AI-based Big Data Risk Assessment System

Abstract

This invention relates to the field of artificial intelligence technology, specifically to an AI-based big data risk assessment system, comprising the following modules: a time offset correction module, a risk factor modeling module, a mutation factor monitoring module, a risk cross-derivation module, and a prediction optimization and adjustment module. In this invention, by aligning the time density of multi-source data and unifying the reference baseline of time-series data, the temporal consistency of subsequent factor modeling is improved. By combining the trend offset rate and co-occurrence progression relationship of risk factors, the ability to identify the linkage effect between factors is enhanced. By using the trend deviation degree of cross points and the continuity of residuals, the identification of the risk impact range under the influence of multiple variables is optimized. By adopting a prediction error deviation trend combined with a weight self-adjustment mechanism, dynamic correction of model parameters is achieved, ensuring that the model adapts to the dynamic changes of the risk environment during long-term operation, enabling the risk prediction path to have adaptive repair capabilities, and improving the stability and accuracy of big data risk prediction.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an AI-based big data risk assessment system. Background Technology

[0002] The field of artificial intelligence technology encompasses multiple subfields, including machine learning, deep learning, natural language processing, and computer vision. It aims to use computers to simulate human intelligence, enabling tasks such as perception, reasoning, and decision-making. The core components of this technology include data acquisition and preprocessing, feature engineering, model training and optimization, and inference computation. It relies on large-scale datasets to build intelligent algorithms, achieving pattern recognition, automatic reasoning, and adaptive learning. Artificial intelligence technology is widely applied in fields such as speech recognition, image analysis, autonomous driving, medical diagnosis, and financial risk control. By utilizing neural networks, statistical learning, and reinforcement learning, it enhances the decision-making capabilities of computers and improves the accuracy of prediction and analysis through continuous learning and optimization.

[0003] Among them, the AI-based big data risk assessment system refers to the use of artificial intelligence technology to analyze, identify, and assess risks in batches of data, discover potential risk factors, and generate assessment results. The technical aspects of this system cover multiple stages, including data acquisition, data preprocessing, risk modeling, risk classification, and risk prediction. Data acquisition typically involves collecting multi-source heterogeneous data through web scraping, database connections, and API calls, and preprocessing the data by combining steps such as data cleaning, deduplication, and format conversion. Risk modeling uses statistical analysis, cluster analysis, and feature extraction to construct a risk assessment indicator system, and trains classification models based on machine learning, such as decision trees, support vector machines, and neural networks, to classify risk categories. Risk prediction uses time series analysis and regression analysis to construct trend prediction models, and identifies potential risks by combining threshold setting and rule matching, and uses real-time monitoring mechanisms to label and track abnormal data.

[0004] Traditional big data risk assessment techniques suffer from several drawbacks during risk prediction. First, asynchronous timestamps from multiple data sources lead to temporal misalignment after data integration, resulting in risk assessments based on incorrect time series and impacting prediction accuracy. Second, the failure to calculate trend shifts across multiple time windows for risk factor changes can cause short-term fluctuations to be misinterpreted as long-term trends, reducing assessment accuracy. Third, anomaly monitoring, typically based on static thresholds, cannot adapt to dynamic changes in mutation factors, leading to false alarms or missed alarms and reduced responsiveness to sudden risks. Fourth, risk crossover identification relies on threshold comparisons of single variables, failing to effectively measure complex interactions between multiple variables and resulting in incomplete risk transmission paths. Fifth, prediction error correction depends on fixed weight adjustments, unable to adaptively optimize based on error trends, causing prediction models to degrade in accuracy over long-term operation and making them ill-equipped to cope with continuous changes in the risk environment. Summary of the Invention

[0005] To address the technical problems existing in the prior art, this invention provides an AI-based big data risk assessment system. The technical solution is as follows:

[0006] On the one hand, it provides an AI-based big data risk assessment system, which includes:

[0007] The time offset correction module collects the timestamp field of multi-source data, identifies the position of the benchmark aligned data point by comparing the difference in the distribution density of data points in adjacent segments, corrects the timestamp sequence of the data source, and establishes the data time series offset value.

[0008] Based on the data time series offset value, the risk factor modeling module extracts the volatility rate of risk factor data for each time window, calculates the trend offset rate of risk factor data in adjacent windows, sets risk factor association rules according to the co-occurrence and time series correlation of risk factors, analyzes the progressive influence chain of risk factors, constructs a risk factor regression model, and establishes the risk factor influence gradient value.

[0009] The mutation factor monitoring module, based on the risk factor influence gradient value, monitors the fluctuation range of abnormal data in the risk factor data, detects the trajectory deviation direction of abnormal data, predicts the risk trigger time point, and generates the mutation factor trigger time point value.

[0010] The risk crossover derivation module calls the gradient value of the risk factor influence, calculates the difference in parameter offset rate, identifies candidate positions of risk crossover points, analyzes the continuity of crossover point trends by fitting the residuals of data trends and the deviation of current crossover point trends, and generates the range value of risk crossover point impact.

[0011] As a further embodiment of the present invention, the data time series offset value includes timestamp difference distribution information, alignment reference point position, and fragment density offset; the risk factor influence gradient value includes factor fluctuation rate, trend offset rate, and association rule group; the mutation factor trigger point value specifically includes abnormal fluctuation range, trajectory offset direction, and trigger prediction time; and the risk crossover point impact range value specifically refers to rate difference, trend deviation degree, and trend continuity level.

[0012] As a further aspect of the present invention, the time offset correction module includes:

[0013] The timestamp extraction submodule collects timestamp fields from multi-source data, extracts the timestamps of data points within each data segment, records the start time, end time, and time interval of the timestamps in each data segment, and obtains the recorded values ​​of the timestamp interval.

[0014] The data alignment determination submodule obtains the data point distribution density of each data segment based on the timestamp interval record value, calculates the distribution density difference between adjacent segments, and sets the reference alignment point position based on the distribution density difference.

[0015] The timing correction submodule calls the reference alignment point position, analyzes the offset between the timestamp of each data segment and the reference alignment point, corrects the timestamp of each data segment based on the difference between the timestamp intervals, and establishes the data timing offset value.

[0016] As a further aspect of the present invention, the risk factor modeling module includes:

[0017] The fluctuation feature extraction submodule obtains risk factor data within each time window based on the data time series offset value, detects the magnitude of the change in the value of the risk factor and the rate of change of adjacent data points within each time window, calculates the change gradient of the data points within each time window, and obtains the fluctuation rate value of the risk factor.

[0018] The trend offset calculation submodule calls the risk factor volatility rate value, calculates the difference in the rate of change of risk factor data between adjacent windows, analyzes the trend of risk factor change in the time dimension, and obtains the risk factor trend offset value.

[0019] The regression model construction submodule calls the trend offset value of the risk factors, analyzes the co-occurrence characteristics and temporal correlation characteristics of multiple risk factors, sets the correlation rules between risk factors, analyzes the progressive influence chain by combining the trend change magnitude of risk factors, constructs a risk factor regression model, and obtains the gradient value of the influence of risk factors.

[0020] As a further aspect of the present invention, the mutation factor monitoring module includes:

[0021] The anomaly detection submodule, based on the risk factor influence gradient value, analyzes the numerical fluctuations in the risk factor data, detects the magnitude of change of each data point, calculates the deviation of each data point, detects abnormal data points, and obtains the abnormal data deviation value.

[0022] The trajectory analysis submodule calls the abnormal data deviation value to obtain the time series information of the abnormal data points, calculates the change direction of the abnormal data at adjacent time points, analyzes the change trend of the abnormal data on the time axis, extracts the continuous offset path of the data change trend, and obtains the trajectory offset value of the abnormal data.

[0023] The timing prediction submodule calls the abnormal data trajectory offset value, analyzes the trajectory change trend of the abnormal data, calculates the trajectory extension rate, analyzes the trend change magnitude, predicts the risk trigger time point, and obtains the mutation factor trigger time point value.

[0024] As a further aspect of the present invention, the specific formula for calculating the trajectory extension rate is as follows:

[0025]

[0026] Calculate the trajectory extension rate;

[0027] Among them, V TRP X represents the trajectory extension rate. end The value representing the endpoint of the trajectory, X start The value representing the starting point of the trajectory, X i′ X represents the value of the i′-th data point. i′+1 T represents the value of the (i′+1)th data point, n′ represents the total number of outlier data points in the trajectory, and T end T represents the time at which the trajectory terminates. start The time ΔT represents the starting point of the trajectory. i′ represents the time interval between adjacent data points, n′ represents the total number of outlier data points in the time series, i′ represents the index number of the outlier data point in the time series, and i′+1 represents the index of the next data point after the current data point i′ in the trajectory sequence.

[0028] As a further aspect of the present invention, the risk cross-deduction module includes:

[0029] The parameter offset calculation submodule collects the change data of risk factors in a continuous time series based on the risk factor influence gradient value, calculates the parameter change rate of each risk factor between adjacent time nodes, obtains the change difference of parameter rate between risk factors, and establishes the parameter rate offset difference.

[0030] The intersection identification submodule calls the parameter rate offset difference, predicts the location of the intersection area of ​​the risk factor parameter change paths according to the set rate offset threshold, and establishes a risk intersection candidate area.

[0031] The trend continuity analysis submodule calls the risk crossover candidate region, obtains the parameter trend data within the crossover region, calculates the trend fitting residual value and trend deviation magnitude, analyzes the continuity of the crossover trend, and establishes the risk crossover range value.

[0032] As a further aspect of the present invention, the system further includes:

[0033] The prediction optimization and adjustment module calls the trigger point value of the mutation factor and the range value of the risk crossover point, combines the deviation between the actual risk event backtracking data and the prediction results, calculates the prediction error deviation trend, adjusts the model parameters, including the weight coefficient of each risk factor, and generates a risk prediction correction value.

[0034] The risk prediction correction value includes the error deviation trend, risk weight adjustment value, and model parameter correction coefficient.

[0035] As a further aspect of the present invention, the prediction optimization adjustment module includes:

[0036] The error calculation submodule calculates the time deviation between the predicted risk triggering time and the actual risk event occurrence time based on the triggering time value of the mutation factor and the ripple range value of the risk crossover point, and establishes the prediction error time deviation value.

[0037] The trend analysis submodule calls the prediction error time deviation value to obtain the change data of the error value within a continuous time window, calculates the trend rate of error change, analyzes the direction of error change, and establishes the error change trend value.

[0038] The parameter adjustment submodule calls the error change trend value and optimizes multiple parameters of the risk factor regression model through iterative adjustment, including the weight coefficient of each risk factor, and establishes a risk prediction correction value.

[0039] As a further aspect of the present invention, the specific formula for optimizing multiple parameters of the risk factor regression model is as follows:

[0040]

[0041] Calculate the adjusted risk factor weights;

[0042] Among them, W new W represents the adjusted risk factor weights. old V represents the original risk factor weights, k represents the risk adjustment factor, and V represents the risk factor weights. errV represents the rate of change of error. base The baseline rate of change of the risk factor, ΔT err This represents the trend of error change.

[0043] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0044] By aligning the temporal density of multi-source data and unifying the reference baseline of time-series data, the temporal consistency of subsequent factor modeling is improved. By combining the trend deviation rate and co-occurrence progression relationship of risk factors, the ability to identify the linkage effect between factors is enhanced. By using the trend deviation degree of crossover points and the continuity of residuals, the identification of the risk ripple range under the influence of multiple variables is optimized. By adopting the prediction error deviation trend combined with the weight self-adjustment mechanism, the dynamic correction of model parameters is realized, ensuring that the model adapts to the dynamic changes of the risk environment during long-term operation, enabling the risk prediction path to have adaptive repair capabilities, and improving the stability and accuracy of big data risk prediction. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a system flowchart of the present invention;

[0047] Figure 2 This is a schematic diagram of the system framework of the present invention;

[0048] Figure 3 This is a flowchart of the time offset correction module of the present invention;

[0049] Figure 4 This is a flowchart of the risk factor modeling module of the present invention;

[0050] Figure 5 This is a flowchart of the mutation factor monitoring module of the present invention;

[0051] Figure 6 This is a flowchart of the risk cross-deduction module of the present invention;

[0052] Figure 7 This is a flowchart of the prediction, optimization, and adjustment module of the present invention. Detailed Implementation

[0053] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0054] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0055] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent.

[0056] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0057] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0058] This invention provides an AI-based big data risk assessment system. Please refer to [link / reference]. Figures 1 to 2 This invention provides a technical solution: an AI-based big data risk assessment system comprising:

[0059] The time offset correction module collects the timestamp field of multi-source data, identifies the position of the benchmark aligned data point by comparing the difference in the distribution density of data points in adjacent segments, corrects the timestamp sequence of the data source, and establishes the data time series offset value.

[0060] The risk factor modeling module extracts the volatility rate of risk factor data for each time window based on the data time series offset value, calculates the trend offset rate of risk factor data in adjacent windows, sets risk factor association rules according to the co-occurrence and time series correlation of risk factors, analyzes the progressive influence chain of risk factors, constructs a risk factor regression model, and establishes the risk factor influence gradient value.

[0061] The mutation factor monitoring module predicts the risk trigger time point and generates mutation factor trigger time point value by monitoring the fluctuation range of abnormal data in the risk factor data, detecting the trajectory deviation direction of abnormal data, and based on the gradient value of the risk factor influence.

[0062] The risk crossover derivation module calls the gradient value of the risk factor influence, calculates the difference in parameter offset rate, identifies candidate positions of risk crossover points, analyzes the trend continuity of crossover points by fitting the residuals of data trends and the deviation of the current crossover point trend, and generates the range value of risk crossover point impact.

[0063] The prediction optimization and adjustment module calls the trigger point value of the mutation factor and the impact range value of the risk crossover point. It combines the deviation between the actual risk event backtracking data and the prediction results to calculate the prediction error deviation trend, adjust the model parameters, including the weight coefficient of each risk factor, and generate the risk prediction correction value.

[0064] Data time series offset values ​​include timestamp difference distribution information, alignment reference point position, and fragment density offset. Risk factor impact gradient values ​​include factor fluctuation rate, trend offset rate, and association rule group. The trigger point value of mutation factor specifically includes abnormal fluctuation range, trajectory offset direction, and trigger prediction time. The risk crossover range value specifically refers to rate difference, trend deviation degree, and trend continuity level. Risk prediction correction values ​​include error deviation trend, risk weight adjustment value, and model parameter correction coefficient.

[0065] Please see Figure 2 and Figure 3 The time offset correction module includes:

[0066] The timestamp extraction submodule collects timestamp fields from multi-source data, extracts the timestamps of data points within each data segment, records the start time, end time, and time interval of the timestamps in each data segment, and obtains the recorded values ​​of the timestamp interval.

[0067] The timestamp extraction submodule collects the timestamps of data points from each segment of multi-source data, obtains the timestamp fields from each data source, and extracts the time information for each data segment, recording the timestamp value of each data point one by one. For each data segment, it records the start and end times of its timestamp and calculates the time interval. Taking bank transaction data, online lending platform data, and third-party payment platform data as examples, each platform's data segment contains consecutive data points. For bank transaction data, if a segment contains 200 transaction data points, its start time is the timestamp of the first data point, and its end time is the timestamp of the last data point. The module uses the start and end times of the data segment to calculate the time interval and time span for subsequent data alignment and time-series correction operations. The formula used is:

[0068] ΔT=T end -T start ;

[0069] Calculate the time interval, where ΔT is the time interval of the time segment, and T end T represents the end time of the data segment.start This represents the start time of the data segment.

[0070] Set T start =12:00:00, T end =12:03:19, substitute the set value into the calculation:

[0071] ΔT = 12:03:19 - 12:00:00 = 199 seconds;

[0072] The calculation results show that the duration of the data segment is 199 seconds, and this time span is used for subsequent distribution density calculation and data alignment processing. After obtaining the start time, end time, and interval time of each data segment, timestamp intervals are recorded to ensure the integrity of subsequent calculations.

[0073] The data alignment determination submodule obtains the data point distribution density of each data segment based on the timestamp interval record value, calculates the distribution density difference between adjacent segments, and sets the benchmark alignment point position based on the distribution density difference.

[0074] The data alignment determination submodule calls the timestamp interval record values ​​mentioned above to calculate the data point distribution density of each data segment and evaluate the data distribution characteristics between different data segments. It obtains the total number of data points within a data segment and calculates the distribution density of each segment, using the ratio of the total number of data points to the time interval. Based on the density characteristics of the data segments, it calculates the density difference between adjacent segments and determines the position of the baseline alignment point based on the density difference for subsequent time series correction. The formula used is:

[0075]

[0076] Calculate the distribution density of the data points, where ρ is the data distribution density, N is the total number of data points in the data segment, and ΔT is the time interval of the data segment.

[0077] Set N = 200 and ΔT = 199, substitute these values ​​into the calculation:

[0078]

[0079] The calculation results show that the data segment distribution density is 1.005 data points per second. This value is used to calculate the density difference to determine the data alignment point. The calculated data distribution density is then used to calculate the density difference between adjacent data segments. A density difference threshold is set; data segments with a density difference lower than this threshold are selected as the baseline alignment point to ensure the stability and consistency of data alignment.

[0080] The timing correction submodule calls the baseline alignment point position, analyzes the offset between the timestamp of each data segment and the baseline alignment point, corrects the timestamp of each data segment based on the difference between the timestamp intervals, and establishes the data timing offset value.

[0081] The timing correction submodule calls the baseline alignment point position, calculates the time offset of the data segment, and performs timing correction operations. The time offset calculation is based on the difference between the end time of the data segment and the timestamp of the baseline alignment point. For different data segments, the individual time offset is calculated, and this offset is applied to correct the timestamps. By adjusting the timestamps, the time of all data segments is aligned, ensuring data synchronization and reducing data errors caused by time misalignment. The formula used is:

[0082] ΔT offset =T end -T align ;

[0083] Calculate the time offset, where ΔT offset T is the time offset. end T represents the end time of the data segment. align The reference alignment point position.

[0084] Set T end =12:03:09, T align =12:03:19, substitute the set value into the calculation:

[0085] ΔT offset =12:03:09-12:03:19=-10 seconds;

[0086] The calculation results show a time offset of -10 seconds, indicating that the timestamps of the data segment are advanced by 10 seconds. The calculated offset is then used to correct all timestamps in the data segment using the following correction formula:

[0087] T corrected =T original +|ΔT offset |;

[0088] Among them, T corrected For the corrected timestamp, T original For the original data point timestamps, |ΔT offset | represents the absolute value of the time offset.

[0089] Set T original =12:00:05, ΔT offset =-10, substitute the set value into the calculation:

[0090] T corrected=12:00:05+10=12:00:15;

[0091] Calculation results show that the timestamp correction to 12:00:15 ensures the consistency of the time series. After correcting the timestamps of all data segments, a final data time series offset value is established to provide accurate data input and ensure the stability of subsequent analysis.

[0092] Please see Figure 2 and Figure 4 The risk factor modeling module includes:

[0093] The volatility feature extraction submodule obtains risk factor data within each time window based on the data time series offset value, detects the magnitude of the change in the value of the risk factor and the rate of change of adjacent data points within each time window, calculates the change gradient of the data points within each time window, and obtains the volatility rate value of the risk factor.

[0094] The volatility feature extraction submodule acquires risk factor data within each time window based on the data time series offset value and extracts the variation characteristics of the risk factor data. For financial risk control scenarios, each time window can be set to a day, an hour, or a shorter period, divided according to business needs. After acquiring the data within the time window, the magnitude of the numerical change of the risk factor within each time window is calculated. This magnitude can be calculated by the difference between the maximum and minimum values ​​and is used to characterize the volatility within the time window. Further, the rate of change between adjacent data points is extracted, calculated as the ratio of the numerical change of adjacent data points to the time interval. This rate is used to assess the short-term volatility of the risk factor. Finally, the gradient of change for all data points within the time window is calculated. This gradient can be used to identify data change trends and establish the volatility rate value of the risk factor. The formula used is:

[0095]

[0096] Among them, G BW For the changing gradient, |X q+1 -X q | represents the change in value between adjacent data points, N is the total number of data points within the time window, and ΔT is the total duration of the time window.

[0097] Given a time window containing N = 5 data points with values ​​[10, 15, 13, 18, 20], and a total time window duration ΔT = 60 seconds, calculate the gradient of change:

[0098]

[0099] The calculation results show that the volatility rate of the risk factor in this time window is 0.058, and the volatility rate of the risk factor is obtained.

[0100] The trend offset calculation submodule calls the risk factor volatility rate value, calculates the difference in the rate of change of risk factor data between adjacent windows, analyzes the trend of risk factors in the time dimension, and obtains the trend offset value of risk factors.

[0101] The trend offset calculation submodule calls the risk factor volatility rate value to calculate the difference in the rate of change of risk factor data between adjacent time windows, analyzing the trend of change over time. In financial risk control applications, if a company's financial health is used as a risk factor, the rate of change of financial data within different time windows is calculated, and the rate of change between adjacent time windows is compared. The rate of change between adjacent windows is calculated as the difference between the rate of change of the risk factor in the preceding and following windows, used to characterize the direction of the trend change of the risk factor. For a specific risk factor, within time window T... m -1 and T m The difference in the rate of change is calculated internally, and this difference is used to construct a trend offset value. The formula used is:

[0102] D TR =G BW,m -G BW,m-1 ;

[0103] Among them, D TR G represents the trend offset value. BW,m G represents the gradient of change within the current time window. BW,m-1 This represents the gradient of change in the previous time window.

[0104] Setting G BW,m =0.058, G BW,m-1 =0.040, calculate the trend offset value:

[0105] D TR =0.058 - 0.040 = 0.018;

[0106] The calculation results show that the trend offset value of this risk factor is 0.018, indicating that the risk volatility in the current window has increased compared to the previous window. Obtain the trend offset value of the risk factor.

[0107] The regression model construction submodule calls the trend offset value of risk factors, analyzes the co-occurrence characteristics and time-series correlation characteristics of multiple risk factors, sets the correlation rules between risk factors, combines the trend change magnitude of risk factors, analyzes the progressive influence chain, constructs a risk factor regression model, and obtains the gradient value of the influence of risk factors.

[0108] The regression model construction submodule calls the trend offset values ​​of risk factors and sets association rules between risk factors by analyzing the co-occurrence and temporal correlation characteristics of multiple risk factors. For credit risk assessment in the financial industry, AI is used to extract the correlation between multiple factors, such as corporate cash flow, debt ratio, and accounts receivable, which may be correlated in time. Therefore, when calculating the trend changes of each risk factor, it is necessary to combine their co-occurrence and analyze the progressive impact between different risk factors. The magnitude of the trend changes of risk factors can be used as a weight input to measure the contribution of each factor to the overall risk impact and to calculate the impact gradient of risk factors. The formula used is:

[0109]

[0110] Among them, W FI Let |D be the weight of the i-th risk factor. TR,i | represents the trend offset value of this risk factor. M is the sum of the trend offset values ​​of all risk factors, where M is the total number of risk factors.

[0111] Set the trend offset values ​​for the three risk factors as D. TR,1 =0.018, D TR,2 =0.025, D TR,3 =0.015, calculate the weight of each factor:

[0112]

[0113] The calculation results show that risk factor 2 has a greater impact, accounting for approximately 43.1%, while risk factors 1 and 3 account for 31.0% and 25.9%, respectively. Based on the calculated weight values, a risk factor regression model is constructed to obtain the gradient values ​​of the risk factor impact.

[0114] Please see Figure 2 and Figure 5 The mutation factor monitoring module includes:

[0115] The anomaly detection submodule is based on the gradient value of the impact of risk factors. By analyzing the numerical fluctuations in the risk factor data, it detects the magnitude of change of each data point, calculates the deviation of each data point, detects abnormal data points, and obtains the abnormal data deviation value.

[0116] The anomaly detection submodule, based on the risk factor impact gradient value, analyzes the numerical fluctuations in risk factor data, extracts the value of each data point, and calculates its fluctuation amplitude. First, it acquires the value of each data point and records its time position. Then, it calculates the difference between adjacent data points, forming a data fluctuation amplitude matrix. This matrix characterizes the fluctuation state of risk factors at each time point. For each value in this matrix, it further calculates its deviation from the average value within its time window using the formula:

[0117]

[0118] Among them, PD ip Let X be the deviation of the ipth data point. ip Let be the value of the ipth data point, μ be the mean within the current time window, and σ be the standard deviation within the current time window.

[0119] Suppose the numerical data within a certain time window is [20, 22, 18, 21, 23], with a mean μ = 20.8 and a standard deviation σ ≈ 1.79. Calculate the deviation of the first data point:

[0120]

[0121] For each data point, the deviation is calculated sequentially, and it is determined whether the deviation exceeds a preset deviation threshold, for example, a deviation threshold of 1.5. If the deviation exceeds the threshold, the data point is marked as an outlier, and the outlier deviation value is obtained.

[0122] The trajectory analysis submodule calls the abnormal data deviation value to obtain the time series information of the abnormal data points, calculates the change direction of the abnormal data at adjacent time points, analyzes the change trend of the abnormal data on the time axis, extracts the continuous offset path of the data change trend, and obtains the trajectory offset value of the abnormal data.

[0123] The trajectory analysis submodule calls the abnormal data deviation value to extract the time series information of each abnormal data point, including the time point identifier, the data point value, and its corresponding deviation. By analyzing the changes in abnormal data at adjacent time points, it calculates the direction of change of each abnormal data point relative to the previous data point, and uses this direction of change to track the offset trajectory of the abnormal data on the time axis. The method for calculating the direction of change is the ratio of the change in value of adjacent data points to the time interval, using the formula:

[0124]

[0125] Among them, D ip X represents the direction of change for the ipth data point. ip+1 X is the value of the (ip+1)th data point. ipLet be the value of the ipth data point, and ΔT be the time interval between the two points.

[0126] Assuming the outlier data points are [25, 28, 30, 33, 40], and the time interval between the two points is 5 seconds, calculate the first direction of change value:

[0127]

[0128] Calculate all change direction values ​​sequentially, select data points with continuous change directions and large deviations, further analyze the fluctuation trend between data points, mark the trend change path as the data offset path, and obtain the abnormal data trajectory offset value.

[0129] The timing prediction submodule calls the trajectory offset value of abnormal data, analyzes the trajectory change trend of abnormal data, calculates the trajectory extension rate, analyzes the trend change magnitude, predicts the risk trigger time point, and obtains the trigger time point value of the mutation factor.

[0130] The specific formula for calculating the trajectory extension rate is:

[0131]

[0132] Calculate the trajectory extension rate;

[0133] Among them, V TRP X represents the trajectory extension rate. end The value representing the endpoint of the trajectory, X start The value representing the starting point of the trajectory, X i′ X represents the value of the i′-th data point. i′+1 T represents the value of the (i′+1)th data point, n′ represents the total number of outlier data points in the trajectory, and T end T represents the time at which the trajectory terminates. start The time ΔT represents the starting point of the trajectory. i′ represents the time interval between adjacent data points, n′ represents the total number of outlier data points in the time series, i′ represents the index number of the outlier data point in the time series, and i′+1 represents the index of the next data point after the current data point i′ in the trajectory sequence.

[0134] formula:

[0135]

[0136] Detailed explanation of the formula and its calculation derivation:

[0137] The formula is used to calculate the trajectory weighted extension rate, and the result is used to predict the trajectory development trend of abnormal data points, and further derive the trigger point value of the mutation factor.

[0138] Parameter meanings and settings:

[0139] X end The value representing the trajectory termination point is set to 85, reflecting the value of abnormal data points at the trajectory termination point;

[0140] X start The value representing the starting point of the trajectory is set to 30, reflecting the value of the abnormal data point at the starting point of the trajectory;

[0141] X i′ The value representing the i′th data point is set to [35, 50, 65, 75], which reflects the values ​​of abnormal data points in the trajectory.

[0142] n′ represents the total number of outlier data points within the trajectory, set to 4, reflecting the number of outlier points contained within the data trajectory in the time series.

[0143] T end The time representing the trajectory termination point is set to 25 seconds, reflecting the timestamp of abnormal data points reaching the endpoint.

[0144] T start The time representing the starting point of the trajectory is set to 5 seconds, reflecting the initial timestamp of the abnormal data point;

[0145] ΔT i′ This represents the time interval between adjacent data points, set to [4,6,5,5], reflecting the time interval of data point changes on the time axis;

[0146] This represents the cumulative change of all outlier data points within the trajectory.

[0147] This represents the square root of the sum of the squares of all time intervals within the trajectory.

[0148] Substitute the parameters into the formula to calculate:

[0149]

[0150] Analysis of calculation results:

[0151] The calculated result 2.74 represents the weighted extension rate of the outlier data point in the trajectory. This value reflects the growth trend of the outlier data trajectory and is used to predict the future development direction of the data trajectory. A higher value indicates that the trajectory of the outlier data point changes faster and may trigger the mutation factor in a short period of time, while a lower value indicates that the trajectory changes slower and the mutation factor may be triggered after a longer period of time. The result will be further used to calculate the trigger point value of the mutation factor and identify the time node when the risk occurs.

[0152] Please see Figure 2 and Figure 6 The risk cross-deduction module includes:

[0153] The parameter offset calculation submodule collects the change data of risk factors in continuous time series based on the gradient value of the risk factor influence, calculates the parameter change rate of each risk factor between adjacent time nodes, obtains the change difference of parameter rate between risk factors, and establishes the parameter rate offset difference.

[0154] The parameter offset calculation submodule, based on the gradient values ​​of risk factor influence, collects the change data of risk factors in continuous time series, extracts the values ​​of risk factors at different time points, and calculates the rate of parameter change between adjacent time points. By calculating the rate of change of risk factor values ​​within each time window, a risk factor rate sequence is constructed. For the rate of change of risk factors in adjacent time windows, the difference in the rate of change within adjacent time windows is calculated using the formula:

[0155] ΔS pq =S pq+1 -S pq ;

[0156] Where, ΔS pq S represents the difference in the rate of change of the parameter within the pq-th time window. pq+1 S is the rate of change in the (pq+1)th time window. pq Let be the rate of change of the pq-th time window.

[0157] Suppose that the rate of change of a certain risk factor is 2.5, 3.0, and 4.2 in three consecutive time windows, respectively. Calculate the rate difference between the first and second time windows:

[0158] ΔS1 = 3.0 - 2.5 = 0.5;

[0159] The rate change values ​​of all adjacent time windows are calculated sequentially, and the time windows with larger change rates are extracted to screen out the key change nodes of risk factors and establish parameter rate offset differences.

[0160] The intersection identification submodule calls the parameter rate offset difference, predicts the intersection area of ​​the risk factor parameter change paths based on the set rate offset threshold, and establishes risk intersection candidate areas.

[0161] The intersection point identification submodule calls the parameter rate offset difference to analyze the change pattern of risk factor change rates within all time windows, and selects time windows where the rate change exceeds the set rate offset threshold. For all selected time windows, it further analyzes the risk factor parameter change paths and finds the intersection regions of different risk factor parameter change paths. The determination of the intersection region is based on the intersection points of multiple risk factor change paths, using the formula:

[0162]

[0163] Among them, P int S represents the intersection of the paths of change of risk factors. A and S B T represents the rate of change of the two risk factors within the cross region. A and T B This corresponds to the time interval.

[0164] Assuming two risk factors change at rates of 2.8 and 3.4 within an intersection region, corresponding to time intervals of 6 seconds and 4 seconds respectively, calculate the intersection point:

[0165]

[0166] Select a time window where the intersection point values ​​are relatively stable, and establish a candidate region for risky intersections.

[0167] The trend continuity analysis submodule calls the risk crossover candidate region, obtains the parameter trend data within the crossover region, calculates the trend fitting residual value and trend deviation magnitude, analyzes the continuity of the crossover trend, and establishes the risk crossover range value.

[0168] The trend continuity analysis submodule calls the risk crossover candidate region, analyzes the parameter trend data of all risk factors within the crossover region, and calculates the trend fitting residual value. The trend fitting residual value is calculated based on historical trend data using the formula:

[0169]

[0170] Among them, R fit X represents the trend fitting residual. ia Let X be the actual value of the ia-th data point. fit is the fitted value for that point, and na is the total number of data points.

[0171] Assuming the risk factor data points within the cross region are [5.2, 4.8, 5.1, 4.9, 5.0], and the fitted data point values ​​are [5.0, 5.0, 5.0, 5.0, 5.0], calculate the trend fitting residual:

[0172]

[0173] Further calculate the trend deviation magnitude and analyze the continuity of the crossover trend to finally establish the risk crossover range value.

[0174] Please see Figure 2 and Figure 7 The prediction, optimization, and adjustment module includes:

[0175] The error calculation submodule calculates the time deviation between the predicted risk triggering time and the actual risk event occurrence time based on the triggering time value of the mutation factor and the ripple range value of the risk crossover point, and establishes the prediction error time deviation value.

[0176] The error calculation submodule, based on the mutation factor trigger time point value and the risk crossover range value, collects the predicted risk trigger time point and the actual risk event occurrence time, and calculates the time deviation. It extracts the mutation factor trigger time point value, determines the corresponding time position of the mutation factor, obtains the risk crossover range value, filters key time points within the crossover area, arranges all time points in chronological order, and calculates the time difference between the predicted time point and the actual event occurrence time using the formula:

[0177] ΔT=T pred -T actual ;

[0178] Where ΔT is the prediction error time deviation value, T pred To predict the timing of risk triggering, T actual This refers to the actual time when the risk event occurred.

[0179] Assuming a certain mutation factor triggers at 12.5 seconds, and the actual event occurrence time within the risk crossover area is 10.8 seconds, the prediction error is calculated as follows:

[0180] ΔT = 12.5 - 10.8 = 1.7 seconds;

[0181] Calculate the error values ​​for all prediction time points and establish the prediction error time deviation value.

[0182] The trend analysis submodule calls the prediction error time deviation value to obtain the change data of the error value within a continuous time window, calculates the trend rate of error change, analyzes the direction of error change, and establishes the error change trend value.

[0183] The trend analysis submodule calls the prediction error time deviation value to obtain the error value's change data within a continuous time window, calculates the trend rate of error change, and analyzes the direction of error change. For error values ​​in multiple time windows, it calculates the rate of change of error values ​​within adjacent time windows and constructs an error change trend sequence. The error change trend rate is calculated using the formula:

[0184]

[0185] Among them, V err Let ΔT be the rate of change of the error trend. ib+1 and ΔT ib These are the error values ​​for two adjacent time windows, T. w This represents the size of the time window.

[0186] Assuming the error values ​​are 1.7 seconds and 2.4 seconds within time windows T1 and T2 respectively, and the window size is 5 seconds, the rate of error change is calculated as follows:

[0187]

[0188] Calculate the rate of error change trend for all time windows and establish the error change trend value.

[0189] The parameter adjustment submodule calls the error change trend value and optimizes multiple parameters of the risk factor regression model through iterative adjustment, including the weight coefficient of each risk factor, and establishes the risk prediction correction value.

[0190] The specific formula for optimizing multiple parameters of the risk factor regression model is as follows:

[0191]

[0192] Calculate the adjusted risk factor weights;

[0193] Among them, W new W represents the adjusted risk factor weights. old V represents the original risk factor weights, k represents the risk adjustment factor, and V represents the risk factor weights. err V represents the rate of change of error. base The baseline rate of change of the risk factor, ΔT err This represents the trend of error change.

[0194] formula:

[0195]

[0196] Detailed explanation of the formula and its calculation derivation:

[0197] The formula is used to calculate the adjusted risk factor weights, and the results are used to optimize the parameter adjustment in the risk factor regression model.

[0198] Parameter meanings and settings:

[0199] W new The adjusted risk factor weights are used to correct the model's risk assessment capabilities;

[0200] W old The original risk factor weights represent the relative importance of the risk factors in the model calculation before error correction, and are set to 0.75.

[0201] k is the risk adjustment factor, set at 0.3;

[0202] V err Let V be the rate of change of the error trend, representing the rate of change of the prediction error. err =0.5;

[0203] V base The baseline rate of change for risk factors represents the rate of change of the standard error without additional influence, with a set value of 0.25.

[0204] ΔT err Let ΔT be the error trend value, that is, the trend of error value within a continuous time window. err =0.0625;

[0205] Substitute the parameters into the formula to calculate:

[0206]

[0207] W new =0.75×(1-0.3×0.2427);

[0208] W new =0.75×(1-0.0728);

[0209] W new =0.75 × 0.9272 = 0.6954;

[0210] Calculation result W new =0.6954 indicates the degree of adjustment of the weighting factor of the risk factor after error correction, which has been reduced from the original 0.75 to 0.6954. This value is used to correct the weighting parameters of the risk factor in the risk assessment model in order to optimize the overall assessment accuracy and adjust the importance distribution of each risk factor.

[0211] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0212] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0213] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0214] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0215] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0216] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0217] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0218] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0219] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0220] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0221] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An AI-based big data risk assessment system, characterized in that, The system includes: The time offset correction module collects the timestamp field of multi-source data, identifies the position of the benchmark aligned data point by comparing the difference in the distribution density of data points in adjacent segments, corrects the timestamp sequence of the data source, and establishes the data time series offset value. Based on the data time series offset value, the risk factor modeling module extracts the volatility rate of risk factor data for each time window, calculates the trend offset rate of risk factor data in adjacent windows, sets risk factor association rules according to the co-occurrence and time series correlation of risk factors, analyzes the progressive influence chain of risk factors, constructs a risk factor regression model, and establishes the risk factor influence gradient value. The mutation factor monitoring module, based on the risk factor influence gradient value, monitors the fluctuation range of abnormal data in the risk factor data, detects the trajectory deviation direction of abnormal data, predicts the risk trigger time point, and generates the mutation factor trigger time point value. The risk crossover derivation module calls the gradient value of the risk factor influence, calculates the difference in parameter offset rate, identifies candidate positions of risk crossover points, analyzes the continuity of crossover point trend by fitting the residual of data trend and the deviation of the current crossover point trend, and generates the range value of risk crossover point impact. The data time series offset value includes timestamp difference distribution information, alignment reference point position, and fragment density offset. The risk factor influence gradient value includes factor fluctuation rate, trend offset rate, and association rule group. The mutation factor trigger point value specifically includes abnormal fluctuation range, trajectory offset direction, and trigger prediction time. The risk crossover point impact range value specifically refers to rate difference, trend deviation degree, and trend continuity level.

2. The AI-based big data risk assessment system according to claim 1, characterized in that, The time offset correction module includes: The timestamp extraction submodule collects timestamp fields from multi-source data, extracts the timestamps of data points within each data segment, records the start time, end time, and time interval of the timestamps in each data segment, and obtains the recorded values ​​of the timestamp interval. The data alignment determination submodule obtains the data point distribution density of each data segment based on the timestamp interval record value, calculates the distribution density difference between adjacent segments, and sets the reference alignment point position based on the distribution density difference. The timing correction submodule calls the reference alignment point position, analyzes the offset between the timestamp of each data segment and the reference alignment point, corrects the timestamp of each data segment based on the difference between the timestamp intervals, and establishes the data timing offset value.

3. The AI-based big data risk assessment system according to claim 1, characterized in that, The risk factor modeling module includes: The fluctuation feature extraction submodule obtains risk factor data within each time window based on the data time series offset value, detects the magnitude of the change in the value of the risk factor and the rate of change of adjacent data points within each time window, calculates the change gradient of the data points within each time window, and obtains the fluctuation rate value of the risk factor. The trend offset calculation submodule calls the risk factor volatility rate value, calculates the difference in the rate of change of risk factor data between adjacent windows, analyzes the trend of risk factor change in the time dimension, and obtains the risk factor trend offset value. The regression model construction submodule calls the trend offset value of the risk factors, analyzes the co-occurrence characteristics and temporal correlation characteristics of multiple risk factors, sets the correlation rules between risk factors, analyzes the progressive influence chain by combining the trend change magnitude of risk factors, constructs a risk factor regression model, and obtains the gradient value of the influence of risk factors.

4. The AI-based big data risk assessment system according to claim 1, characterized in that, The mutation factor monitoring module includes: The anomaly detection submodule, based on the risk factor influence gradient value, analyzes the numerical fluctuations in the risk factor data, detects the magnitude of change of each data point, calculates the deviation of each data point, detects abnormal data points, and obtains the abnormal data deviation value. The trajectory analysis submodule calls the abnormal data deviation value to obtain the time series information of the abnormal data points, calculates the change direction of the abnormal data at adjacent time points, analyzes the change trend of the abnormal data on the time axis, extracts the continuous offset path of the data change trend, and obtains the trajectory offset value of the abnormal data. The timing prediction submodule calls the abnormal data trajectory offset value, analyzes the trajectory change trend of the abnormal data, calculates the trajectory extension rate, analyzes the trend change magnitude, predicts the risk trigger time point, and obtains the mutation factor trigger time point value.

5. The AI-based big data risk assessment system according to claim 4, characterized in that, The specific formula for calculating the trajectory extension rate is as follows: ； Calculate the trajectory extension rate; in, Represents the trajectory extension rate. The value representing the endpoint of the trajectory. The value representing the starting point of the trajectory. Representing the The value of each data point. Representing the The value of each data point. The time representing the end point of the trajectory. The time representing the starting point of the trajectory. Represents the time interval between adjacent data points. This represents the total number of outlier data points in the time series. This represents the index number of the outlier data point in the time series. Represents the current data point The index of the next data point.

6. The AI-based big data risk assessment system according to claim 1, characterized in that, The risk cross-deduction module includes: The parameter offset calculation submodule collects the change data of risk factors in a continuous time series based on the risk factor influence gradient value, calculates the parameter change rate of each risk factor between adjacent time nodes, obtains the change difference of parameter rate between risk factors, and establishes the parameter rate offset difference. The intersection identification submodule calls the parameter rate offset difference, predicts the location of the intersection area of ​​the risk factor parameter change paths according to the set rate offset threshold, and establishes a risk intersection candidate area. The trend continuity analysis submodule calls the risk crossover candidate region, obtains the parameter trend data within the crossover region, calculates the trend fitting residual value and trend deviation magnitude, analyzes the continuity of the crossover trend, and establishes the risk crossover range value.

7. The AI-based big data risk assessment system according to claim 1, characterized in that, The system also includes: The prediction optimization and adjustment module calls the trigger point value of the mutation factor and the range value of the risk crossover point, combines the deviation between the actual risk event backtracking data and the prediction results, calculates the prediction error deviation trend, adjusts the model parameters, including the weight coefficient of each risk factor, and generates a risk prediction correction value. The risk prediction correction value includes the error deviation trend, risk weight adjustment value, and model parameter correction coefficient.

8. The AI-based big data risk assessment system according to claim 7, characterized in that, The prediction optimization and adjustment module includes: The error calculation submodule calculates the time deviation between the predicted risk triggering time and the actual risk event occurrence time based on the triggering time value of the mutation factor and the ripple range value of the risk crossover point, and establishes the prediction error time deviation value. The trend analysis submodule calls the prediction error time deviation value to obtain the change data of the error value within a continuous time window, calculates the trend rate of error change, analyzes the direction of error change, and establishes the error change trend value. The parameter adjustment submodule calls the error change trend value and optimizes multiple parameters of the risk factor regression model through iterative adjustment, including the weight coefficient of each risk factor, and establishes a risk prediction correction value.

9. The AI-based big data risk assessment system according to claim 8, characterized in that, The specific formula for optimizing multiple parameters of the risk factor regression model is as follows: ； Calculate the adjusted risk factor weights; in, Represents the adjusted risk factor weights. Represents the weight of the original risk factors. Represents the risk adjustment factor. This represents the rate of change of the error trend. The benchmark rate of change representing the risk factor, This represents the trend of error change.

Images