Model training method, system change risk control method, device and electronic equipment

Abstract

The specification discloses a model training method, a system change risk control method and device, and electronic equipment. The method comprises: obtaining system change samples, the system change samples comprising reference system change samples and comparison system change samples, the reference system change samples being system change samples with an existing change risk, and the comparison system change samples being labeled with risk classification labels. A similarity between the reference system change samples and the comparison system change samples in time sequence feature data in a performance change dimension is calculated by a comparison learning network based on a risk identification model, to obtain feature similarity. A classification calculation network based on the risk identification model is used to perform classification calculation on the feature similarity and time sequence feature data of the comparison system change samples in a change influence dimension, to obtain a predicted risk classification result. A training loss is determined by calculating the predicted risk classification result and an actual risk classification result, and parameters of the comparison learning network and the classification calculation network are adjusted according to the training loss.

Description

Technical Field

[0001] This document belongs to the field of artificial intelligence technology, and in particular relates to a model training method, a risk control method for system changes, and related devices and electronic equipment. Background Technology

[0002] With the development of emerging information technologies such as big data and cloud computing, the architecture of online service systems is becoming increasingly complex. Any changes to such systems, such as code or configuration modifications, may trigger a series of problems that could lead to production accidents and ultimately cause significant losses to the enterprise.

[0003] Therefore, how to accurately and reliably identify the risks of system changes is a technical problem that urgently needs to be solved. Summary of the Invention

[0004] This specification provides a model training method, a risk control method for system changes, an apparatus, and an electronic device that can mechanically identify the risks brought about by system changes and promptly adopt corresponding risk control measures to avoid production accidents after identifying the risks.

[0005] To solve the above-mentioned technical problems, the embodiments in this specification are implemented as follows:

[0006] Firstly, a model training method is proposed, including:

[0007] Obtain system change samples, which include baseline system change samples and comparison system change samples. The baseline system change samples are system change samples with predetermined change risks, and the comparison system change samples are labeled with risk classification tags. The risk classification tags are used to characterize the actual risk classification results of the comparison system change samples.

[0008] A contrastive learning network based on a risk identification model calculates the similarity of the temporal feature data of the baseline system change sample and the contrast system change sample in the dimension of performance change, and obtains the feature similarity.

[0009] The classification calculation network based on the risk identification model performs classification calculations on the feature similarity and the temporal feature data of the comparison system change sample in the dimension of change impact, and obtains the predicted risk classification result of the comparison system change sample;

[0010] Based on the predicted risk classification results of the modified samples in the comparison system and the actual risk classification results of the corresponding labels, the training loss of the risk identification model is determined, and the parameters of the comparison learning network and the classification calculation network are adjusted according to the training loss.

[0011] Secondly, a risk control method for system changes is proposed, including:

[0012] Acquire the time-series characteristic data of the target system change and the baseline system change sample in the performance change dimension, as well as the time-series characteristic data of the target system change in the change impact dimension. The target system change refers to the system change that has been executed or is to be executed, and the baseline system change sample is a system change sample that is known to have change risk.

[0013] The time-series feature data of the target system change and the baseline system change samples on the performance change dimension, as well as the time-series feature data of the target system change on the change impact dimension, are input into a risk identification model to obtain a predicted risk classification result for the target system change. The risk identification model includes a contrastive learning network and a classification calculation network. The contrastive learning network is used to calculate the similarity between the time-series feature data of the target system change and the baseline system change samples on the performance change dimension to obtain feature similarity. The classification calculation network is used to perform classification calculations on the feature similarity and the time-series feature data of the target system change on the change impact dimension to obtain the predicted risk classification result.

[0014] For changes to the target system, risk control decisions are made that match the predicted risk classification results.

[0015] Thirdly, a model training device is proposed, comprising:

[0016] The sample acquisition module acquires system change samples, which include a baseline system change sample and a comparison system change sample. The baseline system change sample is a system change sample with predetermined change risks, and the comparison system change sample is labeled with a risk classification label. The risk classification label is used to characterize the actual risk classification result of the comparison system change sample.

[0017] The contrastive learning module uses a contrastive learning network based on a risk identification model to calculate the similarity of the temporal feature data of the baseline system change sample and the contrastive system change sample in the performance change dimension, and obtains the feature similarity.

[0018] The classification calculation module performs classification calculations on the feature similarity and the temporal feature data of the comparison system change sample in the dimension of change impact based on the classification calculation network of the risk identification model, and obtains the predicted risk classification result of the comparison system change sample;

[0019] The parameter adjustment module determines the training loss of the risk identification model based on the predicted risk classification results of the comparison system changed samples and the corresponding labeled actual risk classification results, and adjusts the parameters of the comparison learning network and the classification calculation network according to the training loss.

[0020] Fourthly, an electronic device is proposed, comprising: a processor; and a memory arranged to store computer-executable instructions, which, when executed, cause the processor to perform the following operations:

[0021] Obtain system change samples, which include baseline system change samples and comparison system change samples. The baseline system change samples are system change samples with predetermined change risks, and the comparison system change samples are labeled with risk classification tags. The risk classification tags are used to characterize the actual risk classification results of the comparison system change samples.

[0022] A contrastive learning network based on a risk identification model calculates the similarity of the temporal feature data of the baseline system change sample and the contrast system change sample in the dimension of performance change, and obtains the feature similarity.

[0023] The classification calculation network based on the risk identification model performs classification calculations on the feature similarity and the temporal feature data of the comparison system change sample in the dimension of change impact, and obtains the predicted risk classification result of the comparison system change sample;

[0024] Based on the predicted risk classification results of the modified samples in the comparison system and the actual risk classification results of the corresponding labels, the training loss of the risk identification model is determined, and the parameters of the comparison learning network and the classification calculation network are adjusted according to the training loss.

[0025] Fifthly, a computer-readable storage medium is provided that stores one or more programs, which, when executed by an electronic device including multiple applications, cause the electronic device to perform the following operations:

[0026] Obtain system change samples, which include baseline system change samples and comparison system change samples. The baseline system change samples are system change samples with predetermined change risks, and the comparison system change samples are labeled with risk classification tags. The risk classification tags are used to characterize the actual risk classification results of the comparison system change samples.

[0027] A contrastive learning network based on a risk identification model calculates the similarity of the temporal feature data of the baseline system change sample and the contrast system change sample in the dimension of performance change, and obtains the feature similarity.

[0028] The classification calculation network based on the risk identification model performs classification calculations on the feature similarity and the temporal feature data of the comparison system change sample in the dimension of change impact, and obtains the predicted risk classification result of the comparison system change sample;

[0029] Based on the predicted risk classification results of the modified samples in the comparison system and the actual risk classification results of the corresponding labels, the training loss of the risk identification model is determined, and the parameters of the comparison learning network and the classification calculation network are adjusted according to the training loss.

[0030] Sixthly, a risk control device for system changes is proposed, including:

[0031] The data acquisition module acquires the time-series characteristic data of the target system change and the baseline system change sample in terms of performance change dimension, as well as the time-series characteristic data of the target system change in terms of change impact dimension. The target system change refers to the system change that has been executed or is to be executed, and the baseline system change sample is a system change sample that is known to have change risk.

[0032] The risk identification module inputs the time-series feature data of the target system change and the baseline system change sample on the performance change dimension, as well as the time-series feature data of the target system change on the change impact dimension, into the risk identification model to obtain the predicted risk classification result of the target system change; wherein, the target system change refers to a system change that has been executed or is to be executed, and the baseline system change sample is a system change sample with a predetermined change risk; the risk identification model includes a contrastive learning network and a classification calculation network, the learning network is used to calculate the similarity of the time-series feature data of the target system change and the baseline system change sample on the performance change dimension to obtain feature similarity; the classification calculation network is used to perform classification calculation on the feature similarity and the time-series feature data of the target system change on the change impact dimension to obtain the predicted risk classification result;

[0033] Risk control execution module. Executes risk control decisions that match the predicted risk classification results in response to the changes to the target system.

[0034] In a seventh aspect, an electronic device is proposed, comprising: a processor; and a memory arranged to store computer-executable instructions, which, when executed, cause the processor to perform the following operations:

[0035] Acquire the time-series characteristic data of the target system change and the baseline system change sample in the performance change dimension, as well as the time-series characteristic data of the target system change in the change impact dimension. The target system change refers to the system change that has been executed or is to be executed, and the baseline system change sample is a system change sample that is known to have change risk.

[0036] The time-series feature data of the target system change and the baseline system change samples on the performance change dimension, as well as the time-series feature data of the target system change on the change impact dimension, are input into a risk identification model to obtain a predicted risk classification result for the target system change. The risk identification model includes a contrastive learning network and a classification calculation network. The contrastive learning network is used to calculate the similarity between the time-series feature data of the target system change and the baseline system change samples on the performance change dimension to obtain feature similarity. The classification calculation network is used to perform classification calculations on the feature similarity and the time-series feature data of the target system change on the change impact dimension to obtain the predicted risk classification result.

[0037] For changes to the target system, risk control decisions are made that match the predicted risk classification results.

[0038] In an eighth aspect, a computer-readable storage medium is provided that stores one or more programs, which, when executed by an electronic device including multiple applications, cause the electronic device to perform the following operations:

[0039] Acquire the time-series characteristic data of the target system change and the baseline system change sample in the performance change dimension, as well as the time-series characteristic data of the target system change in the change impact dimension. The target system change refers to the system change that has been executed or is to be executed, and the baseline system change sample is a system change sample that is known to have change risk.

[0040] The time-series feature data of the target system change and the baseline system change samples on the performance change dimension, as well as the time-series feature data of the target system change on the change impact dimension, are input into a risk identification model to obtain a predicted risk classification result for the target system change. The risk identification model includes a contrastive learning network and a classification calculation network. The contrastive learning network is used to calculate the similarity between the time-series feature data of the target system change and the baseline system change samples on the performance change dimension to obtain feature similarity. The classification calculation network is used to perform classification calculations on the feature similarity and the time-series feature data of the target system change on the change impact dimension to obtain the predicted risk classification result.

[0041] For changes to the target system, risk control decisions are made that match the predicted risk classification results.

[0042] The embodiments in this specification construct a risk identification model consisting of a contrastive learning network and a classification computation network. During the training phase of the risk identification model, a predetermined system change sample with known risk is used as a baseline system change sample. First, the contrastive learning network calculates the feature similarity between the baseline system change sample and the contrastive system change sample in the performance change dimension. Then, based on the classification computation network, the temporal feature data of the feature similarity and the contrastive system change sample in the change impact dimension are classified to obtain the predicted risk classification result for the contrastive system change sample. The training loss is then calculated based on the predicted risk classification result and the actual risk classification result corresponding to the label. The parameters of the contrastive learning network and the classification computation network are adjusted based on the training loss to optimize their computational power. In the deployment phase after the risk identification model is trained, whenever the system has just completed a change or is about to implement a change, the risk identification model can be used to attempt to predict potential risks. Upon predicting the existence of a risk, corresponding risk control decisions are promptly executed to prevent production accidents. In the entire scheme, the computational factor in the classification computation network is the feature comparison result between the baseline system change sample and the contrast system change sample provided by the contrast learning network, i.e., feature similarity. Feature comparison only needs to determine whether the two samples are similar in features, rather than determining the correspondence between the features and the probability of risk at different levels. Therefore, the computational training of the classification computation network does not rely too much on a large amount of data. Thus, given the provision of baseline system change samples, only a small number of labeled contrast system change samples are needed to complete the training of the risk identification model. Attached Figure Description

[0043] The accompanying drawings, which are included to provide a further understanding of this specification and form part of this specification, illustrate exemplary embodiments and are used to explain this specification, but do not constitute an undue limitation thereof. In the drawings:

[0044] Figure 1 This is a schematic diagram of the structure of the risk identification model trained by the model training method provided in the embodiments of this specification.

[0045] Figure 2 This is a flowchart illustrating the model training method provided in the embodiments of this specification.

[0046] Figure 3 This is a schematic diagram of the structure of the contrastive learning network in the risk identification model provided in the embodiments of this specification.

[0047] Figure 4This is a schematic diagram of the classification calculation network in the risk identification model provided in the embodiments of this specification.

[0048] Figure 5 This is a flowchart illustrating the risk control method for system changes provided in the embodiments of this specification.

[0049] Figure 6 This is a schematic diagram illustrating the first application of the risk control method for system changes provided in the embodiments of this specification.

[0050] Figure 7 This is a schematic diagram illustrating a second application of the risk control method for system changes provided in the embodiments of this specification.

[0051] Figure 8 This is a schematic diagram of the structure of the model training device provided in the embodiments of this specification.

[0052] Figure 9 This is a schematic diagram of the structure of the risk control device for system modification provided in the embodiments of this specification.

[0053] Figure 10 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this specification. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this document clearer, the technical solutions of this specification will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments in this document, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort are within the scope of protection of this document.

[0055] As mentioned earlier, with the development of emerging information technologies such as big data and cloud computing, the architecture of online service systems is becoming increasingly complex. Any changes to such systems, such as code or configuration modifications, may trigger a series of problems that could lead to production accidents and ultimately cause significant losses to the enterprise.

[0056] Therefore, this specification aims to propose a machine learning-based risk identification scheme for system changes, based on deep learning technology. This scheme can provide automated risk warnings for system changes, thereby preventing production accidents caused by system changes. Specifically, this specification uses system change samples to train a risk identification model. When a change has just been implemented or is about to be implemented, the risk identification model is used to predict potential risks. Once a risk is predicted, corresponding risk control decisions can be made promptly to resolve it.

[0057] Figure 1This is a schematic diagram of the risk identification model in an embodiment of this specification. The risk identification model specifically includes a contrastive learning network and a classification computation network. Here, the target change is defined as a system change that has been executed or is yet to be executed, and the baseline system change sample is defined as a system change sample with a predetermined risk of change. Specifically, the contrastive learning network in the risk identification model is used to calculate the similarity between the time-series feature data of the target system change and the baseline system change sample in the performance change dimension, obtaining feature similarity; the classification computation network in the risk identification model is used to analyze the feature similarity and the target system change sample.

[0058] The time-series characteristic data of the target system change in the dimension of the change impact are classified and calculated to obtain the predicted risk classification results of the target system change.

[0059] As can be seen, in the embodiments of this specification, when pre-estimating the risks of changes to the target system, a sample of system changes with predetermined risks is used as a baseline sample of system changes. First, comparative learning is used to determine the similarity of features between the target system change and the baseline sample of system changes in the dimension of performance changes; that is, whether the target system change is similar to the baseline sample of system changes in terms of performance indicators.

[0060] The system is then compared to systems with potential risks. Subsequently, by combining the feature similarity provided by contrastive learning with the temporal characteristics of the target system change itself in the dimension of its impact, the system is predicted to be risky.

[0061] It greatly improves the accuracy of predictions.

[0062] The embodiments of this specification will be described in detail below.

[0063] Figure 2 This is a flowchart illustrating a model training method provided in one embodiment of this specification. Figure 2 The method shown can

[0064] Performed by the device described below, the specific steps include: 5S202, obtaining a system change sample, which includes a baseline system change sample and a comparison system change sample.

[0065] In this document, the baseline system change sample is a system change sample with predetermined change risks, and the comparison system change sample is labeled with a risk classification label, which is used to characterize the actual risk classification result of the comparison system change sample.

[0066] In this embodiment, the system change samples serve as training samples for the risk identification model. The system change samples are further subdivided into baseline system change samples and comparison system change samples.

[0067] 0. Among them, the baseline system change sample, as mentioned above, refers to the sample that is determined to have risks after the system change, in this implementation...

[0068] In the example, the black sample serves as the reference benchmark; the comparison system change sample is the object of comparison relative to the benchmark system change sample. The comparison system change sample is further subdivided into two categories: those with confirmed risks and those with confirmed non-risks. The comparison system change sample with confirmed risks is labeled "black sample" according to its risk classification label, while the comparison system change sample with confirmed non-risks is labeled "white sample" according to its risk classification label.

[0069] 5. Correspondingly, the application principle of contrastive learning networks is to calculate the difference between the changed samples of the contrastive system and the changed samples of the baseline system.

[0070] Similarity across multiple feature dimensions is used to identify risk-related feature dimensions between risky comparative system change samples and baseline system change samples, and these dimensions are encoded using the same vector encoding method. Simultaneously, the feature dimensions showing high similarity between risk-free comparative system change samples and baseline system change samples are also considered.

[0071] The feature dimension that is irrelevant to risk is identified and encoded using a different vector encoding method. Ultimately, the contrastive learning network can distinguish between the encoding methods of the risk-related feature dimension and the encoding methods of the risk-related feature dimension.

[0072] In this embodiment, the learning network encodes the risk-related feature dimensions as valid encoding results, thereby enabling it to output the similarity between the comparison system change sample and the baseline system change sample on the risk-related feature dimensions.

[0073] S204, a contrastive learning network based on a risk identification model calculates the similarity of the temporal feature data of the baseline system change sample and the contrast system change sample in the dimension of performance change, and obtains the feature similarity.

[0074] 5. In this embodiment, the time-series feature data on the performance change dimension reflects the characteristic changes in performance indicators before and after the system change. It should be understood that the performance change dimension is not limited to one type; it can be the performance change dimension of the application layer, the performance change dimension of the business layer, the performance change dimension of the interface layer, etc., and this article does not make specific limitations.

[0075] refer to Figure 3As shown, the contrastive learning network in this embodiment specifically includes a first neural subnetwork, a second neural subnetwork, and a similarity calculation subnetwork. The first and second neural subnetworks constitute a Siamese network with shared parameters. This step involves vector encoding the temporal feature data of the benchmark system change sample in the performance change dimension based on the first neural subnetwork to obtain a first feature vector; and vector encoding the temporal feature data of the contrast system change sample in the performance change dimension based on the second neural subnetwork to obtain a second feature vector. It should be understood that the first and second neural subnetworks share the same parameters, therefore the temporal feature data of the benchmark system change sample and the contrast system change sample can be converted into corresponding feature vectors according to the same encoding logic. Then, the similarity calculation subnetwork calculates the similarity between the first and second feature vectors to obtain the feature similarity between the benchmark system change sample and the contrast system change sample in the performance change dimension.

[0076] In practical applications, this embodiment can monitor the performance indicators after system changes, thereby obtaining performance time-series characteristic data. If the performance time-series characteristic data of the baseline system change sample and the comparison system change sample are inconsistent in timing—for example, the performance time-series characteristic data of the baseline system change sample is obtained by monitoring at a granularity of "3 hours," while the performance time-series characteristic data of the comparison system change sample is obtained by monitoring at a granularity of "1 hour"—then... Figure 3 As shown, the contrastive learning network in this embodiment can also add a weight calculation subnetwork at the input end of the first neural subnetwork and the second neural subnetwork. This weight calculation subnetwork is used to perform time slot length weighted normalization calculation on the time-series feature data of the baseline system change sample and the contrast system change sample in the performance change dimension, thereby improving the accuracy of feature similarity calculation.

[0077] Furthermore, the similarity calculation subnetwork in this embodiment can also employ a distance function constructed using the Dynamic Time Warping algorithm to calculate the similarity between the first feature vector and the second feature vector, thereby eliminating the temporal phase difference between the performance time-series feature data of the first and second feature vectors. The Dynamic Time Warping algorithm can stretch or shorten time-series data, thus maintaining consistent time-series lengths for different time-series data. It should be noted that the Dynamic Time Warping algorithm is existing technology, and the distance function based on it is not unique; therefore, this paper does not impose a specific limitation.

[0078] It should be understood that for risk identification models that have not been fully trained, the feature similarity calculated by the contrastive learning network may contain some errors. Subsequent adjustments to the parameters of the contrastive learning network are needed to optimize its computational power. Specifically, the parameters that need adjustment include those shared by the first and second neural sub-networks, as well as the parameters of the similarity calculation sub-network.

[0079] S206, the classification calculation network based on the risk identification model performs classification calculations on the temporal feature data of the comparison system change samples in the dimension of change impact, and obtains the predicted risk classification results of the comparison system change samples.

[0080] The classification calculation network in this embodiment classifies the presence or absence of risk for the comparison system change samples based on two calculation factors. One factor is the feature similarity between the baseline system change sample and the comparison system change sample provided by the comparison learning network in the performance change dimension; the other factor is the time-series feature data of the comparison system change sample in the change impact dimension.

[0081] The temporal characteristic data of the change impact surface dimension reflects the impact surface of the system change before and after the change. The change impact surface dimension can include the impact surface dimension of proactive changes and / or the impact surface dimension of reactive changes; the impact surface dimension of proactive changes includes the change operations that need to be performed for the system change and the entities that perform the change operations, etc., while the impact surface dimension of reactive changes includes the businesses, interfaces, logs, etc. that are affected after the system change.

[0082] Specifically, refer to Figure 4 As shown, the classification calculation network includes a third neural subnetwork, a fourth neural subnetwork, and a classification calculation subnetwork. This step involves vector encoding the feature similarity provided by the contrastive learning network based on the third neural subnetwork to obtain a third feature vector; and vector encoding the temporal feature data of the contrastive system change sample in the dimension of change impact based on the fourth neural subnetwork to obtain a fourth feature vector. Then, the classification calculation network performs classification calculations on the third feature vector and the fourth neural subnetwork to obtain the predicted risk classification result for the first system change sample.

[0083] S208. Based on the predicted risk classification results of the comparison system changed samples and the actual risk classification results of the corresponding labels, determine the training loss of the risk identification model, and adjust the parameters of the comparison learning network and the classification calculation network according to the training loss.

[0084] In this embodiment, the actual risk classification results of the system change sample annotations are used as the expected output of the risk identification model. During the training phase, the predicted risk classification results output by the risk identification model are not necessarily accurate, and therefore will have an error compared with the true actual risk classification results. This error is the training loss of the risk identification model. This step adjusts the parameters of the contrastive learning network and the classification computation network in the risk identification model by reducing the training loss as the gradient direction, so that the predicted risk classification results output by the risk identification model gradually approach the actual risk classification results, thereby enabling it to identify system change risks.

[0085] In practical applications, the error between the predicted risk classification result and the actual risk classification result can be calculated using a loss function. The specific implementation of the loss function is not unique, and this article does not make a specific limitation here.

[0086] As described above, the model training method in this specification constructs a risk identification model composed of a contrastive learning network and a classification computation network. During the training phase of the risk identification model, a predetermined system change sample with known risk is used as a baseline system change sample. First, the contrastive learning network calculates the feature similarity between the baseline system change sample and the contrastive system change sample in the performance change dimension. Then, based on the classification computation network, the feature similarity and the temporal feature data of the contrastive system change sample in the change impact dimension are classified to obtain the predicted risk classification result of the contrastive system change sample. The training loss is then calculated based on the predicted risk classification result and the actual risk classification result corresponding to the label. Finally, the parameters of the contrastive learning network and the classification computation network are adjusted based on the training loss to optimize their computational power. In the entire scheme, the computational factor in the classification computation network is the feature comparison result between the baseline system change samples and the comparison system change samples provided by the contrastive learning network, i.e., feature similarity. Feature comparison only needs to determine whether the two samples are similar in features, rather than determining the correspondence between features and the probability of risk at different levels. Therefore, the computational training of the classification computation network does not rely too much on a large number of system change samples as data support. Thus, given the provision of baseline system change samples, only a small number of labeled comparison system change samples are needed to complete the training of the risk identification model.

[0087] Figure 5 This is a flowchart of a system change risk control method provided in one embodiment of this specification. Figure 5 The method shown can be performed by the apparatus described below, and specifically includes the following steps:

[0088] S502, obtain the time-series characteristic data of the target system change and the baseline system change sample in the performance change dimension, as well as the time-series characteristic data of the target system change in the change impact dimension. The target system change refers to the system change that has been executed or is to be executed, and the baseline system change sample is a system change sample that is known to have change risk.

[0089] Specifically, the time-series characteristic data of the performance change dimension includes the performance time-series characteristic data after the system change. The change impact dimension includes the active change impact dimension and / or the passive change impact dimension. Among them, the active change impact dimension includes at least one of the change operation and the subject executing the change operation, and the passive change impact dimension includes at least one of the business, interface, and log affected by the change.

[0090] S504, the time-series feature data of the target system change and the baseline system change samples in the performance change dimension, as well as the time-series feature data of the target system change in the change impact dimension, are input into the risk identification model to obtain the predicted risk classification result of the target system change; wherein, the risk identification model includes a contrastive learning network and a classification calculation network. The learning network is used to calculate the similarity of the time-series feature data of the target system change and the baseline system change samples in the performance change dimension to obtain the feature similarity; the classification calculation network is used to perform classification calculation on the feature similarity and the time-series feature data of the target system change in the change impact dimension to obtain the predicted risk classification result.

[0091] In this embodiment, the risk identification model is based on Figure 2 The method shown is used for training.

[0092] The risk classification result indicates whether the target system change falls under the risk classification of the baseline system change sample. If the target system change falls under the risk classification of the baseline system change sample, it means that the target system change carries the same risks as the baseline system change sample; similarly, if the target system change does not fall under the risk classification of the baseline system change sample, it means that the target system change does not carry the same risks as the baseline system change sample.

[0093] S506, for target system changes, implement risk control decisions that match the predicted risk classification results.

[0094] For example, if the target system change has already been executed and falls under the risk classification of the baseline system change sample, the risk control decision may include rolling back the target system change to restore the system to its state before the target system change was executed. If the target system change is pending execution and falls under the risk classification of the baseline system change sample, the risk control decision may include intercepting subsequent requests to initiate the target system change to prevent the target system change from occurring and to provide early warning of change risks.

[0095] The risk control method of this embodiment will be introduced below in conjunction with actual application scenarios.

[0096] This application scenario exemplifies a business system that can adaptively predict change risks.

[0097] refer to Figure 6 As shown, during the training phase of the risk identification model, the business system mainly involves: a change engine, sample annotation nodes, a sample database, and model training nodes.

[0098] in:

[0099] The change engine uses changes to the business system as samples, sending time-series characteristic data on both the performance change dimension and the impact surface dimension to the sample annotation nodes. Specifically, the time-series characteristic data on the impact surface dimension can be extracted from the business system's change logs, changed interfaces, etc. The impact surface dimension reflects the areas affected by the changes to the business system, such as the application layer, containers, and interfaces. The time-series characteristic data on the performance change dimension, based on the impact surface dimension, collects the changes in metrics in the areas affected by the changes to the business system, such as the time-series performance metrics of the application layer, containers, and interfaces.

[0100] The sample annotation node is responsible for determining whether a sample carries risk. Based on the temporal characteristic data of risk in both the performance change dimension and the impact dimension, it performs comparative annotations to subdivide the samples into baseline system change samples and comparison system change samples. Finally, these samples are uploaded to the sample database as the training set for the risk model. The baseline system change samples are those where risk is determined to exist after the system change, serving as a reference benchmark for black samples. The comparison system change samples are the objects of comparison relative to the baseline system change samples. The comparison system change samples are further subdivided into two categories: those with confirmed risk and those without confirmed risk. Comparison system change samples with confirmed risk are labeled "black samples," while those without confirmed risk are labeled "white samples."

[0101] The model training node is responsible for retrieving the training set from the sample database to train the risk identification model. The training process is as described above and will not be repeated here.

[0102] refer to Figure 7 ,refer to Figure 7 As shown, during the training phase of the risk identification model, the business system mainly involves: change engine, streaming database, change monitoring service, and system-as-operation support platform.

[0103] in:

[0104] The change engine is used to execute system changes and make risk control decisions based on these changes. In this application scenario, after the change engine completes the target system change, it sends a notification of the target system change to the streaming database and the change monitoring service.

[0105] Streaming databases are used to dynamically record feature data of the performance change dimension and change impact dimension of a target system in a streaming manner. In this application scenario, after receiving a notification of a change to the target system, the database determines the time-series feature data of the real-time performance change dimension and the real-time change impact dimension corresponding to the current change, as well as the time-series feature data of the historical performance change dimension. The time-series feature data of the real-time performance change dimension corresponding to the current change can be obtained by combining the feature data of the performance change dimension corresponding to the target system before and after the current change; similarly, the time-series feature data of the real-time change impact dimension corresponding to the current change can be obtained by combining the feature data of the change impact dimension corresponding to the target system before and after the current change. The time-series feature data of the historical performance change dimension can refer to the time-series feature data of performance change dimensions that have been previously identified as having risks, that is, the historical performance change dimensions of the baseline system change sample.

[0106] The change monitoring service initiates a change monitoring task upon receiving a notification of a target system change. It retrieves time-series characteristic data for the current real-time performance change dimension, historical performance change dimension, and real-time change impact dimension from the database, and sends these three sets of time-series characteristic data to the System as Operation Support Platform. The System as Operation Support Platform is responsible for predicting the risks of system changes. In this application scenario, the System as Operation Support Platform invokes a risk identification model, inputting the received historical performance change dimension time-series characteristic data (serving as a baseline system change sample), the real-time performance change dimension time-series characteristic data corresponding to the current change in the target system, and the real-time change impact dimension time-series characteristic data into the risk identification model to predict the risk classification result of this change. Subsequently, the System as Operation Support Platform sends the risk classification result of the target system change output by the risk identification model to the change engine.

[0107] Furthermore, once the change engine obtains the risk classification results for the target system change provided by the system, i.e., the operation support platform, it executes risk control decisions based on the risk classification results for this change. For example, if the risk classification results indicate that this change belongs to the risk type of the baseline system change sample, then the change is rolled back so that the target system can resume loading before this change; or, if the risk classification results indicate that this change does not belong to the risk type of the baseline system change sample, then the change engine approves the change.

[0108] Furthermore, if the change engine approves the change, it can also notify the change monitoring service, which will then upload the time-series characteristic data of the real-time change impact dimension and the real-time performance change dimension of the change. Figure 6 The sample database shown is used as a sample for target system changes, which will be labeled before the risk identification model is updated and trained.

[0109] Figure 8 This is a schematic diagram of the structure of a model training device 800 provided in one embodiment of this specification, including:

[0110] The sample acquisition module 810 acquires system change samples, which include a baseline system change sample and a comparison system change sample. The baseline system change sample is a system change sample with predetermined change risks, and the comparison system change sample is labeled with a risk classification label. The risk classification label is used to characterize the actual risk classification result of the comparison system change sample.

[0111] The contrastive learning module 820, based on the risk identification model, performs similarity calculation on the temporal feature data of the baseline system change sample and the contrastive system change sample in the performance change dimension to obtain feature similarity.

[0112] The classification calculation module 830 performs classification calculations on the feature similarity and the temporal feature data of the comparison system change sample in the dimension of change impact based on the classification calculation network of the risk identification model, and obtains the predicted risk classification result of the comparison system change sample.

[0113] The parameter adjustment module 840 determines the training loss of the risk identification model based on the predicted risk classification results of the comparison system change samples and the corresponding labeled actual risk classification results, and adjusts the parameters of the comparison learning network and the classification calculation network according to the training loss.

[0114] The apparatus in this embodiment constructs a risk identification model consisting of a contrastive learning network and a classification computation network. During the training phase of the risk identification model, a predetermined system change sample with known risk is used as a baseline system change sample. First, the contrastive learning network calculates the feature similarity between the baseline system change sample and the contrastive system change sample in the performance change dimension. Then, based on the classification computation network, the temporal feature data of the feature similarity and the contrastive system change sample in the change impact dimension are classified to obtain the predicted risk classification result for the contrastive system change sample. The training loss is calculated based on the predicted risk classification result and the actual risk classification result corresponding to the label. The parameters of the contrastive learning network and the classification computation network are then adjusted based on the training loss to optimize their computational power. In the deployment phase after the risk identification model is trained, whenever the system has just completed a change or is about to implement a change, the risk identification model can be used to attempt to predict potential risks. Upon predicting the existence of a risk, corresponding risk control decisions are promptly executed to prevent production accidents. In the entire scheme, the computational factor in the classification computation network is the feature comparison result between the baseline system change sample and the contrast system change sample provided by the contrast learning network, i.e., feature similarity. Feature comparison only needs to determine whether the two samples are similar in features, rather than determining the correspondence between the features and the probability of risk at different levels. Therefore, the computational training of the classification computation network does not rely too much on a large amount of data. Thus, given the provision of baseline system change samples, only a small number of labeled contrast system change samples are needed to complete the training of the risk identification model.

[0115] Optionally, the comparison learning network includes: a first neural subnetwork, a second neural subnetwork, and a similarity calculation subnetwork, wherein the first neural subnetwork and the second neural subnetwork constitute a Siamese network with shared parameters; the comparison learning module 820 specifically performs vector encoding on the temporal feature data of the benchmark system change sample in the performance change dimension based on the first neural subnetwork to obtain a first feature vector; and performs vector encoding on the temporal feature data of the comparison system change sample in the performance change dimension based on the second neural subnetwork to obtain a second feature vector; and performs similarity calculation on the first feature vector and the second feature vector based on the similarity calculation subnetwork to obtain feature similarity.

[0116] Optionally, the contrastive learning network further includes a weight calculation sub-network; before the contrastive learning module 820 performs vector encoding on the temporal feature data of the benchmark system change sample in the performance change dimension based on the first neural sub-network, and on the temporal feature data of the contrastive system change sample in the performance change dimension based on the second neural sub-network, it also performs time slot length weighted normalization calculation on the temporal feature data of the benchmark system change sample and the contrastive system change sample in the performance change dimension based on the weight calculation sub-network.

[0117] Optionally, the comparison learning module 820 is specifically based on the similarity calculation sub-network, and uses a distance function constructed by the dynamic time warping algorithm to calculate the similarity between the first feature vector and the second feature to obtain the feature similarity.

[0118] Optionally, the classification calculation network includes: a third neural subnetwork, a fourth neural subnetwork, and a classification calculation subnetwork; the classification calculation module 830 specifically performs vector encoding on the feature similarity based on the third neural subnetwork to obtain a third feature vector; and performs vector encoding on the temporal feature data of the comparison system change sample in the dimension of change impact based on the fourth neural subnetwork to obtain a fourth feature vector; and performs classification calculation on the third feature vector and the fourth neural subnetwork based on the classification calculation network to obtain the predicted risk classification result of the first system change sample.

[0119] Optionally, the change impact dimension includes an active change impact dimension and / or a passive change impact dimension; wherein, the active change impact dimension includes at least one of the change operation and the entity executing the change operation, and the passive change impact dimension includes at least one of the services, interfaces, and logs affected by the change.

[0120] Obviously, the model training device in the embodiments of this specification can achieve... Figure 2 The steps and functions in the illustrated embodiments will not be described in detail here.

[0121] Figure 9 This is a schematic diagram of the structure of a system modification risk control device provided in one embodiment of this specification, including:

[0122] The data acquisition module 910 acquires the time-series characteristic data of the target system change and the baseline system change sample in the performance change dimension, as well as the time-series characteristic data of the target system change in the change impact dimension. The target system change refers to the system change that has been executed or is to be executed, and the baseline system change sample is a system change sample that is known to have change risk.

[0123] The risk identification module 920 inputs the temporal feature data of the target system change and the baseline system change sample on the performance change dimension, as well as the feature data of the target system change on the change impact dimension, into the risk identification model to obtain the predicted risk classification result of the target system change; wherein, the target system change refers to a system change that has been executed or is to be executed, and the baseline system change sample is a system change sample with a predetermined change risk; the risk identification model includes a contrastive learning network and a classification calculation network, the learning network is used to calculate the similarity of the temporal feature data of the target system change and the baseline system change sample on the performance change dimension to obtain feature similarity; the classification calculation network is used to perform classification calculation on the feature similarity and the temporal feature data of the target system change on the change impact dimension to obtain the predicted risk classification result;

[0124] Risk control execution module 930. Executes risk control decisions that match the predicted risk classification results in response to the target system changes.

[0125] Obviously, the system modification risk control device in the embodiments of this specification can achieve... Figure 5 The steps and functions in the illustrated embodiments will not be described in detail here.

[0126] Figure 10 This is a schematic diagram of the structure of an electronic device provided in one embodiment of this specification. Please refer to it. Figure 10 At the hardware level, the electronic device includes a processor, and optionally also includes an internal bus, a network interface, and memory. The memory may include main memory, such as high-speed random-access memory (RAM), or non-volatile memory, such as at least one disk drive. Of course, the electronic device may also include other hardware required for other business operations.

[0127] The processor, network interface, and memory can be interconnected via an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. This bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 10 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0128] Memory is used to store programs. Specifically, programs may include program code, which includes computer operation instructions. Memory may include main memory and non-volatile memory, and provides instructions and data to the processor.

[0129] The processor reads the corresponding computer program from non-volatile memory into main memory and then runs it, forming the aforementioned model training device at the logical level. The processor executes the program stored in memory and specifically performs the following operations:

[0130] Obtain system change samples, which include a baseline system change sample and a comparison system change sample. The baseline system change sample is a system change sample with predetermined change risks, and the comparison system change sample is labeled with a risk classification tag. The risk classification tag is used to characterize the actual risk classification result of the comparison system change sample.

[0131] A contrastive learning network based on a risk identification model calculates the similarity of the temporal feature data of the baseline system change sample and the contrast system change sample in the dimension of performance change, and obtains the feature similarity.

[0132] The classification calculation network based on the risk identification model performs classification calculations on the feature similarity and the temporal feature data of the comparison system change sample in the dimension of change impact, and obtains the predicted risk classification result of the comparison system change sample.

[0133] Based on the predicted risk classification results of the modified samples in the comparison system and the actual risk classification results of the corresponding labels, the training loss of the risk identification model is determined, and the parameters of the comparison learning network and the classification calculation network are adjusted according to the training loss.

[0134] Alternatively, the processor reads the corresponding computer program from non-volatile memory into memory and runs it, logically forming the risk control device for the aforementioned system changes. The processor executes the program stored in memory and specifically performs the following operations:

[0135] Acquire the time-series characteristic data of the target system change and the baseline system change sample in terms of performance change dimension, as well as the time-series characteristic data of the target system change in terms of change impact dimension. The target system change refers to the system change that has been executed or is to be executed, and the baseline system change sample is a system change sample that is known to have change risk.

[0136] The time-series feature data of the target system change and the baseline system change samples on the performance change dimension, as well as the time-series feature data of the target system change on the change impact surface dimension, are input into the risk identification model to obtain the predicted risk classification result of the target system change. The risk identification model includes a contrastive learning network and a classification calculation network. The learning network is used to calculate the similarity between the time-series feature data of the target system change and the baseline system change samples on the performance change dimension to obtain feature similarity. The classification calculation network is used to perform classification calculations on the feature similarity and the time-series feature data of the target system change on the change impact surface dimension to obtain the predicted risk classification result.

[0137] For changes to the target system, risk control decisions are made that match the predicted risk classification results.

[0138] The above is as described in this instruction manual. Figure 2 or Figure 5 The methods disclosed in the illustrated embodiments can be applied to or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above methods can be completed by integrated logic circuits in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The methods, steps, and logic block diagrams disclosed in one or more embodiments of this specification can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in one or more embodiments of this specification can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software module can reside in a mature storage medium in the field, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0139] Of course, in addition to software implementation, the electronic device described in this specification does not exclude other implementation methods, such as logic devices or a combination of hardware and software. In other words, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0140] This specification also provides an embodiment of a computer-readable storage medium that stores one or more programs.

[0141] The program includes instructions that, when executed by a portable electronic device comprising multiple applications, enable the portable electronic device to perform... Figure 2 The method of the illustrated embodiment is specifically used to perform the following operations:

[0142] Obtain system change samples, which include a baseline system change sample and a comparison system change sample. The baseline system change sample is a system change sample with predetermined change risks, and the comparison system change sample is labeled with a risk classification tag. The risk classification tag is used to characterize the actual risk classification result of the comparison system change sample.

[0143] A contrastive learning network based on a risk identification model calculates the similarity of the temporal feature data of the baseline system change sample and the contrast system change sample in the dimension of performance change, and obtains the feature similarity.

[0144] The classification calculation network based on the risk identification model performs classification calculations on the feature similarity and the temporal feature data of the comparison system change sample in the dimension of change impact, and obtains the predicted risk classification result of the comparison system change sample.

[0145] Based on the predicted risk classification results of the modified samples in the comparison system and the actual risk classification results of the corresponding labels, the training loss of the risk identification model is determined, and the parameters of the comparison learning network and the classification calculation network are adjusted according to the training loss.

[0146] Alternatively, the one or more programs include instructions that, when executed by a portable electronic device including multiple applications, enable the portable electronic device to perform... Figure 5 The method of the illustrated embodiment is specifically used to perform the following operations:

[0147] Acquire the time-series characteristic data of the target system change and the baseline system change sample in terms of performance change dimension, as well as the time-series characteristic data of the target system change in terms of change impact dimension. The target system change refers to the system change that has been executed or is to be executed, and the baseline system change sample is a system change sample that is known to have change risk.

[0148] The time-series feature data of the target system change and the baseline system change samples on the performance change dimension, as well as the time-series feature data of the target system change on the change impact surface dimension, are input into the risk identification model to obtain the predicted risk classification result of the target system change. The risk identification model includes a contrastive learning network and a classification calculation network. The learning network is used to calculate the similarity between the time-series feature data of the target system change and the baseline system change samples on the performance change dimension to obtain feature similarity. The classification calculation network is used to perform classification calculations on the feature similarity and the time-series feature data of the target system change on the change impact surface dimension to obtain the predicted risk classification result.

[0149] For changes to the target system, risk control decisions are made that match the predicted risk classification results.

[0150] In summary, the above description is merely a preferred embodiment of this specification and is not intended to limit the scope of protection of this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments of this specification should be included within the scope of protection of one or more embodiments of this specification.

[0151] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0152] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0153] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0154] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

Claims

1. A model training method, comprising: Obtain system change samples, which include baseline system change samples and comparison system change samples. The baseline system change samples are system change samples with predetermined change risks, and the comparison system change samples are labeled with risk classification tags. The risk classification tags are used to characterize the actual risk classification results of the comparison system change samples. A contrastive learning network based on a risk identification model calculates the similarity of the temporal feature data of the baseline system change sample and the comparison system change sample in the performance change dimension to obtain feature similarity. The feature similarity reflects whether the system change sample is similar to the baseline system change sample with previous risks in terms of performance indicators. There are no less than two performance change dimensions, and the temporal feature data of the performance change dimensions reflect the characteristic changes of performance indicators before and after the system change. The classification calculation network based on the risk identification model performs classification calculations on the feature similarity and the temporal feature data of the comparison system change sample in the dimension of change impact, and obtains the predicted risk classification result of the comparison system change sample; Based on the predicted risk classification results of the comparison system change samples and the corresponding labeled actual risk classification results, the training loss of the risk identification model is determined, and the parameters of the comparison learning network and the classification calculation network are adjusted according to the training loss, so that the risk identification model can be used to determine the predicted risk classification results of the target system change and the related risk control decisions to be executed for the target system change, the risk control decisions including intercepting the request for the target system change.

2. The method according to claim 1, The contrastive learning network includes: The network consists of a first neural subnetwork, a second neural subnetwork, and a similarity calculation subnetwork, wherein the first neural subnetwork and the second neural subnetwork constitute a twin network with shared parameters; A contrastive learning network based on a risk identification model calculates the similarity between the time-series feature data of the baseline system change sample and the contrast system change sample in the performance change dimension, obtaining feature similarity, including: Based on the first neural subnetwork, the time-series feature data of the benchmark system change sample in the performance change dimension is vector-encoded to obtain a first feature vector; and based on the second neural subnetwork, the time-series feature data of the comparison system change sample in the performance change dimension is vector-encoded to obtain a second feature vector. The similarity calculation subnetwork is used to calculate the similarity between the first feature vector and the second feature vector to obtain the feature similarity.

3. The method according to claim 2, The contrastive learning network also includes a weight calculation subnetwork; Before performing vector encoding on the temporal feature data of the benchmark system change sample in the performance change dimension based on the first neural subnetwork, and before performing vector encoding on the temporal feature data of the comparison system change sample in the performance change dimension based on the second neural subnetwork, the method further includes: Based on the weighted calculation subnetwork, the time-slot length-weighted normalization calculation is performed on the time-series feature data of the baseline system change sample and the comparison system change sample on the performance change dimension.

4. The method according to claim 3, Based on the similarity calculation subnetwork, the first feature vector and the second feature vector are similar to calculate the feature similarity, including: Based on the similarity calculation sub-network, the distance function constructed using the dynamic time warping algorithm is used to calculate the similarity between the first feature vector and the second feature, thereby obtaining the feature similarity.

5. The method according to claim 1, The classification computing network includes: The third neural subnetwork, the fourth neural subnetwork, and the classification computation subnetwork; A classification computation network based on a risk identification model performs classification computation on the feature similarity and the temporal feature data of the comparison system change samples in the dimension of change impact, to obtain the predicted risk classification results of the comparison system change samples, including: The feature similarity is vector-encoded based on the third neural subnetwork to obtain a third feature vector; and the temporal feature data of the comparison system change sample in the dimension of change influence is vector-encoded based on the fourth neural subnetwork to obtain a fourth feature vector. Based on the classification calculation network, the third feature vector and the fourth neural subnetwork are classified to obtain the predicted risk classification result of the first system change sample.

6. The method according to any one of claims 1-5, The dimensions of the impact of the change include the dimensions of the impact of proactive changes and / or the dimensions of the impact of reactive changes; wherein... The active change impact dimension includes at least one of the change operation and the entity executing the change operation, while the passive change impact dimension includes at least one of the business, interface, and log affected by the change.

7. A risk control method for system changes, comprising: Acquire time-series characteristic data of the target system change and the baseline system change sample in terms of performance change dimension, as well as time-series characteristic data of the target system change in terms of change impact dimension. The target system change refers to a system change that has been executed or is to be executed, and the baseline system change sample is a system change sample that is known to have change risk. There are no less than two performance change dimensions, and the time-series characteristic data of the performance change dimensions reflect the characteristic changes of performance indicators before and after the system change. The time-series feature data of the target system change and the baseline system change samples on the performance change dimension, as well as the time-series feature data of the target system change on the change impact dimension, are input into a risk identification model to obtain a predicted risk classification result for the target system change. The risk identification model includes a contrastive learning network and a classification calculation network. The learning network is used to calculate the similarity between the time-series feature data of the target system change and the baseline system change samples on the performance change dimension, obtaining a feature similarity. This feature similarity reflects whether the system change sample is similar to a previously risky baseline system change sample in terms of performance indicators. The classification calculation network is used to perform classification calculations on the feature similarity and the time-series feature data of the target system change on the change impact dimension to obtain the predicted risk classification result. For the target system change, a risk control decision is executed that matches the predicted risk classification result; the risk control decision includes intercepting the request for the target system change.

8. A model training device, comprising: The sample acquisition module acquires system change samples, which include a baseline system change sample and a comparison system change sample. The baseline system change sample is a system change sample with predetermined change risks, and the comparison system change sample is labeled with a risk classification label. The risk classification label is used to characterize the actual risk classification result of the comparison system change sample. The contrastive learning module, based on a risk identification model, uses a contrastive learning network to calculate the similarity of the temporal feature data of the baseline system change sample and the contrast system change sample in the performance change dimension, thus obtaining feature similarity. The feature similarity reflects whether the system change sample is similar to the baseline system change sample with previous risks in terms of performance indicators. There are at least two performance change dimensions, and the temporal feature data of the performance change dimensions reflect the characteristic changes in performance indicators before and after the system change. The classification calculation module performs classification calculations on the feature similarity and the temporal feature data of the comparison system change sample in the dimension of change impact based on the classification calculation network of the risk identification model, and obtains the predicted risk classification result of the comparison system change sample; The parameter adjustment module determines the training loss of the risk identification model based on the predicted risk classification results of the comparison system change samples and the corresponding labeled actual risk classification results. It then adjusts the parameters of the comparison learning network and the classification calculation network according to the training loss, so that the risk identification model can be used to determine the predicted risk classification results of the target system change and the related risk control decisions to be executed for the target system change. The risk control decisions include intercepting requests for target system changes.

9. An electronic device, comprising: processor; and memory configured to store computer-executable instructions, which, when executed, cause the processor to perform the following operations: Obtain system change samples, which include baseline system change samples and comparison system change samples. The baseline system change samples are system change samples with predetermined change risks, and the comparison system change samples are labeled with risk classification tags. The risk classification tags are used to characterize the actual risk classification results of the comparison system change samples. A contrastive learning network based on a risk identification model calculates the similarity of the temporal feature data of the baseline system change sample and the contrastive system change sample in the performance change dimension, obtaining feature similarity. The feature similarity reflects whether the system change sample is similar to the baseline system change sample with previous risks in terms of performance indicators. There are at least two performance change dimensions, and the temporal feature data of the performance change dimensions reflect the characteristic changes in performance indicators before and after the system change. A classification calculation network based on the risk identification model performs classification calculation on the feature similarity and the temporal feature data of the contrastive system change sample in the change impact dimension, obtaining the predicted risk classification result of the contrastive system change sample. Based on the predicted risk classification results of the comparison system change samples and the corresponding labeled actual risk classification results, the training loss of the risk identification model is determined, and the parameters of the comparison learning network and the classification calculation network are adjusted according to the training loss, so that the risk identification model can be used to determine the predicted risk classification results of the target system change and the related risk control decisions to be executed for the target system change, the risk control decisions including intercepting the request for the target system change.

10. A computer-readable storage medium storing one or more programs, which, when executed by an electronic device including multiple applications, cause the electronic device to perform the following operations: Obtain system change samples, which include a baseline system change sample and a comparison system change sample, wherein, The baseline system change sample is a system change sample with predetermined change risks. The comparison system change sample is labeled with a risk classification label, which is used to characterize the actual risk classification result of the comparison system change sample. A contrastive learning network based on a risk identification model calculates the similarity of the temporal feature data of the baseline system change sample and the contrast system change sample in the dimension of performance change, and obtains the feature similarity. The feature similarity reflects whether the performance indicators of the system change sample are similar to those of previous risk benchmark system change samples. The performance change dimension is no less than two, and the time-series feature data of the performance change dimension reflects the characteristic changes of the performance indicators before and after the system change. The classification calculation network based on the risk identification model performs classification calculations on the feature similarity and the temporal feature data of the comparison system change sample in the dimension of change impact, and obtains the predicted risk classification result of the comparison system change sample; Based on the predicted risk classification results of the comparison system change samples and the corresponding labeled actual risk classification results, the training loss of the risk identification model is determined, and the parameters of the comparison learning network and the classification calculation network are adjusted according to the training loss, so that the risk identification model can be used to determine the predicted risk classification results of the target system change and the related risk control decisions to be executed for the target system change, the risk control decisions including intercepting the request for the target system change.

11. A risk control device for system changes, comprising: The data acquisition module acquires the time-series characteristic data of the target system change and the baseline system change sample in terms of performance change dimension, as well as the time-series characteristic data of the target system change in terms of change impact dimension. The target system change refers to the system change that has been executed or is to be executed, and the baseline system change sample is a system change sample that is known to have change risk. The risk identification module inputs the time-series feature data of the target system change and the baseline system change sample on the performance change dimension, as well as the time-series feature data of the target system change on the change impact dimension, into the risk identification model to obtain the predicted risk classification result of the target system change. Here, the target system change refers to a system change that has been executed or is to be executed, and the baseline system change sample is a system change sample with a predetermined change risk. The risk identification model includes a contrastive learning network and a classification calculation network. The learning network is used to calculate the similarity between the time-series feature data of the target system change and the baseline system change sample on the performance change dimension to obtain feature similarity. The feature similarity reflects whether the system change sample is similar to the baseline system change sample with previous risks in terms of performance indicators. There are at least two performance change dimensions, and the time-series feature data of the performance change dimensions reflect the characteristic changes in performance indicators before and after the system change. The classification calculation network is used to perform classification calculation on the feature similarity and the time-series feature data of the target system change on the change impact dimension to obtain the predicted risk classification result. The risk control execution module executes risk control decisions that match the predicted risk classification results in response to the target system change; the risk control decisions include intercepting requests for target system changes.

12. An electronic device, comprising: processor; and memory configured to store computer-executable instructions, which, when executed, cause the processor to perform the following operations: Acquire the time-series characteristic data of the target system change and the baseline system change sample in the performance change dimension, as well as the time-series characteristic data of the target system change in the change impact dimension. The target system change refers to the system change that has been executed or is to be executed, and the baseline system change sample is a system change sample that is known to have change risk. The time-series feature data of the target system change and the baseline system change samples on the performance change dimension, as well as the time-series feature data of the target system change on the change impact dimension, are input into a risk identification model to obtain a predicted risk classification result for the target system change. The risk identification model includes a contrastive learning network and a classification computation network. The learning network is used to calculate the similarity between the time-series feature data of the target system change and the baseline system change samples on the performance change dimension to obtain feature similarity. The feature similarity reflects whether the system change sample is similar to a previously risky baseline system change sample in terms of performance indicators. There are at least two performance change dimensions, and the time-series feature data of the performance change dimensions reflect the characteristic changes in performance indicators before and after the system change. The classification computation network is used to perform classification computation on the feature similarity and the time-series feature data of the target system change on the change impact dimension to obtain the predicted risk classification result. For the target system change, a risk control decision is executed that matches the predicted risk classification result; the risk control decision includes intercepting the request for the target system change.

13. A computer-readable storage medium storing one or more programs, which, when executed by an electronic device including multiple applications, cause the electronic device to perform the following operations: Obtain the time-series characteristic data of the target system change and the baseline system change samples in terms of performance change dimension, as well as the time-series characteristic data of the target system change in terms of change impact surface dimension, wherein, The target system change refers to a system change that has been executed or is to be executed, and the baseline system change sample is a sample of system changes that are known to have change risks. The time-series feature data of the target system change and the baseline system change samples on the performance change dimension, as well as the time-series feature data of the target system change on the change impact dimension, are input into a risk identification model to obtain a predicted risk classification result for the target system change. The risk identification model includes a contrastive learning network and a classification computation network. The learning network is used to calculate the similarity between the time-series feature data of the target system change and the baseline system change samples on the performance change dimension to obtain feature similarity. The feature similarity reflects whether the system change sample is similar to a previously risky baseline system change sample in terms of performance indicators. There are at least two performance change dimensions, and the time-series feature data of the performance change dimensions reflect the characteristic changes in performance indicators before and after the system change. The classification computation network is used to perform classification computation on the feature similarity and the time-series feature data of the target system change on the change impact dimension to obtain the predicted risk classification result. For the target system change, a risk control decision is executed that matches the predicted risk classification result; the risk control decision includes intercepting the request for the target system change.

Images