregressive coefficient, which indicates the relationship between the current value and the values ​​of the previous p moments. p is the order of the autoregressive term, which indicates how many past values ​​are used to predict the current value. i is the autoregressive coefficient, which indicates the relationship between the current value and the error term at the previous q moments. q is the order of the moving average term, which indicates how many past error terms are used to predict the current value. t is the random error term, which represents the part not explained by the model and is usually assumed to be white noise;

[0045] Estimate model parameters using minimization of residual sum of squares or maximum likelihood estimation methods;

[0046] A risk scoring model is set to convert the predicted value into a credit risk score, wherein the risk scoring model is a linear relationship model or a nonlinear relationship model.

[0047] Preferably, the electricity data includes electricity consumption, electricity payment habits, and electricity bills, and the financial data includes the company's financial statements, credit records, and loan repayment history.

[0048] Preferably, the method further preprocesses the power data or financial data, and the preprocessing includes data cleaning and data normalization.

[0049] Another aspect of the present invention further provides an enterprise credit risk assessment system integrating power data and financial data. The system is used to implement the above-mentioned enterprise credit risk prevention and control method, including:

[0050] Data collection module, used to collect the company's power data and financial data. Power data includes power consumption, electricity payment habits and electricity bills. Financial data includes the company's financial statements, credit records and loan repayment history.

[0051] Data processing module, used to process the collected power data and financial data. The processing methods include data cleaning, data normalization, and feature interaction;

[0052] The credit risk assessment model is used to assess the customer's credit risk based on the characteristic data output by the data processing model and output the customer's credit risk assessment value.

[0053] Compared with the prior art, the present invention has the following technical effects:

[0054] 1. The risk prevention and control method proposed in the present invention combines power data and financial data, so that financial institutions can more comprehensively and accurately evaluate the operating conditions and financial health of enterprises, improve the stability of enterprise credit risk assessment, thereby reducing credit risks and improving credit approval efficiency. It not only improves the effectiveness of risk control, but also provides stronger data-driven support for financial institutions.

[0055] 2. The risk prevention and control method proposed in the present invention converts power data and financial data into the same time frequency, thereby ensuring the time consistency during fusion, and can effectively align real-time or near real-time power data with financial data based on the end of the day or other frequencies, thereby achieving time consistency. This provides a basis for subsequent data analysis and credit risk assessment, and improves the accuracy and reliability of decision-making.

[0056] 3. The risk prevention and control method proposed in the present invention can better capture the complex relationship between the two by creating new feature items, thereby improving the performance of the model. It not only enhances the expressive power of data, but also provides richer information for credit risk assessment.

[0057] 4. The risk prevention and control method proposed in the present invention uses more common enterprise power data to avoid the impact of fluctuations in photovoltaic data due to seasonal, climate and policy changes, and improves the applicability of the risk prevention and control method in different scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1It is a schematic diagram of the process of the prevention and control method of the present invention. DETAILED DESCRIPTION

[0059] In order to make the objectives, technical solutions and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in combination with specific embodiments of the present application and with reference to the accompanying drawings.

[0060] Embodiment 1

[0061] See also Figure 1 This embodiment is a method for preventing and controlling corporate credit risks by integrating power data and financial data, including the following steps 1 to 3:

[0062] Step 1: Obtain the power data and financial data of the enterprise, and integrate the power data and financial data to generate the integrated data of the enterprise. The power data is obtained from the power supply unit according to the authorization of the enterprise, and the financial data is collected through the financial statements, credit records and loan repayment history provided by the enterprise, and also includes all the corresponding enterprise data of the unit. Establish a data platform to integrate the power data and financial data to form a comprehensive data file of the enterprise. The power data includes power consumption, electricity payment habits, and electricity bills, and the financial data includes the financial statements, credit records, and loan repayment history of the enterprise.

[0063] The method also preprocesses the power data or financial data, and the preprocessing includes data cleaning and data normalization.

[0064] In this step, the method for generating the fusion data is specifically as follows:

[0065] Set the target frequency and resample the power data according to the target frequency. The target frequency can determine an appropriate time frequency, such as hourly, daily or weekly frequency, based on the analysis requirements and data characteristics. Usually, power data is more real-time, and choosing a daily frequency may be a more appropriate choice. Resample the power data according to the target frequency (such as daily), calculate the total power consumption, average power consumption, maximum power consumption, etc., and generate relevant indicators of electricity bill payment habits, such as payment days, number of overdue payments, etc.

[0066] Set a financial data time window and generate average financial indicators during the financial data time window and target frequency. For example, process the data according to the time frequency of financial data (such as end of day): if the financial data is end of day data, it can be directly summarized by date; or the average financial indicators of the past N days can be generated to better reflect the financial health of the enterprise.

[0067] The resampled power data and average financial indicators are aligned according to the timestamp corresponding to the target frequency. In this embodiment, the processed power data and financial data are aligned according to the date to ensure that there is corresponding power and financial data for each date. For the case of missing values, the following interpolation method or previous value filling method can be used to supplement, where the interpolation method performs linear interpolation or other interpolation methods on the missing power data, and the previous value filling method can use the data of the previous day to fill the missing values ​​for financial data.

[0068] The aligned data are combined together to form a time series data set, which contains power indicators and financial indicators at each timestamp. Table 1 is a time series data set of some power data and financial data of an enterprise.

[0069]

[0070] Table 1. Examples of time series datasets

[0071] Step 2: perform feature scaling based on the fused data and create new feature items to generate interactive features.

[0072] Feature scaling involves standardizing or normalizing numerical features to ensure that different features are on the same scale, improving the convergence speed and stability of the model in subsequent steps. Create new feature items and consider creating new features (such as the interaction term between electricity consumption and debt-to-asset ratio) to capture potential nonlinear relationships, thereby improving the performance of the model. Taking electricity consumption and debt-to-asset ratio as an example, in some cases, changes in electricity consumption may interact with the level of debt-to-asset ratio, that is, companies with high electricity consumption may perform differently under high debt-to-asset ratios than companies under low debt-to-asset ratios. Directly using these two features may not be able to fully capture this complex relationship.

[0073] The interaction term can be generated by multiplying two features. Two features in the fused data are selected and the initial interaction feature is obtained by multiplying the two features; it is expressed as follows:

[0074] Interaction characteristic = electricity consumption × debt-to-asset ratio

[0075] The interaction features are calculated for each row in the dataset and added as new columns. Taking Table 1 as an example, the new interaction term representation is generated as shown in Table 2.

[0076]

[0077] Table 2 Interaction item examples The initial interaction features are standardized to obtain the interaction features. The standardization method is as follows:

[0078]

[0079] In the formula, F i is the calculated interaction feature, F o is the initial interaction feature, is the mean of all initial interaction features, and is the standard deviation of all initial interaction features.

[0080] Step three: construct a credit risk assessment model, input the feature-scaled fusion data and interactive features into the credit risk assessment model, and obtain the customer's credit risk assessment value.

[0081] In one embodiment of the present invention, the credit risk assessment model is a decision tree model, which uses entropy to assess the importance of features:

[0082]

[0083] In the formula, Entropy(S) is entropy, c is the number of categories, and p is i is the probability of category i, S is the sample set;

[0084] Then calculate the information gain of each feature to select the best split feature:

[0085]

[0086] Where Gain(S, A) is the information gain of feature A in sample set S, S v It is the subset when feature A takes the value of v, and the feature with the largest information gain is selected as the best split feature.

[0087] In this embodiment, the performance of the decision tree model can be evaluated using the following indicators:

[0088] Confusion matrix, which represents the comparison between the true category and the predicted category;

[0089] Precision: the ratio of predicted positive classes to actual positive classes;

[0090]

[0091] Recall is the ratio of positive class to positive class.

[0092]

[0093] F1 score, the harmonic mean of precision and recall;

[0094]

[0095] In the above formula, TP means that the predicted positive class is actually positive, FP means that the predicted positive class is actually negative, and FN means that the predicted negative class is actually positive. The above evaluation indicators can be selected according to the actual needs of the project during implementation.

[0096] Those skilled in the art will appreciate that, for decision trees, it is usually not necessary to standardize features, but if continuous features are included and other model comparisons are required, standardization may be selected. Therefore, the value input to the decision tree model can be feature data before standardization or feature data after standardization.

[0097] In another embodiment of the present invention, the decision tree model calculates the Gini index of each feature, and selects the feature with the smallest weighted Gini index as the best split feature. The calculation method of the Gini index is:

[0098] Calculate all possible values ​​of a feature and divide them;

[0099] Calculate the Gini index of the subset for each value:

[0100]

[0101] In the formula, S v is the subset when feature A takes value v, Gini(S v ) is the corresponding Gini index, c is the number of categories, p i|v It is a subset S v The proportion of category i in ;

[0102] Calculate the weighted average of the Gini index:

[0103]

[0104] Where Gain w (S, A) is the weighted Gini index of feature A, and S is the sample set.

[0105] In another embodiment of the present invention, the method for the credit risk assessment model to obtain the credit risk assessment value of the customer is specifically:

[0106] Divide the dataset into training and testing sets;

[0107] Constructing a credit risk assessment model:

[0108]

[0109] Where P(Y=1|X) is the probability of default of a customer given feature X, β 0 The intercept term, β 1 , β 2 , …, β nis the corresponding feature X 1 , X 2 , …, X n The regression coefficient of

[0110] Estimate the parameter β using the maximum likelihood estimation method using the training set 0 , β 1 , β 2 , …, β n ;

[0111] Use the testing machine to test the constructed credit risk assessment model and select the best model;

[0112] Use the best model to obtain the customer's default probability as the customer's credit risk assessment value.

[0113] In another embodiment of the present invention, the credit risk assessment model calculation method is as follows:

[0114] Divide the time series data into training set and test set, usually the first 70%-80% as training set and the last 20%-30% as test set;

[0115] Y t =μ+φ 1 Y t-1 +φ 2 Y t-2 +…+φ p Y t-p +θ 1 ε t-1 +θ 2 ε t-2 +…+θ q ε t-q +ε t

[0116] Where Y t is the predicted value of the time series at time t, μ is the intercept term, φ i is the autoregressive coefficient, which indicates the relationship between the current value and the values ​​of the previous p moments. p is the order of the autoregressive term, which indicates how many past values ​​are used to predict the current value. i is the autoregressive coefficient, which indicates the relationship between the current value and the error term at the previous q moments. q is the order of the moving average term, which indicates how many past error terms are used to predict the current value. t is a random error term, representing the part not explained by the model, and is usually assumed to be white noise.

[0117] The model parameters are estimated using the minimization of the residual sum of squares or the maximum likelihood estimation method; the performance of the model is evaluated using the root mean square error (RMSE) or the mean absolute percentage error (MAPE):

[0118]

[0119]

[0120] A risk scoring model is set to convert the predicted value into a credit risk score. The risk scoring model is a linear relationship model or a nonlinear relationship model. The risk level can be determined by setting a threshold. If the risk score is higher than a certain threshold, it is judged as high risk, otherwise it is judged as low risk. This method can capture trends and seasonal changes in time series and help financial institutions make more accurate credit decisions.

[0121] Embodiment 2

[0122] This embodiment is a corporate credit risk assessment system integrating power data and financial data. The system is used to implement the corporate credit risk prevention and control method as described in the first embodiment, including:

[0123] Data collection module, used to collect the company's power data and financial data. Power data includes power consumption, electricity payment habits and electricity bills. Financial data includes the company's financial statements, credit records and loan repayment history.

[0124] Data processing module, used to process the collected power data and financial data. The processing methods include data cleaning, data normalization, and feature interaction;

[0125] The credit risk assessment model is used to assess the customer's credit risk based on the characteristic data output by the data processing model and output the customer's credit risk assessment value.

[0126] The above is only a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several modifications and improvements without departing from the inventive concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. An enterprise credit risk prevention and control method integrating power data and financial data, characterized in that: The following steps are involved: Acquire power data and financial data of the enterprise, and fuse the power data and financial data to generate fused data of the enterprise; Perform feature scaling according to the fused data, and create new feature items to generate interactive features; A credit risk assessment model is constructed, and the fused data and interactive features after feature scaling are input into the credit risk assessment model to obtain the credit risk assessment value of the customer.

2. The enterprise credit risk prevention and control method integrating power data and financial data according to claim 1 is characterized in that: The method for generating the fusion data is specifically as follows: Set the target frequency and resample the power data according to the target frequency; Setting a financial data time window, and generating an average financial indicator during the period according to the financial data time window and the target frequency; Align the resampled power data and average financial indicators according to the timestamps corresponding to the target frequency; The aligned data are combined together to form a time series dataset containing power indicators and financial indicators at each timestamp.

3. The enterprise credit risk prevention and control method integrating power data and financial data according to claim 1 is characterized in that: The method for generating the interactive feature is specifically as follows: Select two features in the fused data, and multiply the two features to obtain an initial interaction feature; The initial interaction features are standardized to obtain the interaction features. The standardization method is as follows: In the formula, F i is the calculated interaction feature, F o is the initial interaction feature, is the mean of all initial interaction features, and σ is the standard deviation of all initial interaction features.

4. The enterprise credit risk prevention and control method integrating power data and financial data according to claim 1 is characterized in that: The credit risk assessment model is a decision tree model, which uses entropy to assess the importance of features: In the formula, Entropy(S) is entropy, c is the number of categories, and p is i is the probability of category i, S is the sample set; Then calculate the information gain of each feature to select the best split feature: Where Gain(S, A) is the information gain of feature A in sample set S, S v It is the subset when feature A takes the value of v, and the feature with the largest information gain is selected as the best split feature.

5. The enterprise credit risk prevention and control method integrating power data and financial data according to claim 4 is characterized in that: The decision tree model calculates the Gini index of each feature and selects the feature with the smallest weighted Gini index as the best split feature. The calculation method of the Gini index is: Calculate all possible values ​​of a feature and divide them; Calculate the Gini index of the subset for each value: In the formula, S v is the subset when feature A takes value v, Gini(S v ) is the corresponding Gini index, c is the number of categories, p i|v It is a subset S v The proportion of category i in ; Calculate the weighted average of the Gini index: Where Gain w (S, A) is the weighted Gini index of feature A, and S is the sample set.

6. The enterprise credit risk prevention and control method integrating power data and financial data according to claim 1 is characterized in that: The method for the credit risk assessment model to obtain the credit risk assessment value of the customer is specifically as follows: Divide the dataset into training and testing sets; Constructing a credit risk assessment model: Where P(Y=1|X) is the default probability of a customer given feature X, β0 is the intercept term, β1, β2, …, β n is the corresponding feature X1, X2, ..., X n The regression coefficient of Use the training set to estimate the parameters β0, β1, β2, …, β by maximum likelihood estimation. n ; Use the testing machine to test the constructed credit risk assessment model and select the best model; Use the best model to obtain the customer's default probability as the customer's credit risk assessment value.

7. The enterprise credit risk prevention and control method integrating power data and financial data according to claim 1 is characterized in that: The calculation method of the credit risk assessment model is as follows: Y t =μ+φ1Y t-1 +φ2Y t-2 +…+φ p Y t-p +θ1ε t-1 +θ2ε t-2 +…+θ q e t-q +e t Where Y t is the predicted value of the time series at time t, μ is the intercept term, φ i is the autoregressive coefficient, which indicates the relationship between the current value and the values ​​of the previous p moments. p is the order of the autoregressive term, which indicates how many past values ​​are used to predict the current value. i is the autoregressive coefficient, which indicates the relationship between the current value and the error term at the previous q moments. q is the order of the moving average term, which indicates how many past error terms are used to predict the current value. t is the random error term, which represents the part not explained by the model and is usually assumed to be white noise; Estimate model parameters using minimization of residual sum of squares or maximum likelihood estimation methods; A risk scoring model is set to convert the predicted value into a credit risk score, wherein the risk scoring model is a linear relationship model or a nonlinear relationship model.

8. The enterprise credit risk prevention and control method integrating power data and financial data according to claim 1 is characterized in that: The electricity data includes electricity consumption, electricity payment habits, and electricity bills, and the financial data includes the company's financial statements, credit records, and loan repayment history.

9. The enterprise credit risk prevention and control method integrating power data and financial data according to claim 8 is characterized in that: The method also preprocesses the power data or financial data, and the preprocessing includes data cleaning and data normalization.

10. An enterprise credit risk assessment system integrating power data and financial data, characterized in that: The system is used to implement the enterprise credit risk prevention and control method according to any one of claims 1 to 9, comprising: Data collection module, used to collect the company's power data and financial data. Power data includes power consumption, electricity payment habits and electricity bills. Financial data includes the company's financial statements, credit records and loan repayment history. Data processing module, used to process the collected power data and financial data. The processing methods include data cleaning, data normalization, and feature interaction; The credit risk assessment model is used to assess the customer's credit risk based on the characteristic data output by the data processing model and output the customer's credit risk assessment value.

Images