Test flow generation method and device, equipment, medium and program product

Abstract

The invention provides a test flow generation method which can be applied to the technical field of artificial intelligence. The method comprises the following steps: acquiring basic traffic of a service module, wherein the basic traffic is recorded in a corresponding service module based on a pre-configured test process template; variation learning is carried out on the basic traffic, and extended traffic is generated based on a breadth-depth joint search algorithm; obtaining the production flow of the remote bank system in the production environment; analyzing the remote bank system code library to identify a system function scene; and inputting the basic traffic, the extended traffic, the generated traffic and the system function scene into a large language model to generate a target test traffic. The invention further provides a test traffic generation device and equipment, a storage medium and a program product.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to the technical field of automated testing, and more particularly to a test traffic generation method, device, equipment, medium and program product. BACKGROUND

[0002] With the increasing online degree of banking business, remote banking systems have become the key channel for financial institutions to provide customers with multi-service such as artificial seat service, telephone voice, text chat, audio and video business handling, etc. However, when the remote banking system is in version iteration, function upgrade or emergency patch online, it needs to comprehensively test these business modules and their cross-module interaction links. In the related art, manual testing methods are generally used. Test personnel manually design test cases according to requirement documents, simulate the operation path of customers in each business module, and then manually confirm whether the result meets the expectation. For the linkage scene across business modules, test personnel need to sort out the interfaces between modules, data formats and calling sequences, and the workload increases exponentially with the number of modules and the complexity of interaction. This leads to high testing cost, incomplete coverage of function scenarios, and insufficient testing of cross-module interaction. Therefore, there is an urgent need for a test traffic generation scheme that is automatic, efficient, and fully covers business scenarios and cross-module interaction. SUMMARY

[0003] In view of the above problems, the present application provides a test traffic generation method, device, equipment, medium and program product that is automatic, efficient, and fully covers business scenarios and cross-module interaction.

[0004] According to a first aspect of the present application, a test traffic generation method is provided, applied to a remote banking system, the remote banking system including at least two business modules among an artificial seat service module, a telephone voice interaction module, a text chat consultation module and an audio and video business handling module, comprising: obtaining basic traffic of a business module, the basic traffic being obtained by recording in the corresponding business module based on a pre-configured test process template; performing variation learning on the basic traffic, and generating extended traffic based on a breadth-depth joint search algorithm; obtaining production traffic of the remote banking system in a production environment; parsing a remote banking system code library to identify system function scenarios; inputting the basic traffic, the extended traffic, the generated traffic and the system function scenarios into a large language model to generate target test traffic; wherein when generating the extended traffic based on the breadth-depth joint search algorithm, sibling branches within the same business module and joint calling chains across the at least two business modules are traversed at the same time to cover the test requirements of the remote banking system in the multi-module interaction scene.

[0005] The method according to the embodiments of the present application comprises: inputting the basic traffic, the extended traffic, the generated traffic and the system function scenario into a large language model to generate target test traffic.

[0006] According to the embodiments of the present application, the fine-tuning training of the initial large language model by the basic traffic, the extended traffic, the generated traffic and the system function scenario to obtain the fine-tuned large language model comprises: performing data normalization processing on the basic traffic, the extended traffic, the generated traffic and the system function scenario to output a normalized training set; mapping samples of the normalized training set to corresponding function branches with the system function scenario as an index to generate a scene sample alignment table; using a breadth-depth joint search algorithm to search and complete the normalized training set with the scene sample alignment table as a search space to output a completed candidate input set; and inputting the normalized training set and the completed candidate input set into the initial large language model as training corpus, fine-tuning using a reinforcement learning algorithm to obtain the fine-tuned large language model.

[0007] According to the embodiments of the present application, the fine-tuning training of the initial large language model by the basic traffic, the extended traffic, the generated traffic and the system function scenario to obtain the fine-tuned large language model comprises: for each sample in the training corpus, generating test traffic by the initial large language model and executing the test traffic in an isolated sandbox to obtain coverage improvement value, redundancy value and abnormality rate value; converting the coverage improvement value, the redundancy value and the abnormality rate value into reward scores according to a preset reward rule, wherein the coverage improvement value generates positive reward and the redundancy value and the abnormality rate value generate negative reward; and using a proximal policy optimization algorithm to update parameters of the initial large language model according to the reward scores until the reward scores converge to obtain the fine-tuned large language model.

[0008] According to an embodiment of the present application, the generating target test traffic using the fine-tuned large language model comprises: obtaining a function scenario list, the function scenario list comprising business module information, interface information corresponding to the business module, and parameter range information; converting scenario information of the function scenario list into a structured prompt, the structured prompt containing a business module identifier; generating a candidate test input set according to the structured prompt using the fine-tuned large language model; filtering the candidate test input set according to coverage, redundancy, and abnormality rate; instantiating the filtered candidate test input set into real test data according to field types, wherein the instantiation comprises one or more of text field filling, audio file generation, and image file generation; and packaging the real test data into a target traffic package, the target traffic package containing a number, a corresponding business module, a scenario name, a priority, and dependency service information.

[0009] According to an embodiment of the present application, the generating extended traffic based on the breadth-depth joint search algorithm comprises: parsing the basic traffic to extract business scenario nodes and call chain information to construct a search space; using a breadth-first strategy to traverse all branches in parallel at the same level to generate a first candidate input set; using a depth-first strategy to explore layers along the call chain based on the first candidate input set until a leaf node is reached to generate a second candidate input set; and performing deduplication and effectiveness filtering on the first candidate input set and the second candidate input set to remove invalid or duplicate inputs to form the extended traffic.

[0010] A second aspect of the present application provides a test traffic generation device, the device comprising: a first obtaining module for obtaining basic traffic of a business module, the basic traffic being recorded based on a pre-configured test process template in a corresponding business module; an extension module for performing variation learning on the basic traffic and generating extended traffic based on a breadth-depth joint search algorithm; a second obtaining module for obtaining production traffic of the remote banking system in a production environment; a parsing module for parsing a remote banking system code library to identify system function scenarios; a generating module for inputting the basic traffic, the extended traffic, the generated traffic, and the system function scenarios into a large language model to generate target test traffic; wherein the generating extended traffic based on the breadth-depth joint search algorithm simultaneously traverses sibling branches within the same business module and joint call chains across the at least two business modules to cover test requirements of the remote banking system in multi-module interaction scenarios.

[0011] According to an embodiment of the present application, the generating module comprises a training submodule and a generating submodule.

[0012] The training submodule is configured to fine-tune the initial large model by using the basic traffic, the extended traffic, the generated traffic, and the system function scenario as training corpus, so as to obtain a fine-tuned large language model; and the generation submodule is configured to generate target test traffic by using the fine-tuned large language model.

[0013] According to an embodiment of the present application, the training submodule comprises a data normalization processing unit, a first generation unit, an output unit, and a training unit.

[0014] The data normalization processing unit is configured to perform data normalization processing on the basic traffic, the extended traffic, the generated traffic, and the system function scenario, so as to output a normalized training set; the first generation unit is configured to map samples of the normalized training set to corresponding function branches by using the system function scenario as an index, so as to generate a scene sample alignment table; the first output unit is configured to search and complete the normalized training set by using a breadth-depth joint search algorithm, so as to output a completed candidate input set, by using the scene sample alignment table as a search space; and the training unit is configured to input the normalized training set and the completed candidate input set into an initial large language model as training corpus, and fine-tune the initial large language model by using a reinforcement learning algorithm, so as to obtain a fine-tuned large language model.

[0015] According to an embodiment of the present application, the training unit is specifically configured to, for each sample in the training corpus, generate test traffic by using the initial large language model, and execute the test traffic in an isolated sandbox, so as to obtain a coverage improvement value, a redundancy value, and an abnormality rate value; convert the coverage improvement value, the redundancy value, and the abnormality rate value into a reward score according to a preset reward rule, wherein the coverage improvement value generates positive reward, and the redundancy value and the abnormality rate value generate negative reward; and update parameters of the initial large language model according to the reward score by using a proximal policy optimization algorithm, until the reward score converges, so as to obtain the fine-tuned large language model.

[0016] According to an embodiment of the present application, the generation submodule comprises an acquisition unit, a conversion unit, a second generation unit, a filtering unit, an instantiation unit, and an encapsulation unit.

[0017] The acquisition unit is configured to acquire a function scenario list, the function scenario list including business module information, interface information corresponding to the business module, and parameter range information; the conversion unit is configured to convert scenario information of the function scenario list into a structured prompt, the structured prompt including a business module identifier; the second generation unit is configured to generate a candidate test input set according to the structured prompt by using a fine-tuned large language model; the filtering unit is configured to filter the candidate test input set according to coverage, redundancy, and abnormality rate; the instantiation unit is configured to instantiate the filtered candidate test input set into real test data according to field types, the instantiation including one or more of text field filling, audio file generation, and image file generation; and the packaging unit is configured to package the real test data into a target traffic packet, the target traffic packet including a number, a corresponding business module, a scenario name, a priority, and dependency service information.

[0018] According to an embodiment of the present application, the extension module includes a parsing submodule, a first candidate input set generation submodule, a second candidate input set submodule, and an extended traffic generation submodule.

[0019] The parsing submodule is configured to parse the basic traffic, extract business scenario nodes and call chain information, and construct a search space; the first candidate input set generation submodule is configured to use a breadth-first strategy to traverse all branches in parallel at the same level to generate a first candidate input set; the second candidate input set submodule is configured to use a depth-first strategy to explore layer by layer along the call chain until a leaf node based on the first candidate input set to generate a second candidate input set; and the extended traffic generation submodule is configured to remove invalid or duplicate inputs by performing deduplication and validity filtering on the first candidate input set and the second candidate input set to form extended traffic.

[0020] A third aspect of the present application provides an electronic device, comprising: one or more processors; a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the above method.

[0021] A fourth aspect of the present application further provides a computer-readable storage medium having a computer program or instructions stored thereon, wherein the computer program or instructions are executed by a processor to implement the steps of the above method.

[0022] A fifth aspect of the present application further provides a computer program product comprising a computer program or instructions, wherein the computer program or instructions are executed by a processor to implement the steps of the above method. BRIEF DESCRIPTION OF DRAWINGS

[0023] The above and other objects, features and advantages of the present application will become more apparent from the following description when taken in conjunction with the accompanying drawings, in which:

[0024] Figure 1 An application scenario diagram of a test traffic generation method, apparatus, device, medium and program product according to an embodiment of the present application is schematically shown;

[0025] Figure 2 A flowchart of a test traffic generation method according to an embodiment of the present application is schematically shown;

[0026] Figure 3 A flowchart of a test traffic generation method according to another embodiment of the present application is schematically shown;

[0027] Figure 4 A structural block diagram of a test traffic generation apparatus according to an embodiment of the present application is schematically shown; and

[0028] Figure 5 A block diagram of an electronic device suitable for implementing a test traffic generation method according to an embodiment of the present application is schematically shown. DETAILED DESCRIPTION

[0029] Hereinafter, embodiments of the present application will be described with reference to the accompanying drawings. It is to be understood, however, that the description is merely exemplary of the present application, and is not intended to limit the scope of the present application. In the following detailed description of the embodiments of the present application, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be apparent to one skilled in the art that the present application can be practiced without these specific details. In other instances, well-known structures and functions have not been described in detail in order to avoid obscuring aspects of the present application.

[0030] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the present application. As used herein, the term "includes" and tautological expressions thereof, means the inclusion of but not limited to.

[0031] All terms used herein including technical and scientific terms have the same meanings as commonly understood by one of ordinary skill in the art unless otherwise defined herein. It should be noted that the terms used herein are merely specific examples for the described present application, and should not be interpreted in an idealized or overly formal way.

[0032] In the case of using expressions such as "at least one of A, B, and C", it is generally intended that the inclusion of at least one of A or B or C in a given claim is satisfied when at least one of the following is true: A is included in a given claim; B is included in the given claim; C is included in the given claim; A, B and C are included in the given claim; if in the above examples, A to be considered to encompass at least one of the possible set of alternatives of A; similarly for B; similarly for C and for any combinations of these.

[0033] With the deepening of the online degree of banking business, the remote banking system has become a key channel for financial institutions to provide customers with artificial seats, telephone voice, text chat, audio and video business handling and other diversified services. The system is usually composed of multiple business modules such as artificial seat service module, telephone voice interaction module, text chat consultation module and audio and video business handling module, covering core business processes such as opening an account, transferring money, financial management, loss reporting and password resetting.

[0034] However, when the remote banking system is version iteration, function upgrade or emergency patch online, it needs to comprehensively test these business modules and their cross-module interaction links. At present, the industry generally adopts manual testing method: test personnel manually design test cases according to requirement documents, simulate customer operation paths in each business module, and manually confirm whether the results meet the expectations. For cross-business module linkage scenarios, such as "customer first consults credit card limit through text chat, and then applies for limit adjustment through telephone voice", test personnel need to sort out the interfaces between modules, data formats and calling sequence, and the workload increases exponentially with the number of modules and interaction complexity.

[0035] The related technical solutions have the following defects: high testing cost: manual design, execution and regression testing requires a large number of test personnel, and the labor cost is high. Function scenarios are difficult to cover: there are many business modules in the remote banking system, and the interaction links are complex. Manual methods cannot exhaust all branches and boundary conditions, resulting in high risk of test omission. Cross-module interaction testing is insufficient: the data formats and interface specifications of different business modules differ, and manual methods cannot systematically simulate cross-module calling chains, which are prone to omit defects caused by data loss, format errors or timing problems. Low efficiency of repeated testing of multiple versions: the same function point needs to be tested manually in each version iteration, which is low in efficiency and cannot meet the needs of rapid iteration and high-quality delivery. Limited performance and security testing capabilities: manual testing is not sufficient in high concurrency, abnormal input, security penetration and other special scenarios, and it is difficult to fully evaluate system performance bottlenecks and potential security vulnerabilities.

[0036] To solve the technical problem above, the embodiments of the present application provide a test traffic generation method, applied to a remote banking system, the remote banking system comprising at least two service modules among an artificial seat service module, a telephone voice interaction module, a text chat consultation module and an audio / video service handling module, comprising: obtaining basic traffic of a service module, the basic traffic being obtained by recording in the corresponding service module based on a pre-configured test process template; performing variation learning on the basic traffic, generating extended traffic based on a breadth-depth joint search algorithm; obtaining production traffic of the remote banking system in a production environment; parsing a remote banking system code library to identify system function scenarios; inputting the basic traffic, the extended traffic, the generated traffic and the system function scenarios into a large language model to generate target test traffic; wherein when generating the extended traffic based on the breadth-depth joint search algorithm, sibling branches within the same service module and joint call chains across the at least two service modules are traversed at the same time to cover test requirements of the remote banking system in multi-module interaction scenarios.

[0037] Figure 1 An application scenario diagram of the test traffic generation method, apparatus, device, medium and program product according to the embodiments of the present application is schematically shown.

[0038] As shown in Figure 1 The application scenario 100 according to the embodiments can include an automated test traffic generation scenario. The network 104 is a medium for providing communication links between the first terminal device 101, the second terminal device 102, the third terminal device 103 and the server 105. The network 104 can include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0039] The user can use the first terminal device 101, the second terminal device 102, the third terminal device 103 to interact with the server 105 through the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only as examples).

[0040] The first terminal device 101, the second terminal device 102, the third terminal device 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smartphones, tablet computers, laptop computers and desktop computers, etc.

[0041] The server 105 can be a server providing various services, such as a background management server (only as an example) providing support for a website browsed by a user using the first terminal device 101, the second terminal device 102, and the third terminal device 103. The background management server can perform analysis and other processing on received user requests and other data, and feed back the processing results (such as a webpage, information, or data generated or obtained according to a user request) to the terminal device.

[0042] It should be noted that the test traffic generation method provided in the embodiments of the present application can generally be executed by the server 105. Correspondingly, the test traffic generation apparatus provided in the embodiments of the present application can generally be arranged in the server 105. The test traffic generation method provided in the embodiments of the present application can also be executed by a server or a server cluster different from the server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or the server 105. Correspondingly, the test traffic generation apparatus provided in the embodiments of the present application can also be arranged in a server or a server cluster different from the server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or the server 105.

[0043] It should be understood that Figure 1 The number of terminal devices, networks, and servers in the above scenario is only illustrative. According to the needs of implementation, there can be any number of terminal devices, networks, and servers.

[0044] The test traffic generation method according to the embodiments of the present application will be described in detail below based on the scenario described above. Figure 1 Figures 2-5 The test traffic generation method according to the embodiments of the present application will be described in detail below based on the scenario described above.

[0045] Figure 2 A flowchart of the test traffic generation method according to the embodiments of the present application is schematically shown.

[0046] As shown in Figure 2 The test traffic generation method of this embodiment includes operations S210-S250, which can be executed by a server or other computing device.

[0047] The test traffic generation method provided in the embodiments of the present application is applied to a remote banking system including at least two service modules of an artificial seat service module, a telephone voice interaction module, a text chat consultation module, and an audio / video service handling module.

[0048] In operation S210, the basic traffic of the service module is obtained.

[0049] According to the embodiments of the present application, the basic traffic is obtained based on a pre-configured test flow template in the corresponding service module.​

[0050] In one example, a test flow template is preset, including test steps, input types, and expected results; test data such as voice, text, and pictures are uploaded, basic traffic is executed and recorded; the basic traffic is manually verified, and after passing, it is stored in the warehouse. Since the basic traffic in the present application is obtained by real recording on the specified business module after manually configuring the test flow template, each basic traffic naturally has a "business module" attribute and can be marked by metadata. Specifically, during the configuration stage, the operator selects the target business module (human operator, telephone voice, text chat, audio and video handling, etc.) on the interface; the system writes the selection into the metadata of the test flow template to form a corresponding module identification field. During the traffic recording stage, the recording script starts the test stub or sandbox of the corresponding business module according to the module identification field; the module identification field is written in the header of the generated traffic file, so that the basic traffic corresponds to the business module one by one.

[0051] In operation S220, the basic traffic is subjected to variation learning, and the extended traffic is generated based on a breadth-depth joint search algorithm.

[0052] According to the embodiments of the present application, the variation learning of the basic traffic and the generation of the extended traffic based on the breadth-depth joint search algorithm include: analyzing the basic traffic, extracting business scenario nodes and call chain information, and constructing a search space; using a breadth-first strategy, all branches at the same level are traversed in parallel to generate a first candidate input set; based on the first candidate input set, a depth-first strategy is used to explore layer by layer along the call chain until the leaf node to generate a second candidate input set; the first candidate input set and the second candidate input set are de-duplicated and filtered for effectiveness to eliminate invalid or duplicate inputs to form the extended traffic.

[0053] According to the embodiments of the present application, when the extended traffic is generated based on the breadth-depth joint search algorithm, the sibling branches in the same business module and the joint call chains across the at least two business modules are traversed at the same time to cover the test requirements of the remote banking system in the multi-module interaction scenario.

[0054] In one example, in the complete implementation process of operation S220, the system first takes the basic traffic that has completed manual verification as the only input source; these traffic respectively comes from the manual seat service module, the telephone voice interaction module, the text chat consultation module and the audio and video business handling module in the remote bank system, covering typical business scenarios such as opening an account, transferring money, balance inquiry, password modification, credit card limit adjustment, etc. In order to convert discrete real interaction records into a search space that can be extended by algorithms, the business scenario nodes are extracted by static scanning of the code repository, and the node granularity is accurate to the interface level; through the protocol parser, the basic traffic is parsed at the message level to generate multiple call chains, which not only record the calling order between modules, but also retain data dependencies, exception branches, and historical defect density and other weight information. Then, taking the node as the vertex and the calling relationship as the edge, a directed acyclic graph with configurable edge weights is constructed; the edges that cross different business modules are marked so that the subsequent search stage can identify and preferentially expand the cross-module links.

[0055] In one example, in the breadth-first search stage, a parallel framework is adopted, and the root node set is evenly distributed to multiple Map tasks; each Map task parallelly expands all brother branches at the same level, and calls the interface contract verifier in real time during the expansion process to intercept the parameter legality, field missing, and type error in time, and prunes the illegal path immediately to avoid invalid expansion. After multiple iterations, the system finds multiple sibling branches in the telephone voice module, the text chat module and the audio and video module, and generates a first candidate input set; in order to ensure that the data size is controllable, the depth of these inputs can be limited within 3 layers. Then enter the depth-first search stage, the system starts a thread pool based on the work-stealing mechanism, and the number of threads is consistent with the number of logical CPU cores to maximize CPU utilization. Each thread starts from the end node of the first candidate input and explores layer by layer along the cross-module call chain, and the termination condition is to reach the leaf node or the depth reaches the preset number of layers; in this process, the system focuses on excavating typical cross-module links, such as "text chat uploads identity card picture → audio and video live detection → account core opening → SMS gateway result notification", and finally generates a second candidate input set, with a depth distribution between 4 and 7 layers, and each input clearly carries the source module and target module identifier.

[0056] In one example, after completing the breadth and depth two-layer search, the system needs to perform deduplication and validity filtering on the candidate input. First, the SHA-256 hash algorithm is used to calculate the fingerprint of the request body, and the collision rate is controlled within 0.01% to eliminate duplicate traffic; then the remaining input is verified by the checker and the business rule engine, the checker is responsible for format problems such as field missing, type error, length overrun, and the business rule engine is responsible for business logic problems such as card number length, amount range, and verification code length; finally, each candidate input is sent to the isolated sandbox for lightweight empty running, and if there is a timeout, an abnormal return code or resource leakage, it is marked as invalid and discarded. After multiple rounds of filtering, the system finally obtains the extended traffic without redundancy, dead chain, and format compliance, and writes it into the extended traffic library.

[0057] In one example, in order to verify the coverage effect of the cross-module interaction scenario, the system takes the "module-module" binary tuple as the statistical dimension to perform backtracking analysis on the extended traffic: for example, 28 new complete paths are added between telephone voice and text chat, 19 new complete paths are added between telephone voice and audio video, and 23 new complete paths are added between text chat and audio video, and the cross-module interaction scenario coverage rate is improved from 34% in traditional manual testing to 87%; in the sandbox playback phase. The generated extended traffic can be directly sent to the large language model together with the basic traffic, production traffic and function scenario file, laying a data foundation for subsequent fine-tuning and target test traffic generation.

[0058] In operation S230, production traffic of the remote banking system in a production environment is obtained.

[0059] In operation S240, the remote banking system code library is parsed to identify system function scenarios.

[0060] In operation S250, the basic traffic, the extended traffic, the generated traffic, and the system function scenarios are input into a large language model to generate target test traffic.

[0061] In one example, during the implementation phase of operations S230 to S250, the system sequentially completes three interrelated tasks: first, it collects business traffic from the real production environment; second, it performs white-box analysis on the codebase to extract functional scenarios; and finally, it injects four types of information (basic traffic, extended traffic, production traffic, and functional scenarios) into a large language model, thereby outputting test traffic that can be directly replayed. Specifically, in the production traffic acquisition phase, the system uses traffic mirroring to capture the real interactions of the remote banking system throughout the day without loss. Then, it performs anonymization and format normalization, retaining only business fields and interaction features to ensure the model can learn the real business distribution without infringing on user privacy. In the functional scenario identification phase, based on static code scanning, the system extracts interface definitions, call chains, and exception branches from the remote banking codebase to form a "functional scenario-call chain" knowledge graph. Nodes in the graph correspond to business operations, and edges correspond to module or cross-module call relationships, providing structured background knowledge for subsequent model training. After vectorizing the four types of data—basic traffic, extended traffic, production traffic, and functional scenarios—into a unified pattern, they are fed into a domain-fine-tuned large language model. Guided by the triple rewards of "increased coverage, reduced redundancy, and anomaly suppression," the model generates test inputs one by one through a constrained decoding process. The generated inputs are first validated in a sandbox to verify their format and business validity, and then packaged into reusable traffic packages according to modules, scenarios, and priorities. At the same time, a test report without conclusions is provided, realizing a closed loop from data to executable test assets.

[0062] The method provided in this application ensures that the test input remains consistent with the semantics of online business by recording real interaction sequences within each business module based on a pre-configured test process template. Based on the constructed search space, a breadth-first strategy enumerates all sibling branches in parallel at the same level, while a depth-first strategy explores down the cross-module call chain layer by layer to the leaf nodes. The two strategies work together to expand the originally discrete single-module scenario into a multi-module interaction link, extending the test scope from vertical depth to a combined coverage of horizontal breadth and vertical depth. Production traffic provides real runtime data, while codebase parsing provides static functional scenarios; the two complement each other, both supplementing boundary conditions and introducing abnormal branches to avoid the model missing potential defects. Basic traffic, extended traffic, production traffic, and system functional scenarios are uniformly formatted and input into a large language model. Based on the integration of the "real-extended-boundary" ternary knowledge, the model outputs directly replayable test traffic, completing a closed loop from data collection and scenario expansion to automated output. By leveraging the mutation learning capabilities of semi-intelligent large models and the scene recognition capabilities of fully intelligent large models, the depth and breadth of test scenarios can be expanded, effectively solving the problem of insufficient coverage of functional scenarios in existing technologies and improving the comprehensiveness and accuracy of testing.

[0063] Figure 3A flowchart illustrating a test traffic generation method according to another embodiment of this application is shown schematically.

[0064] like Figure 3 As shown, operation S250 includes operation S251 and operation S252.

[0065] In operation S251, the basic traffic, the extended traffic, the generated traffic, and the system functional scenario are used to fine-tune the initial large model to obtain the fine-tuned large language model.

[0066] According to an embodiment of this application, the basic traffic, the extended traffic, the generated traffic, and the system functional scenarios are normalized to output a normalized training set; using the system functional scenarios as indexes, the samples in the normalized training set are mapped to corresponding functional branches to generate a scenario sample alignment table; using the scenario sample alignment table as the search space, a breadth-depth joint search algorithm is used to search and complete the normalized training set to output a completion candidate input set; the normalized training set and the completion candidate input set are used as training corpus inputs to the initial large language model, and a reinforcement learning algorithm is used for fine-tuning to obtain the fine-tuned large language model.

[0067] In one example, in operation S251, before fine-tuning the initial large model with the basic traffic, extended traffic, generated traffic, and system functional scenarios, the input data is first normalized. This transforms the four types of original input data (basic traffic, extended traffic, generated traffic, and system functional scenarios) into a continuous vector representation recognizable by the model: text fields are uniformly encoded, and audio and image fields are uniformly sampled and sized. Subsequently, each sample is appended with a source label and module identifier, forming a complete and unambiguous normalized training set. Using functional scenarios as an index, a hash mapping table of scenarios and samples is established: each functional branch corresponds to several normalized samples, ensuring that the model can accurately locate which samples belong to which business link during subsequent training; missing branches are explicitly marked, providing a clear target for search completion. Using the scene-sample alignment table as the search space, a joint breadth-depth search strategy is executed: breadth-first search enumerates sibling branches at the same level to ensure horizontal coverage; depth-first search explores down the call chain layer by layer to the leaf nodes to ensure vertical depth; cross-module links are traversed synchronously to complete interaction paths that are easily missed by manual and static analysis. After deduplication and validity filtering, a complete candidate input set is generated, forming a complementary corpus with the original training set.

[0068] By adding a three-stage preprocessing step—normalization, alignment, and completion—before fine-tuning, normalization ensures consistent data formats across multiple sources, preventing model bias due to heterogeneous inputs; the alignment table precisely binds samples to functional branches, preventing "learning the wrong scenario"; and the completion stage utilizes a breadth-depth joint search to actively fill in blank branches, maintaining the logical integrity of the training corpus. These three steps together ensure that the fine-tuned corpus is both realistic and complete, providing a high-quality baseline for subsequent reinforcement learning.

[0069] According to an embodiment of this application, for each sample in the training corpus, the initial large language model generates test traffic, and the test traffic is executed in an isolated sandbox to obtain coverage improvement value, redundancy value, and anomaly rate value; according to a preset reward rule, the coverage improvement value, the redundancy value, and the anomaly rate value are converted into reward scores, wherein the coverage improvement value generates positive rewards, and the redundancy value and the anomaly rate value generate negative rewards; a near-end policy optimization algorithm is used to update the parameters of the initial large language model according to the reward scores until the reward scores converge to obtain a fine-tuned large language model.

[0070] In one example, after generating a normalized training set and a complete candidate input set, both are used together as training corpus. The following closed-loop iteration is performed: the initial large language model (policy network) generates candidate test traffic for each corpus; the candidate traffic is executed in an isolated sandbox, and feedback data on coverage improvement, redundancy, and anomaly rate are collected in real time; reward scores are calculated according to preset reward rules. Positive rewards are given for improved coverage, while negative rewards are given for redundancy and anomalies. Specifically, the preset reward rules could be, for example, that if the traffic generated by the model can make untested branches become tested, then the representation coverage improvement is given a positive score; if the generated traffic is too similar to existing samples, then the representation redundancy is high and a negative score is given; if system errors, timeouts, or crashes occur while running the traffic in the sandbox, then the representation anomaly rate is high and a negative score is given. A proximal policy optimization algorithm is used to update the model parameters, and constraints are used to prevent the policy from deviating too far; training stops when the reward signal shows no significant fluctuations for several consecutive rounds, resulting in a fine-tuned large language model. Through a four-level progression of normalization, alignment, completion, and fine-tuning, the model not only grasps the real business distribution but also fills in potential defect scenarios. The reinforcement learning closed loop enables the generation strategy to continuously converge towards high coverage, low redundancy, and low anomaly. The final output target test traffic can be directly used for automated playback in remote banking systems without manual enumeration.

[0071] A policy-environment-reward-update closed loop is embedded at the training corpus level. The policy network generates traffic, and the sandbox environment provides real-time feedback on coverage, redundancy, and anomaly rate. The reward function guides the model parameters to converge towards "high coverage, low redundancy, and low anomaly." Proximal policy optimization ensures the stability of the update process and prevents policy drift. This achieves the technical effect of "learning and testing simultaneously, and optimizing while testing," enabling the model output to continuously approach the ideal policy.

[0072] When operating S252, target test traffic is generated using a fine-tuned large language model.

[0073] According to an embodiment of this application, a list of functional scenarios is obtained, which includes business module information, interface information corresponding to the business modules, and parameter range information; the scenario information in the list of functional scenarios is converted into structured prompts, which include business module identifiers; a candidate test input set is generated using a fine-tuned large language model according to the structured prompts; the candidate test input set is filtered according to coverage, redundancy, and anomaly rate; the filtered candidate test input set is instantiated into real test data according to field type, wherein instantiation includes one or more of text field filling, audio file generation, and image file generation; the real test data is encapsulated into a target traffic package, which includes a number, the business module to which it belongs, the scenario name, priority, and dependent service information.

[0074] In one example, based on the results of previous code parsing, the system summarizes all business scenario nodes to form a list of functional scenarios. The list includes business module identifiers, interface paths, parameter constraints, boundary conditions, and cross-module call chain information, ensuring that subsequent prompts have a complete semantic context. Each scenario entry is serialized into a prompt text of a unified format, with the business module identifier embedded at the beginning, the interface path and parameter range listed in the body, and boundary conditions added at the end. This structured prompt serves as the sole input format for the fine-tuned large language model, ensuring that the model is always constrained by the established semantics and format specifications during the generation process. The fine-tuned large language model receives the structured prompts and uses a constrained decoding strategy to output candidate test inputs that conform to the interface contract, token by token. At each step, the model verifies the data type, value range, and cross-module call order in real time to ensure that candidate inputs are legal and cover potential branches. Candidate inputs first enter the coverage evaluation module to calculate the proportion of new functional branches they can activate; then, through redundancy measurement, inputs highly similar to existing use cases are eliminated; finally, they are sent to an isolation sandbox for execution, and the anomaly rate is calculated. Only when coverage improves and redundancy and anomaly rate are both below set thresholds can the input proceed to the target traffic generation operation. Based on field type, the filtered candidate test input set is instantiated into real test data. Specifically, text fields are directly filled in; speech fields use a speech synthesis engine to generate audio files conforming to the sampling rate and bit depth; image fields use a diffusion model to generate documents or screenshots suitable for the business scenario. The instantiation process maintains a one-to-one correspondence between fields and files, ensuring complete and executable input during playback. The instantiated real test data is packaged into a target traffic package according to a unified format. The target traffic package includes a number, the business module it belongs to, the scenario name, priority, and dependent service information. The number is a globally unique identifier; the business module it belongs to is consistent with the module identifier in the structured prompts; the scenario name is the corresponding functional scenario entry; the priority is calculated based on a combination of coverage and anomaly rate; and the dependent services list the external services and versions required to execute the traffic. After packaging, the traffic package and the inconclusive test report are output synchronously for direct use by the continuous integration pipeline.

[0075] By setting up a three-level pipeline of "scenario prompting → instantiation → encapsulation" in the target test traffic generation stage, scenario prompting ensures that the model always outputs based on the established business semantics and avoids going out of bounds; instantiation converts the abstract template into text, audio, and image entities that can be directly played back, eliminating the "script secondary development" step and improving testing efficiency; unified encapsulation makes each traffic item come with its own number, module, scenario, priority, and dependency information, which can be directly connected to the continuous integration / continuous delivery pipeline, realizing zero-manual conversion from scenario description to executable assets.

[0076] Based on the above-described test traffic generation method, this application also provides a test traffic generation device. The following will be combined with... Figure 4The device is described in detail.

[0077] Figure 4 A schematic block diagram of a test flow generation apparatus according to an embodiment of this application is shown.

[0078] like Figure 4 As shown, the test traffic generation device 400 of this embodiment includes a first acquisition module 410, an extension module 420, a second acquisition module 430, a parsing module 440, and a generation module 450.

[0079] The first acquisition module 410 is used to acquire the basic traffic of the business module, which is obtained by recording in the corresponding business module based on a pre-configured test process template. In one embodiment, the first acquisition module 410 can be used to perform the operation S210 described above, which will not be repeated here.

[0080] The extension module 420 is used to perform mutation learning on the basic traffic and generate extended traffic based on a breadth-depth joint search algorithm. In one embodiment, the extension module 420 can be used to perform the operation S220 described above, which will not be repeated here.

[0081] The second acquisition module 430 is used to acquire the production traffic of the remote banking system in a production environment. In one embodiment, the second acquisition module 430 can be used to perform the operation S230 described above, which will not be repeated here.

[0082] The parsing module 440 is used to parse the remote banking system codebase to identify system functional scenarios. In one embodiment, the parsing module 440 can be used to perform the operation S240 described above, which will not be repeated here.

[0083] The generation module 450 is used to input the basic traffic, the extended traffic, the generated traffic, and the system functional scenario into a large language model to generate target test traffic. Specifically, when generating extended traffic based on a breadth-depth joint search algorithm, it simultaneously traverses sibling branches within the same business module and joint call chains across at least two business modules to cover the testing requirements of the remote banking system in multi-module interaction scenarios. In one embodiment, the generation module 450 can be used to execute the operation S250 described above, which will not be repeated here.

[0084] According to an embodiment of this application, the generation module includes a training submodule and a generation submodule.

[0085] The training submodule is used to fine-tune the initial large model using the basic traffic, the extended traffic, the generated traffic, and the system functional scenarios to obtain a fine-tuned large language model; and the generation submodule is used to generate target test traffic using the fine-tuned large language model.

[0086] According to an embodiment of this application, the training submodule includes a data normalization processing unit, a first generation unit, an output unit, and a training unit.

[0087] A data normalization processing unit is used to normalize the basic traffic, the extended traffic, the generated traffic, and the system functional scenarios to output a normalized training set; a first generation unit is used to map the samples of the normalized training set to corresponding functional branches using the system functional scenarios as indexes to generate a scenario sample alignment table; a first output unit is used to use the scenario sample alignment table as the search space and perform search and completion on the normalized training set using a breadth-depth joint search algorithm to output a completion candidate input set; and a training unit is used to input the normalized training set and the completion candidate input set as training corpus into the initial large language model, and use a reinforcement learning algorithm to fine-tune it to obtain a fine-tuned large language model.

[0088] According to an embodiment of this application, the training unit is specifically used to generate test traffic for each sample in the training corpus using the initial large language model, and execute the test traffic in an isolated sandbox to obtain coverage improvement value, redundancy value, and anomaly rate value; according to a preset reward rule, the coverage improvement value, the redundancy value, and the anomaly rate value are converted into reward scores, wherein the coverage improvement value generates positive rewards, and the redundancy value and the anomaly rate value generate negative rewards; and a near-end policy optimization algorithm is used to update the parameters of the initial large language model according to the reward scores until the reward scores converge to obtain a fine-tuned large language model.

[0089] According to an embodiment of this application, the generation submodule includes an acquisition unit, a conversion unit, a second generation unit, a filtering unit, an instantiation unit, and an encapsulation unit.

[0090] The system comprises the following components: an acquisition unit for acquiring a list of functional scenarios, including business module information, corresponding interface information, and parameter range information; a conversion unit for converting the scenario information in the functional scenario list into structured prompts, the structured prompts containing business module identifiers; a second generation unit for generating a candidate test input set based on the structured prompts using a fine-tuned large language model; a filtering unit for filtering the candidate test input set according to coverage, redundancy, and anomaly rate; an instantiation unit for instantiating the filtered candidate test input set into real test data based on field types, wherein instantiation includes one or more of text field filling, audio file generation, and image file generation; and an encapsulation unit for encapsulating the real test data into a target traffic package, the target traffic package containing a number, the business module to which it belongs, the scenario name, priority, and dependent service information.

[0091] According to an embodiment of this application, the extension module includes a parsing submodule, a first candidate input set generation submodule, a second candidate input set submodule, and an extended traffic generation submodule.

[0092] The parsing submodule is used to parse the basic traffic, extract business scenario nodes and call chain information to construct a search space; the first candidate input set generation submodule is used to generate the first candidate input set by using a breadth-first strategy to traverse all branches in parallel at the same level; the second candidate input set submodule is used to generate the second candidate input set by using a depth-first strategy to explore down the call chain layer by layer until the leaf node; and the extended traffic generation submodule is used to perform deduplication and validity filtering on the first candidate input set and the second candidate input set, eliminating invalid or duplicate inputs to form extended traffic.

[0093] According to embodiments of this application, any plurality of modules among the parsing module 410, query module 420, and calling module 430 can be merged into one module, or any one of these modules can be split into multiple modules. Alternatively, at least a portion of the functionality of one or more of these modules can be combined with at least a portion of the functionality of other modules and implemented in one module. According to embodiments of this application, at least one of the parsing module 410, query module 420, and calling module 430 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or implemented in hardware or firmware by any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any appropriate combination of any of these three implementation methods. Alternatively, at least one of the parsing module 410, query module 420, and calling module 430 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0094] Figure 5 A block diagram schematically illustrates an electronic device suitable for implementing a test traffic generation method according to an embodiment of this application.

[0095] like Figure 5As shown, an electronic device 500 according to an embodiment of this application includes a processor 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage portion 508 into a random access memory (RAM) 503. The processor 501 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 501 may also include onboard memory for caching purposes. The processor 501 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0096] RAM 503 stores various programs and data required for the operation of electronic device 500. Processor 501, ROM 502, and RAM 503 are interconnected via bus 504. Processor 501 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 502 and / or RAM 503. It should be noted that the programs may also be stored in one or more memories other than ROM 502 and RAM 503. Processor 501 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0097] According to embodiments of this application, the electronic device 500 may further include an input / output (I / O) interface 505, which is also connected to a bus 504. The electronic device 500 may also include one or more of the following components connected to the input / output (I / O) interface 505: an input section 506 including a keyboard, mouse, etc.; an output section 507 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 508 including a hard disk, etc.; and a communication section 509 including a network interface card such as a LAN card, modem, etc. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to the input / output (I / O) interface 505 as needed. A removable medium 511, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 510 as needed so that computer programs read from it can be installed into the storage section 508 as needed.

[0098] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0099] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 502 and / or RAM 503 and / or one or more memories other than ROM 502 and RAM 503 described above.

[0100] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to enable the computer system to implement the test traffic generation method provided in the embodiments of this application.

[0101] When the computer program is executed by the processor 501, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0102] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 509, and / or installed from a removable medium 511. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0103] In such an embodiment, the computer program can be downloaded and installed from a network via communication section 509, and / or installed from removable medium 511. When the computer program is executed by processor 501, it performs the functions defined in the system of this application embodiment. According to embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0104] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0105] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0106] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

Claims

1. A method for generating test traffic, applied to a remote banking system, said remote banking system comprising at least two business modules selected from a human agent service module, a telephone voice interaction module, a text chat consultation module, and an audio / video service processing module, characterized in that, The method includes: Obtain the basic traffic of the business module, which is recorded in the corresponding business module based on a pre-configured test process template; The basic traffic is subjected to mutation learning, and extended traffic is generated based on a breadth-depth joint search algorithm; Obtain the production traffic of the remote banking system in the production environment; The codebase of the remote banking system is parsed to identify system functional scenarios; The basic traffic, the extended traffic, the generated traffic, and the system functional scenario are input into the large language model to generate the target test traffic; When generating extended traffic based on the breadth-depth joint search algorithm, the algorithm simultaneously traverses sibling branches within the same business module and joint call chains across at least two business modules to cover the testing requirements of the remote banking system in multi-module interaction scenarios.

2. The method according to claim 1, characterized in that, The step of inputting the basic traffic, the extended traffic, the generated traffic, and the system functional scenario into the large language model to generate the target test traffic includes: The initial large model is fine-tuned using the basic traffic, the extended traffic, the generated traffic, and the system functional scenarios to obtain a fine-tuned large language model; and The target test traffic is generated using a fine-tuned large language model.

3. The method according to claim 2, characterized in that, The step of fine-tuning the initial large model using the basic traffic, the extended traffic, the generated traffic, and the system functional scenarios to obtain the fine-tuned large language model includes: The basic traffic, the extended traffic, the generated traffic, and the system functional scenarios are normalized to output a normalized training set. Using the system functional scenarios as indexes, the samples of the normalized training set are mapped to the corresponding functional branches to generate a scenario sample alignment table. Using the scene sample alignment table as the search space, a breadth-depth joint search algorithm is used to search and complete the normalized training set, outputting a candidate input set for completion; and The normalized training set and the completion candidate input set are used as training corpus inputs to the initial large language model, and then fine-tuned using a reinforcement learning algorithm to obtain the fine-tuned large language model.

4. The method according to claim 3, characterized in that, The step of using the normalized training set and the completion candidate input set as training corpus to input the initial large language model, and then using a reinforcement learning algorithm to fine-tune it to obtain the fine-tuned large language model includes: For each sample in the training corpus, the initial large language model generates test traffic, and the test traffic is executed in an isolation sandbox to obtain coverage improvement value, redundancy value and anomaly rate value; According to preset reward rules, the coverage improvement value, the redundancy value, and the anomaly rate value are converted into reward scores, wherein the coverage improvement value generates a positive reward, and the redundancy value and the anomaly rate value generate negative rewards; and A proximal strategy optimization algorithm is used to update the parameters of the initial large language model based on the reward score until the reward score converges, so as to obtain the fine-tuned large language model.

5. The method according to claim 2, characterized in that, The generation of target test traffic using the fine-tuned large language model includes: Obtain a list of functional scenarios, which includes business module information, interface information corresponding to the business modules, and parameter range information; The scenario information in the functional scenario list is converted into structured prompts, which include business module identifiers; The candidate test input set is generated using the fine-tuned large language model according to the structured prompts; The candidate test input set is filtered according to coverage, redundancy, and anomaly rate; The filtered candidate test input set is instantiated into real test data according to the field type, wherein the instantiation includes one or more of the following: text field filling, audio file generation, and image file generation; The real test data is packaged into a target traffic package, which includes a number, the business module to which it belongs, the scenario name, the priority, and the dependent service information.

6. The method according to claim 1, characterized in that, The step of performing mutation learning on the basic traffic and generating extended traffic based on a breadth-depth joint search algorithm includes: The basic traffic is analyzed to extract business scenario nodes and call chain information in order to construct a search space; A breadth-first search strategy is adopted to traverse all branches in parallel at the same level to generate the first candidate input set; Based on the first candidate input set, a depth-first strategy is used to explore the call chain layer by layer until the leaf node is reached, in order to generate a second candidate input set; and The first and second candidate input sets are deduplicated and filtered for validity to remove invalid or duplicate inputs, thus forming extended traffic.

7. A test traffic generation device, applied to a remote banking system, said remote banking system comprising at least two business modules selected from a human agent service module, a telephone voice interaction module, a text chat consultation module, and an audio / video business processing module, characterized in that, The device includes: The first acquisition module is used to acquire the basic traffic of the business module. The basic traffic is recorded in the corresponding business module based on a pre-configured test process template. An extension module is used to perform mutation learning on the basic traffic and generate extended traffic based on a breadth-depth joint search algorithm; The second acquisition module is used to acquire the production flow of the remote banking system in the production environment; The parsing module is used to parse the codebase of the remote banking system in order to identify system functional scenarios; The generation module is used to input the basic traffic, the extended traffic, the generated traffic, and the system functional scenario into the large language model to generate target test traffic; When generating extended traffic based on the breadth-depth joint search algorithm, the algorithm simultaneously traverses sibling branches within the same business module and joint call chains across at least two business modules to cover the testing requirements of the remote banking system in multi-module interaction scenarios.

8. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 6.

Images