A method and system for monitoring the entire financial data chain
Through the full-link monitoring method of financial data, changing nodes and abnormal nodes in the financial data link are identified, and the change threshold is set for risk assessment, which solves the complex problems of link diagnosis in the existing technology and improves the accuracy of financial data monitoring and efficiency to help users identify potential risks.
Patent Information
- Application Number
- CN202411026496.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-30
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-07-30
AI Technical Summary
Existing financial data link monitoring methods are difficult to accurately identify abnormal points, resulting in complex link diagnosis and reducing monitoring accuracy, and unable to effectively screen out useful abnormal points, affecting the authenticity and reliability of financial data.
Through system initialization, data collection, feature extraction and cleaning, monitoring and analysis models, we identify the changing nodes and abnormal nodes of financial data relative to indicator data, set changes thresholds, conduct abnormal risk assessments, and generate reports to enable users to identify potential risks .
It realizes the identification of changing nodes and abnormal nodes in the financial data link, and reversely traces the possible risks at the front end of the link, which facilitates users to identify potential financial data risks and improves monitoring accuracy and efficiency.
Smart Images

Figure CN118552330B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of intelligent data monitoring and analysis, and in particular to a full-link monitoring method and system for financial data. Background Art
[0002] Financial data is important information that reflects the business and financial status of an enterprise, and is of great reference value to stakeholders such as enterprise managers and investors. However, abnormal or erroneous financial data will affect the authenticity and reliability of financial data, bringing risks to enterprise management and decision-making. Therefore, timely detection and handling of financial data anomalies is an important task to improve the quality of financial data and protect the interests of enterprises.
[0003] However, due to the complexity of financial business systems and the excessive number of links used for data integration, there are too many parameters for the operation of financial data links. The existing problems with financial data link monitoring mainly include that the cause of the problem link diagnosis needs to be analyzed one by one. However, there are many anomalies in the financial data that can be ignored. Analyzing the anomalies of the link one by one cannot help customers screen out useful anomalies, thereby reducing the accuracy of financial data monitoring. Summary of the invention
[0004] The purpose of the present invention is to overcome the deficiencies of the prior art and provide a method and system for full-link monitoring of financial data. The specific process of the method is as follows:
[0005] S1. System initialization, configuration of data collection sources, monitoring and analysis models, and user permissions;
[0006] S2. Set the rules and frequency of financial data collection according to the configured data collection source;
[0007] S3. Extract feature data from the collected multi-source heterogeneous financial data, clean, convert and integrate the feature data to obtain a data set with a unified structure, and store the data set; input the processed data into the monitoring and analysis model; compare and analyze the output results of the monitoring and analysis model with the preset financial indicator full-link data, and the results of the comparative analysis can audit and display the dynamic changes and trends of financial data relative to financial indicator data;
[0008] S4. Set a change threshold for the dynamic change area of the financial data relative to the financial indicator data; identify the first node that exceeds the change threshold and define it as an abnormal node; identify the link of the abnormal node, extract the nodes with dynamic changes between the starting end of the link and the abnormal node, perform abnormal risk assessment on the extracted nodes, and identify potential risks;
[0009] S5. Generate corresponding reports based on abnormal risk assessment results;
[0010] S6. The user reviews the report and takes corresponding risk management measures for potential risks.
[0011] The beneficial effects achieved by the present application are as follows: the present application can identify the change nodes and abnormal nodes in the financial data link relative to the indicator data, and by ignoring the change nodes, reversely trace the possible risks at the front end of the link for the abnormal nodes, thereby facilitating users to identify potential financial data risks. BRIEF DESCRIPTION OF THE DRAWINGS
[0012] In order to more clearly illustrate the embodiments of the present application or the technical solutions in 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 described below are only some embodiments recorded in the present application. For those skilled in the art, other drawings can also be obtained based on these drawings.
[0013] Figure 1 The flowchart of the steps of the full-link monitoring method of financial data shown;
[0014] Figure 2 The connection diagram of the full-link monitoring system for financial data is shown.
[0015] in,
[0016] 1. Data acquisition module; 2. Data processing module; 3. Data analysis module; 4. Risk assessment module; 5. Report generation module; 6. User interaction module. DETAILED DESCRIPTION
[0017] In order to enable those skilled in the art to better understand the technical solution of the present application, the present invention is further described in detail below in conjunction with the accompanying drawings and the best embodiment.
[0018] Embodiment 1, as Figure 1 As shown:
[0019] The present application provides a method for monitoring the entire financial data chain, the method comprising the following steps:
[0020] S1. System initialization, configuration of data collection sources, monitoring and analysis models, and user permissions; data collection sources include multi-source performance data, which include multi-source performance subjects, which include budgets, costs, funds, fixed assets or investments, etc.; multi-source performance data covers various financial-related data such as financial transactions, capital flows, and financial statements;
[0021] S2. According to the configured data collection source, set the rules and frequency of financial data collection; the rules of financial data collection include: collecting budget data, cost data, capital data, fixed asset data or investment data for multiple source performance entities.
[0022] The rules for financial data collection have the following properties:
[0023] Accuracy: Ensure that the data collected is accurate, from reliable sources, and has been properly verified and audited. Completeness: Cover all relevant financial data, including income, expenses, assets, liabilities, profits, etc.
[0024] Consistency: The format and definition of collected data remain consistent throughout the entire chain, making it easier to integrate and analyze.
[0025] Compliance: Follow relevant laws, regulations and financial guidelines to ensure the legality of data collection.
[0026] Security: Take appropriate security measures to protect the confidentiality, integrity and availability of data. The frequency of data collection can be set to real-time, daily, weekly, monthly or quarterly collection according to the actual financial flow situation.
[0027] The specific rules and frequency should be customized according to the company's size, business nature, industry characteristics and management needs to achieve effective financial management and decision support.
[0028] S3. Extract feature data from the collected multi-source heterogeneous financial data, clean, convert and integrate the feature data to obtain a data set with a unified structure, and store the data set;
[0029] The collected multi-source heterogeneous financial data are preprocessed separately, and the data information is cleaned and processed into dimensionless data through normalization. Indicators of different links usually have different dimensions. Data normalization can effectively reduce the dimensionality impact between the indicator data.
[0030] Data of different magnitudes are normalized and then converted to a unified magnitude. The processed data is represented by the Z-score. The multidimensional data is reduced in dimension by principal component analysis. While reducing the data dimension, the information and features contained in the original data are maintained as much as possible.
[0031] The processed data is input into the monitoring and analysis model; the output results of the monitoring and analysis model are compared and analyzed with the full-link data of the preset financial indicators. The results of the comparative analysis can audit and display the dynamic changes and trends of the financial data relative to the financial indicator data.
[0032] Compare and analyze the output results of the monitoring and analysis model with the full-link data of the preset financial indicators, including visually displaying the results of the comparative analysis. The display method includes displaying the collected data information and the indicator data information through a curve chart or a rectangular chart.
[0033] The monitoring and analysis model is a back propagation (BP) neural network model, which includes: a network input layer, an output layer, a hidden layer, an optimizer, an activation function, a classifier and a loss function.
[0034] The optimizer uses the adaptive moment estimation algorithm (Adma) as the basic optimization algorithm. Adma combined with momentum and adaptive learning rate can accelerate the convergence speed.
[0035] The BP neural network model is designed as a classification model, which is used for data link diagnosis, so the loss function adopts the cross entropy loss function. The training method of the BP neural network model includes:
[0036] 1) Input training samples in batches. The training samples contain feature data and corresponding label values.
[0037] 2) The training samples include the planned full-year financial data, and data preprocessing is performed to form a financial data monitoring sample data set.
[0038] 3) Forward propagation to obtain the error between the predicted output and the actual output;
[0039] According to the error back propagation, the weights and biases of each hidden layer are corrected;
[0040] 4) Repeat steps 2) to 3) until the calculated result of the loss function is lower than the initially set threshold or the number of algorithm iterations reaches the initially set number, and the training ends.
[0041] The training samples are link sample data, which are divided into training set, validation set and test set according to the proportion. After building the neural network model, the training set is used to train common parameters such as weights and biases, and the test results of the validation set are used to adjust hyperparameters such as the number of hidden layer nodes. After multiple adjustments, the optimal model is obtained, and finally a test set that has never been used is used to verify the accuracy of the model.
[0042] The acquisition of feature data includes selecting features based on the random forest algorithm, evaluating the importance of the selected features, obtaining feature importance scores, and obtaining feature data based on the importance scores.
[0043] Feature importance score F i The solution formula is as follows:
[0044]
[0045] Where n is the total number of samples, I() represents the function transformation, when the two values are equal, it takes 1, when they are not equal, it takes 0, Y′ p Represents the predicted value after adding noise to the feature, Yt represents the selected feature, Y p It represents the predicted value before adding noise to the feature, and i represents any sample.
[0046] Feature selection through the random forest algorithm is to first input the data into the model for training, and when a certain test accuracy is reached, the feature importance of each feature is output, and the importance of each feature in the random forest is evaluated to obtain the feature importance score.
[0047] Correct the weights of each hidden layer, including using a dynamic adaptive particle swarm optimization algorithm for adaptive solution. The adaptive solution of the weight w is as follows:
[0048]
[0049] In the formula, w min Indicates the minimum inertia weight, w max Indicates the maximum inertia weight, f min represents the minimum fitness of the current group, f avg represents the average fitness value of all particles in the current group, and f represents the fitness value of the current particle.
[0050] S4. Set a change threshold for the dynamic change area of financial data relative to financial indicator data; identify the first node that exceeds the change threshold and define it as an abnormal node; identify the link of the abnormal node, extract the nodes with dynamic changes between the starting end of the link and the abnormal node, perform abnormal risk assessment on the extracted nodes, and identify potential risks;
[0051] An abnormal risk assessment is performed on the extracted nodes, including dividing the change threshold into several equal intervals, and upgrading the risk level of the intervals in order from low to high, and comparing the difference between the extracted node and the corresponding node of the indicator data chain with the equal intervals to determine the risk level of the node.
[0052] Since the actual financial data cannot be completely implemented against the indicator data, and the changing data may not have a big impact on the actual data chain, at the same time, the financial data may have positive and negative deviations relative to the dynamic change area of the financial indicator data. By offsetting the positive and negative deviations, there is no need to monitor the changing data. Therefore, a change threshold is set. The change threshold can reflect the maximum degree of the data change interval. When the node data of the data link exceeds this threshold, it will cause essential abnormalities in the financial data. Therefore, the setting of the change threshold needs to be determined manually based on the data of the financial indicator.
[0053] The purpose of extracting nodes with dynamic changes between the starting end of the link and the abnormal node is to regionally determine the possible range of the abnormality. The data at the back end of the abnormal node in the link will be affected by the data of the abnormal node. Therefore, the data at the back end of the abnormal node in the link will be ignored.
[0054] The extracted nodes are evaluated for abnormal risks and potential risks are identified because the abnormalities in the abnormal node data are usually stacked up from the front data. Therefore, intuitively viewing the changes in the front data can make it easier for users to monitor link abnormalities.
[0055] S5. Generate corresponding reports based on abnormal risk assessment results;
[0056] S6. The user reviews the report and takes appropriate risk management measures for potential risks.
[0057] Embodiment 2, as Figure 2 As shown:
[0058] The present application provides a financial data full-link monitoring system, the system comprising:
[0059] Data collection module 1 can collect financial data from the internal financial system, business system and external financial market data sources of the enterprise, covering various financial information such as capital flow, income and expenditure details, assets and liabilities, etc. The data collection module has the ability to automatically adapt to different data source formats and interfaces.
[0060] Data processing module 2 extracts feature data from the collected multi-source heterogeneous financial data, cleans, converts and integrates the feature data to obtain a data set with a unified structure, and stores the data set in a central database.
[0061] Data analysis module 3: Data analysis module 3 inputs the processed data into the monitoring and analysis model; compares and analyzes the output of the monitoring and analysis model with the preset full-link data of financial indicators. The results of the comparative analysis can audit and display the dynamic changes and trends of financial data relative to financial indicator data; set a change threshold for the dynamic change area of financial data relative to financial indicator data; identify the first node that exceeds the change threshold and define it as an abnormal node; identify the link of the abnormal node, and extract the node with dynamic changes between the starting end of the link and the abnormal node;
[0062] Risk assessment module 4 uses preset risk models and algorithms to conduct abnormal risk assessment on the extracted nodes and identify potential financial risks, such as liquidity risk, credit risk and market risk.
[0063] The report generation module 5 automatically generates various financial reports and risk assessment reports based on the analysis results.
[0064] The report generation module supports multi-language and multi-format output to meet the needs of different users.
[0065] The user interaction module 6 provides users with an intuitive and convenient operation interface and supports users to customize monitoring indicators, risk thresholds and report formats.
[0066] The user interaction module has permission management function to ensure that users can only access data and functions within their authorized scope.
[0067] Those skilled in the art can understand that the implementation of all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program. The program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD), etc.; the storage medium can also include a combination of the above-mentioned types of memory.
[0068] Obviously, the above embodiments are merely examples for clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived from these are still within the protection scope of the invention.
Claims
1. A full-link monitoring method for financial data, characterized in that: The steps include: S1. System initialization, configuration of data collection sources, monitoring and analysis models, and user permissions; S2. Set the rules and frequency of financial data collection according to the configured data collection source; S3. Extract feature data from the collected multi-source heterogeneous financial data, clean, convert and integrate the feature data to obtain a data set with a unified structure, and store the data set; input the processed data into the monitoring and analysis model; compare and analyze the output results of the monitoring and analysis model with the preset financial indicator full-link data, and the results of the comparative analysis can audit and display the dynamic changes and trends of financial data relative to financial indicator data; S4. Setting a change threshold for the dynamic change area of the financial data relative to the financial indicator data; identifying the first node exceeding the change threshold and defining it as an abnormal node; Identify the link of the abnormal node, extract the nodes with dynamic changes between the starting end of the link and the abnormal node, perform abnormal risk assessment on the extracted nodes, and identify potential risks; S5. Generate corresponding reports based on abnormal risk assessment results; S6. The user reviews the report and takes corresponding risk management measures for potential risks; The monitoring and analysis model construction method comprises: 1) Construct the basic model of BP neural network structure, including setting the nodes and activation functions of the input layer, hidden layer, and output layer; 2) Input training samples in batches, wherein the training samples include feature data and corresponding label values; 3) Forward propagation to obtain the error between the predicted output and the actual output; 4) According to the error back propagation, correct the weights and bias of each hidden layer; 5) Repeat the process from 2) to 4) until the calculated result of the loss function is lower than the initially set threshold or the number of algorithm iterations reaches the initially set number, and then the training ends; The obtaining of the feature data includes selecting features based on a random forest algorithm, evaluating the importance of the selected features to obtain feature importance scores, and selecting feature data based on the importance scores; The feature importance score F i The solution formula is as follows: Where n is the total number of samples, I() represents the function transformation, and when the two values in I() are equal, they are 1, and when they are not equal, they are 0. p Represents the predicted value after adding noise to the feature, Y t represents the selected feature, Y p Express special The predicted value before adding noise; i represents any sample; The method of modifying the weights of each hidden layer includes using a dynamic adaptive particle swarm optimization algorithm to adaptively modify the weights. The adaptive solution formula of the weight w is as follows: In the formula, w min Indicates the minimum inertia weight, w max Indicates the maximum inertia weight, f min represents the minimum fitness of the current group, f avg represents the average fitness value of all particles in the current group, and f represents the fitness value of the current particle.
2. The method for full-link monitoring of financial data according to claim 1, characterized in that: The data collection sources in step S1 include multi-source performance entities, and the multi-source performance entities include budget, cost, capital, fixed assets and investment.
3. The method for full-link monitoring of financial data according to claim 1, characterized in that: In step S2, the rules for collecting financial data include: collecting budget data, cost data, capital data, fixed asset data or investment data for multiple source performance entities respectively.
4. The method for full-link monitoring of financial data according to claim 1, characterized in that: The abnormal risk assessment for the extracted nodes includes dividing the change threshold into a number of equal intervals, and upgrading the risk level of the intervals in order from low to high, and comparing the difference between the extracted node and the corresponding node of the indicator data chain with the equal intervals to determine the risk level of the node.
5. The method for full-link monitoring of financial data according to claim 1, characterized in that: Compare and analyze the output results of the monitoring and analysis model with the full-link data of preset financial indicators, including visually displaying the results of the comparative analysis. The method of visual display includes displaying the collected data information and indicator data information through a curve chart or a rectangular chart.
6. A system using the financial data full-link monitoring method according to any one of claims 1 to 5, characterized in that: include: A data collection module (1), the data collection module is used to collect and set financial data; A data processing module (2), wherein the data processing module extracts characteristic data from the collected multi-source heterogeneous financial data, cleans, converts and integrates the characteristic data to obtain a data set with a unified structure, and stores the data set; A data analysis module (3), wherein the data analysis module inputs the processed data into a monitoring and analysis model; compares and analyzes the output of the monitoring and analysis model with the preset full-link data of financial indicators, wherein the result of the comparative analysis can audit and display the dynamic changes and trends of the financial data relative to the financial indicator data; sets a change threshold for the dynamic change area of the financial data relative to the financial indicator data; identifies the first node that exceeds the change threshold and defines it as an abnormal node; identifies the link of the abnormal node, and extracts the node with dynamic changes between the starting end of the link and the abnormal node; The risk assessment module (4) performs abnormal risk assessment on the extracted nodes and identifies potential risks; A report generation module (5), wherein the report generation module generates a corresponding report based on the abnormal risk assessment result; A user interaction module (6), wherein the user interaction module is used to display the corresponding report to the user.
