Informational data processing method and system

Abstract

The application provides an information data processing method and system. It relates to the technical field of computers and solves the problem of high code complexity. The method comprises the following steps: when a target system arranged on a first PaaS platform is running, obtaining running data of a plurality of services in the target system in a plurality of time periods, using a feature extraction model to extract features of the running data of the plurality of services in the target system in the plurality of time periods to obtain grouping features of each service, using a clustering algorithm to process the grouping features of each service to obtain at least one service group, for each service group, arranging a monitoring tool according to the information of each service in the service group, obtaining the dependency relationship of each service in the service group monitored by the monitoring tool, generating a migration strategy according to the dependency relationship of each service in the at least one service group, and migrating the target system from the first PaaS platform to a second PaaS platform according to the migration strategy.

Description

Technical Field

[0001] This application relates to the field of computer system technology, and in particular to an information data processing method and system. Background Technology

[0002] With the rapid development of cloud computing, Platform-as-a-Service (PaaS) platforms are becoming increasingly important as a bridge connecting software developers and underlying infrastructure.

[0003] Due to technological updates or evolving user needs, it may be necessary to migrate a software system deployed on one PaaS platform to another. Before migrating services, it is typically necessary to manually deploy monitoring tools within the software system to monitor the dependencies between various services.

[0004] However, since the software system contains many services, the way the monitoring tools are deployed needs to be determined by considering the mutual influence between these services, which will increase the complexity of the monitoring code. Summary of the Invention

[0005] This application provides an information data processing method and system to reduce the complexity of monitoring code.

[0006] This application provides an information data processing method, including: When running the target system deployed on the first PaaS platform, obtain the running data of multiple services in the target system over multiple time periods; The feature extraction model is used to extract features from the runtime data of multiple services in the target system over multiple time periods to obtain the grouped features of each service. Clustering algorithms are used to process the grouping characteristics of each service to obtain at least one service group; For each service group, monitoring tools are deployed based on the information of each service in the service group to obtain the dependencies of each service in the service group monitored by the monitoring tools. A migration strategy is generated based on the dependencies of services in at least one service group, and the target system is migrated from the first PaaS platform to the second PaaS platform according to the migration strategy.

[0007] In the above technical solution, feature extraction is performed on the service's operational data to extract features from a large volume of operational data. Then, a clustering algorithm is used to process these features to obtain groups for each service. This reduces the data processing workload of the clustering algorithm and improves its accuracy. The clustering algorithm ensures that the service groups output by the algorithm are independent of each other. Monitoring code is then deployed within each group as a unit. Since the monitoring code is deployed based on the information of each service within each group, the complexity of the monitoring code is reduced. Furthermore, the improved accuracy of the clustering algorithm and the independence between the service groups output by the algorithm improve the accuracy of the obtained dependency relationships and reduce the probability of system migration failure.

[0008] In one possible implementation, runtime data includes resource usage data and access data; Accordingly, a feature extraction model is used to extract features from the runtime data of multiple services in the target system over multiple time periods to obtain grouped features for each service, specifically including: The first extraction model is used to extract features from the resource usage data of the service over multiple time periods to obtain the first feature of the service; The second extraction model is used to extract features from the service's access data over multiple time periods to obtain the service's second features; The first feature and the second feature are fused to obtain grouped features; wherein the model architecture of the first extraction model is different from that of the second extraction model.

[0009] In the above technical solution, resource usage data and access data of the target system during runtime are acquired. Analysis of this data reveals the relationships between various services. By analyzing these two sets of data, even if one data source has noise or missing information, high-quality features from the other data source can be preserved through the feature extraction module, thus making the entire system more robust. Furthermore, since resource usage data and access data have different distribution characteristics, using different extraction models to process them separately allows for more accurate feature extraction, more precise grouping, and a reduced probability of migration failure.

[0010] In one possible implementation, a first extraction model is used to extract features from resource usage data of the service over multiple time periods to obtain a first feature of the service, specifically including: The local extraction module is used to extract features from resource usage data and output local features. The global extraction module is used to extract features from local features and output global features. The fusion module is used to fuse local and global features to obtain fused features; The global feature and the fused feature are concatenated and output as the first feature.

[0011] In the above technical solution, resource usage data contains rich instantaneous changes and local patterns. The local feature extraction module can accurately capture these details and understand the immediate state of the data. Global feature extraction focuses on the overall trend and long-term dependence of the data, grasping the long-term periodic characteristics and overall behavior of the data, and realizing the identification of long-term patterns. The fusion module merges local and global features, so that the fused features contain richer information extracted from the immediate state and long-term patterns. Then, the fused features and global features are concatenated and output, so that the concatenated features contain richer information. In addition, the local extraction module, global extraction module, and fusion module are each good at extracting different types of information. Even if one model is biased, other models can supplement it to improve the accuracy of the extracted features.

[0012] In one possible implementation, a fusion module is used to fuse local and global features to obtain fused features, specifically including: For each local feature, a first fusion score for the local feature is calculated based on the local feature, and for each local feature, a second fusion score for the global feature is calculated based on the global feature. For each local feature, the first fusion score of the local feature is normalized to obtain the first fusion weight, and the second fusion score of the global feature is normalized to obtain the second fusion weight. The first fusion weight is used to sum the weighted sums of multiple local features to obtain a local weighted vector, and the second fusion weight is used to sum the weighted sums of multiple global features to obtain a global weighted vector. The fusion features are obtained based on the local weighted vector and the global weighted vector.

[0013] In the above technical solution, a learnable parameter generation module is used to process local features to obtain the fusion score of local features, and a learnable parameter generation module is used to process global features to obtain the fusion score of global features. In this way, the weights of local features and global features are dynamically adjustable, thereby dynamically adjusting the contribution ratio of different local features in the local weight vector and dynamically adjusting the contribution ratio of different global features in the global weight vector. This ensures that the most important information is given the greatest attention, enabling the feature extraction module to accurately capture data information.

[0014] In one possible implementation, for each local feature, a first fusion score of the local feature is calculated based on the local feature, and for each local feature, a second fusion score of the global feature is calculated based on the global feature. Specifically, this includes: performing a linear transformation on the local feature to obtain a first intermediate feature, processing the first intermediate feature using a non-linear activation function to obtain a second intermediate feature, and processing the second intermediate feature using fusion vector parameters to obtain the first fusion score. A linear transformation is applied to the global features to obtain the third intermediate feature. A nonlinear activation function is used to process the third intermediate feature to obtain the fourth intermediate feature. The fusion vector parameters are then used to process the fourth intermediate feature to obtain the second fusion score.

[0015] In the above technical solution, a linear function is used to process the global or local features, mapping them to a more suitable space. Then, a non-linear activation function is used to process the linearly processed features, introducing more variability. Finally, a fusion vector parameter is used to process the features processed by the non-linear activation function to obtain a fusion score, so that the fusion score fully combines the information in the global or local features.

[0016] In one possible implementation, a second extraction model is used to extract features from access data of the service over multiple time periods to obtain a second feature of the service, specifically including: The service access data for each time period is vectorized to obtain the access vector for each time period; the time vector for each time period is also vectorized. Generate the input vector for each time period based on the access vector and the time vector for each time period; The encoding module encodes the input vector of the current time period based on the input vectors of other time periods and outputs the encoded features of the current time period. The second feature is obtained by using a multi-layer perceptron to extract features from the encoded features.

[0017] In the above technical solution, considering that log data typically contains a large number of event records, which may be distributed over a wide time range and have randomness, firstly, when generating input vectors, in addition to encoding based on log data, introducing time vectors allows the model to distinguish the significance of the same event occurring at different times, enhancing its understanding of time series. Secondly, when processing log data using the encoding module, the model is allowed to consider data from all time periods simultaneously when processing data from the current time period, rather than being limited to the current time period, thus ensuring that the model can effectively capture the characteristics of these widely dispersed data.

[0018] In one possible implementation, an encoding module encodes the input vector of the current time period based on the input vectors of other time periods, outputting the encoded features of the current time period. Specifically, this includes: For each input vector, the first linear transformation module is used to perform linear processing on the input vector to obtain the first intermediate vector of the input vector, the second linear transformation module is used to perform linear processing on the input vector to obtain the second intermediate vector, and the third linear transformation module is used to perform linear processing on the input vector to obtain the third intermediate vector. For each input vector, the similarity between the first intermediate vector of the input vector and the second intermediate vector of other input vectors is calculated. The similarity between the first intermediate vector of the input vector and the second intermediate vector of other input vectors is normalized to obtain the weight value of the input vector and other input vectors. The weight values ​​of the input vector and other input vectors are used to perform a weighted summation of the third intermediate vectors of each input vector to obtain the encoded features for a time period.

[0019] In some examples, the encoding module combines the input vector of the current time period with the input vectors of other time periods to obtain the encoded features of the current time period. This ensures that the encoded features of the current time period not only contain information from its own input vector but also from the input vectors of other time periods. This allows the encoding module to focus not only on its own information but also on information from other time periods, enabling the capture of information over long time periods.

[0020] In one possible implementation, the accessed data includes: request frequency, response time, and session duration; Accordingly, the access data for each service in each time period is vectorized to obtain the access vector for each time period, specifically including: The request frequency is normalized, and the first perceptron is used to process the normalized request frequency to output a request frequency vector. The response time is normalized, and the normalized response time is processed using a second perceptron to output a response time vector; The dialogue duration is normalized, and a third perceptron is used to process the normalized dialogue duration to output a dialogue duration vector. The request frequency vector, response time vector, and dialogue duration vector are concatenated to generate the access vector; The first, second, and third sensors have the same structure and share parameters.

[0021] In the above technical solution, a perceptron is used to map request frequency, response time, and dialogue duration into vectors, so that the vectors fully express the information in request frequency, response time, and dialogue duration. The first, second, and third perceptrons have the same structure and share parameters. By sharing parameters, the model can better identify and utilize the interdependencies between features, thereby extracting higher-level abstract features and reducing the number of training parameters.

[0022] In one possible implementation, the input vector for each time period is generated based on the access vector and the time vector of each time period, specifically including: For any given time period, calculate the sum of the access vector and the time vector to obtain the input vector.

[0023] In one possible implementation, the first feature and the second feature are fused to obtain grouping features, specifically including: The first and second features of the service are concatenated to obtain the concatenated features of the service. The fully connected layer is used to process the splicing features of the services to obtain the grouping features of the services.

[0024] In one possible implementation, the method also includes: Construct a loss function, which includes the reconstruction loss of the first extraction model, the reconstruction loss of the second extraction model, the feature consistency loss, and the clustering loss; The parameters of the first extraction model, the second extraction model, and the fully connected layer are optimized using the training data with the goal of minimizing the loss function.

[0025] In the above technical solution, the loss function includes reconstruction loss, consistency loss and clustering loss. The reconstruction loss of the two extraction modules enables the two extraction modules to accurately extract information from the data, making the two features more consistent. Feature consistency improves the comprehensiveness and expressiveness of the features. The clustering loss makes the features generated by the extraction model not only effective at the individual level, but also consistent at the group group level, thereby enhancing the overall effectiveness of the features. The three losses influence each other, thereby improving the accuracy of the model.

[0026] In one possible implementation, a loss function is constructed, which includes the reconstruction loss of the first extraction model, the reconstruction loss of the second extraction model, the feature consistency loss, and the clustering loss, specifically including: Construct a decoder for the first extraction model and a decoder for the second extraction model; The first training data is processed by the first extraction model to output the first training feature, and the first training feature is input into the decoder of the first extraction model to output the first reconstructed feature. The second training data is processed by the encoder to output the second training feature, and the second training feature is input into the decoder of the second extraction model to output the second reconstructed feature. The fused training features and the second training features output by the fusion module are passed through a fully connected layer to output clustering training features. The clustering algorithm is then used to output the probability of each training data point belonging to a class. The Euclidean distance between the first reconstructed feature and the first training data is calculated as the reconstruction loss of the first extraction model, and the Euclidean distance between the second reconstructed feature and the second training data is calculated as the reconstruction loss of the second extraction model. The Euclidean distance between the fused training features and the second training features is calculated as the feature consistency loss. The clustering loss for each group is calculated based on the clustering training features of each training data point and the probability that each training data point belongs to each group.

[0027] In the above technical solution, by reconstructing the decoders of the first extraction model and the second extraction model, and calculating the reconstruction loss based on the decoder output data and the input data of the extraction model, it is ensured that the extraction model can accurately capture the information of the input data. The Euclidean distance between the first and second training features is used as the consistency loss. Since the first and second training features are close, after the clustering algorithm outputs groups that meet the cyclic condition, the loss value of each group is used as the clustering loss. By designing the loss function in this way, it is ensured that the smaller the value of the loss function, the more optimal the features extracted by the model.

[0028] In one possible implementation, a migration strategy is generated based on the dependencies between services in at least one service group, specifically including: Based on the dependencies between services in the service group, the services are divided into multiple batches and the order of each batch is determined. The data volume and importance level of each service in a single batch are obtained. The order of services in each batch is determined based on the data volume of each service in a single batch, the importance level of each service, and the migration bandwidth at each time point. Migrate services according to the order of each batch and the order of each service within each batch.

[0029] In the above technical solution, services in the service group are migrated in multiple batches according to the dependencies of each service. In each batch, the priority of each batch is determined by taking into account the importance level of the service, the amount of data, and the migration bandwidth. In this way, compared with moving the entire service, the batch migration strategy is formulated by taking into account the service dependencies, importance level, data amount, and migration bandwidth, which reduces the impact on users and the system during the migration process.

[0030] In one possible implementation, services are divided into multiple batches based on their dependencies within the service group, and the order of these batches is determined. Specifically, this includes: Generate a directed dependency graph based on the dependencies of each service in the service group, group services at the same level in the directed dependency graph as a batch of services, and determine the order of services in each batch based on the level of services in the directed dependency graph. Accordingly, the order of services in each batch is determined based on the data volume of each service in a single batch, the importance level of each service, and the migration bandwidth at each time point, specifically including: The time period covered by the migration for each batch is determined based on the amount of data served in each batch and the migration bandwidth at each time point. In each batch of migration, select the time period corresponding to the maximum migration bandwidth to migrate the service with the largest data volume in that batch; in the remaining time period, migrate services according to their importance level from low to high.

[0031] In the above technical solution, services in the service group are migrated in multiple batches according to the dependencies of each service. In each batch, the priority of each batch is determined by considering the importance level of the service, the amount of data, and the migration bandwidth. In this way, the infrastructure on which the service depends has been migrated, and the migration is carried out according to the importance level. The services with lower importance levels confirm whether the second PaaS platform has the basic environment for migration before migrating the services with higher importance levels, thereby reducing the impact on users and the system during the migration process.

[0032] This application provides an electronic device, including: a memory and a processor; The memory stores the instructions that the computer executes; The processor executes computer execution instructions stored in memory, causing the processor to perform the various possible implementations described above.

[0033] This application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the various possible implementations described above.

[0034] This application provides a computer program product, including a computer program that, when executed by a processor, implements the various possible implementations described above. Attached Figure Description

[0035] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0036] Figure 1 A flowchart illustrating an information data processing method provided in this application; Figure 2 This is a schematic diagram of the architecture of a first extraction module provided in this application.

[0037] The accompanying drawings have illustrated specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concepts of this application to those skilled in the art through reference to specific embodiments. Detailed Description Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application.

[0038] Because software systems contain numerous services, the deployment of monitoring tools needs to be determined based on the interactions between these services, which increases the complexity of the monitoring code. To address this issue, this application provides an information data processing method. By extracting features from the service's runtime data and using a clustering algorithm to process the grouping features, groups of each service are obtained. Then, monitoring code is deployed in each group as a unit, thereby reducing the complexity of the monitoring code.

[0039] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.

[0040] Figure 1 An information data processing method provided in some embodiments of this application specifically includes the following steps: S101. When running the target system deployed on the first PaaS platform, obtain the running data of multiple services in the target system over multiple time periods.

[0041] The target system is deployed on the first PaaS platform and needs to be migrated to the second PaaS platform. The platform framework of the first PaaS platform and the platform framework of the second PaaS platform can be different or the same. The programming language of the first PaaS platform and the programming language of the second PaaS platform can be different or the same.

[0042] The target system comprises multiple services. While the target system is running on the first PaaS platform, operational data for each service over multiple time periods is obtained by reading monitoring data from the monitor on the first PaaS platform or by reading log data from the target system.

[0043] S102. Use a feature extraction model to extract features from the running data of multiple services in the target system over multiple time periods to obtain the grouped features of each service.

[0044] Specifically, for the operational data of each service across multiple time periods, the feature extraction model extracts features from the operational data to obtain grouped features for each service.

[0045] S103. Use a clustering algorithm to process the grouping features of each service to obtain at least one service group.

[0046] After obtaining the grouping features of each service, the following steps are used to cluster the grouping features of each service to obtain the first classification group: Step 1: Randomly select M service grouping features from the grouping features of multiple services as initial centroids. Initialize the membership matrix, where each element represents the degree to which a data point belongs to a cluster, and the sum of each row of elements is 1.

[0047] Step 2: Calculate and update the membership degree of each service's grouping features to the centroid's grouping features based on the grouping features and centroids of each service, and use a weighted average method to process the membership degree of each service and update the centroid.

[0048] Step 3: Repeat Step 1 and Step 2 until the change in membership is less than the preset threshold, or the number of iterations is greater than the preset threshold, then stop the loop.

[0049] S104. For each service group, deploy monitoring tools based on the information of each service in the service group, and obtain the dependency relationships of each service in the service group monitored by the monitoring tools.

[0050] Specifically, taking each service group as a unit, the system analyzes information such as the interface, deployment location, protocol and serialization method, and external dependencies of each service. It then combines existing monitoring tools to deploy monitoring tools and obtain the dependency relationships of each service in the service group monitored by the monitoring tools.

[0051] For example, a service group includes services A, B, C, and D. Based on the information of services A, B, C, and D, a monitoring tool is deployed to obtain the dependency relationships between services A and D. The monitoring tool outputs monitoring results indicating that service A depends on service C, service B depends on service C, and service C depends on service D.

[0052] Another service group includes services E, F, and G. Based on the information of services E, F, and G, a monitoring tool is deployed to obtain the dependency relationship between services E and G. The monitoring tool outputs the monitoring results indicating that service E depends on service F, and service F depends on service G.

[0053] S105. Generate a migration strategy based on the dependencies of services in at least one service group, and migrate the target system from the first PaaS platform to the second PaaS platform according to the migration strategy.

[0054] Specifically, services without dependencies are identified based on the dependencies of each service in the service group, and these services are migrated first. Then, each service is migrated sequentially based on its dependencies.

[0055] For example, among services A, B, C, and D, service C is dependent on services A and B, and service C in turn depends on service D. Service D has no dependencies. Among services E, F, and G, service G has no dependencies. Prioritize migrating services D and G, then migrate services C and F, followed by services B, A, and E.

[0056] In the above technical solution, feature extraction is performed on the service's operational data to extract features from a large volume of operational data. Then, a clustering algorithm is used to process these features to obtain groups for each service. This reduces the data processing workload of the clustering algorithm and improves its accuracy. The clustering algorithm ensures that the service groups output by the algorithm are independent of each other. Monitoring code is then deployed within each group as a unit. Since the monitoring code is deployed based on the information of each service within each group, the complexity of the monitoring code is reduced. Furthermore, the improved accuracy of the clustering algorithm and the independence between the service groups output by the algorithm improve the accuracy of the obtained dependency relationships and reduce the probability of system migration failure.

[0057] In one possible implementation, S102, a feature extraction model is used to extract features from the runtime data of multiple services in the target system over multiple time periods to obtain grouped features for each service, specifically including: S201. Use the first extraction model to extract features from the resource usage data of the service over multiple time periods to obtain the first feature of the service.

[0058] Operational data includes resource usage data and access data. Resource usage data for the target system at various time periods is obtained by acquiring monitoring data from the target system's monitor. This resource usage data includes CPU utilization, memory consumption, storage space utilization, storage I / O access frequency, and network traffic at various time periods.

[0059] The first extraction model includes multiple feature extraction modules. These modules extract features from the resource usage data of the service over multiple time periods to obtain the first feature of the service.

[0060] S202. Use the second extraction model to extract features from the service's access data over multiple time periods to obtain the service's second features.

[0061] Specifically, the system obtains access data for the target system in various time periods by calling the target system's log data. This access data includes the request frequency, response time, and conversation duration in each time period.

[0062] The second extraction model includes multiple feature extraction modules. These modules extract features from the service's access data over multiple time periods to obtain the service's second features.

[0063] The model architectures of the first and second extraction models differ. This is because service resource usage data is a continuous time series and may exhibit periodicity and trend changes, while service access data is random and prone to sudden changes. Considering the different distribution patterns of the two types of data, two separate extraction models are used to process the two types of data respectively.

[0064] S203. The first and second features of the service are fused to obtain grouping features.

[0065] Specifically, the first and second features of the service are concatenated to obtain the concatenated features of the service. A fully connected layer is then used to fuse the concatenated features of the service to obtain the grouping features of the service.

[0066] Taking any service as an example, the first and second features of the service are concatenated to obtain the concatenated features of the service. A fully connected layer is then used to fuse the concatenated features of the service to obtain the grouped features of the service.

[0067] In the above technical solution, resource usage data and access data of the target system during runtime are acquired. Analysis of this data reveals the relationships between various services. By analyzing these two sets of data, even if one data source has noise or missing information, high-quality features from the other data source can be preserved through the feature extraction module, thus making the entire system more robust. Furthermore, since resource usage data and access data have different distribution characteristics, using different extraction models to process them separately allows for more accurate feature extraction, more precise grouping, and a reduced probability of migration failure.

[0068] In one possible implementation, Figure 2 This is a schematic diagram of the architecture of a first extraction model. S201. Using the first extraction model, feature extraction is performed on the resource usage data of each service across multiple time periods to obtain the first feature of each service, specifically including: S301. Use the local extraction module to extract features from the preprocessed resource usage data and output local features.

[0069] The local extraction module extracts data features over a short period, achieving local feature extraction of resource usage data and outputting these local features. In some examples, the local extraction module is a convolutional neural network, which converts preprocessed resource usage data into image data. When there are multiple types of resource usage data, multi-channel image data can be generated, with each channel representing one type. In some examples, the local extraction module is a recurrent neural network.

[0070] In some examples, before using the local extraction module to process the resource usage data, a preprocessing module can be used to preprocess the resource usage data of the service in multiple time periods to obtain preprocessed resource usage data.

[0071] The preprocessing process includes handling missing values, outliers, and numerical normalization. Missing data for a given time period is imputed using the average of data from adjacent time periods. Outliers for a given time period are corrected using resource usage data from adjacent time periods. The resource usage data, after imputed and corrected for outliers, is then normalized to obtain the preprocessed resource usage data.

[0072] S302. Use the global extraction module to extract features from local features and output global features.

[0073] The global extraction module captures long-term data features to extract global features from the local features output by the local extraction module. In some examples, the global extraction module is a Long Short-Term Memory (LSTM) network or a gating monitoring unit. When the local extraction module is a convolutional neural network (CNN), the local features output by the CNN are transformed into multiple sets of one-dimensional data, allowing the LSTM network or gating monitoring unit to process the local features.

[0074] S303. Use the fusion module to perform feature fusion on local features and global features to obtain fused features, and use the splicing module to splice the global features and fused features as the first feature output.

[0075] Specifically, a fusion module is used to fuse local and global features to obtain a fused feature, which incorporates more information. A concatenation module then concatenates the global and fused features to output the first feature.

[0076] In the above technical solution, resource usage data contains rich instantaneous changes and local patterns. The local feature extraction module can accurately capture these details and understand the immediate state of the data. Global feature extraction focuses on the overall trend and long-term dependence of the data, grasping the long-term periodic characteristics and overall behavior of the data, and realizing the identification of long-term patterns. The fusion module merges local and global features, so that the fused features contain richer information extracted from the immediate state and long-term patterns. Then, the fused features and global features are concatenated and output, so that the concatenated features contain richer information. In addition, the local extraction module, global extraction module, and fusion module are each good at extracting different types of information. Even if one model is biased, other models can supplement it to improve the accuracy of the extracted features.

[0077] In one possible implementation, S304, the fusion module is used to extract features from local and global features to obtain fused features, specifically including: S401. For each local feature, calculate the first fusion score of the local feature based on the local feature, and for each local feature, calculate the second fusion score of the global feature based on the global feature.

[0078] Specifically, for each local feature, a parameter generation module processes it to obtain a first fusion score. For each global feature, the parameter generation module processes it to obtain a second fusion score. The parameters in the parameter generation module are learnable.

[0079] S402. Normalize the first fusion score of the local features to obtain the first fusion weight, and normalize the second fusion score of the global features to obtain the second fusion weight.

[0080] Specifically, the first fusion score of each local feature is normalized using an existing normalization function to obtain multiple first fusion weights, and the second fusion score of each global feature is normalized using an existing normalization function to obtain multiple second fusion weights.

[0081] S403. Use the first fusion weight to perform a weighted summation on multiple local features to obtain a local weighted vector, and use the second fusion weight to perform a weighted summation on multiple global features to obtain a global weighted vector.

[0082] Specifically, a local weighted vector is obtained by weighting and summing the first fusion weight of each local feature with each local feature, and a global weighted vector is obtained by weighting and summing the second fusion weight of each global feature with each global feature.

[0083] S404. Obtain the fused features based on the local weighted vector and the global weighted vector.

[0084] The fusion feature is obtained by adding the local weighted vector and the global weighted vector.

[0085] In the above technical solution, a learnable parameter generation module is used to process local features to obtain the fusion score of local features, and a learnable parameter generation module is used to process global features to obtain the fusion score of global features. In this way, the weights of local features and global features are dynamically adjustable, thereby dynamically adjusting the contribution ratio of different local features in the local weight vector and dynamically adjusting the contribution ratio of different global features in the global weight vector. This ensures that the most important information is given the greatest attention, enabling the feature extraction module to accurately capture data information.

[0086] In one possible implementation, S401, calculating the first fusion score of the local features and the second fusion score of the global features, specifically includes: S501. Perform a linear transformation on the local features to obtain the first intermediate feature, process the first intermediate feature using a nonlinear activation function to obtain the second intermediate feature, and process the second intermediate feature using the fusion vector parameters to obtain the first fusion score.

[0087] Specifically, a linear transformation function is used to perform a linear transformation on the local features to obtain the first intermediate feature. A non-linear activation function, such as the Sigmoid function or the ReLU function, is then used to process the first intermediate feature to obtain the second intermediate feature. The fusion vector parameters are learned, and these parameters are used to process the second intermediate feature to obtain the first fusion score.

[0088] S502. Perform a linear transformation on the global features to obtain the third intermediate feature, process the third intermediate feature using a non-linear activation function to obtain the fourth intermediate feature, and process the fourth intermediate feature using the fusion vector parameters to obtain the second fusion score.

[0089] Specifically, a linear transformation function is used to perform a linear transformation on the global features to obtain the third intermediate feature. A non-linear activation function, such as the Sigmoid function or the ReLU function, is then used to process the third intermediate feature to obtain the fourth intermediate feature. The fusion vector parameters are learned, and these parameters are used to process the fourth intermediate feature to obtain the second fusion score.

[0090] In the above technical solution, a linear function is used to process the global or local features, mapping them to a more suitable space. Then, a non-linear activation function is used to process the linearly processed features, introducing more variability. Finally, a fusion vector parameter is used to process the features processed by the non-linear activation function to obtain a fusion score, so that the fusion score fully combines the information in the global or local features.

[0091] In one possible implementation, S202, the second extraction model is used to extract features from the access data of each service over multiple time periods to obtain the second features of each service, specifically including: S601. Vectorize the access data of each service in each time period to obtain the access vector of each time period, and vectorize each time period to obtain the time vector of each time period.

[0092] Taking any one of these services as an example, for any time period of access data for the service, the access data for that time period is vectorized to obtain the access vector for that time period. Vectorization of any time period yields the time vector for that time period. The dimension of the access vector is the same as the dimension of the time vector.

[0093] The service includes access data over T time periods. By vectorizing the access data across all time periods, T access vectors are obtained. Similarly, by vectorizing the access data across all time periods, T time vectors are obtained.

[0094] S602. Generate the input vector for each time period based on the access vector and the time vector for each time period.

[0095] Specifically, for any given time period's access vector and time vector, the access vector and time vector are added together to obtain the input vector for that time period. Given T access vectors and time vectors for T time periods, the input vectors for T time periods can be obtained.

[0096] S603. Use the encoding module to encode the current time period based on the input vector of the current time period and the input vector of other time periods, and output the encoded features of the current time period.

[0097] In some examples, the encoding module combines the input vector of the current time period with the input vectors of other time periods to obtain the encoded features of the current time period. This ensures that the encoded features of the current time period not only contain information from its own input vector but also from the input vectors of other time periods. This allows the encoding module to focus not only on its own information but also on information from other time periods, enabling the capture of information over long time periods.

[0098] In some examples, an encoding module is used to encode the input vector for each time period to obtain the encoded features for that time period, specifically including: For each input vector, the first linear transformation module is used to perform linear processing on the input vector to obtain the first intermediate vector of the input vector, the second linear transformation module is used to perform linear processing on the input vector to obtain the second intermediate vector, and the third linear transformation module is used to perform linear processing on the input vector to obtain the third intermediate vector.

[0099] For each input vector, the similarity between the first intermediate vector of the input vector and the second intermediate vector of other input vectors is calculated. The similarity between the first intermediate vector of the input vector and the second intermediate vector of other input vectors is normalized to obtain the weight value of the input vector and other input vectors. The weight values ​​of the input vector and other input vectors are used to perform a weighted summation of the third intermediate vectors of each input vector to obtain the encoded features for a time period.

[0100] More specifically, given n input vectors, for the first input vector, a linear transformation is used to obtain the first, second, and third intermediate vectors of the first input vector. For the second input vector, the same linear transformation is used to obtain the first, second, and third intermediate vectors of the second input vector. And so on, for the nth input vector, the same linear transformation is used to obtain the first, second, and third intermediate vectors of the nth input vector.

[0101] For the first input vector, let j iterate from 1 to n, and calculate the similarity between the first input vector and the j-th input vector. Then, for the first input vector, let j iterate from 1 to n, and calculate the first product of the similarity between the first input vector and the j-th input vector and the first product of the first input vector. Repeat this process to obtain n first products, and then sum these n first products to obtain the encoded features for the first time segment.

[0102] Other input vectors are also processed in the same way to obtain the encoded features for n time periods.

[0103] S604. Use a multi-layer perceptron to extract features from the encoded features to obtain the second feature.

[0104] The multilayer perceptron layer includes an input layer, at least one hidden layer, and an output layer. The hidden layer includes a linearization layer and a nonlinearization layer. The nonlinearization layer can use a nonlinear activation function to extract features from the encoded features through the multilayer perceptron layer to obtain the second feature.

[0105] In some examples, layer normalization and residual processing can be performed before the multilayer perceptron layers to improve model performance.

[0106] In the above technical solution, considering that log data typically contains a large number of event records, which may be distributed over a wide time range and have randomness, firstly, when generating input vectors, in addition to encoding based on log data, introducing time vectors allows the model to distinguish the significance of the same event occurring at different times, enhancing its understanding of time series. Secondly, when processing log data using the encoding module, the model is allowed to consider data from all time periods simultaneously when processing data from the current time period, rather than being limited to the current time period, thus ensuring that the model can effectively capture the characteristics of these widely dispersed data.

[0107] In one possible implementation, S601, the access data of each service in each time period is vectorized to obtain the access vector for each time period, specifically including: S701. Normalize the request frequency and use the first perceptron to process the normalized request frequency and output the request frequency vector.

[0108] S702. Normalize the response time and use a second perceptron to process the normalized response time and output the response time vector.

[0109] S703. Normalize the dialogue duration and use a third perceptron to process the normalized dialogue duration and output the dialogue duration vector.

[0110] S704. Concatenate the request frequency vector, response time vector, and dialogue duration vector to generate an access vector.

[0111] The access data includes request frequency, response time, and conversation duration. Taking access data for a specific time period as an example, the request frequency for that period is normalized, and then a first perceptron processes the normalized request frequency to output a request frequency vector for that period. The response time for that period is normalized, and then a second perceptron processes the normalized response time to output a response time vector for that period. The conversation duration for that period is normalized, and then a third perceptron processes the normalized conversation duration to output a conversation duration vector for that period. The request frequency vector, response time vector, and conversation duration vector for that period are concatenated to obtain the access vector for that period.

[0112] The first, second, and third sensors have the same structure and share parameters. They can be either multilayer or single-layer sensors.

[0113] In the above technical solution, a perceptron is used to map request frequency, response time, and dialogue duration into vectors, so that the vectors fully express the information in request frequency, response time, and dialogue duration. The first, second, and third perceptrons have the same structure and share parameters. By sharing parameters, the model can better identify and utilize the interdependencies between features, thereby extracting higher-level abstract features and reducing the number of training parameters.

[0114] Some embodiments of this application also provide an information data processing method, which specifically includes: S801. Construct the loss function, which includes the reconstruction loss of the first extraction model, the reconstruction loss of the second extraction model, the feature consistency loss, and the clustering loss.

[0115] The reconstruction loss of the first extraction model refers to the loss between the data obtained after reconstructing the first feature output by the first extraction model and the input data of the first extraction model. The reconstruction loss of the second extraction model refers to the loss between the data obtained after reconstructing the second feature output by the second extraction model and the input data of the second extraction model. The feature consistency loss refers to the loss between the first feature output by the first extraction model and the second feature output by the second extraction model. The clustering loss is used to calculate the differences between the group features of each training data group.

[0116] The reconstruction loss of the feature extraction model ensures that the features generated by the model can reconstruct the original input data well, and that the features output by the model can capture the main information of the input data. By enabling the output features of the feature extraction model to have good reconstruction capabilities, the quality and reliability of the features are enhanced, making them more representative of the essential attributes of the original data. This results in more robust grouped features when using fully connected layers for fusion.

[0117] Although the first and second features focus on different aspects of the data, the consistency loss ensures that these two sets of features maintain consistency when representing the same data, improving the comprehensiveness and expressiveness of the features. Furthermore, the reconstruction loss further enhances the consistency and richness of the information contained in the first and second features, laying the foundation for further feature consistency losses.

[0118] Clustering loss ensures that the features generated by the extraction model are effective not only at the individual level but also at the group group level, thereby enhancing the overall effectiveness of the features. In addition, clustering loss ensures that the grouped features concatenated from the first and second features are logically consistent, which helps to improve the effect of multimodal feature fusion.

[0119] S802. Optimize the parameters of the first extraction model, the second extraction model, and the fully connected layer using the training data with the goal of minimizing the loss function.

[0120] The training data is propagated forward through the first extraction model, the second extraction model, the fully connected layer, and the clustering algorithm. The loss value is calculated based on the loss function, and backpropagation is performed based on the loss value to optimize the parameters of the first extraction model, the second extraction model, and the fully connected layer.

[0121] In the above technical solution, the loss function includes reconstruction loss, consistency loss and clustering loss. The reconstruction loss of the two extraction modules enables the two extraction modules to accurately extract information from the data, making the two features more consistent. Feature consistency improves the comprehensiveness and expressiveness of the features. The clustering loss makes the features generated by the extraction model not only effective at the individual level, but also consistent at the group group level, thereby enhancing the overall effectiveness of the features. The three losses influence each other, thereby improving the accuracy of the model.

[0122] In one possible implementation, S801, a loss function is constructed, which includes the reconstruction loss of the first extraction model, the reconstruction loss of the second extraction model, the feature consistency loss, and the clustering loss, specifically including: S901, Construct the decoder for the first extraction model and the decoder for the second extraction model.

[0123] In the first extraction model, which includes a convolutional neural network, a long short-term memory network, and a fusion module, since the fusion module fuses two features, the decoder for the first extraction model only needs to be constructed as an inverse recurrent neural network and an inverse convolutional neural network.

[0124] The second extraction module includes an encoding module and a multilayer perceptron, and the construction of the second extraction module also includes an encoding module and a multilayer perceptron.

[0125] S902. The first training data of the service is processed by the first extraction model to output the first training feature. The first training feature is input into the decoder of the first extraction model so that the decoder of the first extraction model outputs the first reconstructed feature.

[0126] The training data set includes first training data and second training data for a service. The first training data is input into a first extraction model, which extracts features from the first training data and outputs first training features. The first training features are then input into the decoder of the first extraction model, which reconstructs the data and outputs first reconstructed features.

[0127] S903. The second training data of the service is processed by the second extraction model to output the second training feature. The second training feature is then input into the decoder of the second extraction model so that the decoder of the second extraction model outputs the second reconstructed feature.

[0128] Specifically, the second training data of the service is input into the second extraction model, which extracts features from the second training data and outputs second training features. The second training features are then input into the decoder of the second extraction model, which reconstructs the data and outputs second reconstructed features.

[0129] S904. After passing the first and second training features through a fully connected layer, output clustering training features. Use a clustering algorithm to output the probability that each training data belongs to each group.

[0130] The process involves using a fully connected layer to fuse the first and second training features to output clustering training features. A clustering algorithm then processes these features to output the probability that each training data point belongs to a specific group. Each group has a centroid.

[0131] S905. Calculate the Euclidean distance between the first reconstructed feature and the first training data as the reconstruction loss of the first extraction model, and calculate the Euclidean distance between the second reconstructed feature and the second training data as the reconstruction loss of the second extraction model.

[0132] Specifically, after the forward propagation of the training data in the model is completed, the loss function for this propagation is calculated. More specifically, the Euclidean distance between the first reconstructed feature and the first training data is calculated, and the calculated Euclidean distance is used as the reconstruction loss of the first extraction model. The Euclidean distance between the second reconstructed feature and the second training data is calculated, and the calculated Euclidean distance is used as the reconstruction loss of the second extraction model.

[0133] S906. Calculate the Euclidean distance between the first training feature and the second training feature as the feature consistency loss.

[0134] Specifically, the Euclidean distance between the first training feature and the second training feature is calculated, and the calculated Euclidean distance is used as the consistency loss.

[0135] S907. Calculate the clustering loss for each group based on the clustering training features of each training data and the probability that each training data belongs to each group.

[0136] Specifically, for each group, the Euclidean distance between the clustering training features of each training data in the group and the centroid of the group is calculated. The probability that the training data belongs to the group is used to correct the Euclidean distance. The sum of all corrected Euclidean distances in the group is calculated as the clustering loss.

[0137] In the above technical solution, by reconstructing the decoders of the first extraction model and the second extraction model, and calculating the reconstruction loss based on the decoder output data and the input data of the extraction model, it is ensured that the extraction model can accurately capture the information of the input data. The Euclidean distance between the first and second training features is used as the consistency loss. Since the first and second training features are close, after the clustering algorithm outputs groups that meet the cyclic condition, the loss value of each group is used as the clustering loss. By designing the loss function in this way, it is ensured that the smaller the value of the loss function, the more optimal the features extracted by the model.

[0138] In one possible implementation, S105, generating a migration strategy based on the dependencies of services in at least one service group, specifically includes: S1001. Divide each service into multiple batches according to the dependency relationship of each service in the service group and determine the order of each batch, and obtain the data volume and importance level of each service in a single batch.

[0139] Specifically, a directed dependency graph is generated based on the dependencies of each service in the service group. Services at the same level in the directed dependency graph are grouped into a batch of services, and the order of services in each batch is determined based on the level of the services in the directed dependency graph.

[0140] For example, a service group includes service 1, service 2, service 3, service 4, and service 5. Service 4 is depended on by service 1, service 2, and service 3, and service 4 in turn depends on service 5. Service 5 has no dependencies. Service 4 and service 5 each form a separate layer, while service 1, service 2, and service 3 all depend on 4, forming another layer. Service 5 is in the first batch, service 4 is in the second batch, and services 1, service 2, and service 3 form the third batch.

[0141] The service level is determined based on the function of each service. If a service performs a core function, its level is relatively high; if a service performs a non-core function, its level is relatively low.

[0142] S1002. Determine the order of services in each batch based on the data volume of each service in a single batch, the importance level of each service, and the migration bandwidth at each time point.

[0143] If the service involves a large amount of data, migration can be carried out during off-peak traffic periods; if the service level is higher, migration can be scheduled for a later time. The bandwidth at each point in the migration period is determined by analyzing historical bandwidth data.

[0144] In some examples, the order of services within the same batch is determined as follows: S111. Determine the time period covered by the migration of services in each batch based on the data volume of services in each batch and the migration bandwidth at each time.

[0145] Specifically, the total amount of data in a batch is calculated based on the amount of data of each service in the batch. The time required to migrate the services in the batch is determined based on the total amount of data and the migration bandwidth at each time point. The time period covered by the migration of the services in the batch is determined based on the pre-set migration start time and the time required to migrate the services in the batch.

[0146] S112. In each batch of migration coverage time period, select the time period corresponding to the maximum migration bandwidth to migrate the service with the largest data volume in that batch; in the remaining time period, migrate the services from low to high according to their importance level.

[0147] Continuing with the above example, in the third batch, the data volume of service 1 is greater than that of service 2, the data volume of service 2 is greater than that of service 3, the level of service 3 is greater than that of service 2, and the level of service 2 is greater than that of service 1.

[0148] In the time period covered by the third batch of services, select the time period with the largest migration bandwidth to migrate Service 1. In the remaining time period, prioritize migrating Service 2, then Service 3. If the largest migration bandwidth is in the early part of the time period covered by the third batch of services, the migration order is Service 1, Service 2, and Service 3. If the largest migration bandwidth is in the middle of the time period covered by the third batch of services, the migration order is Service 2, Service 1, and Service 3. If the largest migration bandwidth is in the late part of the time period covered by the third batch of services, the migration order is Service 2, Service 3, and Service 1.

[0149] S113. Migrate each service according to the order of each batch and the order of each service in each batch.

[0150] After obtaining the order of each batch and the order of each service within each batch, the migration is carried out batch by batch according to the order of each batch, and within each batch, the migration is carried out according to the order of each service.

[0151] In the above technical solution, services in the service group are migrated in multiple batches according to the dependencies of each service. In each batch, the priority of each batch is determined by considering the importance level of the service, the amount of data, and the migration bandwidth. In this way, the infrastructure on which the service depends has been migrated, and the migration is carried out according to the importance level. The services with lower importance levels confirm whether the second PaaS platform has the basic environment for migration before migrating the services with higher importance levels, thereby reducing the impact on users and the system during the migration process.

[0152] This application also provides an information data processing system, including: The acquisition module is used to acquire the running data of multiple services in the target system over multiple time periods when the target system is running on the first PaaS platform. The extraction module is used to extract features from the running data of multiple services in the target system over multiple time periods using a feature extraction model to obtain grouped features for each service. The processing module is used to process the grouping features of each service using a clustering algorithm to obtain at least one service group; The monitoring module is used to deploy monitoring tools for each service group based on the information of each service in the service group, and to obtain the dependency relationships of each service in the service group as monitored by the monitoring tools. The migration module is used to generate a migration strategy based on the dependencies of each service in at least one of the service groups, and to migrate the target system from the first PaaS platform to the second PaaS platform according to the migration strategy.

[0153] The electronic device provided in this embodiment includes at least one processor and a memory. Optionally, the device further includes a communication component. The processor, memory, and communication component are connected via a bus.

[0154] In a specific implementation, at least one processor executes computer execution instructions stored in memory, causing at least one processor to perform the above-described method.

[0155] The specific implementation process of the processor can be found in the above method embodiments, and its implementation principle and technical effect are similar, so it will not be repeated here.

[0156] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0157] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0158] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0159] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0160] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0161] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0162] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0163] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0164] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0165] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0166] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0167] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0168] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. An information-based data processing method, characterized in that, include: When running a target system deployed on a first PaaS platform, obtain the running data of multiple services in the target system over multiple time periods; The feature extraction model is used to extract features from the running data of multiple services in the target system over multiple time periods to obtain the grouped features of each service. Clustering algorithms are used to process the grouping characteristics of each service to obtain at least one service group; For each service group, monitoring tools are deployed based on the information of each service in the service group to obtain the dependency relationships of each service in the service group as monitored by the monitoring tools. A migration strategy is generated based on the dependencies of each service in at least one of the service groups, and the target system is migrated from the first PaaS platform to the second PaaS platform according to the migration strategy.

2. The method according to claim 1, characterized in that, The operational data includes resource usage data and access data; Accordingly, a feature extraction model is used to extract features from the operational data of multiple services in the target system over multiple time periods to obtain grouped features for each service, specifically including: The first extraction model is used to extract features from the resource usage data of the service over multiple time periods to obtain the first feature of the service; The second extraction model is used to extract features from the service's access data over multiple time periods to obtain the service's second features; The first feature and the second feature are fused to obtain the grouped features; wherein the model architecture of the first extraction model and the model architecture of the second extraction model are different.

3. The method according to claim 2, characterized in that, The first extraction model is used to extract features from the resource usage data of the service over multiple time periods to obtain the first feature of the service, specifically including: The local extraction module is used to extract features from the resource usage data and output local features. The global extraction module is used to extract features from the local features and output global features. The fusion module is used to fuse the local features and the global features to obtain fused features; The global feature and the fused feature are concatenated using the splicing module and output as the first feature.

4. The method according to claim 3, characterized in that, The step of using a fusion module to fuse the local features and the global features to obtain fused features specifically includes: For each local feature, a first fusion score of the local feature is calculated based on the local feature; and for each local feature, a second fusion score of the global feature is calculated based on the global feature. For each local feature, the first fusion score of the local feature is normalized to obtain a first fusion weight, and for each local feature, the second fusion score of the global feature is normalized to obtain a second fusion weight. The first fusion weight is used to sum the weighted sums of the multiple local features to obtain a local weighted vector, and the second fusion weight is used to sum the weighted sums of the multiple global features to obtain a global weighted vector. The fusion feature is obtained based on the local weighted vector and the global weighted vector.

5. The method according to any one of claims 2 to 4, characterized in that, The second extraction model is used to extract features from the service's access data over multiple time periods to obtain the service's second features, specifically including: The service access data for each time period is vectorized to obtain the access vector for each time period; the time vector for each time period is also vectorized. Generate the input vector for each time period based on the access vector and the time vector for each time period; The encoding module encodes the input vector of the current time period based on the input vectors of other time periods and outputs the encoded features of the current time period. The second feature is obtained by using a multi-layer perceptron to extract features from the encoded features.

6. The method according to claim 5, characterized in that, The encoding module uses the input vector of the current time period and the input vectors of other time periods to encode and output the encoded features of the current time period, specifically including: For each input vector, a first linear transformation module is used to perform linear processing on the input vector to obtain a first intermediate vector of the input vector, a second linear transformation module is used to perform linear processing on the input vector to obtain a second intermediate vector, and a third linear transformation module is used to perform linear processing on the input vector to obtain a third intermediate vector; For each input vector, the similarity between the first intermediate vector of the input vector and the second intermediate vector of other input vectors is calculated. The similarity between the first intermediate vector of the input vector and the second intermediate vector of other input vectors is normalized to obtain the weight value of the input vector and other input vectors. The weight values ​​of the input vector and other input vectors are used to perform a weighted summation of the third intermediate vectors of each input vector to obtain the encoding features for a time period.

7. The method according to claim 5, characterized in that, The access data includes: request frequency, response time, and session duration; Accordingly, the access data for each service in each time period is vectorized to obtain the access vector for each time period, specifically including: The request frequency is normalized, and the normalized request frequency is processed by the first perceptron to output a request frequency vector. The response time is normalized, and a second perceptron is used to process the normalized response time and output a response time vector. The dialogue duration is normalized, and a third perceptron is used to process the normalized dialogue duration and output a dialogue duration vector. The access vector is generated by concatenating the request frequency vector, the response time vector, and the dialogue duration vector. The first, second, and third sensors have the same structure and share parameters.

8. The method according to claim 5, characterized in that, The input vector for each time period is generated based on the access vector and the time vector for each time period, specifically including: For any given time period, the sum of the access vector and the time vector is calculated to obtain the input vector.

9. The method according to any one of claims 2 to 4, characterized in that, The grouping feature is obtained by fusing the first feature and the second feature, specifically including: The first and second features of the service are concatenated to obtain the concatenated features of the service. The concatenation features of the service are processed using a fully connected layer to obtain the grouping features of the service.

10. An information data processing system, characterized in that, include: The acquisition module is used to acquire the running data of multiple services in the target system over multiple time periods when the target system is running on the first PaaS platform. The extraction module is used to extract features from the running data of multiple services in the target system over multiple time periods using a feature extraction model to obtain grouped features for each service. The processing module is used to process the grouping features of each service using a clustering algorithm to obtain at least one service group; The monitoring module is used to deploy monitoring tools for each service group based on the information of each service in the service group, and to obtain the dependency relationships of each service in the service group as monitored by the monitoring tools. The migration module is used to generate a migration strategy based on the dependencies of each service in at least one of the service groups, and to migrate the target system from the first PaaS platform to the second PaaS platform according to the migration strategy.

Images