Market fluctuation early warning method and system based on multi-dimensional index fusion

Abstract

The invention provides a market fluctuation early warning method and system based on multi-dimensional index fusion, and is applied to the field of data processing. The multi-dimensional index data pre-collected by the transaction terminal is classified to generate the index vectors, the fuzzy membership function is combined with the pre-training reasoning model to calculate the membership degree of each risk level, quantitative grading of market fluctuation is achieved, the risk intensity and the maximum risk level can be clearly identified, and the risk level is accurately identified. The reliability of a risk judgment result is ensured through confidence judgment, misjudgment caused by abnormal data or short-term fluctuation is avoided, when the risk reaches a preset threshold value, according to the precondition of the transaction terminal, a transaction action type is generated from the action matrix and dynamically executed, action protection, hedging strategy and limit adjustment are included, and real-time risk control linkage is achieved.

Description

Technical Field

[0001] This invention relates to the field of data processing, and in particular to a market fluctuation early warning method and system based on multi-dimensional indicator fusion. Background Technology

[0002] With the rapid development of the securities market, real-time monitoring and early warning of market fluctuations are playing an increasingly important role in brokerage platforms and investment advisory platforms. Traditional market fluctuation early warning methods usually rely on a few single indicators, such as price volatility, trading volume, or the fear index (VIX). These methods are simple to calculate, but due to the limited dimensions of the indicators, they often fail to comprehensively depict the market's operating status and are prone to inaccurate early warning signals. Summary of the Invention

[0003] This invention aims to address the problem that existing early warning methods typically only output alerts of market volatility risk, lacking quantitative grading of volatility intensity and corresponding risk control linkage mechanisms, making it difficult to provide direct guidance for risk management of securities firms or investment advisory platforms. The invention provides a market volatility early warning method and system based on the fusion of multi-dimensional indicators.

[0004] The present invention employs the following technical means to solve the technical problem:

[0005] This invention provides a market volatility early warning method based on multi-dimensional indicator fusion, comprising:

[0006] Based on the multi-dimensional indicator data pre-collected by the trading terminal, the corresponding indicator types are classified from the multi-dimensional indicator data to generate the indicator vector at the current moment. Specifically, the indicator types include trading indicators, macroeconomic indicators and text sentiment indicators.

[0007] Determine whether the timestamps of the indicator vectors are consistent;

[0008] If so, then construct the fuzzy membership function corresponding to the indicator vector, input the fuzzy membership function into the pre-trained inference model, calculate the membership degree value of each risk level under the indicator vector, perform a weighted combination of the membership degree values ​​of the same risk level to obtain the comprehensive membership degree of the same risk level, compare the comprehensive membership degree, and select the membership degree value of the highest risk level as the transaction risk judgment result at the current moment. The fuzzy membership function is specifically a fuzzy interval corresponding to different risk fluctuations.

[0009] Determine whether the confidence level of the transaction risk assessment result reaches a preset confidence level threshold;

[0010] If the conditions are met, the preconditions preset by the trading terminal are identified. Based on the preconditions, a corresponding action set is constructed from the action matrix preset by the trading terminal. The action type in the trading process is generated through the action set. The approval response of the trading terminal is obtained. Based on the approval response, the action set is dynamically executed in the preset account. The preconditions specifically include trading time period, trading liquidity value and trading permissions. The action type specifically includes protection action, hedging strategy and limit adjustment.

[0011] Furthermore, the step of calculating the membership value of each risk level under the indicator vector also includes:

[0012] Read the indicator values ​​one by one from the indicator vector at the current moment and identify the preset risk range of the indicator values;

[0013] Determine whether the indicator value falls within the intersection range of the two risk levels;

[0014] If so, the index value is marked as a preset boundary state, the first-level membership degree and the second-level membership degree of the index value are retained, and based on the distance of the index value from the center of the cross interval, the first-level membership degree and the second-level membership degree are assigned different weights. The sum of the membership degrees of the first-level membership degree and the second-level membership degree is detected, and the total membership degree is dynamically adjusted according to the sum of the membership degrees.

[0015] Furthermore, before the step of constructing the fuzzy membership function corresponding to the index vector, the method further includes:

[0016] Based on the risk assessment indicators preset by the trading terminal, the dimensional differences of the risk assessment indicators are obtained, wherein the risk assessment indicators specifically include price volatility, trading volume change rate, bid-ask spread and public sentiment index.

[0017] Determine whether the dimensional differences are within a unified numerical range;

[0018] If not, the numerical range of the risk assessment indicator is detected, the numerical span of the risk assessment indicator is identified based on the numerical range, the deviation difference of the risk assessment indicator is calculated based on the numerical span, and the weight ratio of the risk assessment indicator is dynamically adjusted based on the deviation difference.

[0019] Furthermore, the step of constructing a corresponding action set from the preset action matrix of the trading terminal, and generating the action type in the trading process through the action set, further includes:

[0020] Based on the preset triggering conditions of the action set, the real-time status information of the preset account during the transaction process is identified. The triggering conditions specifically include placing an order, canceling an order, modifying an order, and transferring account funds. The real-time status information specifically includes market conditions, account balance status, and order execution status.

[0021] Determine whether the real-time status information meets the triggering condition;

[0022] If so, then according to the risk control rules preset by the transaction terminal, the verification information of the action set is obtained, and the constraint actions of the action set are dynamically removed based on the verification information. Priority actions are divided in the action set, wherein the priority actions specifically include primary actions and secondary actions.

[0023] Furthermore, the step of determining whether the timestamps of the indicator vectors are consistent also includes:

[0024] Based on the generation period of the indicator vector, the acquisition delay of the indicator vector is obtained, wherein the generation period is specifically from the arrival time to the recording time;

[0025] Determine whether the acquisition delay exceeds a preset delay threshold;

[0026] If so, the sampling period of the indicator vector is identified, and the low-frequency data of the indicator vector is dynamically supplemented according to the sampling period. Based on the preset priority of the account, the confidence score of the indicator vector is generated.

[0027] Furthermore, after the step of determining whether the confidence level of the transaction risk assessment result reaches a preset confidence level threshold, the method further includes:

[0028] Based on the risk control measures for the preset account, a continuous time series of the confidence level is obtained, wherein the risk control measures specifically include automatic execution of position reduction and liquidation, manual review, and risk warning;

[0029] Determine whether the continuous time series continues to decline within a preset time period;

[0030] If so, then the trend data of the continuous time series is collected, and the risk control warning for the preset account is dynamically triggered based on the trend data. Based on the risk control warning, the risk control measures are adaptively executed.

[0031] Furthermore, the step of classifying the corresponding indicator types from the multi-dimensional indicator data pre-collected by the trading terminal and generating the indicator vector for the current moment also includes:

[0032] Based on the data quality score preset by the trading terminal, an indicator mapping table for the indicator vector is constructed, wherein the data quality score specifically includes data integrity, stability, and anomaly frequency;

[0033] Determine whether the vector structure of the index mapping table is stable;

[0034] If not, then based on the data quality score, identify the indicator characteristics of the indicator vector, and dynamically update the indicator mapping table according to the indicator characteristics. Specifically, the indicator characteristics include continuous, discrete, and categorical types, and the dynamic update specifically includes adding new indicators and discarding indicators.

[0035] This invention also provides a market fluctuation early warning system based on multi-dimensional indicator fusion, comprising:

[0036] The generation module is used to classify the corresponding indicator types from the multi-dimensional indicator data pre-collected by the trading terminal and generate the indicator vector at the current moment. Specifically, the indicator types include trading indicators, macroeconomic indicators and text sentiment indicators.

[0037] The judgment module is used to determine whether the timestamps of the indicator vectors are consistent.

[0038] The execution module is used to construct the fuzzy membership function corresponding to the indicator vector if the condition is met, input the fuzzy membership function into the pre-trained inference model, calculate the membership degree value of each risk level under the indicator vector, perform a weighted combination of the membership degree values ​​of the same risk level to obtain the comprehensive membership degree of the same risk level, compare the comprehensive membership degree, and select the membership degree value of the highest risk level as the transaction risk judgment result at the current moment. The fuzzy membership function is specifically a fuzzy interval corresponding to different risk fluctuations.

[0039] The second judgment module is used to determine whether the confidence level of the transaction risk assessment result reaches a preset confidence level threshold.

[0040] The second execution module is used to identify the preset preconditions of the trading terminal if the conditions are met, construct a corresponding action set from the preset action matrix of the trading terminal based on the preconditions, generate the action type in the trading process through the action set, obtain the approval response of the trading terminal, and dynamically execute the action set in the preset account based on the approval response. The preconditions specifically include trading time period, trading liquidity value and trading permissions, and the action type specifically includes protection action, hedging strategy and limit adjustment.

[0041] Furthermore, the execution module also includes:

[0042] The identification unit is used to read the indicator values ​​of the indicator vector at the current moment one by one and identify the preset risk range of the indicator values.

[0043] A judgment unit is used to determine whether the indicator value is in the intersection range of two risk levels;

[0044] The execution unit is configured to, if so, mark the index value as a preset boundary state, retain the first-level membership degree and the second-level membership degree of the index value, assign different weights to the first-level membership degree and the second-level membership degree based on the center position of the index value from the cross interval, detect the sum of the membership degrees of the first-level membership degree and the second-level membership degree, and dynamically adjust the total membership degree based on the sum of the membership degrees.

[0045] Furthermore, it also includes:

[0046] The acquisition module is used to acquire the dimensional differences of the risk assessment indicators based on the risk assessment indicators preset by the trading terminal. Specifically, the risk assessment indicators include price volatility, trading volume change rate, bid-ask spread, and public sentiment index.

[0047] The third judgment module is used to determine whether the difference in dimensions is within a uniform numerical range;

[0048] The third execution module is used to detect the numerical range of the risk assessment indicator if no, identify the numerical span of the risk assessment indicator based on the numerical range, calculate the deviation difference of the risk assessment indicator based on the numerical span, and dynamically adjust the weight ratio of the risk assessment indicator based on the deviation difference.

[0049] This invention provides a market fluctuation early warning method and system based on multi-dimensional indicator fusion, which has the following beneficial effects:

[0050] This invention classifies and generates indicator vectors from multi-dimensional indicator data pre-collected by trading terminals. It then uses fuzzy membership functions combined with a pre-trained inference model to calculate the membership degree of each risk level, achieving quantitative grading of market fluctuations. This clearly identifies risk intensity and the maximum risk level, and ensures the reliability of risk assessment results through confidence level determination, avoiding misjudgments caused by abnormal data or short-term fluctuations. When the risk reaches a preset threshold, it generates and dynamically executes trading action types from the action matrix based on the preconditions of the trading terminal, including protective actions, hedging strategies, and limit adjustments, achieving real-time risk control linkage. Compared with existing methods, this solution not only provides risk warnings but also quantifies risk intensity and directly links trading and risk control operations, providing brokerage platforms and investment advisory platforms with highly operable and intelligent risk management capabilities. Attached Figure Description

[0051] Figure 1 This is a flowchart illustrating an embodiment of the market fluctuation early warning method based on multi-dimensional indicator fusion of the present invention.

[0052] Figure 2 This is a structural block diagram of an embodiment of the market fluctuation early warning system based on multi-dimensional indicator fusion of the present invention. Detailed Implementation

[0053] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. The realization of the purpose, functional features, and advantages of the invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings.

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] Reference Appendix Figure 1 The market fluctuation early warning method based on multi-dimensional indicator fusion in one embodiment of the present invention includes:

[0056] S1: Based on the multi-dimensional indicator data pre-collected by the trading terminal, classify the corresponding indicator types from the multi-dimensional indicator data and generate the indicator vector for the current moment. The indicator types specifically include trading indicators, macroeconomic indicators and text sentiment indicators.

[0057] S2: Determine whether the timestamps of the indicator vectors are consistent;

[0058] S3: If so, construct the fuzzy membership function corresponding to the indicator vector, input the fuzzy membership function into the pre-trained inference model, calculate the membership degree value of each risk level under the indicator vector, perform a weighted combination of the membership degree values ​​of the same risk level to obtain the comprehensive membership degree of the same risk level, compare the comprehensive membership degree, and select the membership degree value of the highest risk level as the transaction risk judgment result at the current moment. The fuzzy membership function is specifically a fuzzy interval corresponding to different risk fluctuations.

[0059] S4: Determine whether the confidence level of the transaction risk assessment result reaches the preset confidence level threshold;

[0060] S5: If the condition is met, identify the preconditions preset by the trading terminal, construct a corresponding action set from the action matrix preset by the trading terminal based on the preconditions, generate the action type in the trading process through the action set, obtain the approval response from the trading terminal, and dynamically execute the action set in the preset account based on the approval response. The preconditions specifically include trading time period, trading liquidity value and trading permissions, and the action type specifically includes protection action, hedging strategy and limit adjustment.

[0061] In this embodiment, the system classifies multi-dimensional indicator data pre-collected by the trading terminal into corresponding indicator types. These indicator types specifically include trading indicators, macroeconomic indicators, and sentiment indicators. The system then generates indicator vectors for the current moment. Next, the system determines whether the timestamps of these indicator vectors are consistent to execute corresponding steps. For example, if the system determines that the indicator vectors for the current moment are inconsistent, it assumes that some indicator vectors originate from different acquisition channels or systems, resulting in delays and misalignment of data at the same time point. The system then enters a short-term waiting state until the timestamps of all indicator data are aligned before continuing to generate indicator vectors. Priority is given to important indicators (such as trading prices), while other indicators are not considered. Secondary indicators (such as some public opinion data) can be replaced or downweighted to avoid interrupting the overall calculation due to the absence of individual non-critical indicators. Furthermore, in subsequent confidence calculations, the weight of results at abnormal times is reduced to ensure robustness of the judgment. For example, when the system determines that the indicator vectors at the current moment are consistent, it assumes that the indicator vectors have no delay and that data at the same time point can be aligned. The system then constructs fuzzy membership functions corresponding to these indicator vectors. Specifically, the fuzzy membership functions are fuzzy intervals corresponding to different risk fluctuations. These fuzzy membership functions are input into a pre-trained inference model to calculate the membership values ​​of each risk level under these indicator vectors. The membership values ​​of the same risk level are then weighted and combined to obtain the comprehensive membership score for the same risk level. The system calculates membership degrees and compares these comprehensive membership degrees, selecting the membership degree value with the highest risk level as the transaction risk judgment result at the current moment. By constructing indicator vectors under the premise of consistent timestamps, the system ensures that all multi-dimensional indicator source data are aligned at the same time, avoiding judgment bias caused by delays or missing data. This synchronization guarantees the integrity and real-time nature of the input data, providing a reliable data foundation for subsequent risk analysis, thereby improving the accuracy of risk judgment. At the same time, the system uses fuzzy membership functions to map different risk fluctuations to fuzzy intervals and inputs them into a pre-trained inference model to calculate the membership degree value of each risk level. This transforms the originally difficult-to-quantify risk intensity into a measurable numerical indicator, which is further weighted and combined. By generating a comprehensive membership degree and then comparing and selecting the maximum value, a specific risk level is finally obtained. This process not only achieves quantitative risk classification but also reflects the differences in risk intensity, elevating risk warnings from "risk present / no risk" to "risk strength classification." Furthermore, through comprehensive calculation of the membership degrees of each risk level and selection of the maximum membership degree, the system can output a clear market risk level as the current transaction risk assessment result. Compared to the traditional method that only indicates "risk exists," the assessment result generated by this solution has a clear level and intensity, which can be directly used as the basis for securities firms or investment advisory platforms to trigger risk control measures and adjust investment strategies, thereby achieving automation and intelligence in risk management and improving market response efficiency.The system then determines whether the confidence level of the transaction risk assessment result reaches a pre-set confidence threshold and executes the corresponding steps accordingly. For example, if the system determines that the confidence level of the transaction risk assessment result does not reach the pre-set confidence threshold, the system will consider that the risk identification lacks sufficient evidence. If the result is directly adopted, it may lead to the false triggering or omission of risk control measures. The system will then reacquire the missing or delayed multi-dimensional indicator data, weight and enhance the key indicators, regenerate the indicator vector, and recalculate the risk assessment result. At the same time, it will introduce historical indicator vectors from previous and subsequent moments to smooth short-term fluctuations and improve the overall stability of the assessment. Furthermore, if the confidence level cannot be improved, the result will be downgraded, for example, marked as "low". The system displays a "confidence level risk warning," but this is only provided to the user as a reference signal and does not trigger automatic risk control measures. For example, when the system determines that the confidence level of the transaction risk assessment result has reached a pre-set confidence level threshold, the system will consider the risk identification to be compliant. The system will then identify the pre-set preconditions of the trading terminal, which specifically include the trading period, trading liquidity value, and trading permissions. Based on these preconditions, the system will construct a corresponding action set from the pre-set action matrix of the trading terminal. Different action sets will generate action types in the trading process, specifically including protective actions, hedging strategies, and limit adjustments. The system will obtain the approval response from the trading terminal and, based on the approval response, allocate funds to the pre-set account. These action sets are dynamically executed; the system only enters the subsequent risk control linkage process when the confidence level reaches a preset threshold. This ensures that the risk assessment results have sufficient data support and reliability, avoiding false triggers caused by unstable or abnormal data. By identifying preconditions such as trading time period, trading liquidity value, and trading permissions, the system ensures that the executed risk control measures match the actual trading environment, avoiding operations during inappropriate time periods or beyond authorized limits, thereby improving the compliance and scientific nature of risk control decisions. Simultaneously, based on the preconditions, corresponding action sets are constructed from the action matrix, and different action types are generated through different sets, such as protection actions, hedging strategies, and limit adjustments. This dynamic generation based on condition matching... This approach allows risk control measures to be tailored to different market risk levels and trading environments, avoiding a one-size-fits-all risk response model. This enables more flexible responses to complex and ever-changing market conditions, improving the effectiveness of risk management. Furthermore, after generating an action type, the system also obtains the approval response from the trading terminal and dynamically executes the action set in the preset account, achieving fully automated linkage from risk identification to risk management. Compared to the traditional model relying on manual judgment and intervention, this solution can significantly shorten the response time for risk management, improve the response speed and execution efficiency of brokerage platforms or investment advisory platforms in the face of market fluctuations, thereby reducing potential losses and enhancing overall risk management capabilities.

[0062] It should be noted that multidimensional indicator data specifically refers to heterogeneous data from multiple sources collected and stored by trading terminals at predetermined time intervals to reflect market conditions and trading behavior. This includes: trading indicators that characterize market prices and transaction volume, such as price, trading volume, capital flows, and volatility; macroeconomic indicators that reflect the macroeconomic environment, such as interest rates, exchange rates, inflation rates, and major international stock indices; and textual sentiment indicators extracted through natural language processing technology, such as sentiment bias values ​​in news reports, research reports, or social media sentiment. These multidimensional indicator data collectively constitute a multi-faceted description of the market state, providing a comprehensive data foundation for generating the indicator vector for the current moment.

[0063] It should be added that a fuzzy membership function corresponding to the indicator vector is constructed, and the fuzzy membership function is input into the pre-trained inference model to calculate the membership degree value of each risk level under the indicator vector. The membership degree values ​​of the same risk level are weighted and combined to obtain the comprehensive membership degree of the same risk level. The comprehensive membership degree is then compared, and the membership degree value of the highest risk level is selected as the transaction risk judgment result at the current moment. Specifically:

[0064] 1. Input preparation and standardization: Receive the generated indicator vector for the current moment; confirm that each indicator in the vector has been standardized, normalized or mapped so that all indicators are comparable in the same numerical range (e.g., 0-1); adjust the direction of indicators that need to be reverse mapped (e.g., a positive value of the sentiment indicator indicates optimism but low risk) so that "the larger the value, the higher the risk".

[0065] 2. Membership function library call: Retrieve the predefined fuzzy membership function library from the system. Each type of indicator corresponds to a set of membership functions. The membership functions are divided into three fuzzy intervals: "low risk", "medium risk" and "high risk" according to the risk level. Each interval is represented by a function with continuous transition (such as a triangle or trapezoid).

[0066] 3. Fuzzy single index mapping: Input each standardized index value into its corresponding three membership functions, calculate the membership degree of the index under the three levels of "low / medium / high" (i.e. the degree of membership of the index to each level), and temporarily store each calculation result as the original membership degree set of the index.

[0067] 4. Inference model optimization: Input the above original membership set or original index vector into the pre-trained inference model. The model learns the interaction and nonlinear relationship between indicators based on historical samples, which is used to correct or calibrate the original membership of a single indicator and output the calibrated membership value or membership increment. This step aims to combine the empirical membership function with data-driven results to improve the judgment accuracy.

[0068] 5. Weighting and Weighted Fusion: Assign weights to each indicator (obtained through historical regression, expert experience, or model learning). For the same risk level, sum the (calibrated) membership degrees of all indicators at that level according to their weights to obtain the comprehensive membership degree for that risk level. Normalize the weights if necessary to ensure the sum of the weights is 1.

[0069] 6. Time series smoothing and consistency verification: Perform short-term time series smoothing (e.g., moving average or exponential weighting) on ​​the comprehensive membership degree of each risk level, and verify the consistency with the comprehensive membership degree of the previous time. If the sudden change exceeds the preset fluctuation threshold, trigger an anomaly mark or recalibrate.

[0070] 7. Maximum membership degree determination: Compare the comprehensive membership degree of each risk level, and select the risk level with the largest value as the market risk determination result at the current moment; at the same time, use the maximum comprehensive membership degree as the confidence degree or strength reference value of the determination for subsequent confidence degree determination.

[0071] 8. Output and Recording: Write the determined risk level, the corresponding comprehensive membership value, the key indicators involved in the determination and their weights, and the calibration information into the system log or cache for subsequent risk control linkage, manual review and model iteration.

[0072] Specific examples are as follows:

[0073] Assume the system uses three standardized indicators where "higher values ​​indicate higher risk": price volatility indicator, trading volume change indicator, and negative public opinion indicator; the indicator vector (all mapped to 0-1) is as follows:

[0074] Price fluctuation = 0.70

[0075] Change in trading volume = 0.60

[0076] Negative public opinion (already reflected in reverse) = 0.80

[0077] The membership function library uses a simple triangular / trapezoidal approximation, and the support for the three risk levels in the 0-1 range is as follows:

[0078] Low risk: The support range is approximately [0.00, 0.50], with a membership degree of ~1 at 0.00 and tending to 0 at 0.50 (linear decrease).

[0079] Medium risk: The support range is approximately [0.25, 0.75], with a membership degree of ~1 at 0.50, and a linear decrease to 0 on both sides;

[0080] High risk: The support range is approximately [0.50, 1.00], with a membership degree of ~1 at 1.00 and tending to 0 at 0.50 (linear increase).

[0081] Step A: Calculate the membership degree of each indicator at the three levels.

[0082] Price fluctuation = 0.70

[0083] Low risk: 0.70 > 0.50 → Membership degree = 0.

[0084] Medium risk: 0.70 is located in the right-hand descending zone of medium risk (0.50 to 0.75). The membership degree, calculated using the linear formula, is (0.75-0.70) / (0.75-0.50)=0.05 / 0.25=0.20.

[0085] High risk: 0.70 is in the high-risk rising zone (0.50 to 1.00). The membership degree calculated using the linear formula is (0.70-0.50) / (1.00-0.50)=0.20 / 0.50=0.40.

[0086] Therefore, the membership set of price fluctuations is: Low=0.00, Medium=0.20, High=0.40;

[0087] Change in trading volume = 0.60:

[0088] Low risk: 0.60 > 0.50 → Membership degree = 0.

[0089] Medium risk: 0.60 is to the right of the medium risk zone, membership degree = (0.75-0.60) / 0.25 = 0.15 / 0.25 = 0.60.

[0090] High risk: 0.60 is in the high-risk rising zone, membership degree = (0.60 - 0.50) / 0.50 = 0.10 / 0.50 = 0.20.

[0091] Membership set of changes in trading volume: Low=0.00, Medium=0.60, High=0.20;

[0092] Negative public opinion = 0.80

[0093] Low risk: 0.80 > 0.50 → Membership degree = 0.

[0094] Medium risk: 0.80 > 0.75 → Medium risk membership = 0 (exceeding the right end of medium risk).

[0095] High risk: 0.80 is in the high-risk rising zone, membership degree = (0.80-0.50) / 0.50 = 0.30 / 0.50 = 0.60.

[0096] Membership set of negative public opinion: Low=0.00, Medium=0.00, High=0.60;

[0097] Step B, inference model calibration: Assuming the inference model has a calibrating effect on the interaction between public opinion and price, the original membership degree is multiplied by a learned calibration coefficient: price is multiplied by 1.0 (no change), trading volume is multiplied by 1.0 (no change), and public opinion is multiplied by 1.0 (unchanged in the example).

[0098] Step C involves assigning weights to each indicator and then weighting and merging them. Assume the weights are allocated as follows (weights have been normalized, and the sum is 1): price weight 0.50, trading volume weight 0.30, and public opinion weight 0.20.

[0099] Calculate the overall membership degree for each risk level (weighted summation):

[0100] High overall membership degree = Price_High × Price weight + Volume_High × Volume weight + Public opinion_High × Public opinion weight = 0.40 × 0.50 + 0.20 × 0.30 + 0.60 × 0.20

[0101] Calculate item by item:

[0102] 0.40 × 0.50 = 0.20

[0103] 0.20 × 0.30 = 0.06

[0104] 0.60 × 0.20 = 0.12

[0105] Adding the three terms together: 0.20 + 0.06 + 0.12 = 0.38

[0106] Therefore, the overall membership degree of High is 0.38.

[0107] Medium overall membership degree = Price_Med × Price weight + Volume_Med × Volume weight + Public opinion_Med × Public opinion weight = 0.20 × 0.50 + 0.60 × 0.30 + 0.00 × 0.20

[0108] Calculate item by item:

[0109] 0.20 × 0.50 = 0.10

[0110] 0.60 × 0.30 = 0.18

[0111] 0.00 × 0.20 = 0.00

[0112] Adding the three terms together: 0.10 + 0.18 + 0.00 = 0.28

[0113] Therefore, the overall membership degree of Medium is 0.28.

[0114] Low overall membership degree = all indicators have a Low membership degree of 0, therefore Low = 0.00;

[0115] Step D, time-series smoothing and consistency verification: After obtaining the comprehensive membership degree of each risk level, the system will perform time-series smoothing to avoid excessive impact of abnormal indicators at a single moment or short-term sudden fluctuations on the final risk judgment.

[0116] The system uses a moving average method, which calculates a moving average of the current comprehensive membership degree with the historical comprehensive membership degrees from previous time points. For example: ;

[0117] in, This represents the overall membership degree at the current moment. This represents the smoothed overall membership degree. The length of the sliding window (e.g., 3, 5, or 10).

[0118] Weighted average processing (can be used alternatively or in parallel): When it is necessary to assign greater weight to recent data, the system can use a weighted moving average. ;

[0119] Among them, weight satisfy And it is generally set in a decreasing form (for example, the weight is the largest at the most recent time, and the weight gradually decreases in the past).

[0120] For consistency verification, the system compares the smoothed comprehensive membership degree with the current unsmoothed value. If the difference is less than a preset threshold (e.g., 0.05), the risk assessment is considered stable and reliable. If the difference exceeds the threshold, it indicates short-term fluctuations or data anomalies. In this case, the system will mark the assessment result at that moment as "low stability" and reduce its weight in the final confidence calculation.

[0121] The output of the smoothed result is the comprehensive membership degree after the above smoothing and verification processes. This will serve as the input value for the subsequent "maximum membership determination," ensuring that the output risk level has continuity and stability over time, thereby reducing misjudgments and oversensitive reactions.

[0122] Step E: Maximum membership degree determination and output. Compare the comprehensive membership degrees of the three: High=0.38, Medium=0.28, Low=0.00, and select the level corresponding to the maximum value of 0.38, High. Therefore, the transaction risk determination result at the current moment is "high risk", and the confidence level of this determination can be recorded as 0.38 (or further compare 0.38 with the preset confidence threshold to determine whether to trigger risk control action).

[0123] In summary, in the examples above, the membership function parameters (endpoints of each level interval, function shape) can be initialized based on historical quantiles and expert experience, allowing the system to automatically calibrate. The inference model is used to eliminate the single-indicator membership bias caused by nonlinear coupling between indicators. The weighting coefficients can be obtained from historical regression, principal component analysis, or model learning. When the confidence level is low, multi-model cross-validation or time series smoothing can be triggered to improve stability. The values ​​and intervals in the examples are for demonstration purposes. In actual implementation, the intervals and membership function shapes should be statistically determined based on the historical data of the protected market (e.g., setting triangular / trapezoidal nodes based on the 25%, 50%, and 75% quantiles).

[0124] Identify the preset preconditions of the trading terminal, construct a corresponding action set from the preset action matrix of the trading terminal based on the preconditions, generate the action type in the transaction process through the action set, obtain the approval response of the trading terminal, and dynamically execute the action set in the preset account based on the approval response, specifically as follows:

[0125] Step A: Identify Prerequisites. The system first reads the preset prerequisites from the trading terminal. These prerequisites are used to constrain the triggering scope of risk response actions. The prerequisites mainly include:

[0126] Trading hours: For example, specific risk control actions may be performed only during the opening session (9:30-11:30) or the closing session (13:00-15:00);

[0127] Transaction liquidity value: This is calculated by measuring liquidity indicators such as trading volume and total capital inflow / outflow within a unit of time. For example: L = Average liquidity level of transaction volume ;

[0128] If L>1.5, it indicates high market liquidity, and the system can allow for larger-scale protection or hedging actions.

[0129] Trading permissions: Actions that can be performed are restricted based on account level or compliance requirements. For example, ordinary accounts can only trigger limit adjustments, while institutional accounts can execute hedging strategies.

[0130] Step B involves constructing an action set. Based on the aforementioned prerequisites, the system filters available actions from the action matrix preset by the trading terminal. The action matrix can be understood as a two-dimensional table, with the horizontal axis representing "risk level" and the vertical axis representing "prerequisites." The table content lists "allowed action types." For example, when the risk level is "high," the trading session is "open," and the account permissions are "institutional," the action set might include:

[0131] Protective action: Automatically reduce the single transaction limit by 20%;

[0132] Hedging strategy: Buy short positions in stock index futures as a risk hedge;

[0133] Limit adjustment: Dynamically lower the upper limit of position concentration;

[0134] Step C: Generate Action Types. The system categorizes and combines the various actions in the action set to generate the action types for the transaction process.

[0135] Protective actions include triggering stop-loss orders and freezing trading permissions for certain high-risk assets.

[0136] Hedging strategies: such as calculating the hedging ratio: ;

[0137] If H=0.3, then when the spot market value is 10 million, the system will execute a short futures order of 3 million.

[0138] Limit adjustment: For example, the maximum daily order amount will be reduced from 5 million to 3 million to dynamically reduce risk exposure;

[0139] Step D, obtaining the approval response: Before execution, the system will submit the action set to the trading terminal for approval. If the approval is approved, the system will record the approval number and enter the dynamic execution stage. If the approval is not approved (e.g., insufficient account permissions or rejection by compliance and risk control), the system will revert to the risk assessment stage or trigger a milder alternative action (e.g., only prompting the risk without placing an order).

[0140] Step E: Dynamic execution. The system executes specific actions in the preset account based on the approval response and records the execution results; for example:

[0141] Actual action taken: 30% of the available margin in the account was frozen;

[0142] Order execution: Place a short futures order according to the calculated hedging ratio;

[0143] Limit adjustment: Update the trading system configuration parameters to limit the maximum amount a user can trade in the next step;

[0144] Specific examples are as follows:

[0145] Assume the system determines the risk level to be "high" at 10:00, with a confidence level of 0.87 (exceeding the threshold of 0.8).

[0146] Prerequisites: The current trading session is open, the trading liquidity value is 1.7 (high liquidity), and the account permission is "Institutional".

[0147] Action matrix filtering results: Protective actions (stop loss + reduce quota), hedging strategy (30% short futures position), quota adjustment (reduce concentration);

[0148] The system generates a set of actions, submits it for approval, and the approval is granted.

[0149] Final execution results: The account's position concentration was adjusted from 40% to 30%, and a short position of 3 million was purchased in the futures market as a hedge, while the further purchase permission for some risky assets was frozen.

[0150] In this embodiment, step S3, which calculates the membership value of each risk level under the indicator vector, further includes:

[0151] S31: Read the indicator values ​​of the indicator vector at the current moment one by one and identify the preset risk range of the indicator values;

[0152] S32: Determine whether the value of the indicator is in the intersection range of the two risk levels;

[0153] S33: If so, mark the index value as a preset boundary state, retain the first-level membership degree and the second-level membership degree of the index value, assign different weights to the first-level membership degree and the second-level membership degree based on the distance of the index value from the center of the cross interval, detect the sum of the membership degrees of the first-level membership degree and the second-level membership degree, and dynamically adjust the total membership degree according to the sum of the membership degrees.

[0154] In this embodiment, the system reads the indicator values ​​of the indicator vector at the current moment one by one, identifies the pre-defined risk intervals of the indicator values, and then determines whether these indicator values ​​are in the intersection interval of two risk levels in order to execute the corresponding steps. For example, when the system determines that all indicator values ​​are not in the intersection interval of two risk levels, the system will consider that the current indicator value can completely match a specific risk interval, that is, there is no fuzzy boundary or uncertainty in the level, and the determination result is clear and unambiguous. The system will set the risk level membership degree of each indicator to "fully belong to" the corresponding level, and the membership degree value can be recorded as 1, while the membership degree of other levels is set to 0 to ensure the certainty of the determination process. This approach prioritizes both accuracy and simplicity, improving computational efficiency and avoiding unnecessary complexity arising from multi-level membership. The system uses the deterministic membership results of individual indicators as input for subsequent weighted combinations of comprehensive membership degrees, resulting in a more stable and direct final comprehensive judgment. Furthermore, to ensure the rigor of the entire process, the system records a flag indicating that "the indicator value has not entered the crossover interval" at this step. This flag is then appended to the subsequent risk result output as a transparent explanation of the judgment basis. For example, if the system determines that an indicator value is in the crossover interval of two risk levels, it considers the current state of the indicator to have uncertainty in risk level, meaning that the value simultaneously satisfies two adjacent risk zones. If certain conditions cannot be fully covered by a single level, the system will mark the indicator value as a pre-defined boundary state, retaining the first-level and second-level membership degrees of the indicator value. Based on the indicator value's distance from the center of the crossover interval, different weights are assigned to the first-level and second-level membership degrees. The sum of the first-level and second-level membership degrees is detected, and the total membership degree is dynamically adjusted based on this sum. By retaining the membership degrees of two adjacent risk levels for indicator values ​​within the crossover interval and assigning different weights based on the indicator value's position within the crossover interval, the system can accurately reflect the critical state of the indicator. This processing method overcomes the limitations of traditional hard threshold judgment, enabling... Risk assessment is no longer a simple "either / or" process; it presents the gradual and transitional nature of risk, thereby improving the scientific rigor and precision of the assessment results. Furthermore, when dynamically correcting the total membership degree, the system detects and adjusts the sum of the membership degrees of the first and second levels, making the overall membership degree more reasonable and stable. This method avoids excessive deviation of the final risk assessment by a single indicator under boundary conditions, while maintaining the consistency and controllability of the overall risk calculation. It enhances the robustness of the multi-indicator integrated risk assessment model and provides more accurate risk level information for securities firms or investment advisory platforms, especially at market volatility thresholds or indicator fluctuation edges, enabling the system to identify potential risks in a timely manner.By weighting and dynamically correcting membership, the system can not only output more reliable risk levels, but also provide a scientific basis for triggering subsequent risk control actions, thereby improving the accuracy and effectiveness of risk control linkage and reducing potential losses caused by misjudgment or omission.

[0155] It should be noted that the indicator value is marked as a preset boundary state, the first-level membership degree and the second-level membership degree of the indicator value are retained, and based on the distance of the indicator value from the center of the cross interval, different weights are assigned to the first-level membership degree and the second-level membership degree. The sum of the membership degrees of the first-level membership degree and the second-level membership degree is detected, and the total membership degree is dynamically adjusted according to the sum of the membership degrees. Specifically:

[0156] The boundary state is marked when the system determines that an indicator value is in the intersection of two adjacent risk levels. This mark is used to distinguish ordinary indicator values ​​and remind subsequent processing that dual membership calculation and dynamic weight allocation are required.

[0157] The system retains both membership degrees, meaning it retains the first-level membership degree for this boundary index value. Second-level membership Instead of classifying them into a single level, this step ensures that the critical information of the indicators is fully preserved, providing basic data for subsequent comprehensive membership degree correction.

[0158] Based on the weight allocation according to the center position, the system calculates the distance δ=|XC| between the index value and the center position C of the intersection interval, and assigns different weights to the membership degrees of the first and second levels according to the distance. and The following logic is generally followed: the closer the indicator value is to the center of a certain level, the higher the weight of the corresponding level; the sum of the two weights can be kept at 1 for subsequent weighted fusion; that is: ;

[0159] Calculate the weighted membership degree by weighting the membership degrees of the first and second levels: ;

[0160] The system calculates the membership degree sum. Check whether the preset conditions are met (e.g., not exceeding 1, or satisfying the smoothness constraint) to ensure that no anomalies are introduced in the subsequent comprehensive membership calculation;

[0161] Dynamically adjust the total membership degree, based on Adjusting the overall membership degree ensures that the boundary indicator values ​​reflect the critical state in the final risk level determination while avoiding unreasonable deviations from the overall results; for example, the total membership degree can be proportionally allocated to the first and second levels or smoothed out.

[0162] Specific examples are as follows:

[0163] Assuming the index value X=0.68 falls within the intersection of medium risk (0.5-0.75) and high risk (0.7-1.0), the boundary state is marked, and the system records X=0.68 as a "boundary state," requiring dual membership processing.

[0164] Original membership degree:

[0165] Medium-risk membership =0.6,

[0166] High-risk membership =0.4;

[0167] Calculate distance and weight:

[0168] Medium risk center =0.625 (midpoint)

[0169] High-risk center =0.85 (midpoint);

[0170] distance:

[0171] |0.68-0.625|=0.055, =|0.68-0.85|=0.17;

[0172] Half-interval length:

[0173] Medium risk: 0.75 - 0.5 = 0.25

[0174] High risk: 1.0 - 0.7 = 0.3;

[0175] Weight:

[0176] =1-0.055 / 0.25≈0.78, =1-0.17 / 0.3≈0.43;

[0177] Calculate the weighted membership degree.

[0178] =0.6 × 0.78 ≈ 0.468, =0.4×0.43≈0.172;

[0179] Detect membership degree and sum.

[0180] =0.468+0.172=0.64<1, which meets the smoothing requirement and no anomaly correction is needed;

[0181] Dynamically adjust the total membership degree.

[0182] Will =0.468 and =0.172 is included in the final comprehensive membership degree calculation. After weighted combination, the comprehensive risk level determination can reasonably reflect the state of medium risk as the main risk and high risk as the secondary risk, while retaining boundary information.

[0183] In summary, the examples provided above can scientifically reflect the critical state of indicators, demonstrate the gradual nature of risk changes, avoid misjudgments or biases caused by rigid classifications, enhance the robustness and interpretability of the model when making comprehensive judgments based on multiple indicators, and retain boundary information for risk control decision-making.

[0184] In this embodiment, before step S3 of constructing the fuzzy membership function corresponding to the index vector, the method further includes:

[0185] S301: Based on the risk assessment indicators preset by the trading terminal, obtain the dimensional differences of the risk assessment indicators, wherein the risk assessment indicators specifically include price volatility, trading volume change rate, bid-ask spread and public sentiment index.

[0186] S302: Determine whether the dimensional differences are within a unified numerical range;

[0187] S303: If not, then detect the numerical range of the risk assessment indicator, identify the numerical span of the risk assessment indicator based on the numerical range, calculate the deviation difference of the risk assessment indicator based on the numerical span, and dynamically adjust the weight ratio of the risk assessment indicator based on the deviation difference.

[0188] In this embodiment, the system uses pre-set risk assessment indicators on the trading terminal, specifically including price volatility, volume change rate, bid-ask spread, and public sentiment index. It then obtains the dimensional differences among these risk assessment indicators and determines whether these dimensional differences unify the numerical range to execute corresponding steps. For example, when the system determines that the dimensional differences among these risk assessment indicators can unify the numerical range, different types of indicators, after normalization or standardization, can be mapped to the same numerical scale, for example, all compressed into the interval [0,1][0,1][0,1]. If the risk assessment indicators fall within the range of a standard normal distribution, the system will directly adopt the unified numerical range and use these risk assessment indicators as components of a vector of the same dimension to construct the indicator vector at the current moment. Simultaneously, without additional weight adjustments, it will directly use these indicator vectors to generate corresponding fuzzy membership functions and calculate the membership degree for each risk level. For example, if the system determines that the dimensional differences of a certain risk assessment indicator cannot be unified within a single numerical range, it will assume that different types of indicators cannot be mapped to the same numerical scale. The system will then detect the numerical range of these risk assessment indicators, identify the numerical span of these indicators based on different numerical ranges, and calculate the deviation difference degree of the risk assessment indicators based on this numerical span. Through this deviation difference degree, the system will dynamically adjust the weight ratio of the risk assessment indicators. When the dimensional differences of some risk assessment indicators cannot be unified within a single numerical range, by detecting the numerical range of the indicators and calculating their span, the system can quantify the differences in scale among the indicators and, based on the deviation difference degree... Adjusting weights can prevent a single indicator from gaining an unreasonable advantage in comprehensive calculations due to excessively large dimensions or wide numerical ranges. This ensures a fairer and more objective fusion result of multi-dimensional indicators in risk assessment. Furthermore, by introducing deviation differences and incorporating philosophical principles for dynamic weight adjustment, the system can maintain balance among multi-dimensional indicators. This allows indicators from different sources and with different dimensions to contribute their respective risk information without being excessively amplified or weakened by certain indicators. This method effectively improves the accuracy and stability of risk assessment results, preventing distortion of the overall risk assessment due to extreme fluctuations in a single type of indicator. It also allows the system to automatically re-detect differences in indicator dimensions and dynamically adjust weight ratios when facing newly introduced indicators or changes in the market environment, thus possessing adaptive capabilities. This not only enhances the adaptability of the risk assessment model to complex market environments but also provides compatibility space for the subsequent introduction of more heterogeneous data (such as news sentiment and international market indicators), making the risk management system more scalable and valuable for long-term application.

[0189] It should be noted that the process involves detecting the numerical range of the risk assessment indicator, identifying the numerical span of the risk assessment indicator based on the numerical range, calculating the deviation difference of the risk assessment indicator based on the numerical span, and dynamically adjusting the weight ratio of the risk assessment indicator based on the deviation difference. Specifically:

[0190] When the dimensions / numerical span of certain indicators cannot be directly unified, first measure the typical numerical range (span) of each indicator within the historical window, convert the span into "deviation difference degree", then use a suppression function to map the deviation difference degree into a weight decay factor, and finally use this factor to adjust the basic weight of each indicator and normalize it to ensure that indicators with large spans will not unreasonably amplify their influence in the fusion process.

[0191] The detailed steps are as follows:

[0192] Step 1: Determine the historical window and boundary method.

[0193] Select a historical sample window (e.g., the most recent 90 trading days or the most recent N time windows) to estimate the typical range of values ​​for the indicator; to avoid the influence of extreme values, prioritize using quantile boundaries (e.g., p5 and p95) or IQR (25th and 75th percentiles) to define the upper and lower bounds; if the data is stable, historical minimum / maximum values ​​can also be used; the purpose is to replace extreme maximum and minimum values ​​with robust boundaries to obtain a reasonable "working range";

[0194] Step 2: Calculate the numerical span of each indicator. For each indicator i, calculate the upper bound upper_i (e.g., p95) and the lower bound lower_i (e.g., p5), and calculate the span, span_i = upper_i - lower_i (if using IQR, then span_i = Q3 - Q1). The purpose is to obtain the typical numerical "width" of each indicator within the historical window.

[0195] Step 3: Calculate the overall reference value of the span. Calculate the average span of all indicators, span_mean=(Σspan_i) / M (M is the number of indicators), or the median span_median can be used instead to reduce the influence of extreme values. The purpose is to obtain a scale reference to measure whether the span of a certain indicator is abnormally large / small.

[0196] Step 4—Calculate the relative span (the original measure of deviation from the mean). Calculate the relative span r_i = span_i / span_mean (or r_i = span_i / span_median). Define the "deviation from the mean" PD_i = r_i (or PD_i = max(r_i-1,0) to focus only on the portion that exceeds the mean). The purpose is to represent the degree of deviation of each indicator from the overall scale using a dimensionless relative value.

[0197] Step 5: Map the deviation difference degree to an attenuation factor. Select an attenuation mapping function f(PD), such as a commonly used robust function, f_i = 1 / (1 + PD_i) (simple and monotonically decreasing) or f_i = 1 / (1 + a×PD_i) (a is a smoothing coefficient). To prevent the weight of a certain indicator from being pressed close to 0, a lower limit f_min can be set (e.g., 0.05); if f_i < f_min, then f_i = f_min. The purpose is that the larger the span, the smaller f, thereby reducing the direct impact of this indicator on the final determination, and retaining the threshold to prevent complete loss of information.

[0198] Step 6: Calculate the adjusted (pre-normalized) weight. Set the base weight base_w_i (which can come from experts, historical regression, or prior allocation), satisfying Σbase_w_i = 1, and calculate the pre-normalized weight: w'_i = base_w_i × f_i. The purpose is to directly apply the attenuation factor to the existing confidence level.

[0199] Step 7: Normalize to obtain the final weight. Calculate the weight sum S = Σ_i w'_i, and the final weight w_i = w'_i / S. The purpose is to ensure that the adjusted weights still sum to 1 and can be directly used for comprehensive membership degree weighting.

[0200] Step 8: Smoothing and historical fusion (optional). To avoid sudden changes in weights, use exponential smoothing or weighted average: final_w_i = α × previous_w_i + (1 - α) × w_i (α is, for example, 0.8); update previous_w_i regularly. The purpose is for the weights to gradually adapt to the market and avoid violent fluctuations in weights caused by short-term noise.

[0201] Step 9: Record and audit. Store all of span_i, PD_i, f_i, base_w_i, w'_i, w_i, etc. in the log for easy backtracking and model iteration.

[0202] The specific example is as follows:

[0203] Ⅰ. Assume 4 indicators: price volatility (A), trading volume change rate (B), order imbalance (C), and public opinion sentiment index (D); the historical boundaries estimated by p5 / p95 are as follows (in their respective original units):

[0204] The upper bound of A is 0.15, and the lower bound is 0.01 → span_A = 0.15 - 0.01 = 0.14,

[0205] The upper bound of B is 50000, and the lower bound is 1000 → span_B = 50000 - 1000 = 49000,

[0206] The upper bound of C is 200, and the lower bound is -200 → span_C = 200 - (-200) = 400,

[0207] Upper bound of D is 0.90, lower bound is -0.80 → span_D = 0.90 - (-0.80) = 1.70;

[0208] II, (item by item)

[0209] span_A = 0.15 - 0.01 = 0.14

[0210] span_B = 50000 - 1000 = 49000

[0211] span_C = 200 - (-200) = 400,

[0212] span_D = 0.90 - (-0.80) = 1.70;

[0213] III. (Calculate the average span)

[0214] First, sum the results: 0.14 + 49000 + 400 + 1.70 = 49401.84.

[0215] The average value is span_mean = 49401.84 / 4 = 12350.46;

[0216] Ⅳ, (relative span r_i),

[0217] r_A=span_A / span_mean=0.14 / 12350.46≈0.0000113397,

[0218] r_B = 49000 / 12350.46 ≈ 3.9675 (Note: 12350.46 × 3.9675 ≈ 49000).

[0219] r_C=400 / 12350.46≈0.0324028,

[0220] r_D=1.70 / 12350.46≈0.0001376,

[0221] Treat r_i as the deviation difference PD_i (i.e., PD_i = r_i);

[0222] V, (mapped to the attenuation factor f_i=1 / (1+PD_i)),

[0223] Calculate item by item:

[0224] f_A=1 / (1+0.0000113397)≈1 / 1.0000113397≈0.9999886603,

[0225] f_B=1 / (1+3.9675)=1 / 4.9675≈0.201311,

[0226] f_C=1 / (1+0.0324028)≈1 / 1.0324028≈0.968611,

[0227] f_D=1 / (1+0.0001376)≈0.9998624,

[0228] (If f_min=0.05 is set, then all values ​​will be greater than f_min, and no lower limit processing is needed.)

[0229] VI. (Applying base weights base_w), assuming the base weights are allocated as follows: base_w_A=0.40, base_w_B=0.30, base_w_C=0.20, base_w_D=0.10, calculate the pre-normalized weight w'_i=base_w_i×f_i:

[0230] w'_A=0.40×0.9999886603=0.39999546412→Rounded to 0.399995464,

[0231] w'_B=0.30×0.201311=0.0603933,

[0232] w'_C=0.20×0.968611=0.1937222,

[0233] w'_D=0.10×0.9998624=0.09998624;

[0234] VII. (Normalization), calculate the sum of pre-normalized weights S:

[0235] S=0.39999546412+0.0603933+0.1937222+0.09998624,

[0236] First, add the two terms: 0.39999546412 + 0.0603933 = 0.46038876412.

[0237] Add the third term: 0.46038876412 + 0.1937222 = 0.65411096412.

[0238] Add the fourth term: 0.65411096412 + 0.09998624 = 0.75409720412 (recorded as 0.754097204).

[0239] The final weight w_i = w'_i / S, calculated item by item:

[0240] w_A = 0.39999546412 / 0.75409720412 ≈ 0.53034 (approximately 53.03%).

[0241] Calculation process: If 0.75409720412 × 0.53034 ≈ 0.399995 (verification passed).

[0242] w_B=0.0603933 / 0.75409720412≈0.08007 (approximately 8.01%)

[0243] w_C=0.1937222 / 0.75409720412≈0.25674 (approximately 25.67%)

[0244] w_D=0.09998624 / 0.75409720412≈0.13285 (approximately 13.29%).

[0245] Verification sum: 0.53034 + 0.08007 + 0.25674 + 0.13285 ≈ 1.0000 (satisfies normalization).

[0246] Explanation of example results:

[0247] Because the historical span of indicator B (volume change rate) is very large (49000), its relative span r_B≫1, and the decay factor f_B≈0.201, its original basic weight of 0.30 is compressed to a pre-normalized weight of 0.0604. After normalization, the final weight of B is about 8.0%. On the other hand, the span between indicators A and D is very small, and the decay factor is close to 1. The final weight is slightly higher (A from 40%→53%). This is because the proportion space is allocated after other indicators are suppressed. This avoids the "hegemonic" effect on the fusion result when the volume value span is huge.

[0248] In summary, through the above steps, the system automatically suppresses those indicators that may dominate the fusion result due to excessively large dimensions or spans by following the link of historical span → relative deviation → attenuation mapping → weight redistribution, while ensuring that no information is lost (by setting a lower limit).

[0249] In this embodiment, step S5, which involves constructing a corresponding action set from the preset action matrix of the trading terminal and generating the action type in the trading process using the action set, further includes:

[0250] S51: Based on the preset triggering conditions of the action set, identify the real-time status information of the preset account during the transaction process, wherein the triggering conditions specifically include placing an order, canceling an order, modifying an order, and transferring account funds, and the real-time status information specifically includes market status, account balance status, and order execution status;

[0251] S52: Determine whether the real-time status information meets the triggering condition;

[0252] S53: If so, then according to the risk control rules preset by the transaction terminal, obtain the verification information of the action set, dynamically remove the constrained actions of the action set based on the verification information, and divide the action set into priority actions, wherein the priority actions specifically include primary actions and secondary actions.

[0253] In this embodiment, the system identifies pre-defined trigger conditions based on a set of actions, specifically including order placement, order cancellation, order modification, and account fund transfer. These real-time status information includes market conditions, account balance, and order execution status. The system then determines whether this real-time status information meets the pre-defined trigger conditions and executes the corresponding steps. For example, if the system determines that the account's real-time status information during the transaction does not meet the pre-defined trigger conditions, the system considers that the account's actual operating environment does not match the execution prerequisites required by the action set. Directly executing the action in this case could lead to transaction risks. If execution fails, the system will avoid triggering operations such as order placement, cancellation, modification, or fund transfer that are inconsistent with the account status, thus mitigating the risk of erroneous operations from the source. Furthermore, if the account balance is insufficient, the order execution status is abnormal, or market conditions are volatile, the corresponding triggering conditions and reasons for non-compliance should be recorded in detail in the system log. The system must also provide feedback on the non-compliance conditions and potential impact to the trading terminal user or risk control through the interface or message notification, prompting manual intervention or strategy adjustment. For example, when the system determines that the real-time status information of the account during the trading process meets the pre-set triggering conditions of the action set, the system will recognize that the actual operating environment of the current account matches the execution prerequisites required by the action set, and the system will proceed according to the pre-set conditions of the trading terminal. The system first establishes risk control rules, then obtains verification information for a set of actions. Based on different verification information, it dynamically removes constrained actions from the action set and classifies actions into priority actions, specifically including primary and secondary actions. The system further filters the action set by obtaining verification information after trigger conditions are met, dynamically removing constrained actions that do not comply with risk control rules or may cause risks. This mechanism avoids the drawback of "blindly executing actions as soon as conditions are met," making the execution process of the action set more consistent with the actual market environment and the safety boundaries of the account status, thereby significantly reducing the probability of misoperation or risk escalation. Simultaneously, the system classifies actions into priority actions based on verification information to ensure that primary actions can... Prioritized actions are executed first, while secondary actions are processed only when resources allow or conditions are stable. This priority management strategy enables the system to rationally schedule the execution order in complex trading scenarios, avoiding resource waste or operational conflicts, thereby improving overall execution efficiency and response speed, enhancing system flexibility, and through a dynamic mechanism of "verification-removal-priority allocation," the system can clearly reflect the logical source of each action selection, i.e., why some actions are retained or removed, and why some actions are set as primary actions. This not only improves the interpretability of the risk control process, but also enables the system to quickly adjust the action set according to different market environments or risk control rules, enhancing its adaptability and scalability to complex and ever-changing trading environments.

[0254] It should be noted that, based on the risk control rules preset by the transaction terminal, the verification information of the action set is obtained. Based on the verification information, constrained actions in the action set are dynamically removed, and priority actions are assigned within the action set. Specifically:

[0255] Step 1: Read candidate actions and rules. The system reads the description of each candidate action in the action set (e.g., order placement / cancellation / modification / fund transfer, target quantity, target price, priority metadata, etc.). The system also reads the applicable risk control rule set. Risk control rules are divided into:

[0256] Hard constraints (binary requirements must be met): account permissions, minimum margin requirements, legal compliance restrictions, no trading at night, etc.

[0257] Soft constraints (scoring / threshold assessment): minimum liquidity requirements, maximum single transaction market share, slippage tolerance, cost cap, etc.

[0258] Step 2: Build a checklist for each action. For each action, automatically build checklist items, including specific check elements for hard and soft items (e.g., for the buy action, check the account's buy permission and whether the available margin is greater than or equal to the estimated usage; for the sell action, check the remaining sell limit for the day; for the hedging operation, check the availability of the target futures contract and the margin). Each checklist item is assigned a "type" (hard / soft) and necessary parameters (threshold, calculation formula).

[0259] Step 3: Perform hard constraint judgment (preliminary elimination). For each action, check all hard constraints in turn. If any hard constraint is determined to be unmet, mark the action as "forced elimination" and record the reason for elimination (e.g., insufficient permissions, insufficient margin, compliance prohibition). The eliminated action is immediately removed from the candidate set and will no longer enter the soft scoring. The hard elimination results are logged for auditing purposes.

[0260] Step 4: Calculate the soft validation scores (item-by-item scoring). For the actions that are still retained, calculate the corresponding validation score for each soft item (normalized to 0-1, with higher scores indicating better compliance). Common items include:

[0261] Compliance score (e.g., rule-based risk coefficient mapping).

[0262] Liquidity score (based on estimated trading volume / order size vs. market depth).

[0263] Execution risk score (based on real-time volatility and expected slippage).

[0264] Cost / Expense Score (Mapping of Estimated Fees / Impact Costs to Allowable Cap)

[0265] Each score is calculated according to a preset mapping or model, ensuring that the input and output of each step are supported by a clear calculation formula or mapping table (e.g., "liquidity score = min(1, market_liquidity / required_liquidity)").

[0266] Step 5: Calculate the validation score for the synthesized action. Assign weights to soft components for each action (e.g., w_liq, w_exec, w_cost, etc., with a sum of weights of 1), and calculate the validation score: validation = Σ(w_k × score_k). If validation < preset soft threshold (e.g., 0.4), it is considered "soft failure" and the action is marked as "constraint removal" (or moved to the candidate / manual review queue). If validation ≥ threshold, it is retained as an executable action.

[0267] Step 6: Generate priority scores and prioritize actions. Calculate the priority score for all retained actions. The priority score can be obtained by a linear combination of multiple factors:

[0268] business_importance (business importance, default value, such as high protection action).

[0269] validation (the validation score that was just calculated)

[0270] Urgency (based on a risk level × confidence level mapping).

[0271] Weighting example: priority = α × business_importance + β × validation + γ × urgency (α + β + γ = 1). Normalize and sort the priority, set a primary action threshold (e.g., priority ≥ 0.7 is a primary action), mark those that meet the threshold as "primary actions", and the rest as "secondary actions"; if there are not enough to divide, the first N can be taken as primary actions;

[0272] Step 7: Record, notify, and rollback strategies. Write all removal actions, retention actions, scores for each item, final priority, and removal reasons into the audit log. For removal or downgrade actions, trigger notifications to the trading end or risk control, suggesting feasible alternative actions or manual review. If key actions are removed and the system detects a continuous increase in risk, alternative strategies can be triggered (e.g., changing to gradual reduction of positions instead of one-time settlement), or the manual approval process can be initiated.

[0273] Step 8, Dynamic review and parameter adaptation: The system can continuously review and verify parameters (such as liquidity and slippage) before and after execution and re-score them within a certain window to decide whether to retry or execute in batches. All weights and thresholds can be optimized regularly through historical backtesting and online learning mechanisms (and written into the change record).

[0274] Specific examples are as follows:

[0275] Basic scenario and input; the candidate action set contains 4 actions:

[0276] Action A: Protective Action—Freeze new buy orders and cancel pending buy orders (without order amount); High business importance.

[0277] Action B: Hedging (Buying a short futures position) — Intending to buy a short futures position equivalent to 3 million.

[0278] Action C: Reduce position (market sell) — Intend to sell 5 million units of spot goods at market price.

[0279] Action D: Limit Adjustment - Reduce the single order limit from 5 million to 3 million (system parameter adjustment, no transaction).

[0280] Real-time account / market information (for calculation):

[0281] Account margin available: 1 million (meaning it can be used for futures or margin needs).

[0282] Account permissions: Allows placing orders, canceling orders, modifying orders, and transferring funds.

[0283] Market liquidity index: 0.6 (0-1, the higher the value, the more liquid the market).

[0284] Estimate the required liquidity (for reference in a specific action):

[0285] The liquidity required to hedge B is equivalent to 3 million.

[0286] Reducing holdings in C requires 5 million in liquidity.

[0287] Risk level: High (urgency factor = 1.0), confidence level 0.85;

[0288] Risk control rules (example):

[0289] Mandatory: Opening a futures position requires a margin ≥ the required margin for the futures contract (if insufficient, it is prohibited); market sales cannot exceed the maximum single market share (if exceeded, the volume must be reduced); the account must have the corresponding permissions.

[0290] Soft component weights: w_liq=0.4, w_exec=0.4, w_cost=0.2; Soft threshold=0.45;

[0291] Business importance assumptions (0-1): A=1.0, B=0.9, C=0.8, D=0.4, priority composition weights α=0.5 (business), β=0.3 (validation), γ=0.2 (urgency).

[0292] Hard constraint check (action by action).

[0293] A (Protection): No hard constraints (only system parameter changes, permission allowed) → Pass.

[0294] B (Hedging): The required margin for futures is assumed to be 2.5 million (example), but the account margin is only 1 million → this requirement is not met (insufficient margin) → B is forcibly removed, with the reason recorded as "insufficient margin".

[0295] C (Reducing Position): Selling does not require additional margin, but the market share of a single transaction needs to be checked: If 5 million is too large relative to the average daily trading volume of the market, it needs to be split (this is a soft assessment) → passing the hard assessment.

[0296] D (Quota Adjustment): Permission granted, compliance permitted → Approved through mandatory process;

[0297] Soft sub-item scoring (for A, C, D), example of scoring rule definition (the closer to 1, the more feasible), liquidity score (liquidity_score) = min(1, market_liquidity×scale / required_liquidity_ratio). For simplicity, this example uses an approximation: if required / liquidity_index ratio ≤ 1 → 1, otherwise < 1; simplified estimation:

[0298] A: No market transactions involved → liquidity_score=1.0

[0299] C: Requires 5 million, market liquidity 0.6 → Assuming short-term available funds in the market are approximately = liquidity_index × average daily trading volume (if the average daily trading volume is 8 million, then approximately 4.8 million is available), 5 million / 4.8 million ≈ 1.04 → liquidity_score ≈ 0.96.

[0300] D: System parameter adjustment, independent of liquidity → 1.0;

[0301] The execution risk score is based on the current volatility and the estimated slippage, assuming:

[0302] A: 0.95 (low-risk action)

[0303] C: 0.70 (Market price sell, high risk of slippage).

[0304] D: 0.98 (Internal parameter adjustment, virtually no execution risk);

[0305] The cost score is based on the upper limit of the estimated cost ratio that is acceptable.

[0306] A: 1.0 (no transaction costs)

[0307] C: 0.85 (high slippage cost)

[0308] D: 1.0;

[0309] Calculate the soft validation score for each action:

[0310] validation=0.4liquidity+0.4execution+0.2*cost,

[0311] A: validation_A=0.41.0+0.40.95+0.2*1.0=0.4+0.38+0.2=0.98,

[0312] C: validation_C=0.40.96+0.40.70+0.2*0.85=0.384+0.28+0.17=0.834,

[0313] D: validation_D=0.41.0+0.40.98+0.2*1.0=0.4+0.392+0.2=0.992,

[0314] (Note: B has been forcibly eliminated);

[0315] Prioritization scoring and ranking,

[0316] A:business=1.0,validation=0.98→priority_A=0.5×1.0+0.3×0.98+0.2×1.0=0.5+0.294+0.2=0.994,

[0317] C:business=0.8,validation=0.834→priority_C=0.5×0.8+0.3×0.834+0.2×1.0=0.4+0.2502+0.2=0.8502,

[0318] D:business=0.4,validation=0.992→priority_D=0.5×0.4+0.3×0.992+0.2×1.0=0.2+0.2976+0.2=0.6976;

[0319] Setting the active action threshold to 0.8 → Segmentation results:

[0320] Primary actions: A (0.994), C (0.8502).

[0321] Secondary action: D (0.6976).

[0322] Removed: B (insufficient margin).

[0323] The system will proceed in the following order: "Execute A (protective action) first → Execute C (phased reduction of positions) → Execute D (limit adjustment) in minor cases." If the situation changes during the execution process (e.g., a sudden drop in liquidity), the system will start the review from the "hard constraint check" again.

[0324] Records and notifications: The system records that B is removed and the reason is recorded; the validation and priority values ​​of A, C, and D, the breakdown of each scoring item, etc., are automatically sent to the risk control and trading end and await / record approval (if needed);

[0325] In summary, in the examples above, the system, through a series of clear verification, scoring, elimination, and priority division steps, can automatically select the most feasible and valuable risk control actions while ensuring compliance and security. This avoids failures or increased losses due to blind execution or insufficient permissions / margin / liquidity. At the same time, the quantitative priority scoring enables interpretable sorting of actions, facilitating both automated execution and manual review, thereby improving response speed and robustness.

[0326] In this embodiment, step S2, which determines whether the timestamps of the indicator vectors are consistent, further includes:

[0327] S21: Based on the generation period of the indicator vector, obtain the acquisition delay of the indicator vector, wherein the generation period is specifically from the arrival time to the recording time;

[0328] S22: Determine whether the acquisition delay exceeds a preset delay threshold;

[0329] S23: If so, identify the sampling period of the indicator vector, dynamically supplement the low-frequency data of the indicator vector according to the sampling period, and generate the confidence score of the indicator vector according to the preset account priority.

[0330] In this embodiment, the system obtains the collection delay of the indicator vector based on the generation period of the indicator vector, specifically from the arrival time to the recording time. The system then determines whether this collection delay exceeds a preset delay threshold to execute corresponding steps. For example, if the system determines that the collection delay of the indicator vector does not exceed the preset delay threshold, the system considers the indicator vector generation process to be within the controllable range of the system design. That is, the delay between the "arrival time" and the "recording time" does not exceed the system's acceptable timeliness requirements. The system will then incorporate the indicator vector into the risk control engine to participate in risk level determination, weight allocation, or action triggering, maintaining the current data collection mechanism unchanged without additional compensation or correction measures. The collection delay information is recorded in the log for subsequent delay monitoring and performance optimization. Although the delay does not exceed the threshold, the system... The system can still use the specific value of the delay for fine-grained optimization. If the delay is close to the upper limit of the threshold, the system can issue an early warning to indicate that there may be potential congestion risks in the data channel. When comparing multiple indicators, indicator vectors with smaller delays can be given higher real-time weights, making the risk assessment results more accurate. The trend of delay data can be used as part of the system performance evaluation to optimize the collection frequency, bandwidth allocation, or message queue scheduling strategy. For example, when the system determines that the collection delay of the indicator vector exceeds the preset delay threshold, the system will consider that the generation process of the indicator vector is uncontrollable and will exceed the system's acceptable timeliness requirements. The system will identify the sampling period of these indicator vectors, dynamically supplement the low-frequency data of the indicator vectors according to different sampling periods, and generate confidence scores for these indicator vectors according to the preset account settings priority.When the data acquisition delay exceeds a threshold, the system identifies the sampling period of each indicator vector and dynamically supplements low-frequency data. This effectively avoids gaps or discontinuities in the data stream, ensuring the integrity of the indicator sequence in the time dimension. This allows the subsequent risk assessment model to continuously receive stable and continuous input data, thereby reducing risk assessment interruptions or misjudgments caused by delays. Furthermore, by generating confidence scores for indicator vectors based on the account's preset priorities, the system can differentiate the importance of different indicators even with delays. For example, high-priority indicators can still be assigned higher confidence scores even with delays, while low-priority indicators are assigned lower confidence scores. This adaptive processing mechanism ensures that risk assessment results are not severely distorted by abnormal delays in individual low-priority indicators, improving the reliability and stability of overall risk identification. It also solves the data distortion problem caused by delays exceeding thresholds. Furthermore, by dynamically adjusting confidence scores, the system can flexibly optimize decisions based on actual conditions. For example, in extreme market conditions, the system can prioritize the confidence results of high-priority indicators and quickly take risk control actions. Under normal conditions, the system combines the confidence scores of all indicators for a comprehensive assessment. This adaptive strategy effectively improves the flexibility and real-time nature of risk control responses, making the trading platform more robust in the face of different market environments.

[0331] It should be noted that the sampling period of the indicator vector is identified, and low-frequency data of the indicator vector is dynamically supplemented according to the sampling period. Based on the preset priority of the account, a confidence score of the indicator vector is generated, specifically as follows:

[0332] Sampling period identification: For each risk assessment indicator, read its preset sampling period (e.g., seconds, 10 seconds, 1 minute, etc.), record the most recent sampling timestamp, and output: the sampling period T_i of each indicator and the delay of the last sampling. ;

[0333] To determine the low-frequency indicators that need to be supplemented, for each indicator, compare the delay δ_i with the allowable threshold (usually the coefficient k multiplied by the sampling period, e.g., k=1.5 or 2): if δ_i>k×Ti, then the indicator is considered to be in the state of "needing to be supplemented / cannot be directly used for real-time decision-making", and the set of indicators that need to be supplemented and the set of indicators that can be directly used are determined.

[0334] Choose a completion strategy (decision tree). If a high-frequency surrogate indicator is available for the low-frequency indicator (i.e., a high-frequency indicator that is highly correlated with it is available), prioritize surrogate regression / mapping estimation (linear regression, time series forecasting with error estimation, or Kalman filtering). Otherwise, if the historical series is stable and the most recent historical sample is available, use historical benchmark completion (such as the previous sample value, rolling mean, seasonal mean). If the estimation error or data quality is unacceptable, revert to missing label and trigger manual review or downgrade strategy. Select the completion method and necessary model / parameters for each indicator that needs completion.

[0335] Perform completion and evaluate completion quality. Generate completed values ​​(filled_value) based on the selected method and simultaneously calculate the uncertainty index (e.g., historical RMSE, prediction variance, or confidence interval width) of the estimation method. Record the completion method type along with the uncertainty as input for subsequent confidence calculations. The filled_value for each completed index is compared with the estimation error measure (e.g., RMSE or σ).

[0336] Calculate the confidence score of each indicator (component definition), calculate several sub-items for each indicator i and synthesize the indicator confidence score C_i, completeness score S_{c}, which reflects the proportion of actual observation points to expected points within the observation window (0-1), and timeliness score S_{f}, which is a function based on the delay δ_i and the sampling period T_i, for example, S_f=max(0,1-δ_i / (k× T_i)), with a value of 0-1;

[0337] The method reliability score S_{m} is mapped to [0,1] based on historical validation errors (such as RMSE) if model estimation is used (the smaller the RMSE, the higher the score). If it is direct observation, it is approximately 1. The data source reliability S_{s} is the reputation score of the data provider or channel (0-1), which is preset or adjusted according to operation monitoring.

[0338] In summary: First, synthesize the unshrunken placement reliability by product or weighted product. ;

[0339] The final confidence score is generated by combining account priority. The priority coefficient P_acct,i of this indicator under this account (e.g., high priority 1.2, normal 1.0, low 0.8) is read and applied as a scaling factor. To avoid exceeding 1 after scaling, C_i can be truncated: C_i = min(C_i, 1).

[0340] Adjust the weights of the metrics in the fusion using confidence scores; the initial (base) weights are... Calculate the weights before weighting. Normalize all indicators: This way, the uncertainty caused by filling in the gaps will automatically reduce the influence of the indicator;

[0341] Threshold judgment and operation rules: Calculate the average confidence or weighted confidence of the overall vector; if it is lower than the threshold (e.g., 0.3), it is marked as a "low confidence vector" and a downgrade process is triggered (only prompt, manual review or delayed execution); if it is higher than the threshold, it enters the normal risk judgment process.

[0342] Record and feedback: Write the completion method, estimation error, each sub-score and the final C_i, and the adjusted weight w_i into the log for auditing and model iteration. If the completion method is model estimation, the subsequently observed values ​​will be used to update the model error and S_m online.

[0343] Specific examples are as follows:

[0344] Scene and input, three metrics and their sampling period and most recent time information:

[0345] Indicator A (PriceVol, sampling period T_A=10 seconds), latest sampled at t (delay δ_A=2 seconds);

[0346] Indicator B (Volume, sampling period T_B=1 second), the latest sample was taken at t (delay δ_B=0 seconds);

[0347] The index C (Sentiment, sampling period T_C=60 seconds) was last sampled at t-90 seconds (delay δ_C=90 seconds).

[0348] Basic weights: Account priority coefficients (the account's preference for the three indicators), P_A=1.0, P_B=1.0, P_C=1.0 (this example is for the same priority, the variable cases will be shown later), and the tolerance coefficient k=1.5;

[0349] Step 1, identify the indicators that need to be supplemented:

[0350] δ_A==2,kT_A=1.5×10=15→2≤15→No need to fill in the gaps;

[0351] δ_B=0, kT_B=1.5×1=1.5→0≤1.5→No need to pad;

[0352] δ_C=90, kT_C=1.5×60=90→90≤90→According to strict rules, it can be determined as a boundary. For example, if δ>kT, then fill it; if it is equal, then it is considered to need to be filled. Here, we take the filling (considered as exceeding the threshold); → Set to be filled = {C};

[0353] Step 2, perform completion (regression prediction), assuming the historical regression model (obtained from 30 minutes of historical training) is: Sentiment≈intercept+a×Volume+b×PriceVol; for example: intercept=-0.05, a=0.00001 (Volume is in "lots"), b=0.6 (PriceVol is normalized to 0-1).

[0354] Current high-frequency values ​​(used for regression): Volume (recent window average) = 15,000; PriceVol = 0.65;

[0355] Predicted value: Sent_pred = -0.05 + 0.00001 × 15000 + 0.6 × 0.65 = -0.05 + 0.15 + 0.39 = 0.49;

[0356] Historical validation yielded a model RMSE of 0.08 (example value); the recorded estimation error σ ≈ 0.08.

[0357] Step 3, calculate the confidence score for each component, and the integrity score S_c (sampling integrity): In the most recent 2 × sampling period (120s), the expected number of samples is ≈2, and the actual number of samples is 1 → S_c = 1 / 2 = 0.5;

[0358] Timeliness score S_f: Using S_f=max(0,1-δ / (kT)), and taking k=2 (more conservative) for freshness, kT_C=120: S_f=1-90 / 120=0.25;

[0359] Method reliability score S_m: Using S_m=1 / (1+RMSE) as a simple mapping: RMSE=0.08→S_m=1 / (1+0.08)≈0.9259;

[0360] Data source reliability S_s: Historical reliability of public opinion data source 0.90 → S_s=0.90;

[0361] Synthetic unscaled confidence level: ,

[0362] Calculate: 0.5 × 0.25 = 0.125; 0.125 × 0.9259 ≈ 0.11574; 0.11574 × 0.9 ≈ 0.10417, → ≈0.1042;

[0363] For A and B (not fully completed, direct observation), assume their components are both excellent: S_c=S_f=S_m=S_s≈1 (if the data source is slightly inferior, it can be set to 0.95). Here, we take A: 0.95, B: 0.98 (example). → =0.95, =0.98;

[0364] In this example, the account priority P is 1.0, therefore the final confidence level C_i = min( )for:

[0365] C_A=0.95,C_B=0.98,C_C≈0.1042;

[0366] Step 4: Adjust and normalize the weights using confidence levels.

[0367] Calculate the weights before weighting :

[0368] w'_A=0.50×0.95=0.475,

[0369] w'_B=0.30×0.98=0.294,

[0370] w'_C=0.20×0.1042≈0.02084,

[0371] The summation is S = 0.475 + 0.294 + 0.02084 = 0.78984;

[0372] Final weights after normalization:

[0373] w_A = 0.475 / 0.78984 ≈ 0.6015

[0374] w_B = 0.294 / 0.78984 ≈ 0.3724

[0375] w_C=0.02084 / 0.78984≈0.0264;

[0376] Step 5: Use the imputation value and final weights to enter the subsequent judgment. Fill the index vector with the estimated value of Sentiment, 0.49. In the subsequent comprehensive membership calculation, use each index (the observed values ​​of A and B, and the estimated value of C) with the new weights w_A, w_B, and w_C. This significantly reduces the impact of the unreliable C due to missing / lagging values ​​on the final judgment (to only about 2.6%), ensuring the robustness of the judgment result.

[0377] Step 6, low-confidence overall judgment.

[0378] Calculate the average confidence of the vector / 3≈(0.95+0.98+0.1042) / 3≈0.6787. If the system's preset overall confidence threshold is 0.3, then >0.3→ The vector can be used for automatic judgment. If the threshold is high (e.g., 0.8), the automatic execution will be withdrawn and a manual review will be initiated.

[0379] Step 7, Log and Model Update: Store the imputation method, predicted value, RMSE, sub-scores and final weights in the database. When the actual Sentiment value is reached, update the model RMSE with the actual value, thereby dynamically updating S_m.

[0380] In this embodiment, after step S4 of determining whether the confidence level of the transaction risk assessment result reaches a preset confidence level threshold, the method further includes:

[0381] S401: Based on the risk control measures of the preset account, obtain the continuous time series of the confidence level, wherein the risk control measures specifically include automatic execution of position reduction and closing, manual review and risk warning;

[0382] S402: Determine whether the continuous time series continues to decline within a preset time period;

[0383] S403: If so, then collect the trend data of the continuous time series, dynamically trigger the risk control warning for the preset account based on the trend data, and adaptively execute the risk control measures based on the risk control warning.

[0384] In this embodiment, the system, based on pre-set account risk control measures, specifically including automatic position reduction and liquidation, manual review, and risk alerts, obtains a continuous time series of confidence levels. The system then determines whether these continuous time series continuously decline within a pre-set time period, and executes the corresponding steps accordingly. For example, if the system determines that the continuous time series of confidence levels does not continuously decline within the pre-set time period, the system considers that the current trading account's risk level has not shown a trend of continuous deterioration or rapid decline, meaning the risk remains within a controllable range, and the system maintains real-time monitoring of the confidence level series. The system continues to collect new indicator vector data, keeping the risk alert function available without triggering high-intensity risk control measures. It smooths or weights continuous time series data to more accurately identify potential subsequent downward trends. For example, if the system determines that a continuous time series with a confidence level is continuously declining within a pre-set time period, it considers the current trading account's risk level to be showing a continuous deterioration or rapid decline, making the risk difficult to control. The system collects trend data from these continuous time series and dynamically triggers risk control alerts based on this data. Depending on the specific risk control alert, the system adaptively executes the corresponding risk control measures. The system employs several risk control measures. By analyzing the continuous decline in confidence levels over a preset time period, it can quickly detect the continuous deterioration or rapid decline in the risk level of a trading account. This real-time trend recognition helps to identify potential abnormal trading conditions or the impact of market fluctuations in advance, shifting risk management from a passive response to proactive early warning. This improves the foresight and agility of overall risk control. After acquiring trend data, the system can adaptively trigger corresponding risk control warnings based on different levels of risk severity and execute appropriate measures, such as automatic position reduction, position closing, risk alerts, or manual review. By associating continuously declining confidence levels with preset risk control rules, the system can achieve differentiated responses, ensuring that high-risk accounts are controlled in a timely manner while low-risk accounts are not excessively intervened in. This improves the targeting and efficiency of risk control operations. Furthermore, this dynamic risk control mechanism based on trend data not only guarantees real-time risk intervention but also effectively reduces potential account losses through adaptive execution of risk control measures. The system can adjust the intensity of measures according to trend strength, making the trading platform more robust and controllable in the face of rapid fluctuations or abnormal market conditions. It also provides data support and traceable records for subsequent strategy optimization and risk assessment.

[0385] It should be noted that the continuous time series trend data is collected, and risk control warnings for the preset accounts are dynamically triggered based on the trend data. Based on the risk control warnings, the risk control measures are adaptively executed, specifically as follows:

[0386] Collect the trend data of the continuous time series. The system reads the confidence continuous time series C(t) of the account in real time, records the confidence value C_i corresponding to each time point t_i, and conducts trend analysis on the series. For example, calculate the slope m through linear regression, or use the rolling mean or exponentially weighted moving average (EWMA) to judge the decline rate. Then output the trend parameters (such as slope, change amplitude, weighted average) within each time period for subsequent risk control determination;

[0387] Determine the trend and dynamically trigger risk control warnings. The system sets the trend threshold m_th or the continuous decline amplitude threshold ΔC_th. If the confidence slope m < m_th or within N consecutive time points ,it is determined that the risk is continuously rising, triggering risk control warnings of corresponding levels (such as normal warnings, advanced warnings or emergency warnings), and then output the type of risk control warning and the trigger timestamp as the basis for execution measures;

[0388] Adaptive execution of risk control measures,

[0389] The system selects appropriate risk control measures according to the triggered warning level. For example:

[0390] Normal warning: Send a risk reminder to the user, keep the account operations, but record the logs,

[0391] Advanced warning: Automatically execute position reduction or limit adjustment, and notify the user and the risk control team at the same time,

[0392] Emergency warning: Automatically close positions or freeze account operations, and trigger manual review,

[0393] Each measure dynamically adjusts the intensity according to the account settings, market conditions and real-time confidence,

[0394] Then output the specific operation records executed, including the operation type, time and the affected account funds or trading volume;<​​​​​​​​​​​​​​​​​​​Step 1: Calculate the trend. Fit C against time t by linear regression: obtain the slope m ≈ -0.036 m / min or calculate the total decline: ΔC = 0.92 - 0.68 = 0.24 > 0.15;

[0402] Step 2: Determine risk control warning. Since the slope m < m_th and the total decline ΔC > ΔC_th, trigger a high-level warning.

[0403] The risk control system records the trigger time and warning level:

[0404] Time: The last time point (the 7th minute),

[0405] Warning level: High-level warning;

[0406] Step 3: Adaptively execute risk control measures.

[0407] The system checks the account settings: allowing high-level warnings to execute position reduction and limit adjustment;

[0408] Calculate the operation volume based on the current position and risk exposure:

[0409] Position reduction ratio: Dynamically calculated according to the risk amplitude and the ratio of account funds (for example, the total position is 1 million, and a 20% position reduction → a 200,000 position reduction),

[0410] Limit adjustment: Reduce the upper limit of the single-tradeable fund by 30%;

[0411] Generate a log after the operation is executed: Record the position reduction amount, the adjusted limit, the trigger time, and the warning level;

[0412] To sum up, in the above example content, through continuous time series trend analysis, the system can real-time detect the continuous increase of account risks, avoid losses caused by delayed reactions. At the same time, the risk control measures are adaptively adjusted according to the warning level and the current state of the account, achieving refined and differentiated intervention. And all operation records, trend parameters, and confidence sequences are saved, providing reliable data support for subsequent risk analysis, strategy optimization, and compliance auditing.

[0413] In this embodiment, in step S1 of classifying the corresponding index types from the multi-dimensional index data pre-collected by the trading terminal and generating the index vector at the current moment, it further includes:

[0414] S11: Based on the data quality scores preset by the trading terminal, construct an index mapping table for the index vector, where the data quality scores specifically include data integrity, stability, and anomaly frequency;

[0415] S12: Determine whether the vector structure of the index mapping table is stable;

[0416] S13: If not, then based on the data quality score, identify the indicator characteristics of the indicator vector, and dynamically update the indicator mapping table according to the indicator characteristics. Specifically, the indicator characteristics include continuous, discrete, and categorical types, and the dynamic update specifically includes adding new indicators and discarding indicators.

[0417] In this embodiment, the system constructs an indicator mapping table of indicator vectors based on a pre-set data quality score on the trading terminal. The data quality score specifically includes data integrity, stability, and anomaly frequency. The system then determines whether the vector structure of this indicator mapping table is stable to execute corresponding steps. For example, if the system determines that the vector structure of the indicator mapping table is stable, it considers that the constructed indicator mapping table maintains the expected structural consistency within the current time period, and the mapping relationship between indicators is clear, without significant fluctuations or anomalies. The system uses this stable indicator mapping table to construct the indicator vector for the current moment, calculating the membership degree or confidence level of each risk level. Simultaneously, the indicator vector is input into a pre-trained inference model for comprehensive analysis, generating risk judgment results or trend data. The system also maintains continuous data quality monitoring to ensure that the indicator vector remains stable in subsequent time periods, ready to respond to anomalies. Conversely, if the system determines that the vector structure of the indicator mapping table is unstable, it considers that the mapping relationship between indicators has significant fluctuations or anomalies. The system identifies the indicator characteristics of each indicator vector based on different data quality scores. These indicator characteristics specifically include continuous, discrete, and categorical types. Based on these characteristics, the system dynamically updates the indicator. The mapping table is dynamically updated, specifically including the addition and obsolescence of indicators. By identifying the characteristics of each indicator (continuous, discrete, and categorical) and dynamically updating the mapping table, the system can promptly remove abnormal or invalid indicators and add new ones. This ensures that the indicator vector is always built on high-quality, reliable data. This process avoids risk assessment bias caused by indicator fluctuations or data anomalies, enhancing the completeness and reliability of data input. Furthermore, by dynamically updating the indicator mapping table, the system can adjust indicator usage strategies based on actual data quality, providing the risk assessment model with more stable input. For example, abnormal fluctuations in continuous indicators can be promptly detected. By identifying and adjusting weights, missing or anomaly-related discrete or categorical indicators can be replaced or marked, ensuring that the model is not affected by low-quality data when calculating comprehensive membership and confidence levels. This improves the accuracy and effectiveness of risk control decisions. Furthermore, dynamically updating the mapping table allows the system to quickly adapt to changes in the market environment or the introduction of new indicators. New indicators can be promptly incorporated into risk assessments, while invalid indicators are discarded to avoid interfering with decision-making. This mechanism enhances the robustness and adaptability of the trading platform, enabling the risk control system to operate stably in the long term and continuously provide reliable risk assessment and early warning capabilities, providing continuous data support for subsequent strategy optimization and model iteration.

[0418] It should be noted that, based on the data quality score, the indicator characteristics of the indicator vector are identified, and the indicator mapping table is dynamically updated according to these characteristics, specifically as follows:

[0419] The system reads the data quality score. Based on the data quality score preset by the trading terminal, it evaluates the continuity, completeness, stability and frequency of anomalies of each indicator and outputs the comprehensive data quality score Qi for each indicator, which usually ranges from 0 to 1.

[0420] Identify indicator characteristics and categorize indicators into characteristic categories based on indicator type and data quality score:

[0421] Continuous type: The indicator value can take any real number, such as price volatility or volume change rate.

[0422] Discrete type: The indicator value is an integer or a finite step size, such as the number of orders or transactions.

[0423] Categorization: The indicator value is a category label, such as public sentiment (positive / neutral / negative) or market event type.

[0424] The system combines historical data distribution, missing data, and abnormal fluctuation amplitude to determine whether the indicator characteristics are stable or need adjustment.

[0425] Then the characteristic label and stability score of each indicator will be output;

[0426] The decision-making index update strategy is as follows: for indices with data quality scores below the threshold Q_min, they are marked as "discarded candidates" and will be dynamically removed; for newly introduced indices or indices with quality scores above the threshold, they are marked as "new candidates" and will be included in the mapping table; for indices whose characteristics have changed significantly (such as a large number of discrete anomalies in continuous indices), the mapping method or weights are considered for adjustment.

[0427] Dynamically update the indicator mapping table.

[0428] Handling new metrics: Add new metrics to the mapping table, assign initial weights, and verify the mapping with existing metrics.

[0429] Handling obsolete metrics: Remove metrics that are of low quality, abnormally frequent, or no longer valid from the mapping table.

[0430] Feature adjustment: For metrics whose features have changed, adjust their type label and processing method in the mapping table (e.g., convert continuous to discrete, adjust weighting or normalization strategy).

[0431] To verify the stability of the mapping table, the system performs consistency and integrity checks on the updated indicator mapping table to ensure the stability of the vector structure, which facilitates subsequent risk assessment or risk control operations.

[0432] Specific examples are as follows:

[0433] Input data,

[0434] Three indicators and their data quality scores:

[0435] Indicator A (PriceVol, continuous): Q_A = 0.92,

[0436] Indicator B (OrderCount, discrete type): Q_B=0.65,

[0437] Indicator C (Sentiment, Classification): Q_C=0.95,

[0438] Threshold Q_min = 0.7;

[0439] Step 1: Determine whether to discard or add indicators.

[0440] Indicator A: Q_A = 0.92 > 0.7, retain.

[0441] Indicator B: Q_B = 0.65 < 0.7, marked as a discarded candidate.

[0442] Indicator C: Q_C=0.95>0.7, retain;

[0443] Step 2, dynamically update the mapping table.

[0444] Remove indicator B from the mapping table, retain indicators A and C. If the system adds indicator D (HighFreqVolume, continuous, score 0.88), add it to the mapping table with an initial weight of 0.2 and re-normalize the weights with the existing indicators.

[0445] Step 3, adjust the characteristic processing. If historical statistics show that indicator A has recently shown discrete anomalies (such as short-term extreme jumps), the system can mark it as "continuous + anomaly processing". In the risk calculation, smoothing or correction is applied to the anomaly point. The output result is that the updated indicator mapping table contains indicators A, C and D. Each indicator has a type label (continuous / categorical), data quality score, initial weight and anomaly processing mark.

[0446] In summary, the above examples demonstrate how the system eliminates low-quality indicators to ensure high-quality and stable data vectors for risk assessment. By dynamically adding indicators and adjusting characteristics, the system can adapt to market changes or new data sources, improving the accuracy of risk assessment and risk control response. The updated mapping table undergoes consistency and integrity verification, ensuring the stability of the indicator vector structure and avoiding risk calculation deviations or risk control errors due to data anomalies.

[0447] Reference Appendix Figure 2A market fluctuation early warning system based on multi-dimensional indicator fusion, as described in one embodiment of the present invention, includes:

[0448] The generation module 10 is used to classify the corresponding indicator types from the multi-dimensional indicator data pre-collected by the trading terminal and generate the indicator vector at the current moment. Specifically, the indicator types include trading indicators, macroeconomic indicators and text sentiment indicators.

[0449] The judgment module 20 is used to determine whether the timestamps of the indicator vectors are consistent;

[0450] The execution module 30 is used to construct the fuzzy membership function corresponding to the indicator vector if the condition is met, input the fuzzy membership function into the pre-trained inference model, calculate the membership degree value of each risk level under the indicator vector, perform a weighted combination of the membership degree values ​​of the same risk level to obtain the comprehensive membership degree of the same risk level, compare the comprehensive membership degree, and select the membership degree value of the highest risk level as the transaction risk judgment result at the current moment. The fuzzy membership function is specifically a fuzzy interval corresponding to different risk fluctuations.

[0451] The second judgment module 40 is used to judge whether the confidence level of the transaction risk judgment result reaches the preset confidence level threshold.

[0452] The second execution module 50 is used to identify the preset preconditions of the trading terminal if the conditions are met, construct a corresponding action set from the preset action matrix of the trading terminal based on the preconditions, generate the action type in the trading process through the action set, obtain the approval response of the trading terminal, and dynamically execute the action set in the preset account based on the approval response. The preconditions specifically include trading time period, trading liquidity value and trading permissions, and the action type specifically includes protection action, hedging strategy and limit adjustment.

[0453] In this embodiment, the generation module 10 classifies the corresponding indicator types from the multi-dimensional indicator data pre-collected by the trading terminal. Specifically, the indicator types include trading indicators, macroeconomic indicators, and text sentiment indicators. It then generates an indicator vector for the current moment. The judgment module 20 then determines whether the timestamps of these indicator vectors are consistent, in order to execute the corresponding steps. For example, if the system determines that the indicator vectors for the current moment are inconsistent, it assumes that some indicator vectors originate from different acquisition channels or systems, resulting in delays and misalignment of data at the same time point. The system enters a short-term waiting state until the timestamps of all indicator data are aligned before continuing to generate indicator vectors. Simultaneously, it prioritizes important indicators (such as trading prices). First, wait and allow replacement or weight reduction of secondary indicators (such as some public opinion data) to avoid interruption of the overall calculation due to the absence of individual non-critical indicators. Furthermore, in subsequent confidence calculations, the weight of results at abnormal times is reduced to ensure robustness of the judgment. For example, when the system determines that the indicator vectors at the current time are consistent, the execution module 30 will assume that there is no delay in the indicator vectors and that data at the same time point can be aligned. The system will then construct fuzzy membership functions corresponding to these indicator vectors. Specifically, the fuzzy membership functions are fuzzy intervals corresponding to different risk fluctuations. These fuzzy membership functions are input into a pre-trained inference model to calculate the membership values ​​of each risk level under these indicator vectors. The membership values ​​of the same risk level are then weighted and combined to obtain the membership values ​​for the same risk level. The system calculates the comprehensive membership degree of risk levels and compares these comprehensive membership degrees, selecting the membership degree value of the highest risk level as the transaction risk judgment result at the current moment. By constructing indicator vectors with consistent timestamps, the system ensures that all multi-dimensional indicator source data are aligned at the same time, avoiding judgment bias caused by delays or missing data. This synchronization guarantees the integrity and real-time nature of the input data, providing a reliable data foundation for subsequent risk analysis and thus improving the accuracy of risk judgment. Simultaneously, by using fuzzy membership functions to map different risk fluctuations to fuzzy intervals and inputting them into a pre-trained inference model to calculate the membership degree value of each risk level, the system can transform the originally difficult-to-quantify risk intensity into a measurable numerical indicator. Further, by adding... The system combines weights to form a comprehensive membership degree, then compares and selects the maximum value to obtain a specific risk level. This process not only achieves quantitative risk classification but also reflects the differences in risk intensity, elevating risk warnings from "risk present / no risk" to "risk strength classification." Furthermore, through comprehensive calculation of the membership degrees of each risk level and selection of the maximum membership degree, the system can output a clear market risk level as the current transaction risk assessment result. Compared to the traditional method that only indicates "risk exists," the assessment result generated by this scheme has a clear level and intensity, which can be directly used as the basis for securities firms or investment advisory platforms to trigger risk control measures and adjust investment strategies, thereby achieving automation and intelligence in risk management and improving market response efficiency.The second judgment module 40 then determines whether the confidence level of the transaction risk assessment result reaches a preset confidence threshold, and executes the corresponding steps accordingly. For example, if the system determines that the confidence level of the transaction risk assessment result does not reach the preset confidence threshold, the system will consider that the risk identification lacks sufficient basis. If the result is directly adopted, it may lead to the false triggering or omission of risk control measures. The system will then reacquire the missing or delayed multi-dimensional indicator data, weight and enhance the key indicators, regenerate the indicator vector, and recalculate the risk assessment result. At the same time, it will introduce historical indicator vectors from previous and subsequent moments to smooth short-term fluctuations and improve the overall stability of the assessment. Furthermore, if the confidence level cannot be improved, the result will be downgraded, for example, by marking it. This is a "low-confidence risk warning," but it is only provided to the user as a reference signal and does not trigger automatic risk control measures. For example, when the system determines that the confidence level of the transaction risk assessment result has reached a pre-set confidence threshold, the second execution module 50 will consider the risk identification to be compliant. The system will identify the pre-set preconditions of the trading terminal, which specifically include the trading period, trading liquidity value, and trading permissions. Based on these preconditions, it will construct a corresponding action set from the pre-set action matrix of the trading terminal. Through different action sets, it will generate action types in the trading process, which specifically include protective actions, hedging strategies, and limit adjustments. It will obtain the approval response from the trading terminal and, based on the approval response, will... These action sets are dynamically executed in the established accounts. Only when the confidence level reaches a preset threshold will the system proceed to the subsequent risk control linkage process. This ensures that the risk assessment results have sufficient data support and reliability, avoiding false triggers due to unstable or abnormal data. By identifying preconditions such as trading time periods, trading liquidity values, and trading permissions, the system ensures that the executed risk control measures match the actual trading environment, avoiding operations during inappropriate time periods or beyond authorized limits, thereby improving the compliance and scientific nature of risk control decisions. Simultaneously, based on the preconditions, corresponding action sets are constructed from the action matrix, and different action types are generated through different sets, such as protection actions, hedging strategies, and limit adjustments. This condition-matching dynamic... This dynamic risk generation method allows risk control measures to be refined and differentiated based on different market risk levels and trading environments, avoiding a one-size-fits-all risk response model. This enables more flexible responses to complex and ever-changing market environments, improving the effectiveness of risk management. Furthermore, after generating the action type, the system also obtains the approval response from the trading terminal and dynamically executes the action set in the preset account, achieving fully automated linkage from risk identification to risk management. Compared to the traditional model relying on manual judgment and intervention, this solution can significantly shorten the response time for risk management, improve the response speed and execution efficiency of brokerage platforms or investment advisory platforms in the face of market fluctuations, thereby reducing potential losses and enhancing overall risk management capabilities.

[0454] In this embodiment, the execution module further includes:

[0455] The identification unit is used to read the indicator values ​​of the indicator vector at the current moment one by one and identify the preset risk range of the indicator values.

[0456] A judgment unit is used to determine whether the indicator value is in the intersection range of two risk levels;

[0457] The execution unit is configured to, if so, mark the index value as a preset boundary state, retain the first-level membership degree and the second-level membership degree of the index value, assign different weights to the first-level membership degree and the second-level membership degree based on the center position of the index value from the cross interval, detect the sum of the membership degrees of the first-level membership degree and the second-level membership degree, and dynamically adjust the total membership degree based on the sum of the membership degrees.

[0458] In this embodiment, the system uses pre-set risk assessment indicators on the trading terminal, specifically including price volatility, volume change rate, bid-ask spread, and public sentiment index. It then obtains the dimensional differences among these risk assessment indicators and determines whether these dimensional differences unify the numerical range to execute corresponding steps. For example, when the system determines that the dimensional differences among these risk assessment indicators can unify the numerical range, different types of indicators, after normalization or standardization, can be mapped to the same numerical scale, for example, all compressed into the interval [0,1][0,1][0,1]. If the risk assessment indicators fall within the range of a standard normal distribution, the system will directly adopt the unified numerical range and use these risk assessment indicators as components of a vector of the same dimension to construct the indicator vector at the current moment. Simultaneously, without additional weight adjustments, it will directly use these indicator vectors to generate corresponding fuzzy membership functions and calculate the membership degree for each risk level. For example, if the system determines that the dimensional differences of a certain risk assessment indicator cannot be unified within a single numerical range, it will assume that different types of indicators cannot be mapped to the same numerical scale. The system will then detect the numerical range of these risk assessment indicators, identify the numerical span of these indicators based on different numerical ranges, and calculate the deviation difference degree of the risk assessment indicators based on this numerical span. Through this deviation difference degree, the system will dynamically adjust the weight ratio of the risk assessment indicators. When the dimensional differences of some risk assessment indicators cannot be unified within a single numerical range, by detecting the numerical range of the indicators and calculating their span, the system can quantify the differences in scale among the indicators and, based on the deviation difference degree... Adjusting weights can prevent a single indicator from gaining an unreasonable advantage in comprehensive calculations due to excessively large dimensions or wide numerical ranges. This ensures a fairer and more objective fusion result of multi-dimensional indicators in risk assessment. Furthermore, by introducing deviation differences and incorporating philosophical principles for dynamic weight adjustment, the system can maintain balance among multi-dimensional indicators. This allows indicators from different sources and with different dimensions to contribute their respective risk information without being excessively amplified or weakened by certain indicators. This method effectively improves the accuracy and stability of risk assessment results, preventing distortion of the overall risk assessment due to extreme fluctuations in a single type of indicator. It also allows the system to automatically re-detect differences in indicator dimensions and dynamically adjust weight ratios when facing newly introduced indicators or changes in the market environment, thus possessing adaptive capabilities. This not only enhances the adaptability of the risk assessment model to complex market environments but also provides compatibility space for the subsequent introduction of more heterogeneous data (such as news sentiment and international market indicators), making the risk management system more scalable and valuable for long-term application.

[0459] In this embodiment, it also includes:

[0460] The acquisition module is used to acquire the dimensional differences of the risk assessment indicators based on the risk assessment indicators preset by the trading terminal. Specifically, the risk assessment indicators include price volatility, trading volume change rate, bid-ask spread, and public sentiment index.

[0461] The third judgment module is used to determine whether the difference in dimensions is within a uniform numerical range;

[0462] The third execution module is used to detect the numerical range of the risk assessment indicator if no, identify the numerical span of the risk assessment indicator based on the numerical range, calculate the deviation difference of the risk assessment indicator based on the numerical span, and dynamically adjust the weight ratio of the risk assessment indicator based on the deviation difference.

[0463] In this embodiment, the system uses pre-set risk assessment indicators on the trading terminal, specifically including price volatility, volume change rate, bid-ask spread, and public sentiment index. It then obtains the dimensional differences among these risk assessment indicators and determines whether these dimensional differences unify the numerical range to execute corresponding steps. For example, when the system determines that the dimensional differences among these risk assessment indicators can unify the numerical range, different types of indicators, after normalization or standardization, can be mapped to the same numerical scale, for example, all compressed into the interval [0,1][0,1][0,1]. If the risk assessment indicators fall within the range of a standard normal distribution, the system will directly adopt the unified numerical range and use these risk assessment indicators as components of a vector of the same dimension to construct the indicator vector at the current moment. Simultaneously, without additional weight adjustments, it will directly use these indicator vectors to generate corresponding fuzzy membership functions and calculate the membership degree for each risk level. For example, if the system determines that the dimensional differences of a certain risk assessment indicator cannot be unified within a single numerical range, it will assume that different types of indicators cannot be mapped to the same numerical scale. The system will then detect the numerical range of these risk assessment indicators, identify the numerical span of these indicators based on different numerical ranges, and calculate the deviation difference degree of the risk assessment indicators based on this numerical span. Through this deviation difference degree, the system will dynamically adjust the weight ratio of the risk assessment indicators. When the dimensional differences of some risk assessment indicators cannot be unified within a single numerical range, by detecting the numerical range of the indicators and calculating their span, the system can quantify the differences in scale among the indicators and, based on the deviation difference degree... Adjusting weights can prevent a single indicator from gaining an unreasonable advantage in comprehensive calculations due to excessively large dimensions or wide numerical ranges. This ensures a fairer and more objective fusion result of multi-dimensional indicators in risk assessment. Furthermore, by introducing deviation differences and incorporating philosophical principles for dynamic weight adjustment, the system can maintain balance among multi-dimensional indicators. This allows indicators from different sources and with different dimensions to contribute their respective risk information without being excessively amplified or weakened by certain indicators. This method effectively improves the accuracy and stability of risk assessment results, preventing distortion of the overall risk assessment due to extreme fluctuations in a single type of indicator. It also allows the system to automatically re-detect differences in indicator dimensions and dynamically adjust weight ratios when facing newly introduced indicators or changes in the market environment, thus possessing adaptive capabilities. This not only enhances the adaptability of the risk assessment model to complex market environments but also provides compatibility space for the subsequent introduction of more heterogeneous data (such as news sentiment and international market indicators), making the risk management system more scalable and valuable for long-term application.

[0464] In this embodiment, the second execution module further includes:

[0465] The second identification unit is used to identify the real-time status information of a preset account during the transaction process based on the preset trigger conditions of the action set. The trigger conditions specifically include placing an order, canceling an order, modifying an order, and transferring account funds. The real-time status information specifically includes market status, account balance status, and order execution status.

[0466] The second judgment unit is used to determine whether the real-time status information meets the triggering condition;

[0467] The second execution unit is configured to, if so, obtain the verification information of the action set according to the risk control rules preset by the transaction terminal, dynamically remove the constrained actions of the action set based on the verification information, and divide the action set into priority actions, wherein the priority actions specifically include primary actions and secondary actions.

[0468] In this embodiment, the system identifies pre-defined trigger conditions based on a set of actions, specifically including order placement, order cancellation, order modification, and account fund transfer. These real-time status information includes market conditions, account balance, and order execution status. The system then determines whether this real-time status information meets the pre-defined trigger conditions and executes the corresponding steps. For example, if the system determines that the account's real-time status information during the transaction does not meet the pre-defined trigger conditions, the system considers that the account's actual operating environment does not match the execution prerequisites required by the action set. Directly executing the action in this case could lead to transaction risks. If execution fails, the system will avoid triggering operations such as order placement, cancellation, modification, or fund transfer that are inconsistent with the account status, thus mitigating the risk of erroneous operations from the source. Furthermore, if the account balance is insufficient, the order execution status is abnormal, or market conditions are volatile, the corresponding triggering conditions and reasons for non-compliance should be recorded in detail in the system log. The system must also provide feedback on the non-compliance conditions and potential impact to the trading terminal user or risk control through the interface or message notification, prompting manual intervention or strategy adjustment. For example, when the system determines that the real-time status information of the account during the trading process meets the pre-set triggering conditions of the action set, the system will recognize that the actual operating environment of the current account matches the execution prerequisites required by the action set, and the system will proceed according to the pre-set conditions of the trading terminal. The system first establishes risk control rules, then obtains verification information for a set of actions. Based on different verification information, it dynamically removes constrained actions from the action set and classifies actions into priority actions, specifically including primary and secondary actions. The system further filters the action set by obtaining verification information after trigger conditions are met, dynamically removing constrained actions that do not comply with risk control rules or may cause risks. This mechanism avoids the drawback of "blindly executing actions as soon as conditions are met," making the execution process of the action set more consistent with the actual market environment and the safety boundaries of the account status, thereby significantly reducing the probability of misoperation or risk escalation. Simultaneously, the system classifies actions into priority actions based on verification information to ensure that primary actions can... Prioritized actions are executed first, while secondary actions are processed only when resources allow or conditions are stable. This priority management strategy enables the system to rationally schedule the execution order in complex trading scenarios, avoiding resource waste or operational conflicts, thereby improving overall execution efficiency and response speed, enhancing system flexibility, and through a dynamic mechanism of "verification-removal-priority allocation," the system can clearly reflect the logical source of each action selection, i.e., why some actions are retained or removed, and why some actions are set as primary actions. This not only improves the interpretability of the risk control process, but also enables the system to quickly adjust the action set according to different market environments or risk control rules, enhancing its adaptability and scalability to complex and ever-changing trading environments.

[0469] In this embodiment, the determination module further includes:

[0470] The acquisition unit is used to acquire the acquisition delay of the indicator vector based on the generation period of the indicator vector, wherein the generation period is specifically from the arrival time to the recording time;

[0471] The third judgment unit is used to determine whether the acquisition delay exceeds a preset delay threshold;

[0472] The third execution unit is used to identify the sampling period of the indicator vector if the condition is met, dynamically supplement the low-frequency data of the indicator vector according to the sampling period, and generate the confidence score of the indicator vector according to the preset priority of the account.

[0473] In this embodiment, the system obtains the collection delay of the indicator vector based on the generation period of the indicator vector, specifically from the arrival time to the recording time. The system then determines whether this collection delay exceeds a preset delay threshold to execute corresponding steps. For example, if the system determines that the collection delay of the indicator vector does not exceed the preset delay threshold, the system considers the indicator vector generation process to be within the controllable range of the system design. That is, the delay between the "arrival time" and the "recording time" does not exceed the system's acceptable timeliness requirements. The system will then incorporate the indicator vector into the risk control engine to participate in risk level determination, weight allocation, or action triggering, maintaining the current data collection mechanism unchanged without additional compensation or correction measures. The collection delay information is recorded in the log for subsequent delay monitoring and performance optimization. Although the delay does not exceed the threshold, the system... The system can still use the specific value of the delay for fine-grained optimization. If the delay is close to the upper limit of the threshold, the system can issue an early warning to indicate that there may be potential congestion risks in the data channel. When comparing multiple indicators, indicator vectors with smaller delays can be given higher real-time weights, making the risk assessment results more accurate. The trend of delay data can be used as part of the system performance evaluation to optimize the collection frequency, bandwidth allocation, or message queue scheduling strategy. For example, when the system determines that the collection delay of the indicator vector exceeds the preset delay threshold, the system will consider that the generation process of the indicator vector is uncontrollable and will exceed the system's acceptable timeliness requirements. The system will identify the sampling period of these indicator vectors, dynamically supplement the low-frequency data of the indicator vectors according to different sampling periods, and generate confidence scores for these indicator vectors according to the preset account settings priority.When the data acquisition delay exceeds a threshold, the system identifies the sampling period of each indicator vector and dynamically supplements low-frequency data. This effectively avoids gaps or discontinuities in the data stream, ensuring the integrity of the indicator sequence in the time dimension. This allows the subsequent risk assessment model to continuously receive stable and continuous input data, thereby reducing risk assessment interruptions or misjudgments caused by delays. Furthermore, by generating confidence scores for indicator vectors based on the account's preset priorities, the system can differentiate the importance of different indicators even with delays. For example, high-priority indicators can still be assigned higher confidence scores even with delays, while low-priority indicators are assigned lower confidence scores. This adaptive processing mechanism ensures that risk assessment results are not severely distorted by abnormal delays in individual low-priority indicators, improving the reliability and stability of overall risk identification. It also solves the data distortion problem caused by delays exceeding thresholds. Furthermore, by dynamically adjusting confidence scores, the system can flexibly optimize decisions based on actual conditions. For example, in extreme market conditions, the system can prioritize the confidence results of high-priority indicators and quickly take risk control actions. Under normal conditions, the system combines the confidence scores of all indicators for a comprehensive assessment. This adaptive strategy effectively improves the flexibility and real-time nature of risk control responses, making the trading platform more robust in the face of different market environments.

[0474] In this embodiment, it also includes:

[0475] The second acquisition module is used to acquire the continuous time series of the confidence level based on the risk control measures of the preset account, wherein the risk control measures specifically include automatic execution of position reduction and closing, manual review and risk warning;

[0476] The fourth judgment module is used to determine whether the continuous time series continues to decline within a preset time period;

[0477] The fourth execution module is used to collect trend data of the continuous time series if the condition is met, dynamically trigger risk control warnings for the preset account based on the trend data, and adaptively execute the risk control measures based on the risk control warnings.

[0478] In this embodiment, the system, based on pre-set account risk control measures, specifically including automatic position reduction and liquidation, manual review, and risk alerts, obtains a continuous time series of confidence levels. The system then determines whether these continuous time series continuously decline within a pre-set time period, and executes the corresponding steps accordingly. For example, if the system determines that the continuous time series of confidence levels does not continuously decline within the pre-set time period, the system considers that the current trading account's risk level has not shown a trend of continuous deterioration or rapid decline, meaning the risk remains within a controllable range, and the system maintains real-time monitoring of the confidence level series. The system continues to collect new indicator vector data, keeping the risk alert function available without triggering high-intensity risk control measures. It smooths or weights continuous time series data to more accurately identify potential subsequent downward trends. For example, if the system determines that a continuous time series with a confidence level is continuously declining within a pre-set time period, it considers the current trading account's risk level to be showing a continuous deterioration or rapid decline, making the risk difficult to control. The system collects trend data from these continuous time series and dynamically triggers risk control alerts based on this data. Depending on the specific risk control alert, the system adaptively executes the corresponding risk control measures. The system employs several risk control measures. By analyzing the continuous decline in confidence levels over a preset time period, it can quickly detect the continuous deterioration or rapid decline in the risk level of a trading account. This real-time trend recognition helps to identify potential abnormal trading conditions or the impact of market fluctuations in advance, shifting risk management from a passive response to proactive early warning. This improves the foresight and agility of overall risk control. After acquiring trend data, the system can adaptively trigger corresponding risk control warnings based on different levels of risk severity and execute appropriate measures, such as automatic position reduction, position closing, risk alerts, or manual review. By associating continuously declining confidence levels with preset risk control rules, the system can achieve differentiated responses, ensuring that high-risk accounts are controlled in a timely manner while low-risk accounts are not excessively intervened in. This improves the targeting and efficiency of risk control operations. Furthermore, this dynamic risk control mechanism based on trend data not only guarantees real-time risk intervention but also effectively reduces potential account losses through adaptive execution of risk control measures. The system can adjust the intensity of measures according to trend strength, making the trading platform more robust and controllable in the face of rapid fluctuations or abnormal market conditions. It also provides data support and traceable records for subsequent strategy optimization and risk assessment.

[0479] In this embodiment, the generation module further includes:

[0480] The construction unit is used to construct an indicator mapping table for the indicator vector based on the preset data quality score of the transaction terminal, wherein the data quality score specifically includes data integrity, stability and anomaly frequency;

[0481] The fourth judgment unit is used to determine whether the vector structure of the index mapping table is stable;

[0482] The fourth execution unit is used to identify the indicator characteristics of the indicator vector based on the data quality score if no, and dynamically update the indicator mapping table based on the indicator characteristics. Specifically, the indicator characteristics include continuous, discrete, and categorical types, and the dynamic update specifically includes adding new indicators and discarding indicators.

[0483] In this embodiment, the system constructs an indicator mapping table of indicator vectors based on a pre-set data quality score on the trading terminal. The data quality score specifically includes data integrity, stability, and anomaly frequency. The system then determines whether the vector structure of this indicator mapping table is stable to execute corresponding steps. For example, if the system determines that the vector structure of the indicator mapping table is stable, it considers that the constructed indicator mapping table maintains the expected structural consistency within the current time period, and the mapping relationship between indicators is clear, without significant fluctuations or anomalies. The system uses this stable indicator mapping table to construct the indicator vector for the current moment, calculating the membership degree or confidence level of each risk level. Simultaneously, the indicator vector is input into a pre-trained inference model for comprehensive analysis, generating risk judgment results or trend data. The system also maintains continuous data quality monitoring to ensure that the indicator vector remains stable in subsequent time periods, ready to respond to anomalies. Conversely, if the system determines that the vector structure of the indicator mapping table is unstable, it considers that the mapping relationship between indicators has significant fluctuations or anomalies. The system identifies the indicator characteristics of each indicator vector based on different data quality scores. These indicator characteristics specifically include continuous, discrete, and categorical types. Based on these characteristics, the system dynamically updates the indicator. The mapping table is dynamically updated, specifically including the addition and obsolescence of indicators. By identifying the characteristics of each indicator (continuous, discrete, and categorical) and dynamically updating the mapping table, the system can promptly remove abnormal or invalid indicators and add new ones. This ensures that the indicator vector is always built on high-quality, reliable data. This process avoids risk assessment bias caused by indicator fluctuations or data anomalies, enhancing the completeness and reliability of data input. Furthermore, by dynamically updating the indicator mapping table, the system can adjust indicator usage strategies based on actual data quality, providing the risk assessment model with more stable input. For example, abnormal fluctuations in continuous indicators can be promptly detected. By identifying and adjusting weights, missing or anomaly-related discrete or categorical indicators can be replaced or marked, ensuring that the model is not affected by low-quality data when calculating comprehensive membership and confidence levels. This improves the accuracy and effectiveness of risk control decisions. Furthermore, dynamically updating the mapping table allows the system to quickly adapt to changes in the market environment or the introduction of new indicators. New indicators can be promptly incorporated into risk assessments, while invalid indicators are discarded to avoid interfering with decision-making. This mechanism enhances the robustness and adaptability of the trading platform, enabling the risk control system to operate stably in the long term and continuously provide reliable risk assessment and early warning capabilities, providing continuous data support for subsequent strategy optimization and model iteration.

[0484] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A market fluctuation early warning method based on multi-dimensional indicator fusion, characterized in that, Includes the following steps: Based on the multi-dimensional indicator data pre-collected by the trading terminal, the corresponding indicator types are classified from the multi-dimensional indicator data to generate the indicator vector at the current moment. Specifically, the indicator types include trading indicators, macroeconomic indicators and text sentiment indicators. Determine whether the timestamps of the indicator vectors are consistent; If so, then construct the fuzzy membership function corresponding to the indicator vector, input the fuzzy membership function into the pre-trained inference model, calculate the membership degree value of each risk level under the indicator vector, perform a weighted combination of the membership degree values ​​of the same risk level to obtain the comprehensive membership degree of the same risk level, compare the comprehensive membership degree, and select the membership degree value of the highest risk level as the transaction risk judgment result at the current moment. The fuzzy membership function is specifically a fuzzy interval corresponding to different risk fluctuations. Determine whether the confidence level of the transaction risk assessment result reaches a preset confidence level threshold; If the conditions are met, the preconditions preset by the trading terminal are identified. Based on the preconditions, a corresponding action set is constructed from the action matrix preset by the trading terminal. The action type in the trading process is generated through the action set. The approval response of the trading terminal is obtained. Based on the approval response, the action set is dynamically executed in the preset account. The preconditions specifically include trading time period, trading liquidity value and trading permissions. The action type specifically includes protection action, hedging strategy and limit adjustment.

2. The market fluctuation early warning method based on multi-dimensional indicator fusion according to claim 1, characterized in that, The step of calculating the membership value of each risk level under the indicator vector further includes: Read the indicator values ​​one by one from the indicator vector at the current moment and identify the preset risk range of the indicator values; Determine whether the indicator value falls within the intersection range of the two risk levels; If so, the index value is marked as a preset boundary state, the first-level membership degree and the second-level membership degree of the index value are retained, and based on the distance of the index value from the center of the cross interval, the first-level membership degree and the second-level membership degree are assigned different weights. The sum of the membership degrees of the first-level membership degree and the second-level membership degree is detected, and the total membership degree is dynamically adjusted according to the sum of the membership degrees.

3. The market fluctuation early warning method based on multi-dimensional indicator fusion according to claim 1, characterized in that, Before the step of constructing the fuzzy membership function corresponding to the index vector, the method further includes: Based on the risk assessment indicators preset by the trading terminal, the dimensional differences of the risk assessment indicators are obtained, wherein the risk assessment indicators specifically include price volatility, trading volume change rate, bid-ask spread and public sentiment index. Determine whether the dimensional differences are within a unified numerical range; If not, the numerical range of the risk assessment indicator is detected, the numerical span of the risk assessment indicator is identified based on the numerical range, the deviation difference of the risk assessment indicator is calculated based on the numerical span, and the weight ratio of the risk assessment indicator is dynamically adjusted based on the deviation difference.

4. The market fluctuation early warning method based on multi-dimensional indicator fusion according to claim 1, characterized in that, The step of constructing a corresponding action set from the preset action matrix of the trading terminal, and generating the action type in the trading process through the action set, further includes: Based on the preset triggering conditions of the action set, the real-time status information of the preset account during the transaction process is identified. The triggering conditions specifically include placing an order, canceling an order, modifying an order, and transferring account funds. The real-time status information specifically includes market conditions, account balance status, and order execution status. Determine whether the real-time status information meets the triggering condition; If so, then according to the risk control rules preset by the transaction terminal, the verification information of the action set is obtained, and the constraint actions of the action set are dynamically removed based on the verification information. Priority actions are divided in the action set, wherein the priority actions specifically include primary actions and secondary actions.

5. The market fluctuation early warning method based on multi-dimensional indicator fusion according to claim 1, characterized in that, The step of determining whether the timestamps of the indicator vectors are consistent further includes: Based on the generation period of the indicator vector, the acquisition delay of the indicator vector is obtained, wherein the generation period is specifically from the arrival time to the recording time; Determine whether the acquisition delay exceeds a preset delay threshold; If so, the sampling period of the indicator vector is identified, and the low-frequency data of the indicator vector is dynamically supplemented according to the sampling period. Based on the preset priority of the account, the confidence score of the indicator vector is generated.

6. The market fluctuation early warning method based on multi-dimensional indicator fusion according to claim 1, characterized in that, After the step of determining whether the confidence level of the transaction risk assessment result reaches the preset confidence level threshold, the method further includes: Based on the risk control measures for the preset account, a continuous time series of the confidence level is obtained, wherein the risk control measures specifically include automatic execution of position reduction and liquidation, manual review, and risk warning; Determine whether the continuous time series continues to decline within a preset time period; If so, then the trend data of the continuous time series is collected, and the risk control warning for the preset account is dynamically triggered based on the trend data. Based on the risk control warning, the risk control measures are adaptively executed.

7. The market fluctuation early warning method based on multi-dimensional indicator fusion according to claim 1, characterized in that, The step of classifying corresponding indicator types from the multi-dimensional indicator data pre-collected by the trading terminal and generating the indicator vector for the current moment further includes: Based on the data quality score preset by the trading terminal, an indicator mapping table for the indicator vector is constructed, wherein the data quality score specifically includes data integrity, stability, and anomaly frequency; Determine whether the vector structure of the index mapping table is stable; If not, then based on the data quality score, identify the indicator characteristics of the indicator vector, and dynamically update the indicator mapping table according to the indicator characteristics. Specifically, the indicator characteristics include continuous, discrete, and categorical types, and the dynamic update specifically includes adding new indicators and discarding indicators.

8. A market fluctuation early warning system based on multi-dimensional indicator fusion, characterized in that, include: The generation module is used to classify the corresponding indicator types from the multi-dimensional indicator data pre-collected by the trading terminal and generate the indicator vector at the current moment. Specifically, the indicator types include trading indicators, macroeconomic indicators and text sentiment indicators. The judgment module is used to determine whether the timestamps of the indicator vectors are consistent. The execution module is used to construct the fuzzy membership function corresponding to the indicator vector if the condition is met, input the fuzzy membership function into the pre-trained inference model, calculate the membership degree value of each risk level under the indicator vector, perform a weighted combination of the membership degree values ​​of the same risk level to obtain the comprehensive membership degree of the same risk level, compare the comprehensive membership degree, and select the membership degree value of the highest risk level as the transaction risk judgment result at the current moment. The fuzzy membership function is specifically a fuzzy interval corresponding to different risk fluctuations. The second judgment module is used to determine whether the confidence level of the transaction risk assessment result reaches a preset confidence level threshold. The second execution module is used to identify the preset preconditions of the trading terminal if the conditions are met, construct a corresponding action set from the preset action matrix of the trading terminal based on the preconditions, generate the action type in the trading process through the action set, obtain the approval response of the trading terminal, and dynamically execute the action set in the preset account based on the approval response. The preconditions specifically include trading time period, trading liquidity value and trading permissions, and the action type specifically includes protection action, hedging strategy and limit adjustment.

9. The market fluctuation early warning system based on multi-dimensional indicator fusion according to claim 8, characterized in that, The execution module further includes: The identification unit is used to read the indicator values ​​of the indicator vector at the current moment one by one and identify the preset risk range of the indicator values. A judgment unit is used to determine whether the indicator value is in the intersection range of two risk levels; The execution unit is configured to, if so, mark the index value as a preset boundary state, retain the first-level membership degree and the second-level membership degree of the index value, assign different weights to the first-level membership degree and the second-level membership degree based on the center position of the index value from the cross interval, detect the sum of the membership degrees of the first-level membership degree and the second-level membership degree, and dynamically adjust the total membership degree based on the sum of the membership degrees.

10. The market fluctuation early warning system based on multi-dimensional indicator fusion according to claim 8, characterized in that, Also includes: The acquisition module is used to acquire the dimensional differences of the risk assessment indicators based on the risk assessment indicators preset by the trading terminal. Specifically, the risk assessment indicators include price volatility, trading volume change rate, bid-ask spread, and public sentiment index. The third judgment module is used to determine whether the difference in dimensions is within a uniform numerical range; The third execution module is used to detect the numerical range of the risk assessment indicator if no, identify the numerical span of the risk assessment indicator based on the numerical range, calculate the deviation difference of the risk assessment indicator based on the numerical span, and dynamically adjust the weight ratio of the risk assessment indicator based on the deviation difference.

Images