A big data-driven financial risk identification system

Abstract

The present invention relates to the field of risk identification technology, specifically a big data-driven financial risk identification system, which includes a time data synchronization module, a multidimensional feature mapping module, a financial behavior analysis module, and a risk dynamic identification module. In the present invention, by synchronizing financial data streams in real time and applying multidimensional scaling transformations, the capture and processing of financial data are optimized, and more efficient data stream analysis is achieved. It is possible to accurately capture subtle changes and key forms of time series data in complex financial environments, greatly improving the sensitivity and accuracy of risk identification. Refined similarity calculations and behavior identification help the system discover potential risks earlier, ensuring that financial institutions can respond quickly, thereby effectively preventing and mitigating risk events, and then responding to dynamically changing markets, which can significantly improve the formulation of risk management strategies and the rationality of resource allocation.

Description

Technical Field

[0001] The present invention relates to the field of risk identification technology, and in particular to a big data-driven financial risk identification system. Background Art

[0002] Risk identification technology is a key branch of financial technology. It involves using technology to predict and identify potential financial risks. It extensively applies algorithms, big data analytics, and machine learning techniques to identify risk factors by analyzing historical and real-time data. The goal is to enhance financial institutions' risk management capabilities, optimize resource allocation, and improve decision-making efficiency. Applications include, but are not limited to, credit scoring, market risk analysis, fraud detection, and operational risk management.

[0003] Among them, the financial risk identification system refers to an integrated technical system designed to identify and assess potential risks in the financial environment in real time so that timely measures can be taken to mitigate the risks. It is usually based on advanced data analysis and predictive models, and can process large amounts of data and extract risk signals from them. It has a wide range of uses, such as helping banks and financial institutions assess the risks of loan applications, monitor the market risks of investment portfolios, or provide risk forecasts for insurance companies to optimize their insurance products and pricing strategies.

[0004] While existing technologies employ a variety of data analysis and machine learning techniques for risk management, they often struggle with data synchronization and real-time processing. They frequently encounter delays and inaccuracies when processing synchronization in a big data environment, especially in situations requiring real-time analysis and rapid response. They often fail to provide adequate support. Furthermore, existing systems are often limited in their ability to discern subtle changes in data, hindering their effectiveness in predicting and identifying risks in rapidly changing market environments. For example, during rapid market fluctuations, traditional systems may be unable to capture key signals, resulting in insufficiently timely or accurate risk assessments. These time lags and insensitivity can expose financial institutions to losses from sudden risks. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the existing technology and propose a big data driven financial risk identification system.

[0006] To achieve the above objectives, the present invention adopts the following technical solutions: A big data-driven financial risk identification system includes:

[0007] The time data synchronization module collects real-time financial data streams, synchronizes time tags, calculates the time difference of each data point, performs dynamic time adjustment based on the calculation results, performs frequency correction and time tag calibration on the data, and outputs synchronized time series data;

[0008] The multidimensional feature mapping module converts the time data into a spatial feature vector through multidimensional scaling based on the synchronized time series data, performs dimension mapping on the data at each time point, extracts key spatial features, and generates a mapping feature set;

[0009] The financial behavior analysis module uses feature analysis to extract sequence patterns based on the mapping feature set, identifies key morphological changes in the time series, monitors and compares historical trend data in real time based on the recognition results, calculates the similarity between data points, reveals abnormal behavior of financial data, and outputs behavior recognition results;

[0010] Based on the behavior recognition results, the risk dynamic identification module evaluates the risk level of abnormal behavior by comparing the risk thresholds at different time points, determines potential risk threats based on the evaluation results, classifies the risk level and marks sensitive events, and outputs financial risk warning records.

[0011] As a further solution of the present invention, the step of calculating the time difference is:

[0012] Perform data collection initialization processing, establish a secure connection with the financial market data interface, batch pull real-time transaction data of different financial products, and generate original real-time financial data streams;

[0013] Based on the original real-time financial data stream, adding an exact server reception timestamp to each piece of data, synchronizing the time tag, and obtaining a real-time financial data stream with a time tag;

[0014] Using the real-time financial data stream with time tags, the formula is adopted:

[0015] Δt i =t i+1 -t i

[0016] Calculate the time difference Δt between adjacent data points i , generate time difference data, where t i Represents the timestamp of the i-th data point, t i+1 Represents the timestamp of the i+1th data point.

[0017] As a further solution of the present invention, the step of acquiring the synchronized time series data is:

[0018] Using the time difference data, linear interpolation is performed using the formula:

[0019]

[0020] Calculate the new timestamp t for each data point new,i , get the adjusted timestamp, where t iis the original timestamp, Δt i is the time difference value, g is the normalization constant, and n is the total number of data points;

[0021] Performing frequency correction on the data stream using the adjusted timestamp to ensure that the data acquisition frequency matches a preset standard, analyzing the original data frequency through Fourier transform and applying a filter to adjust it to the target frequency, thereby generating a frequency-corrected data stream;

[0022] Time stamp calibration is performed on the frequency-corrected data stream, and time series analysis is applied to re-mark the timestamp of each data point to output synchronized time series data.

[0023] As a further solution of the present invention, the step of obtaining the mapping feature set is:

[0024] Based on the synchronized time series data, a multidimensional scaling transformation is performed to preliminarily locate the multidimensional spatial position of each time data point and generate a primary spatial feature vector set;

[0025] Using the primary spatial feature vector set, the formula is adopted:

[0026]

[0027] Calculate the Euclidean distance d between vector i and vector j ij , generate the optimized spatial feature vector set, where v ik and v jk are the coordinate values ​​of vectors i and j in the kth dimension, and m represents the coordinate dimension;

[0028] According to the optimized set of spatial feature vectors, the spatial attributes and relationships of the vectors are integrated to reveal the key spatial characteristics of the original data and generate mapping features.

[0029] As a further embodiment of the present invention, the step of identifying the key morphological changes is:

[0030] Extracting key time-intensity feature data based on the mapping feature set and generating windowed feature data using a sliding time window;

[0031] Performing principal component analysis on the windowed feature data to compress data dimensions, extract key information, and generate reduced-dimensional feature data;

[0032] The reduced dimensionality feature data is fed into a deep learning model to identify and classify the key morphological changes of the time series using the formula:

[0033]

[0034] Calculate the pattern probability p(t) of the time series and generate the key morphological change recognition results, where αh is the influence coefficient of the hth mode, γ h Indicates the attenuation parameter corresponding to the corresponding mode, μ h is the central time point of the corresponding pattern, and M represents the number of patterns.

[0035] As a further solution of the present invention, the steps for obtaining the behavior recognition results are:

[0036] Using the key morphological change identification results, new time series data is monitored in real time, compared with historical trend data, and difference analysis is performed to generate real-time monitoring and comparison results;

[0037] Based on the real-time monitoring and comparison results, data points are extracted and the formula is used.

[0038]

[0039] Calculate the similarity coefficient cs, perform similarity evaluation, and generate the data point similarity evaluation results, where A and B represent the feature vectors of the current data point and the historical data point respectively, and ‖A‖ and ‖B‖ represent the modulus respectively;

[0040] Based on the data point similarity evaluation results, a threshold determination is performed to analyze the abnormality of the data points, identify data points that significantly deviate from the normal pattern, reveal potential abnormal financial data behavior, and generate behavior recognition results.

[0041] As a further solution of the present invention, the risk level assessment steps are:

[0042] Utilizing the behavior recognition results, extracting key time points in the behavior data, comparing them with the set baseline risk threshold, and generating behavioral feature comparison results for each time point;

[0043] Analyze the behavioral characteristics comparison results at each time point using the formula:

[0044]

[0045] Calculate the risk index R(t) at each time point to obtain the preliminary risk assessment results, where ΔB(t) represents the deviation of the behavioral data from the baseline threshold, and B base is the baseline behavioral threshold;

[0046] Based on the preliminary risk assessment results, each time point that deviates from the benchmark threshold is classified into a risk level, the risk level of each time point is determined, and a risk level assessment result is generated.

[0047] As a further solution of the present invention, the steps for obtaining the financial risk warning record are:

[0048] Analyze the risk level assessment results, identify abnormal behavior events, and generate a list of potential risk threats;

[0049] Based on the potential risk threat list, classify according to the risk level and mark abnormal events as sensitive events using the formula:

[0050]

[0051] Calculate the percentage S of sensitive events in the total events and get the risk level classification results, where R high is the number of abnormal events, R total is the total number of events;

[0052] According to the risk level classification results, based on the characteristics of sensitive events and historical data analysis, financial risk warning records are output.

[0053] Compared with the prior art, the advantages and positive effects of the present invention are:

[0054] In the present invention, by synchronizing financial data streams in real time and applying multidimensional scaling transformations, the capture and processing of financial data is optimized, and more efficient data stream analysis is achieved. It is possible to accurately capture subtle changes and key forms of time series data in complex financial environments, greatly improving the sensitivity and accuracy of risk identification. Refined similarity calculations and behavior recognition help the system discover potential risks earlier, ensuring that financial institutions can respond quickly, thereby effectively preventing and mitigating risk events, and then coping with dynamically changing markets, which can significantly improve the formulation of risk management strategies and the rationality of resource allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 is a system flow chart of the present invention;

[0056] Figure 2 This is a flow chart for calculating the time difference of the present invention;

[0057] Figure 3 This is a flowchart for obtaining synchronized time series data according to the present invention;

[0058] Figure 4 A flowchart for obtaining a mapping feature set according to the present invention;

[0059] Figure 5 This is a flow chart for identifying key morphological changes of the present invention;

[0060] Figure 6 This is a flow chart for obtaining behavior recognition results of the present invention;

[0061] Figure 7 This is a flow chart for evaluating the risk level of the present invention;

[0062] Figure 8 This is a flowchart for obtaining financial risk warning records of the present invention. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0064] In the description of the present invention, it should be understood that the terms "length," "width," "up," "down," "front," "back," "left," "right," "vertical," "horizontal," "top," "bottom," "inside," "outside," and the like, indicating positions or relationships, are based on the positions or relationships shown in the accompanying drawings and are intended only to facilitate the description of the present invention and simplify the description. They do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limiting the present invention. Furthermore, in the description of the present invention, "plurality" means two or more, unless otherwise expressly and specifically defined.

[0065] See also Figure 1 , a big data driven financial risk identification system includes:

[0066] The time data synchronization module collects real-time financial data streams, synchronizes time tags, calculates the time difference of each data point, performs dynamic time adjustment based on the calculation results, performs frequency correction and time tag calibration on the data, and outputs synchronized time series data;

[0067] The multidimensional feature mapping module converts time data into spatial feature vectors through multidimensional scaling based on synchronized time series data. It then performs dimension mapping on the data at each time point, extracts key spatial features, and generates a mapping feature set.

[0068] The financial behavior analysis module uses feature analysis to extract sequence patterns based on the mapped feature set, identifying key morphological changes in the time series. Based on the identification results, it monitors and compares historical trend data in real time, calculates the similarity between data points, reveals abnormal financial data behavior, and outputs behavior identification results.

[0069] Based on the behavior recognition results, the dynamic risk identification module evaluates the risk level of abnormal behavior by comparing the risk thresholds at different time points. According to the evaluation results, it determines the potential risk threats, classifies the risk level and marks sensitive events, and outputs financial risk warning records.

[0070] Synchronized time series data includes data alignment indicator results, correction frequency values ​​and marked synchronization points. The mapping feature set includes feature vector sequence, dimension identifier and key feature marking record. The behavior recognition results include morphological variation index, sequence similarity score and anomaly indicator set. The financial risk warning record specifically refers to the risk level classification, threat identification mark and event sensitivity marking results.

[0071] See also Figure 2 , the calculation steps of time difference are:

[0072] Perform data collection initialization processing, establish a secure connection with the financial market data interface, batch pull real-time transaction data of different financial products, and generate original real-time financial data streams;

[0073] When initializing the data acquisition module, it is necessary to ensure that the connection with the financial market data interface is secure and reliable. In order to ensure the security and integrity of data transmission, data encryption and authentication protocols are first used to ensure that the data is not stolen or tampered with during transmission. Subsequently, a real-time data monitoring system is used to monitor the quality of the incoming data to filter out any abnormal or damaged data packets, thereby ensuring that the acquired data stream is complete. Through specific operations, the accurate collection of real-time transaction data of financial products is ensured, and the original real-time financial data stream is generated.

[0074] Based on the original real-time financial data stream, add the exact timestamp of server receipt to each data, synchronize the time tag, and obtain the real-time financial data stream with time tag;

[0075] During the execution of time synchronization, the operation of adding a timestamp to each piece of data needs to be accurate to milliseconds to ensure the time accuracy of subsequent data processing. First, the server clock is calibrated to ensure its consistency with the international standard time. Then, at the moment the data arrives at the server, a timestamp is generated by a high-precision clock and attached to each piece of data. The process involves capturing and marking time to ensure that each piece of data can reflect its actual reception time. The processed data stream then has a time tag, which can provide an accurate time reference for subsequent time difference calculation and data analysis.

[0076] Using real-time financial data streams with time tags, the formula is:

[0077] Δt i =t i+1 -t i

[0078] Calculate the time difference Δt between adjacent data points i , generate time difference data, where t i Represents the timestamp of the i-th data point, t i+1Represents the timestamp of the i+1th data point;

[0079] Collection i = 1500000000.123 seconds and t i+1 =1500000000.456 seconds, the calculated time difference is Δt i = 1500000000.456 - 1500000000.123 = 0.333 seconds. The results show that the time interval between these two data points is 0.333 seconds. The time difference will be used in subsequent analysis to identify possible time deviations or make further time corrections.

[0080] See also Figure 3 , the steps for obtaining synchronized time series data are:

[0081] Using time difference data, linear interpolation is performed using the formula:

[0082]

[0083] Calculate the new timestamp t for each data point new,i , get the adjusted timestamp, where t i is the original timestamp, Δt i is the time difference value, g is the normalization constant, and n is the total number of data points;

[0084] In a certain data set, i =1500000000.050,Δt i =0.300, n=100, g=0.1, then the adjusted timestamp t new,i The calculation is as follows: First calculate the square root of each time difference and divide it by the total time differences, that is, Then multiply it by the constant g, and the result is 0.018257 × 0.1 ≈ 0.001826. Finally, after calculation, the adjusted timestamp is 1500000000.050 + 0.001826 = 1500000000.051826. This shows that the adjustment process has indeed reduced the time difference and improved data synchronization accuracy.

[0085] Use the adjusted timestamps to perform frequency correction on the data stream to ensure that the data acquisition frequency matches the preset standard. Analyze the original data frequency through Fourier transform and apply a filter to adjust it to the target frequency to generate a frequency-corrected data stream.

[0086] In the detailed process of data frequency correction, the Fourier transform is first applied to analyze the frequency characteristics of the original data stream in detail, and the time series data is converted into a frequency domain representation, thereby identifying the main frequency components in the data and any abnormal frequency noise. Based on the continuous fast Fourier transform algorithm, it is ensured that the frequency of each data point is accurately measured and recorded. Once the frequency analysis is completed, the data stream is processed using a customized digital filter. The filter is designed to cut or enhance a specific frequency range so that the frequency of the data stream is strictly consistent with the preset operating frequency. Through the steps, the data stream is adjusted to a unified frequency standard, which improves the consistency and accuracy of data processing. Finally, the adjusted data is reconverted into a time series through an inverse Fourier transform, ensuring that all operations maintain the integrity and validity of the data. The correction process is especially important for subsequent data analysis because it directly affects the quality and effect of data analysis.

[0087] Perform time stamp calibration based on the frequency-corrected data stream, apply time series analysis to re-mark the timestamp of each data point, and output synchronized time series data;

[0088] During the execution of time stamp calibration, we first use the frequency-corrected data stream and apply advanced time series analysis methods to accurately calibrate the timestamp of each data point. This process involves complex time series modeling techniques, such as ARIMA models or seasonal decomposition. The model can be used to analyze the temporal structure and pattern of the data in detail to ensure that the time stamp of each data point accurately reflects its original collection time. The calibration process is particularly critical when facing large-scale or high-frequency data, because even a small time error can lead to significant deviations in the analysis results. In addition, time stamp calibration also includes adjusting the data time axis to reflect the actual order and time interval of events. This is usually achieved through interpolation and data alignment techniques to ensure that each time point in the data set is in the correct time series position, so that the final output time series data is not only synchronized in time, but also more stable and reliable, providing a solid foundation for further data analysis.

[0089] See also Figure 4 , the steps to obtain the mapping feature set are:

[0090] Based on the synchronous time series data, multidimensional scaling transformation is performed to preliminarily locate the multidimensional spatial position of each time data point and generate a primary spatial feature vector set;

[0091] Starting with synchronized time series data, multidimensional scaling (MDS) is used to transform the data into a set of primary spatial feature vectors. The key lies in how to process and map the multidimensional time series data. The specific steps include first collecting and synchronizing all relevant time series data. The data may originate from different sensors or measurement devices, and ensuring the temporal consistency and integrity of the data is paramount. Next, an appropriate multidimensional scaling algorithm is selected, which typically depends on the data dimensionality and the expected complexity of the results. The algorithm is then executed to map the data at each time point into a multidimensional space. The position of each data point in the multidimensional space is represented by a set of coordinates, which constitute the primary spatial feature vectors. This process preserves the relative positions and distances between data points for further analysis and processing in subsequent steps. Furthermore, a quality assessment of the initial mapping results is necessary, including checking whether the distribution of data points in the new space is reasonable and whether there are outliers or clustering. This is crucial for ensuring the accuracy of the subsequent data processing. The resulting set of primary spatial feature vectors provides the foundational data and reference framework for subsequent optimization and feature extraction.

[0092] Using the primary spatial eigenvector set, the formula is:

[0093]

[0094] Calculate the Euclidean distance d between vector i and vector j ij , generate the optimized spatial feature vector set, where v ik and v jk are the coordinate values ​​of vectors i and j in the kth dimension, and m represents the coordinate dimension;

[0095] In a three-dimensional space, the coordinates of vector i and vector j are (1, 2, 3) and (4, 5, 6) respectively, then:

[0096]

[0097] The results show that the distance between vector i and vector j in three-dimensional space is approximately 5.20 units. This value can help us understand and evaluate the similarity or difference in attributes between two data points.

[0098] Based on the optimized set of spatial feature vectors, the spatial attributes and relationships of the vectors are integrated to reveal the key spatial characteristics of the original data and generate mapping features;

[0099] After obtaining the optimized set of spatial feature vectors, the spatial properties and relationships of the vectors are further integrated to extract key spatial features. This process carefully examines the contribution of each vector and its relative importance within the entire dataset, using statistical analysis and pattern recognition techniques to identify representative feature vectors. First, the optimized feature vectors are normalized to eliminate the influence of different dimensions. Next, cluster analysis is used to group the vectors based on their distances, with each group representing a potential pattern or trend in the data. The most representative vectors are then selected from each group. These vectors play a key role in defining the structure and distribution of the dataset. Finally, the selected vectors are combined to form the final mapped feature set. This mapped feature set not only reflects the core structure of the original data but also provides an important source of information for subsequent data analysis and applications. In this way, the mapped feature set effectively captures and represents the key spatial characteristics of the data, providing a basis for data interpretation and further data-driven decision-making.

[0100] See also Figure 5 , the identification steps of key morphological changes are:

[0101] Extract key time-intensity feature data based on the mapping feature set, and generate windowed feature data using a sliding time window;

[0102] The process of extracting key time-intensity feature data from the mapping feature set is achieved by screening and analyzing a large amount of time series data. First, representative data points are selected based on the changing characteristics of the time series. The data points can reflect the key trends and fluctuations of the overall time series. Next, the sliding time window technology is used to divide the time series into multiple continuous window segments according to the preset time window length. Each window segment contains a fixed number of data points, which can maintain the continuity of the time series while capturing the local characteristics of the data in each window. Statistical analysis is performed on the data in each time window, and its statistical indicators such as mean, standard deviation, and extreme value are calculated. This helps to extract more accurate and useful feature information from the original time series data, and finally generate windowed feature data.

[0103] Perform principal component analysis on the windowed feature data to compress the data dimension, extract key information, and generate reduced-dimensional feature data;

[0104] Principal component analysis is performed on windowed feature data. Its purpose is to reduce redundant information in the data through mathematical transformation and improve the efficiency and accuracy of data analysis. In the specific operation process, the covariance matrix between each feature is first calculated in order to understand the correlation between different features; next, based on the covariance matrix, the eigenvalues ​​and corresponding eigenvectors are extracted. The eigenvectors represent the most important direction of change in the data; then, according to the size of the eigenvalues, the principal components with high contribution are selected, which can retain the most important information while reducing the data dimension; in this way, the original high-dimensional data is converted into low-dimensional feature data containing key information. This reduced-dimensional data is more suitable for complex pattern analysis and classification processing, and finally generates reduced-dimensional feature data to provide input data for the next deep learning model.

[0105] The reduced dimensionality feature data is fed into the deep learning model to identify and classify the key morphological changes of the time series using the formula:

[0106]

[0107] Calculate the pattern probability p(t) of the time series and generate the key morphological change recognition results, where α h is the influence coefficient of the hth mode, γ h Indicates the attenuation parameter corresponding to the corresponding mode, μ h is the central time point of the corresponding pattern, M represents the number of patterns;

[0108] Assume that M = 3, α1 = 0.5, α2 = 0.3, α3 = 0.2, γ1 = 1.0, γ2 = 0.8, γ3 = 1.2, μ1 = 10, μ2 = 20, μ3 = 15, and calculate the value of p(12) at t = 12: Calculation yields p(12) = 0.5 × e (-4) +0.3×e (-51.2) +0.2×e (-10.8) ≈0.5×0.0183+0.3×1.938×10 -23 +0.2×0.00002≈0.00915+0+0=0.00915. The results show that at t=12, the pattern probability of the time series is 0.00915, indicating that the key morphological changes at this moment are low and it is unlikely that significant events or features will appear.

[0109] The deep learning model is specifically a long short-term memory network (LSTM): LSTM is a special recurrent neural network (RNN) that is suitable for processing time series data. It can capture long-term dependencies and can be used to learn key patterns in time series and predict the probability of future key state changes based on the input sequence.

[0110] A large amount of time series data and its annotation information (such as the occurrence of key moments) are used to train the model so that it can learn the mapping relationship from the original time series data to key morphological changes.

[0111] Classification using probability:

[0112] The trained model can input new time series data, calculate the probability of key state change p(t), and then determine whether the time point belongs to a key morphological change based on the preset threshold, thereby completing the judgment process.

[0113] See also Figure 6 , the steps to obtain behavior recognition results are:

[0114] Using key morphological change identification results, monitor new time series data in real time, compare it with historical trend data, perform difference analysis, and generate real-time monitoring and comparison results;

[0115] The new time series data is monitored in real time using the identified key morphological changes and compared with the archived historical trend data. The process involves extracting relevant time series data from the historical data storage system. The data is usually arranged in chronological order and contains past performance data points and possible trend changes. Then, data mining technology, especially time series analysis technology, is used to conduct a detailed comparative analysis of the new data and historical data. The difference analysis mainly calculates the statistical differences between the new and old data points, such as changes in statistical indicators such as the mean and standard deviation, to determine the similarities or differences between the data points. The final real-time monitoring report will describe these differences in detail to provide a basis for subsequent data processing and decision-making.

[0116] Based on real-time monitoring and comparison results, data points are extracted and the formula is used.

[0117]

[0118] Calculate the similarity coefficient cs, perform similarity evaluation, and generate the data point similarity evaluation results, where A and B represent the feature vectors of the current data point and the historical data point respectively, and ‖A‖ and ‖B‖ represent the modulus respectively;

[0119] Assume that A = (1, 2, 3) and B = (2, 3, 4).

[0120] but And A·B=1×2+2×3+3×4=20. Therefore, the similarity coefficient is calculated as: The result shows that the current data point and the historical data point have a high similarity in the direction of the eigenvector (0.993 is very close to 1), indicating that they may belong to similar data patterns or trends.

[0121] Based on the data point similarity assessment results, threshold determination is performed to analyze the degree of abnormality of the data points, identify data points that significantly deviate from the normal pattern, reveal potential abnormal financial data behavior, and generate behavior recognition results;

[0122] Based on the data point similarity assessment results, the threshold judgment method is applied to analyze the degree of abnormality of the data points. It relies on the comparison of the similarity results with the preset threshold. By setting a threshold, for example, setting the threshold to 0.5, the similarity score of each data point is compared. If the similarity score is lower than 0.5, it is considered that the data point is significantly different from the historical data, indicating an anomaly or a new trend. This method is simple but very effective. It can quickly screen out data points that need further analysis, identify data points that significantly deviate from the normal pattern, further reveal potential abnormal financial data behavior, and ultimately generate behavior recognition results to provide strong data support for decision makers.

[0123] See also Figure 7 , the risk level assessment steps are:

[0124] Using the behavior recognition results, we extract key time points in the behavior data and compare them with the set baseline risk threshold to generate behavioral feature comparison results at each time point.

[0125] Through the behavior recognition results, the key time points in the behavior data are extracted and compared with the set baseline risk threshold. The difference between each time point and the baseline risk threshold is analyzed, and the comparison results of the behavior characteristics are evaluated. The key parameters in the process include time point feature data and baseline risk threshold. The time point feature data is automatically extracted by the behavior recognition system and compared with the preset risk threshold in the database. The comparison algorithm uses a simple difference calculation, that is, the difference between the behavior feature value of each data point and the risk threshold. It can quickly identify behaviors that deviate from the normal range, thereby preliminarily identifying potential risk points and generating behavioral feature comparison results for each time point.

[0126] Analyze the behavioral characteristics and comparison results at each time point using the formula:

[0127]

[0128] Calculate the risk index R(t) at each time point to obtain the preliminary risk assessment results, where ΔB(t) represents the deviation of the behavioral data from the baseline threshold, and B base is the baseline behavioral threshold;

[0129] At time point t, the behavioral data value is 120, and the baseline threshold is 100, then ΔB(t) is 20. If the baseline threshold B base is 100, then the risk index R(t) is The result shows that the risk index at time point t is 0.2, which shows the degree of deviation between the behavioral data at this time point and the benchmark threshold. Based on the degree of deviation, the risk level of the behavior at time point t can be further evaluated.

[0130] Based on the preliminary risk assessment results, classify the risk level of each time point that deviates from the benchmark threshold, determine the risk level of each time point, and generate a risk level assessment result;

[0131] Based on the preliminary risk assessment results, high-risk behaviors are identified and marked as abnormal. In this process, the key steps include comparing the risk level at each time point with the set risk level threshold. The risk level classification uses a simple classification model. The model parameters include risk level thresholds. The thresholds are set based on historical data and the guidance of the risk management department. The risk index at each time point is compared with these thresholds. If the risk index exceeds the threshold, it is marked as abnormal, thereby determining the risk level at each time point and generating the final risk level assessment result.

[0132] See also Figure 8 , the steps to obtain financial risk warning records are:

[0133] Analyze risk level assessment results, identify abnormal behavior events, and generate a list of potential risk threats;

[0134] By analyzing the risk level assessment of the previous key item, behavioral events with high-risk indicators are identified. This process involves an in-depth analysis of the risk assessment results. First, based on the preset risk classification model, behaviors with high-risk indicators are screened out. For example, if the behavioral data shows that the behavioral indicators at a certain time point are far higher than those at other time points, this behavior will be classified as high-risk behavior. This classification is based on data analysis and predefined risk thresholds. Then, these behaviors are further screened to determine which behaviors may lead to serious negative consequences, such as financial losses or safety accidents. Each behavioral event will be recorded in detail and the risk level will be determined. According to its risk level, these behaviors will be added to the potential risk threat list to provide a basis for subsequent risk management measures.

[0135] Based on the potential risk threat list, classify according to the risk level and mark abnormal events as sensitive events using the formula:

[0136]

[0137] Calculate the percentage S of sensitive events in the total events and get the risk level classification results, where R high is the number of abnormal events, R total is the total number of events;

[0138] In a certain period of time, a total of 1000 events were identified, of which 100 were high-risk events. The percentage of sensitive events is calculated as follows: This result shows that 10% of the events were marked as high-risk. This ratio is an important indicator of the current security management status, reflecting the proportion of sensitive events in all identified events, helping decision makers understand the prevalence of high-risk events and adjust risk control strategies accordingly.

[0139] Output financial risk warning records based on risk classification results, characteristics of sensitive events, and historical data analysis;

[0140] Based on the risk level classification results, sensitive events are marked and recorded, which involves marking, processing and recording the classified events. First, the system will classify them according to their risk level and specially mark events classified as high-risk or sensitive. The mark reflects the potential threat level that the event may pose to the organization. Each marked event will be recorded in detail in the financial risk early warning system. Based on the characteristics of these events and comparative analysis with historical data, the system outputs corresponding financial risk early warning records, providing management with the necessary information to formulate effective response measures and prevention strategies to ensure that the organization's risks are under control.

[0141] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A big data driven financial risk identification system, characterized by: The system comprises: The time data synchronization module collects real-time financial data streams, synchronizes time tags, calculates the time difference of each data point, performs dynamic time adjustment based on the calculation results, performs frequency correction and time tag calibration on the data, and outputs synchronized time series data; The multidimensional feature mapping module converts the time data into a spatial feature vector through multidimensional scaling based on the synchronized time series data, performs dimension mapping on the data at each time point, extracts key spatial features, and generates a mapping feature set; The financial behavior analysis module uses feature analysis to extract sequence patterns based on the mapping feature set, identifies key morphological changes in the time series, monitors and compares historical trend data in real time based on the recognition results, calculates the similarity between data points, reveals abnormal behavior of financial data, and outputs behavior recognition results; Based on the behavior recognition results, the risk dynamic identification module evaluates the risk level of abnormal behavior by comparing the risk thresholds at different time points. Based on the evaluation results, it determines the potential risk threats, classifies the risk levels, marks sensitive events, and outputs financial risk warning records. The steps for obtaining the synchronized time series data are as follows: Using the time difference data, linear interpolation is performed using the formula: Calculate a new timestamp for each data point , get the adjusted timestamp, where is the original timestamp, is the time difference value, is the normalization constant, is the total number of data points; Performing frequency correction on the data stream using the adjusted timestamp to ensure that the data acquisition frequency matches a preset standard, analyzing the original data frequency through Fourier transform and applying a filter to adjust it to the target frequency, thereby generating a frequency-corrected data stream; Time stamp calibration is performed on the frequency-corrected data stream, and time series analysis is applied to re-mark the timestamp of each data point to output synchronized time series data.

2. The big data driven financial risk identification system according to claim 1, characterized in that: The steps for calculating the time difference are: Perform data collection initialization processing, establish a secure connection with the financial market data interface, batch pull real-time transaction data of different financial products, and generate original real-time financial data streams; Based on the original real-time financial data stream, adding an exact server reception timestamp to each piece of data, synchronizing the time tag, and obtaining a real-time financial data stream with a time tag; Using the real-time financial data stream with time tags, the formula is adopted: Calculate the time difference between adjacent data points , generate time difference data, where, Representative The timestamp of each data point, Representative The timestamp of the data point.

3. The big data driven financial risk identification system according to claim 1, characterized in that: The steps for obtaining the mapping feature set are: Based on the synchronized time series data, a multidimensional scaling transformation is performed to preliminarily locate the multidimensional spatial position of each time data point and generate a primary spatial feature vector set; Using the primary spatial feature vector set, the formula is adopted: Calculating vectors With vector Euclidean distance between , generate the optimized spatial feature vector set, where and for and Vector in the The coordinate value of the dimension, Indicates the coordinate dimension; According to the optimized set of spatial feature vectors, the spatial attributes and relationships of the vectors are integrated to reveal the key spatial characteristics of the original data and generate mapping features.

4. The big data driven financial risk identification system according to claim 3, characterized in that: The steps for identifying the key morphological changes are: Extracting key time-intensity feature data based on the mapping feature set and generating windowed feature data using a sliding time window; Performing principal component analysis on the windowed feature data to compress data dimensions, extract key information, and generate reduced-dimensional feature data; The reduced dimensionality feature data is fed into a deep learning model to identify and classify the key morphological changes of the time series using the formula: Calculate the probability of a pattern in a time series , generate key morphological change recognition results, where For the The influence coefficient of each mode, Indicates the attenuation parameter corresponding to the corresponding mode, is the central time point of the corresponding pattern, Indicates the number of modes.

5. The big data driven financial risk identification system according to claim 4, characterized in that: The steps for obtaining the behavior recognition results are: Using the key morphological change identification results, new time series data is monitored in real time, compared with historical trend data, and difference analysis is performed to generate real-time monitoring and comparison results; Based on the real-time monitoring and comparison results, data points are extracted and the formula is used. Calculate the similarity coefficient , perform similarity evaluation and generate data point similarity evaluation results, where, and Represent the feature vectors of the current data point and the historical data point respectively, and Respectively represent the module length; Based on the data point similarity evaluation results, a threshold determination is performed to analyze the abnormality of the data points, identify data points that significantly deviate from the normal pattern, reveal potential abnormal financial data behavior, and generate behavior recognition results.

6. The big data driven financial risk identification system according to claim 5, characterized in that: The steps for assessing the risk level are: Utilizing the behavior recognition results, extracting key time points in the behavior data, comparing them with the set baseline risk threshold, and generating behavioral feature comparison results for each time point; Analyze the behavioral characteristics comparison results at each time point using the formula: Calculate the risk index at each time point , and obtain preliminary risk assessment results, among which, represents the deviation of behavioral data from the baseline threshold, is the baseline behavioral threshold; Based on the preliminary risk assessment results, each time point that deviates from the benchmark threshold is classified into a risk level, the risk level of each time point is determined, and a risk level assessment result is generated.

7. The big data driven financial risk identification system according to claim 6, characterized in that: The steps for obtaining the financial risk warning record are: Analyze the risk level assessment results, identify abnormal behavior events, and generate a list of potential risk threats; Based on the potential risk threat list, classify according to the risk level and mark abnormal events as sensitive events using the formula: % Calculate the percentage of sensitive events to total events , and obtain the risk level classification results, among which, is the number of abnormal events, is the total number of events; According to the risk level classification results, based on the characteristics of sensitive events and historical data analysis, financial risk warning records are output.

Images