Change execution method and related product

Abstract

The invention discloses a change execution method and a related product. The method comprises the steps of obtaining an original change request text; performing feature extraction on the original change request text to obtain feature data; constructing a target function based on the feature data; solving the objective function based on constraint conditions to obtain an optimal change execution sequence; and performing a change operation based on the optimal change execution sequence to obtain a change result corresponding to the original change request text. According to the method, the target function is solved in combination with the actual operation and maintenance constraint conditions, so that the optimal change execution sequence is obtained. The change operation is carried out through the execution sequence, so that the scientificity and the execution efficiency of the change process can be effectively improved, the system abnormal risk caused by an improper change sequence is reduced, and the accuracy of the change execution efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a change execution method and related products. BACKGROUND

[0002] In the operation and maintenance work of a bank data center, change is a key task, and its importance is self-evident. However, according to the operation and maintenance report of a large commercial bank in 2022, 15632 change work orders were handled in the whole year, with an average time consumption of 47 minutes per order, 23% of the change delay was caused by manual sorting errors, and 18% of the production failures were due to insufficient change risk assessment.

[0003] From the existing technology, the traditional change system has many technical bottlenecks. The traditional ITSM system has a serious lack of text understanding ability, mainly relying on fixed format forms, such as selecting change type through drop-down menu, and being helpless in the face of unstructured text description in actual work order, resulting in a cumbersome and error-prone process of change request processing. For example, when the work order description is "upgrade the database of the payment system", the traditional system cannot automatically associate the influence of payment clearing, mobile banking and other systems, which greatly limits the efficiency and accuracy of change execution. SUMMARY

[0004] Based on the above problems, the present application provides a change execution method and related products, aiming to improve the efficiency and accuracy of change execution.

[0005] The embodiments of the present application disclose the following technical solutions:

[0006] The first aspect of the present application provides a change execution method, which comprises:

[0007] obtaining an original change request text;

[0008] performing feature extraction on the original change request text to obtain feature data;

[0009] constructing a target function based on the feature data;

[0010] solving the target function based on a constraint condition to obtain an optimal change execution sequence;

[0011] performing change operation based on the optimal change execution sequence to obtain a change result corresponding to the original change request text.

[0012] Optionally, the constructing a target function based on the feature data specifically comprises:

[0013] performing mathematical conversion on the feature data to obtain a mathematical expression corresponding to the feature data;

[0014] constructing a target function based on a mathematical expression corresponding to the feature data.

[0015] Optionally, the solving the target function based on the constraint condition to obtain the optimal change execution sequence specifically includes:

[0016] The improved NSGA-III is adopted to solve the target function based on the constraint condition to obtain the optimal change execution sequence.

[0017] Optionally, the solving the target function based on the constraint condition to obtain the optimal change execution sequence by using the improved NSGA-III specifically includes:

[0018] Let the value of k be 1;

[0019] An initial population is constructed based on the original change request text, and the initial population is taken as the population of the kth iteration;

[0020] The population of the kth iteration is subjected to non-dominated sorting and adaptive normalization processing to obtain the population of the kth iteration after normalization processing;

[0021] The population of the kth iteration after normalization processing is selected by using a niche reservation strategy to obtain the selected population of the kth iteration;

[0022] The selected population of the kth iteration is subjected to cross-over and mutation processing to obtain the sub-population of the kth iteration;

[0023] The final population of the kth iteration is obtained based on the selected population of the kth iteration and the sub-population of the kth iteration by using an elite reservation strategy;

[0024] If the iteration end condition is reached, the individuals in the final population of the kth iteration are output as the optimal change execution sequence;

[0025] If the iteration end condition is not reached, the final population of the kth iteration is taken as the population of the (k+1)th iteration, the value of k is increased by 1, and the step of performing non-dominated sorting and adaptive normalization processing on the population of the kth iteration to obtain the population of the kth iteration after normalization processing is returned to.

[0026] Optionally, the constructing an initial population based on the original change request text and taking the initial population as the population of the kth iteration specifically includes:

[0027] A basic parameter corresponding to the original change request text is determined;

[0028] Random sampling is performed based on the basic parameter to obtain a plurality of individuals; one individual represents a change execution sequence;

[0029] Performing repair processing on the multiple individuals, and filtering and screening the multiple individuals based on the constraint conditions to obtain multiple filtered individuals;

[0030] An initialization population is constructed based on the multiple screened individuals, and the initialization population is used as the population of the kth iteration.

[0031] Optionally, performing a crossover mutation process on the selected population of the k-th iteration to obtain a subpopulation of the k-th iteration specifically includes:

[0032] Performing a crossover process on the selected population of the k-th iteration to obtain a crossover population of the k-th iteration;

[0033] If there are individuals meeting the preset conditions in the cross population of the kth iteration, the individuals meeting the preset conditions are repaired to obtain a repaired cross population of the kth iteration;

[0034] If there is no individual meeting the preset conditions in the crossover population of the kth iteration, the crossover population of the kth iteration is used as the repaired crossover population of the kth iteration;

[0035] Performing mutation processing on the repaired crossover population of the k-th iteration to obtain a mutant population of the k-th iteration;

[0036] If there are individuals meeting the preset conditions in the variant population of the kth iteration, the individuals meeting the preset conditions are repaired or split to obtain a repaired variant population of the kth iteration;

[0037] If there is no individual meeting the preset conditions in the mutant population of the kth iteration, the mutant population of the kth iteration is used as the repaired mutant population of the kth iteration;

[0038] If the repaired mutant population of the k-th iteration does not meet the resource capacity, splitting the repaired mutant population of the k-th iteration to obtain a sub-population of the k-th iteration;

[0039] If the repaired mutant population of the k-th iteration meets the resource capacity, the repaired mutant population of the k-th iteration is used as the subpopulation of the k-th iteration.

[0040] Optionally, the optimal change execution sequence includes multiple change operations; performing the change operation based on the optimal change execution sequence to obtain a change result corresponding to the original change request text specifically includes:

[0041] The cosine similarity algorithm is used to perform script matching on the multiple change operations in the optimal change execution sequence based on the script library, to obtain a matching result corresponding to each change operation; the matching result includes a first result or a second result; the matching success includes a matching script and a matching success; the second result includes a matching failure;

[0042] The change operation is performed based on the risk prediction model and the matching result corresponding to each change operation, to obtain a change result corresponding to the original change request text.

[0043] Optionally, the change operation is performed based on the risk prediction model and the matching result corresponding to each change operation, to obtain a change result corresponding to the original change request text, and specifically includes:

[0044] The change operation is performed based on the matching result corresponding to each change operation, and the risk prediction model is used to predict the risk of each change operation in the process of the change operation, to obtain a prediction result corresponding to each change operation;

[0045] If the prediction result is greater than or equal to a preset threshold, the change operation corresponding to the prediction result greater than the preset threshold is stopped;

[0046] If the prediction result is less than the preset threshold, the change operation corresponding to the prediction result less than the preset threshold is performed, to obtain a change result corresponding to the original change request text.

[0047] The second aspect of the present application provides a change execution device, and the change execution device includes:

[0048] An acquisition module is configured to acquire an original change request text.

[0049] A feature extraction module is configured to perform feature extraction on the original change request text, to obtain feature data.

[0050] A construction module is configured to construct a target function based on the feature data.

[0051] A solution module is configured to solve the target function based on a constraint condition, to obtain an optimal change execution sequence.

[0052] An execution module is configured to perform a change operation based on the optimal change execution sequence, to obtain a change result corresponding to the original change request text.

[0053] The third aspect of the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the change execution method provided in the first aspect.

[0054] Compared with the prior art, the present application has the following beneficial effects:

[0055] The application comprises obtaining original change request text; performing feature extraction on the original change request text to obtain feature data; constructing an objective function based on the feature data; solving the objective function based on constraint conditions to obtain an optimal change execution sequence; and performing a change operation based on the optimal change execution sequence to obtain a change result corresponding to the original change request text.

[0056] The application obtains original change request text, performs feature extraction thereon, constructs feature data reflecting change content and dependency relationship, establishes an objective function based on the feature data, solves the objective function in combination with actual operation and maintenance constraint conditions, and obtains an optimal change execution sequence. Performing a change operation through the execution sequence can effectively improve the scientificity and execution efficiency of a change process, reduce system abnormal risk caused by improper change order, and improve the accuracy of change execution efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0057] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0058] Figure 1 A flowchart of a change execution method provided by an embodiment of the present application;

[0059] Figure 2 A structural diagram of a change execution device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0060] As described above, a traditional ITSM (IT service management) system mainly relies on manual filling of structured fields to process change requests. Operation and maintenance personnel need to manually input relevant information of changes, such as change type (selected from a drop-down menu), estimated time, impact range, etc., according to a predefined form format. The system classifies and sorts the structured data to generate a change scheduling plan. In terms of risk assessment, the traditional ITSM system usually assigns a fixed initial risk value based on some fixed rules, such as change type (such as database change, system configuration change, etc.), and then adjusts it according to manual judgment. For script management, the traditional ITSM system generally requires operation and maintenance personnel to manually associate and select scripts, and lacks intelligent script matching and recommendation functions.

[0061] Traditional ITSM systems cannot handle natural language tickets, and cannot effectively parse and utilize unstructured text descriptions in change requests, such as change requirement descriptions in free text form, resulting in incomplete information extraction and affecting the comprehensiveness and accuracy of change management. Traditional ITSM systems mainly rely on simple rules and manual experience in change scheduling, and cannot comprehensively consider multiple complex factors (such as the difficulty of change content, estimated time consumption, mutual influence between multiple changes, etc.) for intelligent optimization scheduling.

[0062] In view of the above problems, the present application provides a change execution method and related products, which comprises: obtaining an original change request text; performing feature extraction on the original change request text to obtain feature data; constructing a target function based on the feature data; solving the target function based on the constraint condition to obtain an optimal change execution sequence; performing change operation based on the optimal change execution sequence to obtain a change result corresponding to the original change request text.

[0063] The present application obtains the original change request text, performs feature extraction thereon to generate feature data reflecting the change content and its dependency relationship, constructs a target function based thereon, and solves the function in combination with the constraint condition in actual operation and maintenance to obtain an optimal change execution sequence, which can fully consider various factors to generate an optimal change execution sequence, thereby improving the rationality and efficiency of scheduling. Based on the execution sequence, change operation is carried out, which not only improves the scientificity and execution efficiency of the change process, but also effectively reduces the risk of system abnormality caused by unreasonable execution sequence, and significantly improves the accuracy and execution efficiency of the change process.

[0064] Related terms are explained as follows:

[0065] NLP (Natural Language Processing) : is an important direction in the field of computer science and artificial intelligence, which studies various theories and methods that can realize effective communication between man and computer with natural language, involving language understanding, information extraction, semantic analysis and other aspects, just like making computers learn to understand human language.

[0066] Machine learning : is a multi-disciplinary subject, involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines, which is dedicated to studying how computers simulate or implement human learning behavior to acquire new knowledge or skills, reorganize existing knowledge structure to continuously improve their performance, which is used here to analyze the potential impact of changes on the system.

[0067] Intelligentization: Intelligentization is the deep integration of intelligentization and automation. It refers to the use of technologies such as artificial intelligence (AI), machine learning (ML), robotic process automation (RPA), Internet of Things (IoT), and big data to enable systems to not only perform pre-set tasks automatically but also perceive the environment, analyze data in real-time, and dynamically optimize decisions, achieving a closed-loop operation mode of "perception → decision → execution → feedback". The essence is to endow the automated system with a "brain" so that it has the ability of self-learning, self-adaptation, and self-evolution.

[0068] In order to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0069] Figure 1 A flowchart of a change execution method provided by the embodiments of the present application is shown in FIG. 1. The change execution method includes the following steps. Figure 1

[0070] S101: Obtain the original change request text.

[0071] The change request text submitted by the user is automatically obtained from the change management system (such as Jira, ServiceNow, CMDB, etc.). The change request text can be a natural language description, such as "online payment service v2.0" or "upgrade the database version to MySQL 8.0". Structured and unstructured text input is supported, and different change types (such as functional change, configuration change, patch update, etc.) are adapted.

[0072] S102: Feature extraction is performed on the original change request text to obtain feature data.

[0073] NLP technology (such as BERT, TF-IDF, keyword extraction) is used to perform semantic analysis on the text. Key feature fields are extracted, including: change object (such as "payment service", "database connection pool"), change type (such as "upgrade", "configuration modification", "add"), involved system (dependent component), time window (such as start time, end time, preset time, etc.), and risk level (such as "high risk", "medium risk", "low risk") and the like. The structured expression of the change request is realized; quantitative input is provided for subsequent target function modeling; and multi-language and multi-format text analysis is supported.

[0074] ​S103: Construct a target function based on the feature data.

[0075] The target function is constructed to measure the rationality of the change execution sequence. The target can be set according to actual conditions, for example, the target is to minimize business impact (priority processing of changes affecting core systems), maximize resource utilization (avoid resource idling and excessive contention), and minimize risk accumulation (reduce the risk accumulation effect of parallel changes).

[0076] S104: Solve the target function based on the constraint condition to obtain the optimal change execution sequence.

[0077] The construction of the constraint condition is generally set according to the actual situation, for example, through real-time docking with the CMDB system, timely obtaining the configuration and state information of the system, accurately detecting resource conflicts in the change execution process, and providing real-time constraint conditions for intelligent scheduling optimization. The constraint conditions can include: core system change total time limit, time window hard constraint (such as only changing in the early morning), and resource capacity constraint (such as only one high-risk change can be executed at the same time).

[0078] S105: Perform change operation based on the optimal change execution sequence to obtain the change result corresponding to the original change request text.

[0079] The change execution sequence is issued to the automated change platform (such as Ansible, Chef, Jenkins); the change task is executed and the execution status is monitored in real time; the change result is recorded, including: whether it is successful, the actual execution time, whether it causes an exception, the impact range, etc.; the change result is fed back to the model for subsequent model optimization and learning.

[0080] The present application obtains the original change request text and extracts features therefrom to construct feature data reflecting the change content and system dependency relationship. On this basis, a target function is established and solved in combination with operation and maintenance constraint conditions to finally generate an optimal change execution sequence. By executing the change operation through the sequence, the scientificity and automation level of the change process can be effectively improved, the system abnormal risk caused by unreasonable change sequence can be reduced, and the accuracy and efficiency of change execution can be improved. Not only the intelligent capability of complex IT system change management is enhanced, but also the change success rate and system stability are significantly improved.

[0081] The above is the main technical solution of the present application. Now the expansion implementation of the main technical solution of the present application will be introduced. Specifically as follows:

[0082] For S102, the original change request text is extracted to obtain feature data. The present application provides an optional embodiment:

[0083] The ALBERT-based domain adaptation model extracts features from the original change request text. This model can more accurately understand the semantic information in the change request and accurately extract key elements such as change objects, operation types, and time windows, laying the foundation for intelligent processing of change management.

[0084] After obtaining the feature data, considering that the feature data is generally in text form and is not suitable for mathematical calculation, the application also provides an optional embodiment:

[0085] Mathematically transforming the feature data to obtain a mathematical expression corresponding to the feature data.

[0086] For numerical features (such as change time window length, estimated impact range, etc.), Z-Score standardization or Min-Max scaling methods are used to convert them into numerical values on a unified scale to facilitate subsequent calculations. For text features (such as change type, system name involved, etc.), word embedding techniques (such as Word2Vec, GloVe) are used to convert them into fixed-length vector representations for easy processing by machine learning models. If the feature data contains inter-system dependency relationships, graph neural networks (GNNs) or other graph representation learning methods can be used to encode these dependency relationships into node and edge feature vectors, reflecting the interaction strength between components.

[0087] For example: Suppose a change request feature includes "change time window", "change type", and "estimated number of affected users". Among them, "change time window" is standardized; "change type" is converted into a 3-dimensional vector through word embedding; "estimated number of affected users" remains as an input feature.

[0088] Constructing a target function based on the mathematical expression corresponding to the feature data.

[0089] As an optional embodiment, the target function includes:

[0090] ;

[0091] Where f1 represents the business impact cost (the smaller the better); f2 represents the resource utilization rate (the larger the better, taking the negative sign to unify into minimization); f3 represents the time violation penalty (the smaller the better); represents the earliest allowed start time of change operation i; m represents the resource capacity; represents the total available time of resource j (such as the available time of a database cluster within a maintenance window); represents the actual time of change operation i occupying resource j; represents the latest must-complete time of change operation i; represents the estimated time consumption (in minutes) of change operation i; si represents the planned start time (in minutes) of change operation i, typically an integer; e i represents the planned end time (in minutes) of change operation i, e i = s i + ; represents the risk assessment (0-1) of change operation i; represents the business impact weight of change operation i, such as ; represents the operation complexity of change operation i, , represents the number of operation steps. represents the buffer time requirement of change operation i, ; represents the inter-change impact overlap degree of change operation i, , S represents the set of imaging systems.

[0092] The objective of the application is to minimize business impact (prioritize changes that impact core systems), maximize resource utilization (avoid resource idling and excessive contention), and minimize risk accumulation (reduce the cumulative effect of parallel changes).

[0093] The design intent of minimizing the comprehensive cost mainly considers the business impact cost (prioritize changes to core systems (Class A)) and the risk complexity cost (suppress high-risk and complex changes from occupying resources for a long time); for example, a change operation that impacts a Class A system, takes 120 minutes, has a risk value of 0.3, and has 15 steps, its cost is:

[0094] 0.6x0.9x120+0.4x0.3x4=64.8+0.48=65.28;

[0095] The design intent of maximizing resource utilization mainly considers balancing the load (to prevent excessive congestion of certain resources); for example, a database cluster maintenance window is 20:00-06:00 (600 minutes), if it is occupied by 3 changes for 550 minutes, its utilization rate = 550 / 600≈0.917.

[0096] The design intent of minimizing time violations mainly considers strictly ensuring that changes are executed within the time window approved and quantifying the penalty for violations (increase the cost per minute of violation); for example, a change plan is executed in [20:00-23:00], if the actual execution time is 19:50-22:50, then f3=max(0, 19:50-20:00)+max(0, 22:50-23:00)=(-10)+(-10)→take 0 (no penalty).

[0097] By converting the original feature data into mathematical expressions, complex business requirements and constraints can be accurately described in mathematical models, providing a solid foundation for automated decision-making. Mathematical expressions allow flexible adjustment of parameters to adapt to different business scenarios or change requirements, improving the flexibility and adaptability of the system. The objective function constructed based on mathematical expressions can consider multiple dimensions of optimization targets such as time and resources, achieve optimal resource allocation, and reduce change costs.

[0098] The S104 is used to solve the objective function based on the constraint conditions to obtain an optimal change execution sequence. The application also provides an optional embodiment:

[0099] The improved NSGA-III is used to solve the objective function based on the constraint conditions to obtain an optimal change execution sequence.

[0100] As an optional embodiment, the constraint conditions include:

[0101] Time window hard constraint: ;

[0102] Resource capacity constraint: ;

[0103] Core system change total duration limit: ;

[0104] The time window hard constraint is to ensure that each change is completed within the time period approved; in specific applications, the time window hard constraint can be used to perform time window verification on the initial solution, or the time window hard constraint can be used to automatically repair violations (such as splitting overtime changes into multiple segments) when crossover and mutation.

[0105] The resource capacity constraint is to ensure that the capacity is within the allowed range during the change process; for example, the maximum number of connections of a certain database cluster is 500, if change A requires 300 connections and change B requires 250 connections, a conflict is triggered: 300+250=550>500300+250=550>500 → needs to be rescheduled.

[0106] The core system change total duration limit can be set according to actual conditions; in specific applications, the core system change total duration limit can be used to filter illegal solutions in the population initialization stage, or the core system change total duration limit can be used to impose high penalties on illegal solutions during the evolution process.

[0107] The improved NSGA-III algorithm in the application effectively avoids the problem of uneven distribution of solutions in the objective space by adopting the Das-Dennis method to generate uniformly distributed reference points; at the same time, an adaptive crossover strategy is introduced to dynamically adjust the crossover probability (0.7-0.9) according to the population diversity to balance the exploration and development capabilities of the algorithm; in addition, time window sliding and resource sensitive mutation operators are developed for the bank scenario, further improving the adaptability of the algorithm to the specific business environment of the bank, ensuring efficient and stable operation in complex change scheduling.

[0108] The improved NSGA-III algorithm in the application adopts an advanced mixed integer programming algorithm, combines multi-factor grading results and resource conflict detection information to generate the optimal change execution sequence, and realizes the efficiency and intelligence of change management. The improved NSGA-III algorithm considers the influence range of multiple changes, the starting time of change application, the difficulty of change content, the estimated time consumption and other factors, generates the optimal change execution sequence by minimizing business impact, maximizing resource utilization, minimizing risk accumulation and strictly complying with time, etc. Key steps such as population initialization, fitness evaluation, selection operation, crossover and mutation operation, and environment selection cooperate with each other to constantly evolve the population and find the Pareto optimal solution set.

[0109] For different types of change conflicts, the application designs corresponding processing strategies: for completely overlapping conflicts, the method of delaying low-priority changes according to priority is adopted, for example, high-priority core system changes will be executed first; for partially overlapping and resource contention, long-time-consuming changes are split to solve, such as splitting a 4-hour change into two 2-hour segments for execution; and for cross-midnight conflicts, additional buffer time is reserved for cross-midnight changes to ensure smooth execution, such as adding a 10-minute buffer period before and after 00:00. These strategies effectively improve the flexibility and success rate of change operations, ensuring stable operation of the system in complex change scenarios.

[0110] The above roughly outlines the improved NSGA-III and constraints in the application, in order to more clearly illustrate the specific application of the improved NSGA-III and constraints in the application, the application also provides an optional embodiment:

[0111] Let the value of k be 1.

[0112] An initialization population is constructed based on the original change request text, and the initialization population is used as the population of the kth iteration.

[0113] For the construction of the initialization population, the application provides an optional embodiment:

[0114] determine a basic parameter corresponding to the original change request text.

[0115] random sampling is performed based on the basic parameter to obtain a plurality of individuals; one individual represents a change execution sequence.

[0116] Repair processing is performed on the plurality of individuals, and the plurality of individuals are filtered and screened based on the constraint condition to obtain a plurality of screened individuals.

[0117] An initialization population is constructed based on the plurality of screened individuals, and the initialization population is taken as a population of the kth iteration.

[0118] In order to more intuitively illustrate the construction method of the initialization population, a specific application embodiment is further provided in the application:

[0119] In the W start i …W end i -T exec i Uniform sampling in the interval to obtain 150 chromosomes.

[0120] Resource overload → automatically split the super-long task (such as 4h) into two segments 2h.

[0121] Core system total night length >180 min → move or split the low-priority tasks as a whole.

[0122] If the repair still violates the regulatory red line, the individual is eliminated and regenerated to obtain a plurality of screened individuals; an initialization population is constructed based on the plurality of screened individuals, and the initialization population is taken as a population of the kth iteration.

[0123] After determining the population of the kth iteration, non-dominated sorting and adaptive normalization processing are performed on the population of the kth iteration to obtain a normalized population of the kth iteration.

[0124] 150 solutions are divided into a plurality of "frontiers" F1, F2, …, F1 being the Pareto optimal layer according to the dominance relationship. Calculate the ideal point z* = (min f1, min f2, min f3); translate the target value: f' = f - z*; calculate the extreme point (the optimal solution of each target) and the hyperplane intercept a = (a1, a2, a3).

[0125] Normalization: fnorm = f ' / a, ensure that all targets fall within the interval [0, 1], and prepare for subsequent reference point matching.

[0126] After determining the normalized population of the kth iteration, a niche reservation strategy is used to select the normalized population of the kth iteration to obtain a selected population of the kth iteration.

[0127] The Das-Dennis method is used to uniformly scatter 91 reference points on a three-dimensional unit hyperplane to ensure diversity. Each normalized solution is projected to the nearest reference line (the shortest perpendicular distance to the reference point) to establish a "solution ↔ reference point" mapping table.

[0128] The number of individuals associated with each reference point ρ_j is counted; the nearest solution in the reference point with ρ_j=0 is preferentially selected; if all reference points are occupied, a randomly associated solution is selected from the reference point with the smallest ρ_j to join the next generation.

[0129] This ensures that the Pareto front is uniformly distributed in the three-dimensional target space and does not "cluster".

[0130] After determining the selected population of the kth iteration, the selected population of the kth iteration is subjected to crossover and mutation processing to obtain a sub-population of the kth iteration.

[0131] The present application provides an optional embodiment for crossover and mutation:

[0132] The selected population of the kth iteration is subjected to crossover processing to obtain a crossover population of the kth iteration.

[0133] If there is an individual in the crossover population of the kth iteration that meets the preset condition, the individual that meets the preset condition is subjected to repair processing to obtain a repaired crossover population of the kth iteration.

[0134] The present application does not limit the preset condition, for example, the preset condition includes that the start time of the change execution sequence corresponding to the individual is not within the earliest allowed time of the change operation and the latest must complete time of the change operation.

[0135] If there is no individual in the crossover population of the kth iteration that meets the preset condition, the crossover population of the kth iteration is taken as the repaired crossover population of the kth iteration.

[0136] The present application does not limit the repair method, for example, the individual that does not meet the preset condition is translated to within the earliest allowed time of the change operation and the latest must complete time of the change operation.

[0137] The repaired crossover population of the kth iteration is subjected to mutation processing to obtain a mutation population of the kth iteration.

[0138] If there is an individual meeting the preset condition in the variation population of the kth iteration, the individual meeting the preset condition is subjected to repair processing or the individual meeting the preset condition is subjected to splitting processing, to obtain a repaired variation population of the kth iteration.

[0139] The application does not limit the method of splitting processing, for example, splitting an individual into two segments.

[0140] If there is no individual meeting the preset condition in the variation population of the kth iteration, the variation population of the kth iteration is taken as a repaired variation population of the kth iteration.

[0141] If the repaired variation population of the kth iteration does not meet the resource capacity, the repaired variation population of the kth iteration is subjected to splitting processing, to obtain a subpopulation of the kth iteration.

[0142] If the repaired variation population of the kth iteration meets the resource capacity, the repaired variation population of the kth iteration is taken as a subpopulation of the kth iteration.

[0143] The application provides a specific application embodiment for crossover variation:

[0144] Immediately after each crossover / variation, a repair function is called to ensure that all offspring are "legal and feasible solutions".

[0145] After the "selection-elite reservation" of each generation is completed, the "crossover-variation-repair" three-in-one operation is performed on the 150 parent individuals reserved, to generate 150 offspring.

[0146] The crossover probability Pc adopts diversity-feedback self-adaptation instead of a fixed value:

[0147] diversity = average_crowding_distance(pop)

[0148] if diversity < 0.30:

[0149] Pc = 0.9 # high exploration

[0150] elif diversity > 0.70:

[0151] Pc = 0.7 # high development

[0152] else:

[0153] Pc = 0.8

[0154] The mutation probability Pm is two-stage: the first 60% of generations: Pm = 0.05, to ensure sufficient disturbance; the last 40% of generations: Pm = 0.02, to protect the converged area. It can also be refined according to the individual non-dominated level (the higher the level, the smaller the Pm).

[0155] In terms of crossover operators, the improved method uses a hybrid SBX (real number) and time window alignment single-point crossover method. Compared with traditional SBX or single-point crossover, it immediately checks whether `start[i]` falls within the `[W_start, W_end - T_exec]` interval after generating offspring; if it is out of bounds, the "time window alignment" strategy is executed to move the out-of-bound gene to the nearest legal boundary, avoiding a large amount of repair work in the later stage. In terms of mutation operators, the improved method introduces a mutation operator dedicated to bank changes, replacing the traditional polynomial or uniform mutation. This mutation operator contains two key mechanisms: one is time window sliding, which slides 5 minutes or automatically splits into two segments when mutation causes out-of-bound; the other is resource-sensitive mechanism, which delays the change by 10 minutes or splits it if the mutation causes resource overload (such as Oracle connection number exceeding 500), until the capacity requirement is met, thereby ensuring the rationality of change operations and effective use of system resources.

[0156] After the determination of the sub-population of the kth iteration, the final population of the kth iteration is obtained based on the selected population of the kth iteration and the sub-population of the kth iteration using an elite reservation strategy.

[0157] Parent (selected population of the kth iteration) + offspring (sub-population of the kth iteration) -> 300 individuals -> non-dominated sorting -> take the front F1…Fl until the number of individuals >=150.

[0158] If exactly 150 individuals, -> directly as the next generation.

[0159] If >150, use the "reference point-niche" strategy to preferentially select solutions corresponding to "sparse reference points" from the last layer Fl until 150 are supplemented.

[0160] If the iteration end condition is reached, the individuals in the final population of the kth iteration are output as the optimal change execution sequence.

[0161] The present application does not limit the iteration end condition, for example, the iteration end condition includes repeating iterations for 200 generations or 20 consecutive generations without improvement in the Pareto front.

[0162] If the iteration end condition is not reached, the final population of the kth iteration is taken as the population of the k+1th iteration, the value of k is increased by 1, and the step of performing non-dominated sorting and adaptive normalization processing on the population of the kth iteration to obtain the population of the kth iteration after normalization processing is returned.

[0163] The application also provides a corresponding pseudo code for solving the objective function:

[0164] from pymoo.algorithms.moo.nsga3 import NSGA3

[0165] from pymoo.factory import get_reference_directions

[0166] from pymoo.optimize import minimize

[0167] from pymoo.core.problem import ElementwiseProblem

[0168] class ChangeSchedulingProblem(ElementwiseProblem):

[0169] def __init__(self,n_changes,res_cap,res_total):

[0170] super().__init__(n_var=n_changes,n_obj=3,xl=0,xu=1440)

[0171] self.res_cap = res_cap # {resource_id: capacity}

[0172] self.res_total = res_total # {resource_id: available_minutes}

[0173] def _evaluate(self,x,out,*args,**kwargs):

[0174] s = x

[0175] e = s + T_exec

[0176] # Objective 1: Business Impact Cost

[0177] f1 = np.sum(0.6 * C * T_exec + 0.4 * R_risk * D)

[0178] # Objective 2: Resource utilization (maximize by taking negative)

[0179] f2 = -np.mean([np.sum(T_exec[mask]) / self.res_total[j]

[0180] for j,mask in enumerate(res_usage_mask)])

[0181] # Objective 3: Time violations

[0182] f3 = np.sum(np.maximum(0,s - W_start) + np.maximum(0,e - W_end))

[0183] # Hard constraint penalty

[0184] penalty = 0

[0185] for j,cap in self.res_cap.items():

[0186] if np.sum(R[:,j]) > cap:

[0187] penalty += 1e6

[0188] out["F"] = [f1 + penalty,f2 + penalty,f3 + penalty]

[0189] # Reference point generation

[0190] ref_dirs = get_reference_directions("das-dennis",3,n_partitions=12)

[0191] algorithm = NSGA3(

[0192] ref_dirs=ref_dirs,

[0193] pop_size=150,

[0194] crossover = get_crossover("real_sbx", prob=0.8, eta=30),

[0195] mutation = get_mutation("real_pm", eta=20) )

[0197] res = minimize(ChangeSchedulingProblem(n, res_cap, res_total),

[0198] algorithm,

[0199] ('n_gen', 200))

[0200] The improved NSGA-III algorithm of the present application is optimized for the bank change scheduling scenario. Compared with the traditional general continuous optimization NSGA-III, it has achieved significant improvement in multiple dimensions. First, the reference point is changed from the default distribution to the more uniform Das-Dennis generation method, ensuring comprehensive coverage of the solution space. Second, the crossover probability adopts an adaptive mechanism, dynamically adjusting between 0.7 and 0.9 to adapt to the search needs at different stages. The mutation operator combines time window sliding and resource sensitivity, enhancing the algorithm's adaptability to the actual business environment. In terms of constraint handling, through hard constraint repair and high penalty strategy, the generation of infeasible solutions is effectively avoided. Finally, business compliance is fully guaranteed, ensuring 100% satisfaction of regulatory time window requirements, thus exhibiting higher efficiency and reliability in bank change scheduling.

[0201] The above describes the solution method of the objective function in more detail. Now the application method of the obtained optimal change execution sequence will be described, that is, based on the optimal change execution sequence, the change operation is performed on S105 to obtain the change result corresponding to the original change request text. The present application provides an optional embodiment:

[0202] The cosine similarity algorithm is used to match the multiple change operations in the optimal change execution sequence based on the script library to obtain the matching result corresponding to each change operation. The matching result includes a first result or a second result. The matching success includes a matching script and a matching success. The second result includes a matching failure.

[0203] The feature vectors of the change operation and the script are extracted. For the change operation, the key information is extracted using the natural language processing technology of the previous change operation; for the script in the script library, the script content is parsed, and the features such as the change type applicable to the script, the operation object, and the system environment are extracted to form the feature vector of the script. The cosine similarity algorithm is used to calculate the similarity between the feature vectors of the change operation and the script. The value range of the semantic similarity is between 0 and 1. The closer the value is to 1, the higher the semantic similarity between the two.

[0204] Based on the script library (which stores a large number of change implementation scripts, each script has detailed metadata information, including script applicable change type, operation object, system environment, and feature description, so that the script matching recommendation module can be screened and matched), according to the semantic similarity and the historical use of the script (statistical script use frequency in past change tasks, high use frequency of the script indicates that it has good universality and reliability to a certain extent), execution success rate (calculate the success rate of script execution, high success rate of script can reduce the risk of change implementation), and adaptation degree (according to the semantic similarity between the script and the change request and the applicable range of the script, evaluate the adaptation degree of the script to the current change task), etc. Multi-dimensional indexes, a comprehensive evaluation function is used to weight and integrate multi-dimensional indexes such as historical use, execution success rate and adaptation degree.

[0205] According to the score of the comprehensive evaluation function, the matched scripts are sorted. The higher the score, the more priority the script has. Scripts with a similarity higher than a set threshold (such as 0.7) are screened out. The screened scripts are intelligently sorted and evaluated to generate a recommendation list.

[0206] After determining the matching result corresponding to each change operation, the change operation is performed based on the risk prediction model and the matching result corresponding to each change operation, to obtain the change result corresponding to the original change request text.

[0207] To ensure the safety of the change operation, the application also provides an optional embodiment, which performs the change operation based on the matching result corresponding to each change operation, and uses the risk prediction model to predict the risk of each change operation during the change operation to obtain the prediction result corresponding to each change operation.

[0208] The application does not limit the construction method of the risk prediction model. As an optional embodiment:

[0209] A risk prediction model based on a Time-Decay Long Short-Term Memory (LSTM) network, combining dynamic time series modeling with real-time monitoring feedback, accurately predicts change risks. The feature engineering layer extracts 32-dimensional features, including change attributes, system status, historical risks, and time factors. After processing them using a time-decay function, these features are fed into an improved LSTM-TD network for risk prediction. The real-time prevention and control layer, based on dynamic risk thresholds and risk prediction results, triggers a risk warning mechanism, dynamically adjusts the risk level, and automatically performs rollbacks when necessary, ensuring effective risk prevention and control.

[0210] The risk prediction model of this application is built based on the time decay function; the prediction model layer in the risk prediction model adopts the improved LSTM-TD (Time-Decay LSTM) network:

[0211] class RiskLSTM(nn.Module):

[0212] def __init__(self):

[0213] super().__init__()

[0214] self.lstm = nn.LSTM(

[0215] input_size=32,

[0216] hidden_size=128,

[0217] num_layers=2,

[0218] dropout=0.3 )

[0220] self.time_decay = nn.Parameter(torch.exp(-0.1 * torch.arange(10))) #Time decay parameter

[0221] self.fc = nn.Sequential(

[0222] nn.Linear(128, 64),

[0223] nn.ReLU(),

[0224] nn.Linear(64, 1),

[0225] nn.Sigmoid() )

[0227] def forward(self, x):

[0228] x = x * self.time_decay.unsqueeze(-1) # Time decay weighting

[0229] out,_ = self.lstm(x)

[0230] return self.fc(out[:,-1,:])

[0231] This application introduces a learnable time decay parameter to dynamically adjust the weight of historical features; adopts a layered dropout strategy (0.3 for the input layer and 0.2 for the hidden layer); and a dynamic risk threshold.

[0232] After determining the prediction result, if the prediction result is greater than or equal to a preset threshold, stopping the change operation corresponding to the prediction result greater than the preset threshold;

[0233] If the predicted result is smaller than the preset threshold, a change operation corresponding to the predicted result smaller than the preset threshold is performed to obtain a change result corresponding to the original change request text.

[0234] To ensure the validity of the original change request text, this application also provides an optional embodiment for preprocessing the original change request text:

[0235] Utilize a regular expression library containing 200+ financial terminology transformation rules to clean and preprocess the original change request text, remove irrelevant information, and provide cleaner and more standardized input data for subsequent semantic analysis.

[0236] Open-source scheduling tools typically provide basic scheduling functions, such as task sequencing and resource allocation. They typically schedule tasks according to pre-defined scheduling algorithms (such as first-come, first-served and priority scheduling) by defining parameters such as task priority, dependencies, and resource requirements. When processing a change request, operations personnel must define and enter the change task in the format required by the tool, including information such as the task name, execution time, and required resources. The scheduling tool then schedules tasks based on this information and allocates resources according to the pre-defined rules. Regarding script management, open-source scheduling tools are generally only able to execute pre-defined scripts and are unable to intelligently match and recommend scripts based on the semantic information of the change request.

[0237] The open-source scheduling tool lacks understanding and adaptation of banking business rules and cannot automatically avoid special requirements such as regulatory freeze periods. The present application fully considers the business characteristics and regulatory requirements of the banking industry and integrates business impact, risk control and other factors into the multi-objective optimization scheduling model to generate a change scheduling plan that meets regulatory requirements and business objectives. The open-source scheduling tool usually uses a relatively simple scheduling algorithm and cannot handle complex multi-objective optimization problems. The present application uses an improved NSGA-III algorithm that can balance the conflicts between multiple objectives and find the optimal change scheduling solution. The open-source scheduling tool has limited functionality in script management and cannot achieve intelligent matching and recommendation of scripts. The script intelligent matching and recommendation function of the present application can accurately match and recommend appropriate scripts according to the semantic information of the change request and the characteristics of the script, improving the efficiency and accuracy of change implementation.

[0238] The traditional change management system relies on manual filling of structured fields and cannot handle natural language work orders. The present application uses a change request intelligent analysis module to use natural language processing technology to accurately understand and analyze change request text and convert unstructured text information into structured data, improving the efficiency and accuracy of change management. Existing risk rating is based on predefined rules and does not consider real-time system status. The risk prediction and prevention module of the present application uses a Time-Decay LSTM that integrates real-time monitoring data to dynamically assess change risk. The feature engineering layer extracts rich real-time system status features such as CPU usage and memory occupancy, and combines historical risk data to weight historical features using a time decay function, allowing the model to more comprehensively and dynamically assess change risk.

[0239] Through the above innovative means, the present application effectively solves the defects of existing change management technology, improves the intelligent level and risk prevention and control capability of change management, and meets the high requirements of bank data centers for change management in complex business environments.

[0240] The commercial risk analysis platform performs risk assessment based on a static rule base and does not fully combine real-time system status (such as current CPU load and transaction concurrency) to dynamically adjust risk assessment results. The risk prediction and prevention module of the present application uses a Time-Decay LSTM that integrates real-time monitoring data to achieve dynamic risk assessment, allowing more timely and accurate reflection of changes in change risk. The commercial risk analysis platform mainly focuses on risk analysis and does not involve specific operations of change scheduling and script management. The present application not only performs well in risk prediction and prevention, but also covers whole-process functions such as change request intelligent analysis, multi-objective optimization scheduling and script intelligent matching and recommendation, achieving comprehensive intelligentization of change management.

[0241] The application uses a regular expression library containing 200+ financial term transformation rules to clean and preprocess the original change request text, removing irrelevant information to provide clean and standardized input data for subsequent semantic analysis. This innovative approach customizes rules for the financial field, effectively improving the accuracy and efficiency of change request text preprocessing. The ALBERT-based domain adaptation model accurately understands the semantic information in the change request and accurately extracts key elements such as change objects, operation types, and time windows. This innovation enables change management to more accurately grasp the core content of change requests, laying a solid foundation for subsequent intelligent processing.

[0242] By comprehensively considering the influence range of multiple changes, the starting time of change application, the difficulty of change content, the estimated time consumption, and the mutual influence between multiple changes, etc., the application uses an improved NSGA-III algorithm to realize multi-objective optimization scheduling. This algorithm introduces reference point generation, adaptive crossover, and bank mutation operators, such as using the Das-Dennis method to generate uniformly distributed reference points to avoid uneven distribution of solutions in the target space; dynamically adjusting the crossover probability (0.7-0.9) according to population diversity to balance exploration and development capabilities; developing time window sliding and resource-sensitive mutation strategies to improve the algorithm's adaptability to bank scenarios. This improved algorithm can effectively solve the complex scheduling problem in the multi-change scenario of the bank data center, while meeting various independent maintenance window constraints, achieving the core goals of minimizing business impact, maximizing resource utilization, minimizing risk accumulation, and strictly complying with time regulations, and generating the optimal change execution sequence.

[0243] The application extracts 32-dimensional features such as change attributes, system state, historical risk, and time factors, and uses a time decay function to weight the historical features, enabling the model to focus more on recent changes in risk characteristics. This innovative feature processing method, combined with the dual-engine architecture of dynamic time series modeling and real-time monitoring feedback, enables the risk prediction model to more accurately capture dynamic changes in risk. The improved LSTM-TD (Time-Decay LSTM) network in the prediction model layer introduces a learnable time decay parameter to dynamically adjust the weight of historical features, and uses a hierarchical Dropout strategy (input layer 0.3, hidden layer 0.2) to improve the prediction performance and generalization ability of the model. This innovation enables the model to more accurately predict change risks, providing strong support for risk prevention and control.

[0244] The real-time prevention and control layer triggers the risk warning mechanism based on the dynamic risk threshold and risk prediction results, dynamically adjusts the risk level, and automatically executes rollback operations when necessary, effectively achieving risk prevention and control. This dynamic risk threshold-based prevention and control strategy can more timely and intelligently respond to changes in risk, reducing change risk.

[0245] By changing the request intelligent analysis, quickly and accurately extracting key elements, reducing the time and effort of manual understanding and processing change requests, and improving the response speed of change management. Multi-objective optimization scheduling can generate the optimal change execution sequence, reasonably arrange change tasks, avoid resource conflicts and time waste, and improve the overall efficiency of change management. Script intelligent matching recommendation function quickly matches and recommends appropriate scripts, reduces the difficulty and time cost of script selection, and speeds up the preparation process of change implementation.

[0246] Figure 2 The structural diagram of a change execution device provided by an embodiment of the present application is shown in Figure 2 Based on the change execution method provided by the foregoing embodiment, the present application also correspondingly provides a change execution device, which includes:

[0247] The acquisition module is configured to acquire an original change request text.

[0248] The feature extraction module is configured to perform feature extraction on the original change request text to obtain feature data.

[0249] The construction module is configured to construct a target function based on the feature data.

[0250] The solving module is configured to solve the target function based on the constraint condition to obtain an optimal change execution sequence.

[0251] The execution module is configured to perform a change operation based on the optimal change execution sequence to obtain a change result corresponding to the original change request text.

[0252] As an optional embodiment, the construction module is specifically configured to:

[0253] perform mathematical conversion on the feature data to obtain a mathematical expression corresponding to the feature data; and construct the target function based on the mathematical expression corresponding to the feature data.

[0254] As an optional embodiment, the solving module is specifically configured to:

[0255] solve the target function based on the constraint condition by using an improved NSGA-III to obtain the optimal change execution sequence.

[0256] As an optional embodiment, the solving module specifically includes:

[0257] The setting unit is configured to set the value of k to 1.

[0258] The initialization population unit is configured to construct an initialization population based on the original change request text, and use the initialization population as the population of the kth iteration.

[0259] a normalization unit configured to perform non-dominated sorting and adaptive normalization on the population of the kth iteration to obtain a normalized population of the kth iteration.

[0260] a selection unit configured to perform selection on the normalized population of the kth iteration by using a niche preservation strategy to obtain a selected population of the kth iteration.

[0261] a mutation unit configured to perform cross mutation on the selected population of the kth iteration to obtain a sub-population of the kth iteration.

[0262] a screening unit configured to obtain a final population of the kth iteration based on the selected population of the kth iteration and the sub-population of the kth iteration by using an elitist preservation strategy.

[0263] a first judging unit configured to output individuals in the final population of the kth iteration as an optimal change execution sequence if an iteration end condition is reached.

[0264] a second judging unit configured to take the final population of the kth iteration as a population of a (k+1)th iteration and increase the value of k by 1 and return to the normalization unit if the iteration end condition is not reached.

[0265] As an optional embodiment, the initialization population unit specifically comprises:

[0266] a basic parameter determination sub-unit configured to determine a basic parameter corresponding to the original change request text.

[0267] a sampling sub-unit configured to perform random sampling based on the basic parameter to obtain a plurality of individuals. Each individual represents a change execution sequence.

[0268] a first repair sub-unit configured to perform repair processing on the plurality of individuals and filter and screen the plurality of individuals based on the constraint condition to obtain a plurality of screened individuals.

[0269] a construction sub-unit configured to construct an initialization population based on the plurality of screened individuals and take the initialization population as the population of the kth iteration.

[0270] As an optional embodiment, the mutation unit specifically comprises:

[0271] a cross sub-unit configured to perform cross processing on the selected population of the kth iteration to obtain a cross population of the kth iteration.

[0272] The first judging subunit is configured to, if there is an individual meeting the preset condition in the crossover population of the kth iteration, perform a repair process on the individual meeting the preset condition to obtain a repaired crossover population of the kth iteration.

[0273] The second judging subunit is configured to, if there is no individual meeting the preset condition in the crossover population of the kth iteration, take the crossover population of the kth iteration as the repaired crossover population of the kth iteration.

[0274] The mutation subunit is configured to perform a mutation process on the repaired crossover population of the kth iteration to obtain a mutation population of the kth iteration.

[0275] The third judging subunit is configured to, if there is an individual meeting the preset condition in the mutation population of the kth iteration, perform a repair process or a split process on the individual meeting the preset condition to obtain a repaired mutation population of the kth iteration.

[0276] The fourth judging subunit is configured to, if there is no individual meeting the preset condition in the mutation population of the kth iteration, take the mutation population of the kth iteration as the repaired mutation population of the kth iteration.

[0277] The fifth judging subunit is configured to, if the repaired mutation population of the kth iteration does not meet the resource capacity, perform a split process on the repaired mutation population of the kth iteration to obtain a subpopulation of the kth iteration.

[0278] The sixth judging subunit is configured to, if the repaired mutation population of the kth iteration meets the resource capacity, take the repaired mutation population of the kth iteration as the subpopulation of the kth iteration.

[0279] As an optional embodiment, the optimal change execution sequence includes multiple change operations. The execution module specifically includes:

[0280] The similarity calculation unit is configured to perform script matching on the multiple change operations in the optimal change execution sequence based on the script library by using a cosine similarity algorithm to obtain a matching result corresponding to each change operation; the matching result includes a first result or a second result; the matching success includes a matched script and a matching success; and the second result includes a matching failure.

[0281] The execution unit is configured to perform a change operation based on the risk prediction model and the matching result corresponding to each change operation to obtain a change result corresponding to the original change request text.

[0282] As an optional embodiment, the execution unit specifically includes:

[0283] The risk prediction subunit is configured to perform the change operation based on the matching result corresponding to each change operation, and perform risk prediction on each change operation by using a risk prediction model during the change operation to obtain a prediction result corresponding to each change operation.

[0284] The seventh determination subunit is configured to stop the change operation corresponding to the prediction result greater than the preset threshold if the prediction result is greater than or equal to the preset threshold.

[0285] The eighth determination subunit is configured to perform the change operation corresponding to the prediction result less than the preset threshold if the prediction result is less than the preset threshold to obtain a change result corresponding to the original change request text.

[0286] The embodiment of the present application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to implement the change execution method.

[0287] The embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the change execution method.

[0288] The embodiment of the present application provides a computer program product, including a computer program, and the computer program is executed by a processor to implement the change execution method.

[0289] It should be noted that each of the embodiments in the present specification is described in a progressive manner, and the same and similar parts of each embodiment can be referred to each other, and each embodiment mainly describes the difference from other embodiments. Especially, the device and equipment embodiments are basically similar to the method embodiments, so the description is relatively simple, and the related parts can be referred to the part of the method embodiment. The device and equipment embodiments described above are only schematic, and the units described as separate components can be or can not be physically separated, and the components indicated as units can be or can not be physical units, that is, they can be located in one place, or can be distributed on multiple network units. According to the actual needs, part or all of the modules can be selected to achieve the purpose of the embodiment scheme. Those skilled in the art can understand and implement without creative labor.

[0290] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited to this. Any skilled person in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A change execution method, characterized in that: The change execution method includes: Get the original change request text; Performing feature extraction on the original change request text to obtain feature data; constructing an objective function based on the feature data; Solving the objective function based on the constraint conditions to obtain an optimal change execution sequence; A change operation is performed based on the optimal change execution sequence to obtain a change result corresponding to the original change request text.

2. The change execution method according to claim 1, characterized in that: The constructing of the objective function based on the feature data specifically includes: Performing mathematical conversion on the characteristic data to obtain a mathematical expression corresponding to the characteristic data; An objective function is constructed based on the mathematical expression corresponding to the feature data.

3. The change execution method according to claim 1, characterized in that: Solving the objective function based on the constraint conditions to obtain the optimal change execution sequence specifically includes: The improved NSGA-III is used to solve the objective function based on the constraints and obtain the optimal change execution sequence.

4. The change execution method according to claim 3, characterized in that: The improved NSGA-III is used to solve the objective function based on the constraint conditions to obtain the optimal change execution sequence, specifically including: Let the value of k be 1; Constructing an initialization population based on the original change request text, and using the initialization population as the population for the kth iteration; performing non-dominated sorting and adaptive normalization processing on the population of the k-th iteration to obtain a normalized population of the k-th iteration; Selecting the normalized population of the k-th iteration using a niche retention strategy to obtain a selected population of the k-th iteration; Performing a crossover mutation process on the selected population of the k-th iteration to obtain a subpopulation of the k-th iteration; Obtaining a final population of the kth iteration based on the selected population of the kth iteration and the subpopulation of the kth iteration using an elite retention strategy; If the iteration end condition is met, the individuals in the final population of the kth iteration are output as the optimal change execution sequence; If the iteration end condition is not met, the final population of the kth iteration is used as the population of the k+1th iteration, and the value of k is increased by 1, and the step of performing non-dominated sorting and adaptive normalization on the population of the kth iteration is returned to obtain the normalized population of the kth iteration.

5. The change execution method according to claim 4, characterized in that: The step of constructing an initialization population based on the original change request text and using the initialization population as the population for the kth iteration specifically includes: Determine the basic parameters corresponding to the original change request text; Random sampling is performed based on the basic parameters to obtain multiple individuals; each individual represents a change execution sequence; Performing repair processing on the multiple individuals, and filtering and screening the multiple individuals based on the constraint conditions to obtain multiple filtered individuals; An initialization population is constructed based on the multiple screened individuals, and the initialization population is used as the population of the kth iteration.

6. The change execution method according to claim 4, characterized in that: The step of performing crossover mutation processing on the selected population of the k-th iteration to obtain a subpopulation of the k-th iteration specifically includes: Performing a crossover process on the selected population of the k-th iteration to obtain a crossover population of the k-th iteration; If there are individuals meeting the preset conditions in the cross population of the kth iteration, the individuals meeting the preset conditions are repaired to obtain a repaired cross population of the kth iteration; If there is no individual meeting the preset conditions in the crossover population of the kth iteration, the crossover population of the kth iteration is used as the repaired crossover population of the kth iteration; Performing mutation processing on the repaired crossover population of the k-th iteration to obtain a mutant population of the k-th iteration; If there are individuals meeting the preset conditions in the variant population of the kth iteration, the individuals meeting the preset conditions are repaired or split to obtain a repaired variant population of the kth iteration; If there is no individual meeting the preset conditions in the mutant population of the kth iteration, the mutant population of the kth iteration is used as the repaired mutant population of the kth iteration; If the repaired mutant population of the k-th iteration does not meet the resource capacity, splitting the repaired mutant population of the k-th iteration to obtain a sub-population of the k-th iteration; If the repaired mutant population of the k-th iteration meets the resource capacity, the repaired mutant population of the k-th iteration is used as the subpopulation of the k-th iteration.

7. The change execution method according to claim 1, characterized in that: The optimal change execution sequence includes multiple change operations; performing the change operation based on the optimal change execution sequence to obtain a change result corresponding to the original change request text specifically includes: Using a cosine similarity algorithm to perform script matching on multiple change operations in the optimal change execution sequence based on a script library, obtaining a matching result corresponding to each change operation; the matching result includes a first result or a second result; the matching success includes a matching script and a matching success; the second result includes a matching failure; Based on the risk prediction model and the matching results corresponding to each change operation, the change operation is performed to obtain the change result corresponding to the original change request text.

8. The change execution method according to claim 7, characterized in that: The change operation is performed based on the risk prediction model and the matching result corresponding to each change operation to obtain the change result corresponding to the original change request text, specifically including: Perform the change operation based on the matching results corresponding to each change operation, and use the risk prediction model to predict the risk of each change operation during the change operation process to obtain the prediction results corresponding to each change operation; If the prediction result is greater than or equal to the preset threshold, then stop the change operation corresponding to the prediction result greater than the preset threshold; If the predicted result is smaller than the preset threshold, a change operation corresponding to the predicted result smaller than the preset threshold is performed to obtain a change result corresponding to the original change request text.

9. A change execution device, characterized in that: The change execution device includes: The acquisition module is used to obtain the original change request text; A feature extraction module is used to extract features from the original change request text to obtain feature data; A construction module, configured to construct an objective function based on the feature data; A solution module, configured to solve the objective function based on the constraint conditions to obtain an optimal change execution sequence; The execution module is used to perform a change operation based on the optimal change execution sequence to obtain a change result corresponding to the original change request text.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the change execution method according to any one of claims 1 to 8 is implemented.

Images