Quantitative evaluation method for system application change level in securities industry

Abstract

The present invention discloses a method for quantitatively evaluating the level of application changes in securities industry systems, including: reading original change application data, historical change records and market transaction data, respectively generating change feature vectors, historical impact vectors and market feature vectors; performing time window segmentation and weight calculation on the market feature vectors to generate a time series feature matrix; performing nonlinear transformation on the change feature vectors and performing dependency analysis with the historical impact vectors to generate coupling feature tensors; decomposing the coupling feature tensors and performing feature mapping on the time series feature matrix to generate evaluation scores; performing cluster analysis based on the evaluation scores and historical impact vectors to generate level threshold vectors, and then obtaining change level results. The present invention realizes accurate quantitative evaluation of system change risks through dynamic fusion and adaptive optimization of multidimensional data, and improves the accuracy and reliability of evaluation.

Description

Technical Field

[0001] The invention belongs to the field of application change management, and in particular to a method for quantitatively evaluating the level of application changes in a securities industry system. Background Art

[0002] In the securities industry information system, the effectiveness and security of system change management are directly related to the stable operation of the financial market. As the core infrastructure of the financial market, the stability and reliability of the securities trading system are crucial. Therefore, scientific and accurate risk assessment of system changes and the establishment of a reasonable change level classification mechanism are of great significance to ensure the smooth operation of the securities market and prevent systemic risks. Especially in the current market environment, the securities trading system is becoming increasingly complex, the microservice architecture is widely used, and the dependencies between systems are complex, making the accuracy and timeliness of change risk assessment even more important.

[0003] At present, the risk assessment of system changes in the securities industry mainly adopts a combination of empirical assessment and static rules. The typical approach is to use a preset assessment form to score the change scope, the number of affected modules, the number of lines of code, and other dimensions, and then determine the change level based on the cumulative score. Some institutions have introduced static analysis tools based on code complexity to assist in the assessment by calculating indicators such as cyclomatic complexity and dependency. Some institutions also use simple historical data statistics to judge the risk of changes based on the success rate and rollback rate of historical changes. These methods have played a certain role in practice and provided a basic decision-making basis for system change management.

[0004] However, the existing evaluation methods have some obvious technical limitations: First, in terms of complexity evaluation, the existing methods often use the number of lines of code as the main indicator, ignoring deep-level features such as interface complexity and function coupling, resulting in oversimplification of the evaluation results; second, in terms of dependency analysis, the existing methods are mostly limited to the statistics of direct dependencies, and it is difficult to effectively identify and quantify the potential risks brought by indirect dependencies; third, in terms of market environment perception, the existing methods lack the ability to analyze market transaction characteristics in real time, and cannot dynamically adjust the risk assessment strategy according to the market status of different periods; in addition, in terms of historical data application, the existing methods mostly stay at a simple statistical level, and fail to deeply explore the pattern characteristics and risk laws contained in the historical change data; finally, in terms of evaluation threshold setting, the existing methods mostly use fixed threshold standards, lack an adaptive adjustment mechanism, and are difficult to adapt to the dynamic changes of system complexity and market environment. These technical problems seriously affect the accuracy and practicality of change risk assessment, and more advanced technical means need to be introduced to solve them. Summary of the invention

[0005] The purpose of the invention is to provide a method for quantitatively evaluating the level of changes in system application in the securities industry, in order to solve at least one technical problem existing in the prior art.

[0006] Technical solution, a quantitative assessment method for the change level of securities industry system application, including the following steps:

[0007] S1. Read the original change application data, extract key feature information and perform denoising to obtain the change feature vector; read the historical change records, perform statistical analysis to obtain the historical impact vector; read the market transaction data, perform time-division feature extraction to obtain the market feature vector;

[0008] S2. Based on the market feature vector, the time window is segmented to obtain a set of key time points; based on the set of key time points and the market feature vector, weight calculation is performed to obtain a time period weight vector; based on the time period weight vector and the market feature vector, a time series feature matrix is ​​obtained through matrix construction processing;

[0009] S3. Perform nonlinear transformation on the change feature vector to obtain a complexity vector; perform dependency analysis based on the change feature vector and the historical impact vector to obtain an impact range vector; perform comprehensive calculation based on the complexity vector, the impact range vector and the historical impact vector to obtain a risk vector; construct a coupling feature tensor based on the complexity vector, the impact range vector and the risk vector;

[0010] S4. Perform tensor decomposition on the coupling feature tensor to obtain a coupling coefficient matrix; process the coupling coefficient matrix and the timing feature matrix through feature mapping to obtain a timing coupling matrix; perform weighted calculation based on the timing coupling matrix and the time period weight vector to obtain an evaluation score;

[0011] S5. Based on the assessment score and the historical impact vector, a level threshold vector is obtained through cluster analysis. Based on the assessment score and the level threshold vector, threshold grading is performed to obtain a change level result. Based on the change level result, an assessment description document is generated.

[0012] Beneficial effect: The present invention establishes a complete evaluation framework including change characteristics, historical impacts and market environment, realizing dynamic and three-dimensional evaluation of change risks; at the same time, it not only overcomes the defects of traditional evaluation methods such as strong subjectivity and single dimension, but also improves the accuracy and reliability of change risk assessment through dynamic fusion and adaptive optimization of multi-dimensional features, realizes intelligent division of change levels, and provides strong technical support for the safe operation and maintenance of securities industry systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a flow chart of the present invention.

[0014] Figure 2 This is a flow chart of step S1 of the present invention.

[0015] Figure 3 This is a flow chart of step S2 of the present invention.

[0016] Figure 4 This is a flow chart of step S3 of the present invention.

[0017] Figure 5 This is a flow chart of step S4 of the present invention.

[0018] Figure 6 This is a flow chart of step S5 of the present invention. DETAILED DESCRIPTION

[0019] like Figure 1 As shown, this application proposes a quantitative assessment method for the level of application changes in securities industry systems, including the following steps:

[0020] S1. Read the original change application data, extract key feature information and perform denoising to obtain the change feature vector; read the historical change records, perform statistical analysis to obtain the historical impact vector; read the market transaction data, perform time-division feature extraction to obtain the market feature vector;

[0021] S2. Based on the market feature vector, the time window is segmented to obtain a set of key time points; based on the set of key time points and the market feature vector, weight calculation is performed to obtain a time period weight vector; based on the time period weight vector and the market feature vector, a time series feature matrix is ​​obtained through matrix construction processing;

[0022] S3. Perform nonlinear transformation on the change feature vector to obtain a complexity vector; perform dependency analysis based on the change feature vector and the historical impact vector to obtain an impact range vector; perform comprehensive calculation based on the complexity vector, the impact range vector and the historical impact vector to obtain a risk vector; construct a coupling feature tensor based on the complexity vector, the impact range vector and the risk vector;

[0023] S4. Perform tensor decomposition on the coupling feature tensor to obtain a coupling coefficient matrix; process the coupling coefficient matrix and the timing feature matrix through feature mapping to obtain a timing coupling matrix; perform weighted calculation based on the timing coupling matrix and the time period weight vector to obtain an evaluation score;

[0024] S5. Based on the assessment score and the historical impact vector, a level threshold vector is obtained through cluster analysis. Based on the assessment score and the level threshold vector, threshold grading is performed to obtain a change level result. Based on the change level result, an assessment description document is generated.

[0025] like Figure 2As shown, according to one aspect of the present application, step S1 is further:

[0026] S11. Read the original change application data, and obtain module complexity data through module complexity calculation; perform dependency analysis on the original change application data to obtain dependency data; perform code volume statistics on the original change application data to obtain code line data; construct a change feature vector based on the module complexity data, dependency data, and code line data;

[0027] S12, read the historical change records, and obtain the failure rate data through failure statistics analysis; perform performance impact assessment on the historical change records to obtain performance impact data; perform rollback statistics on the historical change records to obtain rollback rate data; construct a historical impact vector based on the failure rate data, performance impact data and rollback rate data;

[0028] S13. Read the market transaction data, and obtain the transaction density data by calculating the transaction density; perform volatility analysis on the market transaction data to obtain volatility index data; perform time period type identification on the market transaction data to obtain time period type data; construct a market feature vector based on the transaction density data, volatility index data and time period type data.

[0029] In one embodiment of the present application, the original change application data V (including change description, affected module, change type, etc.) is obtained, and key features are extracted, including the module complexity index mc, the number of dependencies dr, and the number of code lines cl, to obtain the change feature vector F_v = (mc, dr, cl); the historical change record H (including change results, impact scope, problem report, etc.) is obtained, the impact of historical changes is statistically analyzed, the failure rate fr, performance impact pi, and rollback rate rr are calculated, and the historical impact vector F_h = (fr, pi, rr) is obtained; the market transaction data M (including transaction volume, volatility, timestamp, etc.) is obtained, and the market characteristics of each time period are calculated, including the transaction density td, the volatility index vi, and the time period type tt, to obtain the market feature vector F_m = (td, vi, tt).

[0030] This embodiment achieves a comprehensive assessment of the risk of securities system changes through the collaborative processing of multi-dimensional data sources. Specifically, by extracting key features from the original change application data to construct a change feature vector, the complexity of the change itself is ensured to be accurately quantified; by statistically analyzing historical change records to construct a historical impact vector, a risk prediction model based on historical data is established; by extracting time-segment features of real-time market transaction data to construct a market feature vector, dynamic perception of the sensitivity of the market environment is achieved. This embodiment not only overcomes the problem that traditional single-dimensional evaluation methods are prone to bias, but also improves the accuracy and comprehensiveness of change risk assessment through cross-validation and complementarity of data, so that the system can make a more accurate assessment of change risks while considering multiple dimensions such as technical complexity, historical impact and market environment.

[0031] According to one aspect of the present application, step S11 is further:

[0032] S111. Read the original change application data, and obtain interface list data through interface analysis; perform function complexity calculation on the interface list data to obtain function complexity data; perform module aggregation on the function complexity data to obtain module aggregation data; and obtain module complexity data through weighted calculation based on the module aggregation data;

[0033] S112. Based on the original change application data, obtain module reference data through static scanning; perform call chain analysis on the module reference data to obtain call chain data; perform correlation calculation on the call chain data to obtain correlation data; perform graph structure analysis based on the correlation data to obtain dependency relationship data;

[0034] S113. Based on the original change application data, code type data is obtained through code classification processing; valid lines are extracted from the code type data to obtain valid line data; based on the valid line data, weighted line data is obtained through complexity weight processing; based on the weighted line data, summary statistics are performed to obtain code line number data;

[0035] S114. Normalize the module complexity data, dependency data and code line data respectively to obtain normalized complexity data, normalized dependency data and normalized line data; perform vector concatenation on the normalized complexity data, normalized dependency data and normalized line data to obtain a change feature vector.

[0036] In one embodiment of the present application, a function complexity calculation method is as follows: module complexity MC(m) = λ1·∑(CC(f) / CCmax) + λ2·∑(PC(f) / PCmax) + λ3·∑(HC(f) / HCmax); wherein CC(f) is the cyclomatic complexity of function f, and CCmax is the maximum cyclomatic complexity threshold; PC(f) is the parameter complexity of function f, which is equal to the sum of the input parameter quantity and the output parameter quantity, and PCmax is the maximum parameter complexity threshold; HC(f) is the hierarchical complexity of function f, which is equal to the function call hierarchy depth, and HCmax is the maximum hierarchical complexity threshold; λ1, λ2, and λ3 are weight coefficients and satisfy λ1+λ2+λ3=1; MC(m) is the comprehensive complexity value of module m, and its value range is [0, 1].

[0037] Call chain correlation calculation method: module correlation CR(i, j) = θ1·DC(i, j) + θ2·IC(i, j) +θ3·SC(i, j); DC(i, j) is the direct call correlation between modules i and j, which is equal to the number of direct calls divided by the maximum number of calls; IC(i, j) is the indirect call correlation, which is equal to exp(-k·L(i, j)), L(i, j) is the shortest call path length, and k is the attenuation coefficient; SC(i, j) is the shared resource correlation, which is equal to the number of shared resources divided by the total number of resources; θ1, θ2, and θ3 are weight coefficients and satisfy θ1+θ2+θ3=1; CR(i, j) is the inter-module correlation, with a value range of [0, 1].

[0038] This embodiment achieves accurate quantification of change complexity through multi-level code analysis and feature extraction. First, through interface parsing and function complexity calculation, a fine-grained analysis of the internal structure of the changed module is achieved, and the intrinsic complexity of the module is accurately assessed; secondly, through static scanning and call chain analysis, a complete module dependency graph is established, and the propagation path of the change impact is accurately identified; thirdly, through code classification and effective line extraction, differentiated treatment of different types of code is achieved, avoiding the evaluation bias that may be caused by simple code line statistics; finally, through the normalization and vector splicing of multi-dimensional features, a standardized change feature vector is constructed. This embodiment breaks through the limitations of traditional evaluation that relies solely on the number of code lines. Through a comprehensive analysis of code structure, dependencies and quality characteristics, a more objective and comprehensive evaluation of change complexity is achieved, providing reliable basic data for subsequent risk assessment.

[0039] According to one aspect of the present application, step S13 is further:

[0040] S131, reading market transaction data, classifying the market transaction data to obtain classified transaction data; based on the classified transaction data, dividing and processing by preset time windows to obtain window transaction data; performing frequency statistics on the window transaction data to obtain transaction frequency data; based on the transaction frequency data, calculating and obtaining transaction density data;

[0041] S132, based on the market transaction data, obtain price series data through price extraction; perform trend stripping on the price series data to obtain detrended data; based on the detrended data, calculate and obtain fluctuation range data; based on the fluctuation range data, perform index conversion to obtain fluctuation index data;

[0042] S133, based on the market transaction data, extracting time features to obtain time feature data; performing transaction volume analysis on the time feature data to obtain transaction volume feature data; performing pattern recognition on the transaction volume feature data to obtain transaction pattern data; performing type mapping processing based on the transaction pattern data to obtain time period type data;

[0043] S134. Based on the transaction density data, standardized density data is obtained through standardization processing; based on the volatility index data, standardized volatility data is obtained through standardization processing; based on the time period type data, coding type data is obtained through coding conversion processing; the standardized density data, standardized volatility data and coding type data are vectorized to obtain a market characteristic vector.

[0044] In one embodiment of the present application, the transaction density calculation method is: transaction density TD(t) = α·V(t) / Vmax+ β·F(t) / Fmax + γ·M(t) / Mmax; wherein V(t) is the transaction volume at time t, and Vmax is the historical maximum transaction volume; F(t) is the transaction frequency at time t, and Fmax is the historical maximum transaction frequency; M(t) is the matching quantity at time t, and Mmax is the historical maximum matching quantity; α, β, and γ are weight coefficients and satisfy α+β+γ=1, wherein α represents the transaction volume weight, β represents the transaction frequency weight, and γ represents the matching quantity weight; TD(t) is the normalized transaction density value, and the value range is [0, 1].

[0045] Volatility index calculation method: Market volatility index MV(t) = ω1·PV(t) / PVmax + ω2·TV(t) / TVmax + ω3·CV(t) / CVmax; where PV(t) is the price volatility at time t, equal to |P(t)-P(t-1)| / P(t-1), and PVmax is the maximum price volatility; TV(t) is the trading volume volatility, equal to |V(t)-V(t-1)| / V(t-1), and TVmax is the maximum trading volume volatility; CV(t) is the order cancellation rate, equal to the number of cancelled orders divided by the total number of orders, and CVmax is the maximum order cancellation rate; ω1, ω2, and ω3 are weight coefficients and satisfy ω1+ω2+ω3=1; MV(t) is a comprehensive volatility index with a value range of [0, 1].

[0046] This embodiment realizes the extraction of dynamic features of the market environment through refined market transaction data analysis. First, through transaction classification and time window division, it realizes the accurate identification and time series feature extraction of different types of transaction behaviors, and accurately grasps the dynamic characteristics of market transactions; secondly, through price series analysis and trend stripping, a quantitative model of volatility characteristics is established, and the volatility law of the market is accurately captured; thirdly, through time feature extraction and transaction pattern recognition, the intelligent classification of market status is realized, and the market characteristics of different time periods are accurately identified; finally, through standardization processing and vector combination, a standardized market feature vector is constructed. This embodiment overcomes the limitations of the traditional single indicator evaluation method, and through the comprehensive analysis of transaction density, price fluctuations and time period characteristics, it realizes a more accurate and comprehensive characterization of the market environment, providing a real-time market environment reference for change risk assessment.

[0047] like Figure 3 As shown, according to one aspect of the present application, step S2 is further:

[0048] S21. Based on the market feature vector, segmentation is performed through a preset fixed window to obtain time window data; based on the time window data, feature change rate data is calculated and obtained; threshold screening is performed on the feature change rate data to obtain a set of key time points;

[0049] S22, dividing the key time point set into time periods to obtain time period division data; calculating feature importance data based on the time period division data and the transaction density data, volatility index data and time period type data in the market feature vector; normalizing the feature importance data to obtain a time period weight vector;

[0050] S23. Based on the time period weight vector, matrix initialization is performed to obtain initial matrix data; based on the initial matrix data and the market characteristic vector, feature mapping is performed to obtain feature mapping data; matrix filling is performed on the feature mapping data to obtain a time series feature matrix.

[0051] In one embodiment of the present application, based on the market feature vector F_m, a sliding window method (w=30 minutes) is used to identify time points when features change significantly, and a set of key time points T_k is obtained; based on the key time point set T_k and the market feature vector F_m, a weight coefficient is calculated according to the feature importance of each time period, and a time period weight vector W_t is obtained; based on the time period weight vector W_t and the market feature vector F_m, a matrix reflecting the features of different time periods is constructed to obtain a time series feature matrix T_m.

[0052] This embodiment achieves a dynamic and accurate characterization of the market trading environment by performing time series processing on the market feature vector. The key time points of market fluctuations are accurately captured by using a 30-minute fixed window segmentation and feature change rate calculation method; a market feature quantitative model based on time period weights is established through a comprehensive analysis of transaction density, volatility indicators and time period types in different time periods; a time series feature matrix is ​​formed through matrix construction processing to achieve a continuous characterization of the market status. This embodiment effectively solves the problem that traditional static assessment methods cannot respond to market changes in a timely manner. Through real-time tracking of market status and dynamic weight adjustment, the system can adaptively adjust the change risk assessment strategy according to the market characteristics of different time periods, thereby improving the sensitivity and adaptability of the change risk assessment to changes in the market environment.

[0053] According to one aspect of the present application, step S21 is further:

[0054] S211. Based on the market feature vector, obtain segmented sequence data by performing sequence segmentation processing; extract statistical features from the segmented sequence data to obtain statistical feature data; perform trend analysis on the statistical feature data to obtain trend data; perform window division based on the trend data to obtain time window data;

[0055] S212, performing differential calculation on the time window data to obtain differential feature data; based on the differential feature data, calculating the change rate data; based on the change rate data, calculating the acceleration data; based on the acceleration data, obtaining the characteristic change rate data through comprehensive processing of the change rate;

[0056] S213. Based on the feature change rate data, dynamic threshold data is obtained through dynamic threshold calculation; based on the dynamic threshold data, anomaly point detection is performed to obtain anomaly point data; time point verification is performed on the anomaly point data to obtain verification time data; based on the verification time data, a key time point set is constructed.

[0057] In one embodiment of the present application, the trend analysis method is as follows: characteristic trend degree TF(t) = Δ1·MA(t) + Δ2·ACC(t) + Δ3·RSI(t); wherein MA(t) is a moving average trend factor, which is equal to (SMA(t)-LMA(t)) / LMA(t), SMA is a short-term average, and LMA is a long-term average; ACC(t) is an acceleration factor, which is equal to (V(t)-2V(t-1)+V(t-2)) / V(t-2); RSI(t) is a relative strength factor, which is equal to U(t) / (U(t)+D(t)), U(t) is an increase, and D(t) is a decrease; Δ1, Δ2, and Δ3 are weight coefficients and satisfy Δ1+Δ2+Δ3=1; TF(t) is a trend strength indicator, and its value range is [-1, 1].

[0058] Calculation method of characteristic change rate: change rate VR(t) = μ1·DR(t) + μ2·AR(t) + μ3·SR(t); where DR(t) is the direction change rate, which is equal to the sign function sign(F(t)-F(t-1)); AR(t) is the amplitude change rate, which is equal to |F(t)-F(t-1)| / F(t-1); SR(t) is the morphology change rate, which is equal to 1-cos(F(t), F(t-1)); F(t) is the characteristic vector; μ1, μ2, and μ3 are weight coefficients and satisfy μ1+μ2+μ3=1; VR(t) is the comprehensive change rate, and its value range is [0, 2].

[0059] This embodiment achieves accurate identification of key market time points through adaptive time window analysis. First, through sequence segmentation and statistical feature extraction, a time series feature model of market data is established to accurately capture the time series laws of market changes; secondly, through differential calculation and change rate analysis, accurate quantification of market feature change trends is achieved, and mutation points of market status are accurately identified; thirdly, through acceleration calculation and comprehensive processing of change rate, a dynamic change model of market characteristics is established to accurately depict the acceleration characteristics of market changes; finally, through dynamic threshold calculation and outlier detection, adaptive identification of key time points is achieved. This embodiment breaks through the limitations of traditional fixed time window analysis, and through dynamic tracking and intelligent identification of changes in market characteristics, it achieves more accurate and timely capture of key market time points, providing a reliable basis for the selection of change time windows.

[0060] like Figure 4 As shown, according to one aspect of the present application, step S3 is further:

[0061] S31, based on the changed feature vector, perform feature normalization to obtain normalized feature data; perform weight assignment on the normalized feature data to obtain weighted feature data; based on the weighted feature data, obtain a complexity vector through nonlinear mapping;

[0062] S32, based on the change feature vector, perform module dependency analysis to obtain dependency relationship data; based on the dependency relationship data and the historical impact vector, perform propagation path analysis to obtain impact path data; perform range quantification on the impact path data to obtain an impact range vector;

[0063] S33, based on the complexity vector, through weight configuration processing, obtain complexity weight data; based on the impact range vector, through weight configuration processing, obtain range weight data; based on the historical impact vector, through weight configuration processing, obtain historical weight data; weighted fusion of the complexity weight data, range weight data and historical weight data to obtain a risk vector;

[0064] S34. Based on the complexity vector, the extended complexity data is obtained through dimensional expansion; based on the impact range vector, the extended range data is obtained through dimensional expansion; based on the risk vector, the extended risk data is obtained through dimensional expansion; based on the extended complexity data, the extended range data and the extended risk data, a coupling feature tensor is constructed.

[0065] In one embodiment of the present application, based on the change feature vector F_v, the complexity score is calculated through nonlinear transformation to obtain the complexity vector C_v; based on the change feature vector F_v and the historical impact vector F_h, the impact range is calculated through the module dependency graph to obtain the impact range vector I_v; based on the complexity vector C_v, the impact range vector I_v and the historical impact vector F_h, the risk index is comprehensively calculated to obtain the risk vector R_v; based on the complexity vector C_v, the impact range vector I_v and the risk vector R_v, a three-dimensional feature tensor is constructed to obtain the coupling feature tensor G_t.

[0066] This embodiment realizes a three-dimensional assessment of the complexity, scope of impact, and risk of changes through the construction of a multi-level feature tensor. The inherent complexity of the change is accurately quantified through nonlinear transformation processing of the change feature vector; the scope of impact of the change is accurately assessed through the combination of dependency analysis and historical impact vectors; the risk vector is formed through comprehensive calculation, and the multi-dimensional quantification of the change risk is realized; finally, a coupled feature tensor is formed through tensor construction, and the correlation between the features of each dimension of the change is established. This embodiment breaks through the limitations of the traditional linear assessment model. Through the multi-dimensional coupling analysis of the change features, it can not only fully reflect the complexity and risk of the change, but also reveal the potential correlation between different features, thereby providing a more reliable basis for the accurate assessment of the change risk.

[0067] According to one aspect of the present application, step S31 is further:

[0068] S311, based on the changed feature vector, perform extreme value detection processing to obtain feature extreme value data; based on the feature extreme value data, perform outlier replacement to obtain cleaned feature data; perform interval mapping on the cleaned feature data to obtain interval data; perform normalization processing on the interval data to obtain normalized feature data;

[0069] S312, based on the normalized feature data, calculate feature importance data; perform correlation analysis on the feature importance data to obtain correlation data; based on the correlation data, perform weight optimization to obtain optimized weight data; based on the optimized weight data, perform feature weighting processing to obtain weighted feature data;

[0070] S313, performing feature distribution analysis on the weighted feature data to obtain distribution feature data; processing the distribution feature data through kernel function transformation to obtain transformed feature data; based on the transformed feature data, performing mapping function optimization processing to obtain optimized mapping data; performing nonlinear transformation on the optimized mapping data to obtain a complexity vector.

[0071] In one embodiment of the present application, a feature normalization calculation method is provided: normalized feature NF(x) = ρ1·LP(x)+ ρ2·GP(x) + ρ3·NP(x); wherein LP(x) is a local normalization component, which is equal to (x-min_l) / (max_l-min_l), and min_l and max_l are local minimum and maximum values; GP(x) is a global normalization component, which is equal to (x-min_g) / (max_g-min_g), and min_g and max_g are global minimum and maximum values; NP(x) is a normalization component, which is equal to (x-μ) / (3σ), and μ is the mean and σ is the standard deviation; ρ1, ρ2, and ρ3 are weight coefficients and satisfy ρ1+ρ2+ρ3=1; NF(x) is the normalized feature value, and its value range is [0, 1].

[0072] This embodiment achieves refined quantification of change complexity through nonlinear feature transformation. First, through extreme value detection and outlier replacement, a cleaning mechanism for feature data is established to ensure the reliability of data quality; secondly, through feature importance calculation and correlation analysis, the optimal allocation of feature weights is achieved, accurately reflecting the importance of different features; thirdly, through kernel function transformation and mapping function optimization, a nonlinear transformation model of features is established to accurately characterize the nonlinear relationship between features; finally, through the comprehensive processing of multi-dimensional features, a standardized complexity vector is constructed. This embodiment overcomes the limitations of traditional linear evaluation methods, and through nonlinear transformation and optimal combination of feature data, a more accurate and in-depth characterization of change complexity is achieved, providing a more reliable complexity measurement for risk assessment.

[0073] According to one aspect of the present application, step S32 is further:

[0074] S321. Read the change feature vector, and obtain the module relationship data through module relationship analysis; obtain the direct dependency data through direct dependency extraction; obtain the indirect dependency data through indirect dependency deduction; obtain the dependency network data through dependency network construction of the direct dependency data and the indirect dependency data.

[0075] S322, read the dependent network data, and obtain the node importance data through node importance calculation processing; read the historical influence vector, and obtain the propagation weight data through propagation weight calculation processing based on the node importance data; calculate the propagation weight data through path strength calculation processing to obtain path strength data; and obtain the influence path data through propagation path identification processing on the path strength data.

[0076] S323, read the impact path data, and obtain the path hierarchy data through path layering processing; the path hierarchy data is processed through impact attenuation calculation to obtain the impact attenuation data; the impact attenuation data is processed through range accumulation to obtain the range accumulation data; the range accumulation data is processed through vectorization to obtain the impact range vector.

[0077] In one embodiment of the present application, a dependency network construction method is as follows: dependency strength DS(i, j) = φ1·CD(i, j) + φ2·ID(i, j) + φ3·TD(i, j); wherein CD(i, j) is the code dependency strength, which is equal to the number of common code lines divided by the total number of code lines; ID(i, j) is the interface dependency strength, which is equal to the number of shared interfaces divided by the total number of interfaces; TD(i, j) is the data dependency strength, which is equal to the number of shared table fields divided by the total number of fields; φ1, φ2, and φ3 are weight coefficients and satisfy φ1+φ2+φ3=1; DS(i, j) is the inter-module dependency strength, and its value range is [0, 1].

[0078] Propagation weight calculation method: Propagation weight PW(p) = η1·PL(p) + η2·PS(p) + η3·PI(p); where PL(p) is the path length weight, equal to exp(-α·L(p)), L(p) is the length of path p, and α is the attenuation coefficient of the path length; PS(p) is the path strength weight, equal to ∏DS(i, j), DS(i, j) is the dependency strength between adjacent nodes on the path, and ∏ is the product operator; PI(p) is the path importance weight, equal to ∑Node_imp(i), Node_imp(i) is the importance score of the node on the path; η1, η2, η3 are weight coefficients and satisfy η1+η2+η3=1; PW(p) is the propagation weight of the path, and its value range is [0, 1].

[0079] This embodiment achieves accurate assessment of the scope of change impact through multi-level dependency analysis. First, through module relationship analysis and dependency extraction, a complete dependency network model is established to accurately identify the propagation path of the change impact; secondly, through node importance calculation and propagation weight analysis, a quantitative assessment of the degree of impact is achieved, and the impact intensity of different propagation paths is accurately calculated; thirdly, through path stratification and impact attenuation calculation, an attenuation model of impact propagation is established, which accurately simulates the attenuation law of influence with propagation distance; finally, through range accumulation and vectorization processing, a standardized impact range vector is constructed. This embodiment breaks through the limitations of traditional simple dependency statistics, and through in-depth analysis of dependency relationships and precise simulation of impact propagation, it achieves a more accurate and comprehensive assessment of the scope of change impact, providing a reliable decision-making basis for risk control.

[0080] According to one aspect of the present application, step S33 is further:

[0081] S331, read the complexity vector, and obtain complexity hierarchical data through hierarchical decomposition processing; process the complexity hierarchical data through sensitivity analysis to obtain complexity sensitive data; process the complexity sensitive data through weight learning to obtain complexity learning data; process the complexity learning data through weight optimization to obtain complexity weight data.

[0082] S332, read the influence range vector, obtain range level data through range stratification processing; obtain propagation path data through propagation path analysis processing on the range level data; obtain influence quantification data through influence quantification processing on the propagation path data; obtain range weight data through weight allocation processing on the influence quantification data.

[0083] S333, read the historical impact vector, and obtain the time series feature data through time series decomposition processing; process the time series feature data through pattern matching to obtain pattern matching data; process the pattern matching data through similarity calculation to obtain similarity data; process the similarity data through weight generation to obtain historical weight data.

[0084] S334, read the complexity weight data, and obtain the complexity combination data through feature combination processing; read the range weight data, and obtain the range combination data through feature combination processing; read the historical weight data, and obtain the historical combination data through feature combination processing; combine the complexity combination data, range combination data and historical combination data through multi-dimensional fusion processing to obtain the risk vector.

[0085] In one embodiment of the present application, the complexity weight learning method is: complexity weight CW(t) = ε1·HC(t)+ ε2·TC(t) + ε3·FC(t); wherein HC(t) is the historical correlation weight, equal to exp(-β·Δt), Δt is the historical time difference; TC(t) is the technical correlation weight, equal to cos(V(t), V_h), V(t) is the current feature vector, V_h is the historical feature vector; FC(t) is the fault correlation weight, equal to (1-FR(t)), FR(t) is the historical failure rate; ε1, ε2, ε3 are weight coefficients and satisfy ε1+ε2+ε3=1; CW(t) is the dynamic weight of complexity, and its value range is [0, 1].

[0086] This embodiment realizes the dynamic calculation of change risk through multi-dimensional weight fusion. First, through complexity hierarchical decomposition and sensitivity analysis, a hierarchical evaluation system of complexity features is established, and the sensitivity of features at different levels is accurately quantified; secondly, through scope stratification and propagation path analysis, a refined evaluation of the impact scope is achieved, and the impact weights of different propagation paths are accurately calculated; thirdly, through the time series decomposition and pattern matching of historical data, a weight learning model based on historical experience is established, and the key features of historical changes are accurately extracted; finally, through the combined fusion of multi-dimensional features, a dynamic risk vector is constructed. This embodiment overcomes the limitations of traditional single-dimensional evaluation, and through the dynamic fusion of complexity, impact scope and historical experience, a more comprehensive and accurate evaluation of change risk is achieved, providing a reliable quantitative basis for change decisions.

[0087] like Figure 5 As shown, according to one aspect of the present application, step S4 is further:

[0088] S41, based on the coupling characteristic tensor, obtaining expanded characteristic data through tensor expansion processing; performing characteristic decomposition on the expanded characteristic data to obtain decomposed characteristic data; performing coefficient reconstruction on the decomposed characteristic data to obtain a coupling coefficient matrix;

[0089] S42, based on the coupling coefficient matrix, feature extraction is performed to obtain coupling feature data; based on the coupling feature data and the timing feature matrix, mapping feature data is obtained through timing mapping processing; the mapping feature data is matrix reconstructed to obtain a timing coupling matrix;

[0090] S43. Based on the time series coupling matrix, feature aggregation is performed to obtain aggregated feature data; based on the aggregated feature data and the time period weight vector, weighted calculation is performed to obtain weighted feature data; the weighted feature data is normalized to obtain an evaluation score.

[0091] In one embodiment of the present application, based on the coupling feature tensor G_t, tensor decomposition is used to calculate the coupling coefficient between dimensions to obtain the coupling coefficient matrix K_m; based on the coupling coefficient matrix K_m and the timing feature matrix T_m, the coupling strength of different time periods is calculated to obtain the timing coupling matrix D_m; based on the timing coupling matrix D_m and the time period weight vector W_t, the final evaluation score is weighted and calculated to obtain the evaluation score S_f.

[0092] This embodiment achieves refined processing of time-series coupling assessment through tensor decomposition and feature mapping. By decomposing the coupling feature tensor, the essential correlation between the change features is extracted; the coupling coefficient matrix is ​​associated with the time-series feature matrix through feature mapping, realizing the dynamic fusion of change risk assessment and market time-series characteristics; the final evaluation score is formed through weighted calculation, and a dynamic evaluation system that comprehensively considers the complexity of the change, the scope of impact and the market environment is established. This embodiment effectively solves the problem that traditional evaluation methods are difficult to take into account both the intrinsic characteristics of the change and the external environment. Through dynamic feature mapping and weighted calculation, the dual sensitivity of the evaluation results to the change characteristics and the changes in the market environment is realized, which improves the accuracy and practicality of the evaluation results.

[0093] According to one aspect of the present application, step S41 is further:

[0094] S411, read the coupled feature tensor, and obtain the rearranged tensor data through dimension rearrangement processing; the rearranged tensor data is processed through modal expansion to obtain modal data; the modal data is processed through matrix transformation to obtain transformed matrix data; the transformed matrix data is processed through feature recombination to obtain expanded feature data.

[0095] S412, read the expanded feature data, and obtain orthogonal feature data through feature orthogonalization processing; obtain core feature data through core feature extraction processing on the orthogonal feature data; obtain feature component data through rank decomposition processing on the core feature data; obtain decomposed feature data through importance sorting processing on the feature component data.

[0096] S413, read the decomposed feature data, and obtain the filtered feature data through feature screening processing; obtain the correlation data through correlation calculation processing on the filtered feature data; obtain the coupling coefficient data through coefficient generation processing on the correlation data; obtain the coupling coefficient matrix through matrix reconstruction processing on the coupling coefficient data.

[0097] In one embodiment of the present application, the tensor expansion processing method is: feature importance IE(i) = ξ1·VC(i) +ξ2·CC(i) + ξ3·DC(i); wherein VC(i) is the variance contribution rate of feature i, equal to λi / ∑λj, λi is the eigenvalue corresponding to feature i; CC(i) is the correlation contribution of feature i, equal to |∑r_ij| / n, r_ij is the correlation coefficient between feature i and other features j, and n is the total number of features; DC(i) is the discriminative contribution of feature i, equal to 1-H(i) / log(n), H(i) is the information entropy of feature i; ξ1, ξ2, ξ3 are weight coefficients and satisfy ξ1+ξ2+ξ3=1; IE(i) is the importance of feature i, and its value range is [0, 1].

[0098] Feature orthogonalization processing method: Orthogonalized feature OF(x) = σ1·GS(x) + σ2·PCA(x) + σ3·ICA(x); where GS(x) is the Gram-Schmidt orthogonalized component, equal to x - ∑(proj_u(x)), proj_u(x) is the projection of the input feature vector x on the orthogonalized feature; PCA(x) is the principal component analysis component, equal to W T ·x, W is the eigenvector matrix; ICA(x) is the independent component, which is equal to S·x, S is the separation matrix; σ1, σ2, σ3 are weight coefficients and satisfy σ1+σ2+σ3=1; OF(x) is the orthogonalized feature, and the degree of orthogonalization ranges from [0, 1].

[0099] This embodiment realizes the refined extraction of coupling features through high-dimensional tensor decomposition. First, through dimensional rearrangement and modal expansion, a standardized processing flow of tensor data is established, and the effective expansion of high-dimensional features is realized; secondly, through feature orthogonalization and core feature extraction, an optimization model for feature dimensionality reduction is constructed, and key coupling features are accurately extracted; thirdly, through rank decomposition and importance sorting, a quantitative evaluation of feature importance is realized, and the dominant coupling relationship is accurately identified; finally, through correlation calculation and coefficient generation, a standardized coupling coefficient matrix is ​​constructed. This embodiment breaks through the limitations of traditional matrix analysis, and through the systematic decomposition and optimized reconstruction of high-dimensional features, a more in-depth and accurate characterization of the coupling relationship of changing features is achieved, providing a reliable feature basis for subsequent time series mapping.

[0100] According to one aspect of the present application, step S42 is further:

[0101] S421, read the coupling coefficient matrix, and obtain decomposed characteristic data through feature decomposition processing; the decomposed characteristic data are processed through principal component analysis to obtain principal component data; the principal component data are processed through feature selection to obtain selected characteristic data; the selected characteristic data are processed through feature recombination to obtain coupling characteristic data.

[0102] S422, read the time series feature matrix, and obtain the time series component data through time series decomposition processing; read the coupling feature data, and obtain the corresponding feature data through feature correspondence processing based on the time series component data; obtain the mapping feature data through mapping transformation processing on the corresponding feature data.

[0103] S423, read the mapping feature data, and obtain the alignment feature data through dimension alignment processing; the alignment feature data is processed through matrix construction to obtain the construction matrix data; the construction matrix data is processed through consistency verification to obtain the verification matrix data; the verification matrix data is processed through optimization reconstruction to obtain the timing coupling matrix.

[0104] In one embodiment of the present application, the principal component analysis method: feature reconstruction value RC(x) = γ1·EV(x) +γ2·CV(x) + γ3·RV(x); wherein EV(x) is the eigenvalue contribution, equal to λi / ∑λj, λi is the i-th eigenvalue; CV(x) is the cumulative variance contribution, equal to ∑λk / ∑λj, k is the first n principal components; RV(x) is the reconstruction error, equal to ||x - x'|| / ||x||, x' is the reconstructed feature; γ1, γ2, γ3 are weight coefficients and satisfy γ1+γ2+γ3=1; RC(x) is the feature reconstruction quality, and its value range is [0, 1].

[0105] This embodiment realizes the dynamic evaluation of coupling relationship through time series feature mapping. First, through feature decomposition and principal component analysis, a dimensionality reduction model of coupling features is established, and key coupling modes are accurately extracted; secondly, through time series decomposition and feature correspondence, dynamic mapping of coupling features and time series features is realized, and the time-varying characteristics of coupling relationship are accurately captured; thirdly, through dimension alignment and matrix construction, a standardized processing flow of feature mapping is established to ensure the consistency of mapping results; finally, through consistency verification and optimized reconstruction, a reliable time series coupling matrix is ​​constructed. This embodiment overcomes the limitations of traditional static analysis, and through dynamic mapping and optimized reconstruction of coupling features, it realizes the time series dynamic evaluation of change risks, and provides a reference basis for change decisions in the time series dimension.

[0106] like Figure 6 As shown, according to one aspect of the present application, step S5 is further:

[0107] S51, based on the evaluation score, score distribution data is obtained through distribution analysis; based on the historical impact vector, feature clustering is performed to obtain cluster feature data; based on the score distribution data and the cluster feature data, a level threshold vector is obtained through threshold optimization processing;

[0108] S52, based on the evaluation score, perform numerical comparison to obtain comparison result data; based on the comparison result data and the level threshold vector, perform level mapping to obtain mapping level data; confirm the mapping level data to obtain a level change result;

[0109] S53. Analyze the change level results to obtain analysis result data; based on the analysis result data, obtain key feature data through feature extraction; based on the analysis result data and the key feature data, obtain an assessment description document through document generation processing.

[0110] In one embodiment of the present application, based on the evaluation score S_f and the historical impact vector F_h, adaptive clustering is used to determine the level threshold to obtain the level threshold vector L_t; based on the evaluation score S_f and the level threshold vector L_t, the final change level is determined to obtain the change level result L_r; based on the change level result L_r and all intermediate calculation results, a detailed description of the evaluation basis is generated to obtain the evaluation description document D_e.

[0111] This embodiment achieves intelligent division of change levels through adaptive threshold optimization. A correlation model between evaluation scores and historical impacts is established through cluster analysis, and the level threshold vector is dynamically optimized; adaptive division of change levels is achieved through threshold grading; and a complete evaluation description document is formed through description generation. This embodiment overcomes the rigidity of the traditional fixed threshold division method. Through dynamic learning of historical data and threshold optimization, the system can automatically adjust the grading standards according to historical experience, which not only improves the rationality of the grading results, but also provides detailed basis support for change decisions through the automatic generation of evaluation descriptions.

[0112] According to one aspect of the present application, step S51 is further:

[0113] S511, read the evaluation score, and obtain the quantile data through quantile calculation processing; obtain the density distribution data through density estimation processing on the quantile data; obtain the peak data through peak detection processing on the density distribution data; obtain the score distribution data through distribution fitting processing on the peak data.

[0114] S512, read the historical influence vector, and obtain reduced dimension feature data through feature dimension reduction processing; obtain distance matrix data through distance calculation processing on the reduced dimension feature data; obtain cluster center data through cluster center identification processing on the distance matrix data; obtain cluster center data through feature classification processing on the cluster center data.

[0115] S513, read the score distribution data, and obtain boundary feature data through boundary extraction processing; perform threshold initialization processing on the boundary feature data to obtain initial threshold data; read the cluster feature data, and obtain optimized threshold data through threshold tuning processing based on the initial threshold data; perform vector construction processing on the optimized threshold data to obtain a level threshold vector.

[0116] In one embodiment of the present application, the density distribution calculation method is: score density density(s) = κ1·KD(s) + κ2·HD(s) + κ3·GD(s); KD(s) is the kernel density estimate, which is equal to (1 / nh)∑K((s-si) / h), K is the kernel function, h is the bandwidth, s is the score, and si is the score of the i-th sample; HD(s) is the histogram density estimate, which is equal to ni / (n·Δs), ni is the number of samples in the interval, n is the total number of samples, and Δs is the interval width; GD(s) is the Gaussian mixture density estimate, which is equal to ∑πi·N(s|μi,σi), πi is the mixing coefficient, μi is the center of the i-th class, and σi is the standard deviation of the i-th class; κ1, κ2, and κ3 are weight coefficients and satisfy κ1+κ2+κ3=1; density(s) is the density estimate at score s, and after normalization, it takes the value range [0, 1].

[0117] Cluster feature extraction method: Cluster feature CF(c) = τ1·CS(c) + τ2·CH(c) + τ3·DB(c); where CS(c) is the intra-class compactness, equal to 1 / (1+∑||x-μc|| 2 / nc), μc is the class center, nc is the number of class samples; CH(c) is the inter-class separation, which is equal to min{||μi-μj|| 2}, μi, μj are centers of different classes; DB(c) is the Davies-Bouldin index, which is equal to (1 / n)∑max{(σi+σj) / d(μi, μj)}, σ is the standard deviation within the class, and d is the distance between classes; τ1, τ2, τ3 are weight coefficients and satisfy τ1+τ2+τ3=1; CF(c) is the feature quality evaluation of cluster c, and its value range is [0, 1].

[0118] This embodiment realizes dynamic optimization of level thresholds through adaptive cluster analysis. First, through quantile calculation and density estimation, a distribution model of evaluation scores is established, accurately describing the statistical characteristics of scores; secondly, through feature dimension reduction and distance calculation, cluster analysis of historical data is realized, and characteristic patterns of different risk levels are accurately identified; thirdly, through boundary extraction and threshold initialization, a benchmark model for level division is established, providing an initial reference for threshold optimization; finally, through threshold tuning and vector construction, adaptive optimization of level thresholds is realized. This embodiment breaks through the limitations of traditional fixed thresholds, and realizes adaptive optimization of change level division standards through dynamic learning of historical data and threshold adjustment, providing a reliable quantitative basis for the accurate division of change levels.

[0119] The present invention realizes the accurate quantitative evaluation of system application changes in the securities industry by constructing a dynamic evaluation system driven by multi-dimensional data. First, through the collaborative processing of multi-source data, a complete evaluation framework including change characteristics, historical impacts and market environment is established; secondly, through time series feature analysis and tensor coupling modeling, a dynamic and three-dimensional evaluation of change risks is realized; finally, through adaptive threshold optimization, intelligent division of change levels is realized. The present invention not only overcomes the defects of traditional evaluation methods that are highly subjective and have a single dimension, but also improves the accuracy and reliability of change risk assessment through dynamic fusion and adaptive optimization of multi-dimensional features, providing strong technical support for the safe operation and maintenance of securities industry systems.

[0120] The preferred embodiments of the present invention are described in detail above; however, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all belong to the protection scope of the present invention.

Claims

1. A quantitative assessment method for the level of changes in system application in the securities industry, characterized in that: The steps include: S1. Read the original change application data, extract key feature information and perform denoising to obtain the change feature vector; read the historical change records, perform statistical analysis to obtain the historical impact vector; read the market transaction data, perform time-division feature extraction to obtain the market feature vector; S2. Based on the market feature vector, the time window is segmented to obtain a set of key time points; based on the set of key time points and the market feature vector, weight calculation is performed to obtain a time period weight vector; based on the time period weight vector and the market feature vector, a time series feature matrix is ​​obtained through matrix construction processing; S3. Perform nonlinear transformation on the change feature vector to obtain a complexity vector; perform dependency analysis based on the change feature vector and the historical impact vector to obtain an impact range vector; perform comprehensive calculation based on the complexity vector, the impact range vector and the historical impact vector to obtain a risk vector; Based on the complexity vector, impact range vector and risk vector, a coupling feature tensor is constructed; S4, performing tensor decomposition on the coupling feature tensor to obtain a coupling coefficient matrix; processing the coupling coefficient matrix and the timing feature matrix through feature mapping to obtain a timing coupling matrix; Based on the time series coupling matrix and the time period weight vector, a weighted calculation is performed to obtain an evaluation score; S5. Based on the assessment score and the historical impact vector, a level threshold vector is obtained through cluster analysis; based on the assessment score and the level threshold vector, threshold classification is performed to obtain a change level result; based on the change level result, an assessment description document is generated; Step S4 is further as follows: S41, based on the coupled characteristic tensor, obtaining expanded characteristic data through tensor expansion processing; performing characteristic decomposition on the expanded characteristic data to obtain decomposed characteristic data; Reconstruct the coefficients of the decomposed characteristic data to obtain the coupling coefficient matrix; S42, performing feature extraction based on the coupling coefficient matrix to obtain coupling feature data; Based on the coupling characteristic data and the timing characteristic matrix, mapping characteristic data is obtained through timing mapping processing; Reconstruct the matrix of the mapped feature data to obtain the time series coupling matrix; S43, performing feature aggregation based on the temporal coupling matrix to obtain aggregated feature data; Based on the aggregated feature data and the time period weight vector, weighted calculation is performed to obtain weighted feature data; the weighted feature data is normalized to obtain an evaluation score.

2. The method for quantitatively evaluating the change level of securities industry system application according to claim 1 is characterized in that: Step S1 is further as follows: S11. Read the original change application data, and obtain module complexity data through module complexity calculation; perform dependency analysis on the original change application data to obtain dependency data; Perform code volume statistics on the original change application data to obtain the number of code lines; Construct a change feature vector based on module complexity data, dependency data, and code line data; S12, read the historical change records, and obtain the failure rate data through failure statistics analysis; perform performance impact assessment on the historical change records to obtain performance impact data; perform rollback statistics on the historical change records to obtain rollback rate data; construct a historical impact vector based on the failure rate data, performance impact data and rollback rate data; S13, reading market transaction data, and obtaining transaction density data through transaction density calculation; performing volatility analysis on the market transaction data to obtain volatility index data; performing time period type identification on the market transaction data to obtain time period type data; Based on transaction density data, volatility index data and time period type data, a market feature vector is constructed.

3. The method for quantitatively evaluating the level of application changes in securities industry systems according to claim 2 is characterized in that: Step S2 is further as follows: S21. Based on the market feature vector, segmentation is performed through a preset fixed window to obtain time window data; based on the time window data, characteristic change rate data is calculated; Threshold screening is performed on the feature change rate data to obtain a set of key time points; S22, dividing the key time point set into time periods to obtain time period division data; Based on the time period division data and the transaction density data, volatility index data and time period type data in the market feature vector, the feature importance data is calculated; Normalize the feature importance data to obtain the time period weight vector; S23, based on the time period weight vector, perform matrix initialization to obtain initial matrix data; based on the initial matrix data and the market characteristic vector, perform feature mapping to obtain feature mapping data; Perform matrix filling on the feature map data to obtain the time series feature matrix.

4. The method for quantitatively evaluating the level of application changes in securities industry systems according to claim 3 is characterized in that: Step S3 is further as follows: S31, based on the changed feature vector, perform feature normalization to obtain normalized feature data; perform weight assignment on the normalized feature data to obtain weighted feature data; based on the weighted feature data, obtain a complexity vector through nonlinear mapping; S32, performing module dependency analysis based on the change feature vector to obtain dependency relationship data; Based on dependency data and historical impact vectors, influence path data is obtained through propagation path analysis; the impact path data is range-quantified to obtain an impact range vector; S33, based on the complexity vector, through weight configuration processing, obtain complexity weight data; based on the impact range vector, through weight configuration processing, obtain range weight data; based on the historical impact vector, through weight configuration processing, obtain historical weight data; weighted fusion of the complexity weight data, range weight data and historical weight data to obtain a risk vector; S34, based on the complexity vector, by dimensional expansion, obtain extended complexity data; based on the impact range vector, by dimensional expansion, obtain extended range data; based on the risk vector, by dimensional expansion, obtain extended risk data; A coupling feature tensor is constructed based on the extended complexity data, the extended range data, and the extended risk data.

5. The method for quantitatively evaluating the level of application changes in securities industry systems according to claim 4 is characterized in that: Step S5 is further as follows: S51, based on the evaluation score, score distribution data is obtained through distribution analysis; based on the historical impact vector, feature clustering is performed to obtain cluster feature data; based on the score distribution data and the cluster feature data, a level threshold vector is obtained through threshold optimization processing; S52, based on the evaluation score, perform numerical comparison to obtain comparison result data; based on the comparison result data and the level threshold vector, perform level mapping to obtain mapping level data; confirm the mapping level data to obtain a level change result; S53, analyzing the change level result to obtain analysis result data; based on the analysis result data, obtaining key feature data through feature extraction; Based on the analysis result data and key feature data, an evaluation description document is obtained through document generation processing.

6. The method for quantitatively evaluating the level of application changes in securities industry systems according to claim 5, characterized in that: Step S11 is further as follows: S111. Read the original change application data, and obtain interface list data through interface analysis; perform function complexity calculation on the interface list data to obtain function complexity data; perform module aggregation on the function complexity data to obtain module aggregation data; and obtain module complexity data through weighted calculation based on the module aggregation data; S112. Based on the original change application data, obtain module reference data through static scanning; perform call chain analysis on the module reference data to obtain call chain data; perform correlation calculation on the call chain data to obtain correlation data; perform graph structure analysis based on the correlation data to obtain dependency relationship data; S113. Based on the original change application data, code type data is obtained through code classification processing; Extract valid rows from the code type data to obtain valid row data; based on the valid row data, process the complexity weights to obtain weighted row data; Based on the weighted line data, perform summary statistics to obtain the number of code lines; S114. Normalize the module complexity data, dependency data and code line data respectively to obtain normalized complexity data, normalized dependency data and normalized line data; perform vector concatenation on the normalized complexity data, normalized dependency data and normalized line data to obtain a change feature vector.

7. The method for quantitatively evaluating the change level of securities industry system application according to claim 5 is characterized in that: Step S13 is further as follows: S131, reading market transaction data, classifying the market transaction data to obtain classified transaction data; based on the classified transaction data, dividing and processing by preset time windows to obtain window transaction data; performing frequency statistics on the window transaction data to obtain transaction frequency data; based on the transaction frequency data, calculating and obtaining transaction density data; S132, based on the market transaction data, obtaining price series data through price extraction; performing trend stripping on the price series data to obtain detrended data; Based on the detrended data, the volatility data is calculated; Based on the fluctuation range data, the indicator conversion is performed to obtain the fluctuation indicator data; S133. Based on the market transaction data, extract the time feature to obtain time feature data; perform transaction volume analysis on the time feature data to obtain transaction volume feature data; Perform pattern recognition on the transaction volume feature data to obtain transaction pattern data; perform type mapping processing based on the transaction pattern data to obtain time period type data; S134. Based on the transaction density data, standardized density data is obtained through standardization processing; based on the volatility index data, standardized volatility data is obtained through standardization processing; based on the time period type data, coding type data is obtained through coding conversion processing; the standardized density data, standardized volatility data and coding type data are vectorized to obtain a market characteristic vector.

8. The method for quantitatively evaluating the level of application changes in securities industry systems according to claim 5 is characterized in that: Step S21 is further as follows: S211. Based on the market feature vector, obtain segmented sequence data by performing sequence segmentation processing; extract statistical features from the segmented sequence data to obtain statistical feature data; perform trend analysis on the statistical feature data to obtain trend data; Based on the trend data, the window is divided to obtain the time window data; S212, performing differential calculation on the time window data to obtain differential feature data; Based on the differential characteristic data, the change rate data is calculated; based on the change rate data, the acceleration data is calculated; based on the acceleration data, the characteristic change rate data is obtained through comprehensive processing of the change rate; S213, obtaining dynamic threshold data through dynamic threshold calculation based on the characteristic change rate data; Based on the dynamic threshold data, outlier detection is performed to obtain outlier data; time point verification is performed on the outlier data to obtain verification time data; Based on the verification time data, a set of key time points is constructed.

9. The method for quantitatively evaluating the change level of securities industry system application according to claim 5 is characterized in that: Step S31 is further as follows: S311, based on the changed feature vector, obtain feature extreme value data through extreme value detection processing; Based on the characteristic extreme value data, outlier replacement is performed to obtain cleaned characteristic data; Performing interval mapping on the cleaned feature data to obtain interval data; performing normalization processing on the interval data to obtain normalized feature data; S312, calculating feature importance data based on the normalized feature data; Perform correlation analysis on feature importance data to obtain correlation data; perform weight optimization based on the correlation data to obtain optimized weight data; perform feature weighting processing based on the optimized weight data to obtain weighted feature data; S313, performing feature distribution analysis on the weighted feature data to obtain distribution feature data; processing the distribution feature data through kernel function transformation to obtain transformed feature data; based on the transformed feature data, performing mapping function optimization processing to obtain optimized mapping data; performing nonlinear transformation on the optimized mapping data to obtain a complexity vector.

Images