BiLSTM-FLAML-based enterprise transaction risk identification method for incomplete power data

Abstract

The invention relates to the technical field of intelligent prediction, in particular to a BiLSTM-FLAML-based enterprise transaction risk identification method for incomplete power data, which comprises the following steps of: firstly, identifying and marking abnormal values and vacancy values of a large amount of power data, intercepting the data before and after marking points as a prediction set, and taking the remaining parts as a training set and a test set; secondly, inputting the processed training set and the test set into a BiLSTM model for training; then, inputting a prediction set into the model for prediction, carrying out data filling, and carrying out feature extraction on the filled data according to a set calculation index; and finally, inputting the extracted indexes into the FLAML model to carry out power enterprise transaction risk identification, so that support can be provided for risk judgment of the power enterprise in the transaction, and the accuracy and reliability of risk identification are improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent prediction technology, and in particular to a method for identifying enterprise transaction risks based on incomplete power data using BiLSTM-FLAML. Background Art

[0002] In recent years, with the deepening of the market-oriented reform of electricity, the amount of electricity traded in the national electricity market has continued to rise. Data show that from January to December 2023, the amount of electricity traded in the national electricity market reached 5.7 trillion kWh, a year-on-year increase of 7.9%, accounting for 61.4% of the total electricity consumption in the society, an increase of 0.6 percentage points over the previous year. At the same time, the number of entities registered with trading institutions reached 708,000, indicating that the vitality of the market has been effectively stimulated. However, with the deepening of market-oriented reforms, the diversity and complexity of participants in the electricity market have also continued to increase, and various bad behaviors have emerged in the market, such as arrears, breach of contract, unfair competition and dissemination of false information. These problems not only undermine the fairness and transparency of the market, but also bring significant economic losses and operational risks to enterprises.

[0003] In this context, it is particularly important to establish a sound customer credit evaluation mechanism. Credit evaluation is not only an important means for enterprises to evaluate customer behavior, but also helps enterprises predict and avoid potential credit risks, thereby promoting the healthy development of the market. Financial status, credit records, and data in the power trading process are key indicators for credit scoring. However, due to the existence of bad behavior, missing values ​​and outliers often appear in the data. These data quality issues not only weaken the predictive ability of risk prediction models, but may also lead to incorrect risk judgments, which in turn affect corporate decision-making and market stability.

[0004] Therefore, in response to these shortcomings, how to effectively deal with data missing and outlier problems and improve the accuracy and reliability of risk identification models is a technical problem that technical personnel in this field urgently need to solve. Summary of the invention

[0005] In view of this, the purpose of the present invention is to propose a method for identifying enterprise transaction risks based on incomplete power data based on BiLSTM-FLAML, so as to solve the problem that risk identification cannot be performed accurately and reliably in the prior art.

[0006] Based on the above purpose, the present invention provides a method for identifying enterprise transaction risks based on incomplete power data based on BiLSTM-FLAML, comprising the following steps:

[0007] S1. Identify and mark abnormal values ​​and missing values ​​in historical power data, extract part of the data before and after the marked point as the prediction set, and use the remaining part as the training set and test set;

[0008] S2, input the processed training set and test set into the BiLSTM model for training;

[0009] S3, input the prediction set into the trained BiLSTM model for prediction, perform data filling, and extract features of the filled data according to the established calculation indicators;

[0010] S4. Input the extracted indicators into the FLAML model to identify transaction risks of power enterprises.

[0011] Preferably, step S1 further comprises:

[0012] S11, using the Z-score method to identify abnormal values ​​and missing values ​​of each power data in turn, and record their names, dates, and serial numbers for calibration;

[0013] S12, intercepting the data before and after the calibration value and saving it as a prediction set;

[0014] S13. Pack the remaining power data into training sets and test sets according to a certain ratio.

[0015] Preferably, step S2 further comprises:

[0016] S21, normalize the training set and test set and divide them into different small batches;

[0017] S22, input the data into the BiLSTM model for training in batches, and gradually optimize the model and its parameters through multiple rounds of iterations and back propagation algorithms to minimize the loss function;

[0018] S23. Use the test set batch data to evaluate the performance of the model to monitor the generalization ability of the model and prevent overfitting.

[0019] Preferably, step S3 further comprises:

[0020] S31, normalize the prediction set, and predict missing values ​​and outliers through the model in batches;

[0021] S32, denormalizing the predicted data, and filling and correcting them into the original power data;

[0022] S33. Extract features from the padded data according to given calculation indicators to reduce the data dimension.

[0023] Preferably, step S4 further comprises:

[0024] S41, normalizing the extracted indicators;

[0025] S42, inputting the normalized indicators into the FLAML model for training, and selecting the machine learning model and hyperparameters with the smallest loss function;

[0026] S43. Complete the identification of transaction risks of power companies.

[0027] Beneficial effects of the present invention: In order to solve the problem of missing values ​​and outliers that often appear in data, the present invention adopts a method of identifying and marking outliers and missing values ​​in a large amount of power data, training a BiLSTM model, filling in data, and extracting features. Finally, these extracted indicators are input into the FLAML model to identify transaction risks of power companies. The method can improve the predictive ability of the risk prediction model, provide support for risk judgment of power companies in transactions, and improve the accuracy and reliability of risk identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0029] Figure 1 This is a flow chart of a method for identifying enterprise transaction risks based on incomplete power data based on BiLSTM-FLAML according to an embodiment of the present invention;

[0030] Figure 2 A schematic diagram of a calculation index system according to an embodiment of the present invention;

[0031] Figure 3 Schematic diagram of the FLAML algorithm framework of an embodiment of the present invention. DETAILED DESCRIPTION

[0032] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments.

[0033] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0034] like Figure 1 As shown, the embodiment of this specification provides a method for identifying enterprise transaction risks based on incomplete power data based on BiLSTM-FLAML, including the following steps:

[0035] Step S1, identifying and marking abnormal values ​​and missing values ​​of a large amount of power data, intercepting part of the data before and after the marked point as a prediction set, and using the remaining part as a training set and a test set;

[0036] Step S2, input the processed training set and test set into the BiLSTM model for training;

[0037] Step S3, input the prediction set into the model for prediction, perform data filling, and extract features from the filled data according to the established calculation index;

[0038] Step S4, inputting the extracted indicators into the FLAML model to identify the transaction risks of power enterprises.

[0039] For step S1, first, the Z-score method is used to identify abnormal values ​​and missing values ​​of each power data in turn, and their names, dates, and serial numbers are recorded for calibration. Secondly, the data before and after the calibration value are cut off and saved as the prediction set. If it does not reach ten digits, it is filled with zero values; finally, the remaining power data is packaged into training sets and test sets at a ratio of 8:2.

[0040] For step S2, first, the training set and test set are normalized and divided into different small batches, and the batch size is set to 64. Secondly, the batches are input into the BiLSTM model for training. Through multiple rounds of iterations and back propagation algorithms, the model and its parameters are gradually optimized to minimize the loss function, and the loss function is set to mean square error (MSE). Finally, the performance of the model is evaluated using the test set batch data to monitor the generalization ability of the model and prevent overfitting.

[0041] For step S3, first, the prediction set is normalized to the maximum and minimum, and the maximum and minimum values ​​are recorded. The missing values ​​and abnormal values ​​are predicted in batches through the trained BiLSTM model; secondly, the predicted data is denormalized using the recorded maximum and minimum values, and filled and corrected to the original power data; finally, the filled data is feature extracted according to the given calculation indicators to reduce the data dimension. The specific calculation indicators are as follows:

[0042] 1) Off-site evaluation indicators

[0043] The off-site evaluation mainly evaluates the comprehensive capabilities and credit status of power market entities outside the market. The evaluation cycle is once a year. This time frame allows a comprehensive review of annual financial statements and credit records, so that the indicators are consistent with the fiscal year reports of most companies.

[0044] 1.1) Financial Status

[0045] Financial health is an important indicator for evaluating the credit capacity of market entities. The indicator is calculated by analyzing financial ratios such as debt-to-asset ratio and return on equity.

[0046] 1.2) Credit record

[0047] Credit records are an important basis for evaluating whether market entities have engaged in bad credit behavior in the past. They mainly include corporate credit records, social credit records, etc., and the number of occurrences of bad credit behavior is counted as an indicator.

[0048] 2) On-site evaluation indicators

[0049] On-site evaluation mainly targets the performance of market entities within the electricity market and is divided into full-process evaluation indicators and penalty indicators.

[0050] 2.1) Whole process evaluation indicators

[0051] These indicators directly reflect the credit status of market entities throughout the entire transaction process, and are usually counted on a monthly, quarterly, or annual basis, including the following aspects:

[0052] Operational Capability: refers to the overall ability of market entities to maintain market transactions and management, which is represented by analyzing user stickiness, sustained market share, etc., and the sum of the average monthly transaction rate of the same users in the market is used as an indicator.

[0053] Transaction management: refers to the ability of market entities to conduct and monitor market transactions. By analyzing transaction participation rate, timeliness of interaction and other characteristics, the transactions completed within the specified time are counted as transaction participation rate and used as an indicator.

[0054] Contract management: management of various contracts generated in market transactions, through analysis of contract signing timeliness, contract disputes, etc.;

[0055] Operation management: involves ensuring that all operational activities are carried out in accordance with established procedures and standards, and using the payment of liquidated damages, the number of arrears, and other indicators to measure the number of failure to pay liquidated damages and arrears in a timely manner;

[0056] Settlement management: manage settlement work related to market transactions, and collect statistical indicators by analyzing customer satisfaction, number of complaints, and other indicators after transactions, as well as the number of customer dissatisfaction and complaints;

[0057] Information disclosure: Ensure the transparency of market operations, which is represented by the standardization and timeliness of information disclosure. Full marks will be awarded if the disclosure is timely and comprehensive.

[0058] 2.2) Penalty Indicator

[0059] The penalty index mainly evaluates the illegal and irregular behaviors of market entities and deducts a certain credit score from the final result, including the following aspects:

[0060] False information: Check whether market entities have disclosed false information in their information disclosure, which can be represented by false reporting, misleading statements, etc., and the number of false information counted as an indicator.

[0061] Unfair competition: Evaluate unfair competition behaviors of market players in transaction management, such as false transactions, price manipulation, etc., by counting the number of unfair competition incidents as an indicator.

[0062] Failure to fulfill transaction results: measures whether market entities fulfill transaction results and whether there are any contract breaches or arrears, by counting the number of times transaction results are not fulfilled as an indicator.

[0063] For step S4, first, the extracted indicators are normalized to the maximum and minimum. Secondly, they are input into the FLAML model for training, and the machine learning model and hyperparameters with the smallest loss function (MSE) are selected. The machine learning models include XGBoost, LightGBM, RandomForest, ExtraTree, and KNN. The training time limit for each training is controlled at 900 seconds. Finally, the identification of transaction risks of power companies is completed. The results are shown in the following table:

[0064] The dataset sizes are set to large (17,000 copies), medium (10,000 copies), and small (5,000 copies), and the proposed BiLSTM filling algorithm is compared with other zero-filling, mean-filling, and median-filling methods. The results show that BiLSTM-FLAML achieves good performance on both the training set and the test set, as shown in Table 1, verifying the effectiveness and superiority of the proposed algorithm.

[0065] Table 1 Comparison of BiLSTM-FLAML and other algorithms

[0066]

[0067] Those skilled in the art should understand that the discussion of any of the above embodiments is only exemplary and is not intended to imply that the scope of the present invention is limited to these examples; under the concept of the present invention, the technical features in the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other changes in different aspects of the present invention as described above, which are not provided in detail for the sake of simplicity. Any omissions, modifications, equivalent substitutions, 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 enterprise transaction risks based on incomplete power data using BiLSTM-FLAML, characterized in that: The method comprises the following steps: S1. Identify and mark abnormal values ​​and missing values ​​in historical power data, extract part of the data before and after the marked point as the prediction set, and use the remaining part as the training set and test set; S2, input the processed training set and test set into the BiLSTM model for training; S3, input the prediction set into the trained BiLSTM model for prediction, perform data filling, and extract features of the filled data according to the established calculation indicators; S4. Input the extracted indicators into the FLAML model to identify transaction risks of power enterprises.

2. The enterprise transaction risk identification method based on BiLSTM-FLAML of incomplete power data according to claim 1 is characterized in that: Step S1 further comprises: S11, using the Z-score method to identify abnormal values ​​and missing values ​​of each power data in turn, and record their names, dates, and serial numbers for calibration; S12, intercepting the data before and after the calibration value and saving it as a prediction set; S13. Pack the remaining power data into training sets and test sets according to a certain ratio.

3. The enterprise transaction risk identification method based on BiLSTM-FLAML of incomplete power data according to claim 1 is characterized in that: Step S2 further comprises: S21, normalize the training set and test set and divide them into different small batches; S22, input the data into the BiLSTM model for training in batches, and gradually optimize the model and its parameters through multiple rounds of iterations and back propagation algorithms to minimize the loss function; S23. Use the test set batch data to evaluate the performance of the model to monitor the generalization ability of the model and prevent overfitting.

4. The enterprise transaction risk identification method based on BiLSTM-FLAML of incomplete power data according to claim 1 is characterized in that: Step S3 further comprises: S31, normalize the prediction set, and predict missing values ​​and outliers through the model in batches; S32, denormalizing the predicted data, and filling and correcting them into the original power data; S33. Extract features from the padded data according to given calculation indicators to reduce the data dimension.

5. The enterprise transaction risk identification method based on BiLSTM-FLAML of incomplete power data according to claim 1 is characterized in that: Step S4 further comprises: S41, normalizing the extracted indicators; S42, inputting the normalized indicators into the FLAML model for training, and selecting the machine learning model and hyperparameters with the smallest loss function; S43. Complete the identification of transaction risks of power companies.

Images