A method and system for managing the whole life cycle of aerospace engineering data algorithm models

Abstract

The application provides a kind of aerospace engineering data algorithm model full life cycle management method and system, it is related to aerospace engineering algorithm model management and service technical field.The method provided by the application stores multiple algorithm model files of different categories in the model library, generates the inference flowchart corresponding to the target data analysis process in response to the editing operation and the configuration operation input by the user, generates the algorithm model service file corresponding to the target data analysis process according to the inference flowchart and sends it to the server in response to the release operation input by the user, so that the server deploys the algorithm model service according to the algorithm model service file.The method provided by the application can quickly complete the deployment of the corresponding algorithm model service based on the analysis configuration operation and the algorithm model file set by the user, and then complete the analysis and processing of data through the algorithm model service, reduce the model service construction cost, improve the efficiency of model application, and enhance the data analysis and processing capability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of aerospace engineering algorithm model management and service, and particularly relates to a method and system for managing the whole life cycle of aerospace engineering data algorithm models. BACKGROUND

[0002] In the field of manned space engineering, with the long-period on-orbit operation of space stations and the implementation of subsequent missions, a large amount of engineering data is generated. Based on these data, a large number of algorithm model files for data analysis and processing are obtained. When technical personnel need to analyze and process data by using algorithm model files, the management, deployment, monitoring and evaluation of algorithm model files are complex, and the model management platform in the related art has the following problems: heterogeneous model management is dispersed, and there is a lack of unified coordination mechanism; the model deployment process is complex and time-consuming; the model monitoring mechanism is imperfect, and risk control is difficult. As a result, technical personnel cannot quickly deploy corresponding algorithm model services through algorithm model files, data analysis and processing requires a large amount of time, and there are problems of low data processing efficiency and poor model application effect.

[0003] Therefore, there is an urgent need for a method and system for managing the whole life cycle of aerospace engineering data algorithm models, which can quickly complete the deployment and update of corresponding algorithm model services based on algorithm model files, thereby reducing the cost required for model service construction, improving the efficiency of model application, and enhancing the data analysis and processing capability. SUMMARY

[0004] The embodiments of the present application provide a method and system for managing the whole life cycle of aerospace engineering data algorithm models, which can quickly complete the deployment and update of corresponding algorithm model services based on algorithm model files, thereby reducing the cost required for model service construction, improving the efficiency of model application, and enhancing the data analysis and processing capability.

[0005] To achieve the above-mentioned purpose, the embodiments of the present application adopt the following technical solutions:

[0006] In a first aspect, a method for managing a full life cycle of an aerospace engineering data algorithm model is provided. The method is applied to a first device in communication with a server. The method comprises: storing a plurality of algorithm model files of different categories in a model library; generating an inference flowchart corresponding to a target data analysis process in response to editing operations and configuration operations input by a user on a preset interface; the inference flowchart comprises a plurality of nodes connected in sequence according to the target data analysis process, wherein each node is a model node or a logic node; the editing operations comprise adding a node, deleting a node, and / or connecting nodes; the configuration operations comprise determining an algorithm model file corresponding to each model node from the model library and / or configuring an execution code corresponding to each logic node, wherein the model node is configured to process input data according to the configured algorithm model file, and the logic node is configured to process input data according to the configured execution code; and generating an algorithm model service file corresponding to the target data analysis process according to the inference flowchart and sending the algorithm model service file to the server in response to a publishing operation input by the user on the preset interface, so that the server deploys an algorithm model service according to the algorithm model service file, the algorithm model service is configured to process data according to a data analysis request sent by a second device, the data analysis request carries to-be-analyzed data, and the algorithm model service is configured to obtain a data analysis result by processing the to-be-analyzed data according to the target data analysis process and send the data analysis result to the second device.

[0007] In a possible implementation of the first aspect, the logic node comprises a data processing node, a routing node, and a combiner node; the data processing node comprises at least one input end and at least one output end, and is configured to process data input by the input end according to a configured execution code and output a data analysis result through the output end; the combiner node comprises at least one input end and a plurality of output ends, and is configured to output data input by the input end through the plurality of output ends according to a configured execution code; and the routing node comprises at least one input end and a plurality of output ends, and is configured to determine a target output end according to data input by the input end through a configured execution code and output the input data through the target output end, wherein the target output end is one of the plurality of output ends.

[0008] In a possible implementation of the first aspect, after the server deploys the algorithm model service according to the algorithm model service file, the method further comprises: sending a running state monitoring instruction to the server, wherein the running state monitoring instruction is configured to instruct the server to obtain a running state parameter corresponding to the algorithm model service, the running state parameter comprises a memory utilization rate, a processor utilization rate, and a response time; the server is configured to generate a first alarm information in a case where any running state parameter is greater than or equal to a first threshold value and a duration is greater than or equal to a second threshold value, the first alarm information is configured to prompt an abnormal running state of the algorithm model service; and the server is configured to send the first alarm information to the first device at a preset time interval.

[0009] In a possible implementation of the first aspect, after the server deploys the algorithm model service according to the algorithm model service file, the method further includes: obtaining verification data, the verification data including a plurality of verification samples and a first analysis result of each verification sample obtained by processing the verification sample based on the target data analysis process; sending an analysis result evaluation instruction to the server, the analysis result evaluation instruction carrying the verification data, the analysis result evaluation instruction being used to instruct the server to process the plurality of verification samples based on the algorithm model service according to the target data analysis process to obtain a second analysis result corresponding to each verification sample; determining, by the server, an evaluation index corresponding to the algorithm model service according to the first analysis result and the second analysis result of each verification sample, and generating a second alarm information in a case where any evaluation index is greater than or equal to a preset threshold, the server sending the second alarm information to the first device, the second alarm information being used to prompt an abnormal analysis result of the algorithm model service; wherein, in a case where the category of the algorithm model service is a binary classification model, the evaluation index includes accuracy, precision, recall, specificity, area under ROC curve, K-S test value and F1 measure; in a case where the category of the algorithm model service is a multi-classification model, the evaluation index includes accuracy, weighted precision, weighted recall and weighted F value; in a case where the category of the algorithm model service is a regression model, the evaluation index includes mean absolute error, mean square error, root mean square error and determination coefficient; and in a case where the category of the algorithm model service is a clustering model, the evaluation index includes silhouette coefficient-cosine similarity, silhouette coefficient-Euclidean distance, residual sum of squares and covariance.

[0010] In a possible implementation of the first aspect, the method further includes: in response to an iterative update instruction input by a user, iteratively training, according to the training code, an algorithm model file corresponding to one or more model nodes included in the algorithm model service, to obtain a trained algorithm model file; updating the algorithm model service file according to the trained algorithm model file, to obtain an updated algorithm model service file; and sending the updated algorithm model service file to the server, so that the server deploys the algorithm model service according to the updated algorithm model service file.

[0011] In a possible implementation manner of the first aspect, the method further includes: sending, to the server, data distribution difference monitoring instruction, the data distribution difference monitoring instruction carrying training data used for training each algorithm model file included in the algorithm model service, the data distribution difference monitoring instruction being used for instructing the server to determine a data distribution difference value between the training data and the to-be-analyzed data carried in the data analysis request sent by the second device in a case where the data analysis request is received, the server to generate third alarm information in a case where the data distribution difference value is greater than or equal to a third threshold, and the server to send the third alarm information to the second device, the third alarm information being used for prompting that there is a data distribution difference between the training data and the to-be-analyzed data.

[0012] In a possible implementation manner of the first aspect, the algorithm model service is further used for, in response to the data analysis request sent by the second device, the data analysis request carrying the to-be-analyzed data, adding the to-be-analyzed data into a task queue arranged in sequence, processing a plurality of to-be-analyzed data included in the task queue according to a target data analysis process in sequence to obtain a data analysis result corresponding to each to-be-analyzed data, and sending the data analysis result corresponding to each to-be-analyzed data to the second device.

[0013] In a possible implementation manner of the first aspect, in a case where the server is deployed with a plurality of algorithm model services corresponding to the target data analysis process, the method further includes: sending a publishing mode setting instruction to the server, the publishing mode setting instruction being used to instruct the server to set a publishing mode of the plurality of algorithm model services corresponding to the target data analysis process; wherein the publishing mode includes independent publishing, gray publishing or comparison publishing; in a case where the publishing mode is independent publishing, the publishing mode setting instruction carries an identifier of a target algorithm model service, the target algorithm model service being one of the plurality of algorithm model services, the server responds to a data analysis request sent by the second device, the data analysis request carrying to-be-analyzed data, the server processes the to-be-analyzed data through the target algorithm model service to obtain a data analysis result, and sends the data analysis result to the second device; in a case where the publishing mode is gray publishing, the server responds to a data analysis request sent by the second device, the data analysis request carrying to-be-analyzed data, the server processes the to-be-analyzed data through the plurality of algorithm model services to obtain a data analysis result corresponding to each algorithm model service, and sends the data analysis result corresponding to each algorithm model service to the second device; in a case where the publishing mode is comparison publishing, the server responds to a data analysis request sent by the second device, the data analysis request carrying to-be-analyzed data, the server processes the to-be-analyzed data through the plurality of algorithm model services to obtain a data analysis result corresponding to each algorithm model service, the server determines an evaluation index corresponding to each algorithm model service, and the server sends, to the second device, the data analysis result of the algorithm model service corresponding to a maximum value in the evaluation index corresponding to each algorithm model service.

[0014] The method provided by the application can complete the deployment of the corresponding algorithm model service based on the analysis configuration operation and the algorithm model file set by the user, and then complete the analysis and processing of data through the algorithm model service, reduce the cost required for model service construction, improve the efficiency of model application, and enhance the data analysis and processing capability. It can also be understood that: by uniformly managing a plurality of different types of heterogeneous model files, centralized management and maintenance of algorithm model files are realized, and the management efficiency is improved. In response to the operation configuration of the inference logic diagram corresponding to the target data analysis process input by the user, and the algorithm model service corresponding to the target data analysis process can be quickly sent, which can effectively simplify the deployment process of the service and shorten the cycle from development to deployment of the algorithm model service. Moreover, the method provided by the application can monitor the running state of the algorithm model service in real time, guarantee the stability and reliability of the model in the running process. And the processing result of the algorithm model service is evaluated through multi-dimensional evaluation indexes, which provides comprehensive analysis for the performance of the algorithm model service and accurately grasps the performance of the algorithm model service. According to the monitoring and evaluation results, the iterative update of the algorithm model service is realized through the visual page, the iteration speed of the model is accelerated, and the performance and adaptability of the model are further enhanced. The method provided by the application realizes the management of the whole life cycle of the model, effectively controls the risk of the model in the production environment, improves the comprehensive management level of the model, and ensures the sustainability and convenience of the model application.

[0015] In a second aspect, the present application provides a space engineering data algorithm model full life cycle management system, applied to a first device, the first device being in communication with a server, the system comprising: a model management module, configured to store a plurality of algorithm model files of different categories in a model library; a service configuration module, configured to generate an inference flowchart corresponding to a target data analysis process in response to editing operations and configuration operations input by a user on a preset interface; the inference flowchart comprising a plurality of nodes connected in sequence according to the target data analysis process, the nodes being model nodes or logic nodes; the editing operations comprising adding node operations, deleting node operations and / or connecting node operations; the configuration operations comprising determining an algorithm model file corresponding to each model node from the model library and / or configuring an execution code corresponding to each logic node, the model node being configured to process input data according to the configured algorithm model file, and the logic node being configured to process input data according to the configured execution code; and a service publishing module, configured to generate an algorithm model service file corresponding to the target data analysis process according to the inference flowchart and send the algorithm model service file to the server in response to a publishing operation input by the user on the preset interface, so that the server deploys an algorithm model service according to the algorithm model service file, the algorithm model service being configured to process data to be analyzed according to the target data analysis process in response to a data analysis request sent by a second device, the data analysis request carrying the data to be analyzed, and send a data analysis result to the second device.

[0016] In a third aspect, an electronic device is provided, comprising a memory and one or more processors; the memory is coupled to the processors; wherein the memory stores computer program codes, the computer program codes comprising computer instructions, when the computer instructions are executed by the processors, causing the electronic device to perform the method in any implementation manner of the first aspect.

[0017] In a fourth aspect, a computer readable storage medium is provided, comprising computer instructions, when the computer instructions are run on an electronic device, causing the electronic device to perform the method in any implementation manner of the first aspect.

[0018] In a fifth aspect, a computer program product is provided, when the computer program product is run on a computer, causing the computer to perform the method in any implementation manner of the first aspect.

[0019] It can be understood that the system of the second aspect, the electronic device of the third aspect, the computer readable storage medium of the fourth aspect, and the computer program product of the fifth aspect provided above can achieve the beneficial effects as described in the first aspect and any possible design manner thereof, which will not be described here. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1A structural schematic diagram of a system architecture provided for an embodiment of the present application is shown in FIG. 1.

[0021] Figure 2 A structural schematic diagram of an electronic device provided for an embodiment of the present application is shown in FIG. 2.

[0022] Figure 3 A flowchart of a space engineering data algorithm model full life cycle management method provided for an embodiment of the present application is shown in FIG. 3.

[0023] Figure 4 A schematic diagram of a preset interface provided for an embodiment of the present application is shown in FIG. 4.

[0024] Figure 5 A schematic diagram of a reasoning flowchart provided for an embodiment of the present application is shown in FIG. 5.

[0025] Figure 6 A flowchart of a space engineering data algorithm model full life cycle management method provided for an embodiment of the present application is shown in FIG. 6.

[0026] Figure 7 A flowchart of a space engineering data algorithm model full life cycle management method provided for an embodiment of the present application is shown in FIG. 7.

[0027] Figure 8 A schematic diagram of a node parameter configuration interface provided for an embodiment of the present application is shown in FIG. 8.

[0028] Figure 9 A flowchart of a space engineering data algorithm model full life cycle management method provided for an embodiment of the present application is shown in FIG. 9.

[0029] Figure 10 A structural schematic diagram of a management system provided for an embodiment of the present application is shown in FIG. 10. DETAILED DESCRIPTION

[0030] The technical solutions in the embodiments of the present application will be described below with reference to the accompanying drawings. In the description of the present application, unless otherwise specified, " / " represents an "or" relationship between the objects before and after the " / " symbol, for example, A / B can represent A or B; in the present application, "or" is only a description of the relationship between the associated objects, and can represent three relationships, for example, A or B, which can represent three cases: A alone, A and B together, and B alone, where A and B can be singular or plural. In addition, in the description of the present application, unless otherwise specified, "multiple" means two or more than two. "At least one (one)" or the like means any combination of these items, including any combination of single (one) or multiple items.

[0031] In addition, in order to clearly describe the technical solutions of the embodiments of the present application, in the embodiments of the present application, the terms "first", "second", etc. are used to distinguish the same or similar items with basically the same function and role. Those skilled in the art can understand that the terms "first", "second", etc. do not limit the quantity and execution order, and the terms "first", "second", etc. also do not necessarily mean different.

[0032] Meanwhile, in the embodiments of the present application, the words "exemplary" or "for example" are used to represent as an example, illustration or description. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as more excellent or more advantageous than other embodiments or design schemes. Rather, the words "exemplary" or "for example" are used to present the relevant concept in a specific manner, for the convenience of understanding.

[0033] In the field of manned space engineering, with the long-period on-orbit operation of space stations and the implementation of subsequent tasks, a large amount of engineering data is generated, and based on these data, a large number of algorithm model files for data analysis and processing are obtained. When technical personnel need to analyze and process data by using algorithm model files, due to the complexity of the management, deployment, monitoring and evaluation processes of algorithm model files, the model management platform in the related art has the following problems: heterogeneous model management is dispersed, and there is a lack of unified coordination mechanism; the model deployment process is complex and the time period is long; the model monitoring mechanism is imperfect and risk control is difficult. Therefore, technical personnel cannot quickly deploy corresponding algorithm model services through algorithm model files, data analysis and processing consumes a large amount of time, and there are problems of low data processing efficiency and poor model application effect.

[0034] Therefore, there is an urgent need for a space engineering data algorithm model full life cycle management method and system, which can quickly complete the deployment and update of the corresponding algorithm model service based on the algorithm model file, thereby reducing the cost required for model service construction, improving the efficiency of model application, and enhancing the data analysis and processing capability.

[0035] In view of this, the embodiment of the present application provides a space engineering data algorithm model full life cycle management method and system, which is applied to a first device, the first device communicates with a server, and the method comprises the following steps: storing a plurality of algorithm model files of different categories in a model library; in response to an editing operation and a configuration operation input by a user on a preset interface, generating an inference flowchart corresponding to a target data analysis process; the inference flowchart comprises a plurality of nodes connected in sequence according to the target data analysis process, and the nodes are model nodes or logic nodes; the editing operation comprises a new node operation, a delete node operation and / or a connect node operation; the configuration operation comprises determining an algorithm model file corresponding to each model node from the model library and / or configuring an execution code corresponding to each logic node, the model node is used for processing input data according to the configured algorithm model file, and the logic node is used for processing input data according to the configured execution code; in response to a publishing operation input by the user on the preset interface, generating an algorithm model service file corresponding to the target data analysis process according to the inference flowchart and sending the algorithm model service file to the server, so that the server deploys an algorithm model service according to the algorithm model service file, the algorithm model service is used for responding to a data analysis request sent by a second device, the data analysis request carries to-be-analyzed data, processing the to-be-analyzed data according to the target data analysis process to obtain a data analysis result, and sending the data analysis result to the second device.

[0036] The method provided by the application can complete the deployment of the corresponding algorithm model service based on the analysis configuration operation and the algorithm model file set by the user, and then complete the analysis and processing of data through the algorithm model service, reduce the cost required for model service construction, improve the efficiency of model application, and enhance the data analysis and processing capability. It can also be understood that: by uniformly managing a plurality of different types of heterogeneous model files, centralized management and maintenance of algorithm model files are realized, and the management efficiency is improved. In response to the operation configuration of the inference logic diagram corresponding to the target data analysis process input by the user, and the algorithm model service corresponding to the target data analysis process can be quickly sent, which can effectively simplify the deployment process of the service and shorten the cycle from development to deployment of the algorithm model service. Moreover, the method provided by the application can monitor the running state of the algorithm model service in real time, guarantee the stability and reliability of the model in the running process. And the analysis result of the algorithm model service is evaluated through multi-dimensional evaluation indexes, which provides comprehensive analysis for the performance of the algorithm model service and accurately grasps the performance of the algorithm model service. According to the monitoring and evaluation results, the iterative update of the algorithm model service is realized through the visual page, the iteration speed of the model is accelerated, and the performance and adaptability of the model are further enhanced. The method provided by the application controls the risk of the model in the production environment through the management of the whole life cycle of the model, improves the comprehensive management level of the model, and ensures the sustainability and convenience of the model application.

[0037] Referring to Figure 1 , Figure 1A system architecture diagram shown in an embodiment of the present application includes a first device 100, a second device 200 and a server 300. The first device 100 communicates with the server 300, and the second device 200 communicates with the server 300. The first device 100 is configured to: store a plurality of algorithm model files of different categories in a model library; in response to an editing operation and a configuration operation input by a user on a preset interface, generate an inference flowchart corresponding to a target data analysis process; the inference flowchart includes a plurality of nodes connected in sequence according to the target data analysis process, and the nodes are model nodes or logic nodes; the editing operation includes an adding node operation, a deleting node operation and / or a connecting node operation; the configuration operation includes determining an algorithm model file corresponding to each model node from the model library and / or configuring an execution code corresponding to each logic node, the model node is configured to process input data according to the configured algorithm model file, and the logic node is configured to process input data according to the configured execution code; in response to a publishing operation input by the user on the preset interface, generate an algorithm model service file corresponding to the target data analysis process according to the inference flowchart and send the algorithm model service file to the server. The second device 200 is configured to: send a data analysis request to the server, and the data analysis request carries to-be-analyzed data. The server 300 is configured to: deploy an algorithm model service according to the algorithm model service file, and in response to the data analysis request sent by the second device 200, process the to-be-analyzed data according to the target data analysis process to obtain a data analysis result, and send the data analysis result to the second device 200. The second device 200 is further configured to receive the data analysis result sent by the server 300.

[0038] In one example, the first device 100, the second device 200 and the server 300 can be any electronic device with data processing capability, such as a general-purpose computer, a personal computer, a notebook computer, a switch or a tablet computer, etc. The specific implementation of the first device 100, the second device 200 and the server 300 is not limited here.

[0039] Figure 2 A structure diagram of an electronic device provided by an embodiment of the present application is shown. The electronic device 10 includes a processor 110, a memory 120 and a communication interface 130.

[0040] The processor 110 can include one or more processing cores. The processor 110 connects various parts inside the electronic device 10 with various interfaces and lines, performs various functions of the electronic device 10 and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 120, and calling data stored in the memory 120. Alternatively, the processor 110 can be implemented in at least one hardware form of a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processing (DSP), a field-programmable gate array (FPGA), a programmable logic array (PLA).

[0041] The memory 120 can include a random access memory (RAM) and can also include a read-only memory (ROM). Alternatively, the memory 120 includes a non-transitory computer-readable storage medium. The memory 120 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 120 can include a program storage area. The program storage area can store instructions for implementing an operating system, instructions for implementing at least one function (such as a file storage function, an interface display function and a model training function, etc.), instructions for implementing each of the above method embodiments, etc.

[0042] The communication interface 130 is used to communicate with other devices, equipment or communication networks, such as data storage devices, image processing equipment or Ethernet, radio access network (RAN), wireless local area network (WLAN) and the like. For example, to implement communication between the first device 100 and the server 300, communication between the second device 200 and the server 300.

[0043] In physical implementation, each of the above devices (such as the processor 110, the memory 120 and the communication interface 130) can be a device in the same device (such as a notebook computer). Alternatively, at least two of them can be arranged in the same device as different devices in the device, such as the deployment of devices or components in a distributed system.

[0044] It can be understood that the structure illustrated in the embodiment does not constitute a specific limitation on the electronic device 10. In other embodiments of the application, the electronic device 10 can include more or fewer components than illustrated, or combine certain components, or split certain components, or different component arrangements. The illustrated components can be implemented in hardware, software, or a combination of software and hardware.

[0045] An aerospace engineering data algorithm model full life cycle management method provided by the embodiment of the application is described below in combination with the accompanying drawings of the specification.

[0046] Figure 3 A flowchart of the aerospace engineering data algorithm model full life cycle management method provided by the embodiment of the application. Optionally, the method can be executed by the first device 100 as shown in Figure 1 , that is, by the electronic device 10 as shown in Figure 2 . The method can include the following steps:

[0047] S1, store a plurality of algorithm model files of different categories in a model library.

[0048] Specifically, the algorithm model files of different categories are respectively used to implement different data analysis functions and obtain different data analysis results. It should be understood that the algorithm model files of different categories are obtained based on different training frameworks, wherein the training framework is a software library and tool for building, training, and deploying machine learning and deep learning models, and the training framework includes TensorFlow, PyTorch, MLflow, Scikit-learn, and Spark-MLlib.

[0049] In this way, the method provided by the embodiment of the application can realize unified management of models of different training frameworks, different categories, and different functions, can solve the compatibility and expansibility between models of different training frameworks, different categories, and different functions, and facilitates users to flexibly select models of different training frameworks, different categories, and different functions for use, thereby effectively improving the deployment efficiency of algorithm model services of the user.

[0050] S2, in response to an editing operation and a configuration operation input by a user on a preset interface, generate an inference flowchart corresponding to a target data analysis process, the inference flowchart including a plurality of nodes connected in sequence according to the target data analysis process, the nodes being model nodes or logic nodes.

[0051] Specifically, the editing operation includes a new node operation, a delete node operation and / or a connection node operation; the configuration operation includes determining an algorithm model file corresponding to each model node from a model library and / or configuring an execution code corresponding to each logic node, the model node is used to process input data according to the configured algorithm model file, and the logic node is used to process input data according to the configured execution code.

[0052] The inference flowchart can be understood as a flowchart corresponding to the target data analysis process, and each step in the target data analysis process corresponds to a model node or a logic node, for example, the target data analysis process a is: inputting the data to be analyzed into the trained algorithm model A, then inputting the result output by the algorithm model A into the algorithm model B, and determining the result output by the algorithm model B as the final data analysis result, then the inference flowchart corresponding to the target data analysis process a includes two model nodes, which are model node A and model node B, wherein the input end of the model node A is connected with the input port, used to receive the data to be analyzed and process the data to be analyzed, the output end of the model node A is connected with the input end of the model node B, used to input the output result into the model node B, and then the output end of the model node B is connected with the output port, used to output the final data analysis result. The model node A is configured with an algorithm model file corresponding to the algorithm model A, and the model node B is configured with an algorithm model file corresponding to the algorithm model B.

[0053] In a possible implementation, the logic node includes a data processing node, a routing node and a combiner node; the data processing node includes at least one input end and at least one output end, and the data processing node is used to process data input by the input end according to the configured execution code and output the processing result through the output end; the combiner node includes at least one input end and multiple output ends, and the combiner node is used to output the data input by the input end through the multiple output ends according to the configured execution code; the routing node includes at least one input end and multiple output ends, and the routing node is used to determine a target output end through the configured execution code according to the data input by the input end and output the input data through the target output end, the target output end being one of the multiple output ends.

[0054] For example, Figure 4 , Figure 4A schematic diagram of a preset interface provided by an embodiment of the present application. The preset interface includes an operation area and a canvas area. The operation area is used to display model nodes (which can also be understood as displaying models stored in a model library) corresponding to algorithm model files of different categories and logic nodes corresponding to different execution codes. Specifically, the model nodes displayed by the operation area include Scikit-learn models, PyTorch models, TensorFlow models, XGBoost models, MLflow models, PMML models, Triton models, and LightGBM models. The logic nodes displayed by the operation area include data processing nodes, combiner nodes, and routing nodes. The canvas area is used to display an inference flowchart corresponding to a target data analysis process obtained in response to an editing operation and a configuration operation of a user.

[0055] Further, Figure 5 , Figure 5 A schematic diagram of an inference flowchart provided by an embodiment of the present application. The inference flowchart includes two data processing nodes, two combiner nodes, one routing node, and three model nodes. The two data processing nodes are data processing node 1 and data processing node 2, respectively. The two combiner nodes are combiner node 1 and combiner node 2, respectively. The three model nodes are model node 1, model node 2, and model node 3, respectively. The input end of data processing node 1 is used to receive to-be-analyzed data. The output end of data processing node 1 is connected to the input end of combiner node 1. Two output ends of combiner node 1 are connected to the input ends of model node 1 and the routing node, respectively. The output end of the routing node is connected to the input ends of model node 2 and model node 3, respectively. The output ends of model node 1, model node 2, and model node 3 are used to output data analysis results of the to-be-analyzed data. It should be noted that, Figure 5 The target data analysis process corresponding to the inference flowchart shown in the figure is as follows. The to-be-analyzed data is input into data processing node 1 through an input port. Data processing node 1 processes the to-be-analyzed data and outputs the processed data to combiner node 1. Combiner node 1 outputs the processed data of data processing node 1 to model node 1 and the routing node, respectively. Model node 1 processes the processed data of data processing node 1 and outputs the corresponding processing result to combiner node 2. The routing node determines, according to a preset condition, whether to output the processed data of data processing node 1 to model node 2 or model node 3. Then, model node 2 or model node 3 processes the processed data of data processing node 1 and outputs the corresponding processing result to combiner node 2. Combiner node 2 outputs the processing results output by the output ends of model node 1, model node 2, and model node 3 to data processing node 2. Data processing node 2 processes the processing results output by the output ends of model node 1, model node 2, and model node 3 and outputs the final data analysis result through an output port.

[0056] It should be understood that the above target data analysis process and the corresponding inference flowchart are only illustrative, and the target data analysis process and the corresponding inference flowchart provided by the embodiments of the present application can include more or fewer steps and nodes than the above examples, and the embodiments of the present application do not particularly limit this.

[0057] The specific implementation of the inference flowchart provided by the embodiments of the present application is described below through an example. First, the user adds model nodes, data processing nodes, combiner nodes and routing nodes in the canvas area through the new node operation in the preset interface, and then connects the model nodes, data processing nodes, combiner nodes and routing nodes in sequence according to the target data analysis process through the connection node operation. Among them, the connection of the nodes needs to meet that the data processing nodes are at the top, and at least two output ends included in the combiner nodes and the routing nodes are respectively connected to different model nodes or data processing nodes. Then, in response to the configuration operation input by the user, a corresponding algorithm model file is configured for each model node. A corresponding execution code is configured for each logical node to obtain the inference flowchart corresponding to the target data analysis process.

[0058] As can be seen from the above embodiments, the method provided by the present application can effectively reduce the technical threshold of the user to configure a complex inference process through a graphical preset interface. Through the graphical drag-and-drop method, different logical branches corresponding to different data analysis processes can be conveniently configured, and the result is what you see. Compared with the method of calling multiple interfaces to configure the data analysis process in the traditional way, the method provided by the embodiments of the present application can effectively simplify the operation process of the user and improve the user experience.

[0059] S3, in response to the publishing operation input by the user in the preset interface, generating and sending an algorithm model service file corresponding to the target data analysis process to the server according to the inference flowchart, so that the server deploys the algorithm model service according to the algorithm model service file.

[0060] Specifically, the algorithm model service is used to respond to a data analysis request sent by a second device, the data analysis request carrying to-be-analyzed data, processing the to-be-analyzed data according to the target data analysis process to obtain a data analysis result, and sending the data analysis result to the second device.

[0061] Among them, the second device can display the data analysis result in the visual interface, and then carry out subsequent business applications according to the data analysis result.

[0062] It can also be understood that the algorithm model service is a program running on the server. After deploying the algorithm model service on the server, the server receives the data analysis request sent by the second device, processes the to-be-analyzed data according to the target data analysis process through the algorithm model service, obtains the data analysis result, and sends the data analysis result to the second device.

[0063] It should be noted that the to-be-analyzed data can also be obtained from a database storing the to-be-analyzed data in response to the data analysis request sent by the second device. The specific obtaining method of the to-be-analyzed data is not particularly limited in the embodiment of the application.

[0064] The method provided in the embodiment of the application can respond to the publishing operation of the user, realize the automatic deployment of the model algorithm service, effectively improve the deployment efficiency, and the model algorithm service deployment in the related art needs multiple steps such as packaging the model file, installing the dependent package, and configuring the server environment, which consumes a large amount of time and has a complex process. The user is prone to make errors in the operation process, resulting in failure of the model algorithm service deployment. The method provided in the embodiment of the application can effectively improve the operation and maintenance efficiency of the model algorithm service.

[0065] In one embodiment, the publishing operation input by the user on the preset interface also carries service information and resource configuration information input by the user. The service information is used to describe the function of the algorithm model service, so that the user of the second device selects the algorithm model service corresponding to the data analysis process according to the service information. The resource configuration information is used to describe the resources configured by the algorithm model service, so that the server deploys the algorithm model service based on the resources configured by the resource configuration information.

[0066] In a possible implementation, the algorithm model service is also used to respond to the data analysis request sent by the second device, the data analysis request carries the to-be-analyzed data, the to-be-analyzed data is added to the task queue arranged in sequence, the multiple to-be-analyzed data included in the task queue is processed in sequence according to the target data analysis process, the data analysis result corresponding to each to-be-analyzed data is obtained, and the data analysis result corresponding to each to-be-analyzed data is sent to the second device.

[0067] Specifically, the server adds the to-be-analyzed data to the task queue arranged in sequence, and processes the multiple to-be-analyzed data included in the task queue in sequence according to the target data analysis process. That is, the server asynchronously calls the algorithm model service for the data analysis request sent by the second device, so as to avoid that the to-be-analyzed data exceeding the upper limit of the resources configured by the algorithm model service cannot be processed due to the limited resources, and the user experience is affected.

[0068] In a possible implementation, the algorithm model service provided by the embodiment of the present application supports a model explanation function, the server determines an importance score of each feature parameter included in the to-be-analyzed data on a data analysis result, and displays the importance score of each feature parameter included in the to-be-analyzed data on the data analysis result in a visual chart manner on the second device. The model explanation function is applicable to an SVM / SVC algorithm, a KNN algorithm, a logistic regression algorithm, a MultinomialNB algorithm, and an XGBoost algorithm.

[0069] As can be seen from the above, the method provided by the embodiment of the present application can effectively enhance the explainability of the data analysis result of the algorithm model service by using the model explanation algorithm. The user of the second device can view the influence degree of each feature parameter on the final data analysis result, meet the user experience in different use scenarios, and thus improve the user experience.

[0070] Optionally, the algorithm model service provided by the embodiment of the present application is configured with a dynamic resource configuration strategy, which can dynamically adjust the number of tasks (an example) processed according to the utilization rate of the central processor or the memory. Thus, the number of instances is adjusted according to the utilization rate. The central processor exclusive and sharing strategy is supported.

[0071] In this way, the method provided by the embodiment of the present application can dynamically adjust the number of tasks executed by the algorithm model service according to the utilization rate of the central processor or the memory by using the dynamic resource configuration strategy, which can effectively save computing resources and improve data processing efficiency.

[0072] As can be seen from S1-S3, the method provided by the embodiment of the present application stores a plurality of algorithm model files of different categories in the model library; generates an inference flowchart corresponding to the target data analysis process in response to an editing operation and a configuration operation input by the user on the preset interface; and generates and sends an algorithm model service file corresponding to the target data analysis process to the server in response to a publishing operation input by the user on the preset interface, so that the server deploys the algorithm model service according to the algorithm model service file. The method provided by the present application can quickly complete the deployment of the corresponding algorithm model service based on the analysis configuration operation and the algorithm model file set by the user, and then complete the analysis and processing of data through the algorithm model service, reduce the cost required for model service construction, improve the efficiency of model application, and enhance the data analysis and processing capability.

[0073] In some embodiments, after the server deploys the algorithm model service according to the algorithm model service file, the method provided by the embodiment of the present application further includes:

[0074] The running state monitoring instruction is used to instruct the server to obtain a running state parameter corresponding to the algorithm model service, and the running state parameter includes memory utilization, processor utilization, and response time.

[0075] As can be seen from the above, the method provided by the embodiments of the present application can monitor the running state of the algorithm model service in real time, and when the running state is abnormal, the user of the first device is reminded in time to maintain the deployed algorithm model service, thereby ensuring the stability and reliability of the algorithm model service in the running process.

[0076] In some embodiments, after the server deploys the algorithm model service according to the algorithm model service file, the method provided by the embodiments of the present application further includes:

[0077] Obtaining verification data, the verification data including a plurality of verification samples and a first analysis result of each verification sample obtained by processing the target data analysis process;

[0078] Sending an analysis result evaluation instruction to the server, the analysis result evaluation instruction carrying the verification data, the analysis result evaluation instruction being used to instruct the server to process the plurality of verification samples according to the target data analysis process based on the algorithm model service, to obtain a second analysis result corresponding to each verification sample, and the server to determine an evaluation index corresponding to the algorithm model service according to the first analysis result and the second analysis result of each verification sample, the server to generate a second alarm information in the case that any evaluation index is greater than or equal to a preset threshold, and the server to send the second alarm information to the first device, the second alarm information being used to prompt that the analysis result of the algorithm model service is abnormal.

[0079] In the case that the category of the algorithm model service is a binary classification model, the evaluation index includes accuracy, precision, recall, specificity, area under ROC curve, K-S test value, and F1 measure; in the case that the category of the algorithm model service is a multi-classification model, the evaluation index includes accuracy, weighted precision, weighted recall, and weighted F value; in the case that the category of the algorithm model service is a regression model, the evaluation index includes mean absolute error, mean square error, root mean square error, and determination coefficient; and in the case that the category of the algorithm model service is a clustering model, the evaluation index includes profile coefficient-cosine similarity, profile coefficient-Euclidean distance, residual sum of squares, and covariance.

[0080] Specifically, the ROC curve is a receiver operating characteristic curve. The K-S test value is a Kolmogorov-Smirnov test value.

[0081] It should be understood that the categories of the above algorithm model services and the corresponding evaluation indexes are only exemplary and are not particularly limited by the embodiments of the present application.

[0082] In some embodiments, the first device receives the first analysis result, the second analysis result, the second alarm information corresponding to each verification sample and the evaluation index corresponding to the algorithm model service sent by the server, and displays the first analysis result, the second analysis result, the second alarm information corresponding to each verification sample and the evaluation index corresponding to the algorithm model service on the visual interface.

[0083] In this way, the method provided by the embodiments of the present application evaluates the data analysis result of the algorithm model service through multi-dimensional evaluation indexes, provides comprehensive analysis of the performance of the algorithm model service, and accurately grasps the performance of the algorithm model service.

[0084] In some embodiments, referring to Figure 6 , the method provided by the embodiments of the present application further includes:

[0085] S610, in response to the iterative update instruction input by the user, iteratively training the algorithm model file corresponding to one or more model nodes included in the algorithm model service according to the training code, to obtain a trained algorithm model file.

[0086] The training code is an execution code pre-configured for iteratively training the algorithm model file corresponding to one or more model nodes included in the algorithm model service according to the training sample. The iterative update instruction carries a preset threshold of each evaluation index corresponding to the algorithm model service in the plurality of evaluation indexes, and each evaluation index of the algorithm model service configured with the trained algorithm model file is greater than or equal to the preset threshold.

[0087] S620, updating the algorithm model service file according to the trained algorithm model file to obtain an updated algorithm model service file.

[0088] S630, sending the updated algorithm model service file to the server, so that the server deploys the algorithm model service according to the updated algorithm model service file.

[0089] From the above, the method provided by the embodiment of the application can realize real-time updating of the algorithm model service in the case that the algorithm model service does not meet the actual use requirements, guarantee the accuracy and stability of the algorithm model service, and meet different use requirements of the user in different use scenarios by iteratively updating the instruction.

[0090] The specific implementation of the iterative updating of the algorithm model service provided by the embodiment of the application is explained and described below in combination with an example.

[0091] In an example, referring to Figure 7 The S610 specifically includes the following steps:

[0092] S6101, in response to the iterative updating instruction input by the user, creating a continuous training task in the case that the pre-operation is completed, the continuous training task being used for executing a workflow.

[0093] Specifically, the pre-operation includes that the model training code has been acquired, the model file shelving operation has been completed, the model service deployment operation has been completed, the model evaluation dataset and the model evaluation script have been acquired. The workflow includes a plurality of configurable nodes, and the configurable nodes include a code node, a model shelving node, a model evaluation node, a judgment node and a model deployment node.

[0094] The code node is used for executing the corresponding operation according to the configured execution code file, so as to realize the links of data preprocessing, feature processing, model training and the like. The judgment node is used for judging whether the next process node is continued to be executed according to the self-defined script. The model shelving node is used for executing the model shelving operation, and the algorithm model file generated by the previous node training is stored in the model library as a new version of the algorithm model file. Under the premise of the existing algorithm model service, it is defaulted that the algorithm model file has at least one version stored in the model library. The model evaluation node is used for executing the model evaluation operation on the model node which has been deployed and shelved, and the data required by the evaluation task needs to be configured in the database in advance, including the evaluation dataset and the evaluation script. The model deployment node is used for executing the deployment operation of the algorithm model service, and the updated algorithm model service file is sent to the server and added to the deployed algorithm model service as a new service version. Under the premise of the deployed algorithm model service, it is defaulted that the algorithm model service has at least one online version deployed on the server.

[0095] S6102, configuring the parameters corresponding to each configurable node included in the workflow.

[0096] The parameters of the code node select the instance environment, the code file, and the model saving path used by the model training code. The parameters of the model onboarding node select the model that has been completed model onboarding, and the input parameter source is the output parameter of the previous "code node". The parameters of the model evaluation node select the model evaluation dataset and the model evaluation script obtained in the above steps, select the framework type on which the model training depends, configure the running resources required by the model evaluation task, and configure the input parameter source as the output parameter of the previous "model onboarding" node. The parameters of the judgment node can be configured to customize the content of the judgment logic script, such as "whether the value of the accuracy rate indicator in the output result of the previous model evaluation node is greater than a preset threshold", and the input parameter source is the output parameter of the previous "model evaluation" node.

[0097] For example, referring to Figure 8 , Figure 8 A schematic diagram of a node parameter configuration interface provided by an embodiment of the present application includes a node type display area, a workflow display area, and a parameter setting area. The node type display area is used to display different kinds of configurable nodes, the workflow area is used to show the workflow corresponding to the continuous training task, and the parameter setting area is used to configure the parameters of each configurable node included in the workflow. The parameters include version information, input parameters, output parameters, and scheduling settings, wherein the scheduling settings include manual triggering and real-time scheduling.

[0098] It should be understood that the above parameters are only exemplary, and the configurable nodes provided by the embodiment of the present application can include more or fewer parameters than the above examples. The embodiment of the present application does not particularly limit the specific types and quantities of the parameters of the configurable nodes.

[0099] S6103, performing the continuous training task to iteratively train the algorithm model files corresponding to one or more model nodes included in the algorithm model service according to the training code, and obtaining the trained algorithm model files.

[0100] In a possible implementation, the continuous training task is configured to be executed once or periodically in response to a manual execution instruction or a timing execution instruction input by a user, and to display the task log and the evaluation result corresponding to the continuous training task in response to a result display instruction input by the user.

[0101] In order to better illustrate and understand the principles of the method provided by the present application, the scheme of the present application will be described below in combination with an optional specific embodiment. It should be noted that the specific implementation of each step in the specific embodiment should not be understood as a limitation on the scheme of the present application. On the basis of the principles of the scheme provided by the present application, other implementation manners that can be thought of by those skilled in the art should also be regarded as within the protection scope of the present application.

[0102] In this embodiment, as shown in Figure 9 The continuous iterative optimization scenario of the algorithm model service corresponding to the manned space engineering data prediction algorithm model is a specific embodiment, and in this embodiment, the operation is implemented in the following steps:

[0103] S710, store the algorithm model files corresponding to different algorithm models in the model library.

[0104] Among them, Table 1 is the saving method and file type of the algorithm model file supported in the model library.

[0105] Table 1

[0106]

[0107] S720, obtain incremental data.

[0108] Specifically, a script file for reading incremental data is written, the task output is configured to be the read incremental data file and the number of incremental data, and the automatic acquisition of incremental data is realized.

[0109] The data used in this embodiment is the manned space engineering telemetry data. First, based on the defined start and end time, the number of data partitions contained in the time difference is calculated. Combined with the query time point and the partition, the SQL query filtering condition is constructed, and the incremental data is queried. The result set returned by this SQL query is cached on the NFS in the format of a CSV file. The next time the task runs, the last time of the result set is queried as the start time, and the task running time is used as the end time.

[0110] S730, convert the incremental data into a verification data set corresponding to the algorithm model service.

[0111] By configuring the judgment node input, the output of step S720 is read, it is judged whether the data meets the model input, and TRUE or FALSE is returned. If it is TRUE, the subsequent step is performed, and if it is FALSE, the task is terminated.

[0112] A script file for data preprocessing is written, the data features are extracted, and the task output is configured to be the verification data set after preprocessing.

[0113] S740, input the verification data set into the algorithm model service to obtain the corresponding data analysis result.

[0114] A script file for data prediction is written, a task input is configured as the verification data set output in the previous step, and a task output is configured as a data analysis result. A script file for evaluation index calculation is written, a task input is configured as the verification data set and the data analysis result output in the previous step, and a task output is configured as a model evaluation index. The evaluation indexes of different types of algorithm models are described in the above embodiments and will not be repeated here.

[0115] The verification data set and the data analysis result are saved in two ways, NFS storage and database storage. The NFS storage writes data to a fixed file directory, and the result file can be downloaded after the task is run. The database storage reads the database through a custom function and queries the data by passing in the database name, table name, and column name.

[0116] S750, based on the evaluation index of the algorithm model service, whether to retrain the algorithm model file included in the algorithm model service is judged, and the algorithm model file stored in the model library is iterated.

[0117] By configuring a judgment node, the evaluation index output in the previous step is read, and whether the evaluation index meets the preset condition is judged. If the model index does not meet the preset condition, a verification data set is regenerated, the algorithm model corresponding to the algorithm model file is trained, and TRUE or FALSE is returned. If it is TRUE, the subsequent step is performed, and if it is FALSE, the task is terminated.

[0118] A script file for model training is written, and the task output is configured as the new version of the algorithm model file produced by this model training task. The new version of the algorithm model file is stored in the model library, and the name of the new version of the algorithm model file is set based on the storage time.

[0119] S760, in the case of iterating the algorithm model file stored in the model library according to the trained algorithm model file, updating the deployed algorithm model service based on the iterated algorithm model file.

[0120] It should be understood that the server can deploy multiple different versions of algorithm model services at the same time. When receiving a data analysis request sent by a second device, the algorithm model service determines the called algorithm model file according to the pre-set release mode and outputs the corresponding data analysis result. The description of the release mode is described in the following embodiments.

[0121] Through the method, the problems that various heterogeneous models are difficult to be uniformly managed, model deployment period is long, and a unified standardized process is lacked in the prior art are solved, and the full life cycle management of unified management, unified operation and maintenance, unified application, unified monitoring and unified evaluation of algorithm models is provided, and easy-to-use, efficient, safe and reliable model operation and management are realized.

[0122] In some embodiments, the method provided by the embodiments of the present application further includes:

[0123] The data distribution difference monitoring instruction is sent to the server, the training data for each algorithm model file included in the algorithm model service is carried in the data distribution difference monitoring instruction, and the data distribution difference monitoring instruction is used to instruct the server to determine a data distribution difference value between the training data and the to-be-analyzed data carried in the data analysis request sent by the second device in a case where the data analysis request is received, the server to generate third alarm information in a case where the data distribution difference value is greater than or equal to a third threshold value, and the server to send the third alarm information to the second device, the third alarm information being used to prompt that there is a data distribution difference between the training data and the to-be-analyzed data.

[0124] The embodiments of the present application can adaptively adjust the training data by monitoring the data distribution difference value between the to-be-analyzed data and the training data, retrain the algorithm model service that meets the actual application scenario, avoid obtaining a data analysis result with poor accuracy, and thus improve the stability of the algorithm model service.

[0125] In some embodiments, in a case where the server is deployed with a plurality of algorithm model services corresponding to the target data analysis process, the method provided by the embodiments of the present application further includes:

[0126] The release mode setting instruction is sent to the server, the release mode setting instruction being used to instruct the server to set a release mode of the plurality of algorithm model services corresponding to the target data analysis process, and the release mode including independent release, gray release or comparison release;

[0127] In a case where the release mode is independent release, the identification of the target algorithm model service is carried in the release mode setting instruction, the target algorithm model service being one of the plurality of algorithm model services, the server responding to the data analysis request sent by the second device, the data analysis request carrying the to-be-analyzed data, the server processing the to-be-analyzed data through the target algorithm model service to obtain a data analysis result, and the server sending the data analysis result to the second device;

[0128] In the case that the publishing mode is gray publishing, the server responds to a data analysis request sent by the second device, the data analysis request carrying to-be-analyzed data, the server processes the to-be-analyzed data through the multiple algorithm model services to obtain a data analysis result corresponding to each algorithm model service, and the server sends the data analysis result corresponding to each algorithm model service to the second device.

[0129] In the case that the publishing mode is contrast publishing, the server responds to a data analysis request sent by the second device, the data analysis request carrying to-be-analyzed data, the server processes the to-be-analyzed data through the multiple algorithm model services to obtain a data analysis result corresponding to each algorithm model service, the server determines an evaluation index corresponding to each algorithm model service, and the server sends the data analysis result of the algorithm model service corresponding to the maximum value in the evaluation index corresponding to each algorithm model service to the second device.

[0130] Specifically, the evaluation index is positively correlated with the performance of the algorithm model service.

[0131] In some embodiments, in the case that the evaluation index is negatively correlated with the performance of the algorithm model service, in the case that the publishing mode is contrast publishing, the server sends the data analysis result of the algorithm model service corresponding to the minimum value in the evaluation index corresponding to each algorithm model service to the second device.

[0132] It should be understood that the publishing mode of the algorithm model service is sent to the server by an algorithm model service file corresponding to the algorithm model service before the server deploys the algorithm model service, and the specific setting manner of the publishing mode of the algorithm model service is not particularly limited in the embodiments of the present application.

[0133] As can be seen from the above, the method provided in the embodiments of the present application can flexibly set different publishing modes of the algorithm model service, in the case that the server deploys multiple algorithm model services (which can also be understood as different versions of the same algorithm model service) corresponding to a target data analysis process, different versions of the same algorithm model service can be used to analyze to-be-analyzed data based on different publishing modes, to obtain a data analysis result, thereby meeting the use requirements of users in different use scenarios.

[0134] It should be noted that the first device 100 comprises at least one of the hardware structure and the software module corresponding to each function in order to realize the above functions. Those skilled in the art should easily realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in the present application can be realized in the form of hardware or the combination of hardware and computer software. Whether a certain function is realized in the form of hardware or computer software driven hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the embodiments of the present application.

[0135] The embodiments of the present application can divide the first device 100 into functional units according to the above method examples. For example, the first device 100 can be divided into functional units corresponding to each function, or two or more functions can be integrated into one processing unit. The integrated unit can be realized in the form of hardware or software functional unit. It should be noted that the division of units in the embodiments of the present application is illustrative and is only a logical functional division. There can be another division method in actual implementation.

[0136] For example, Figure 10A structural schematic diagram of a space engineering data algorithm model full life cycle management system provided by an embodiment of the present application is shown. The management system 400 is applied to a first device, the first device communicates with a server, and comprises: a model management module 410, configured to store a plurality of algorithm model files of different categories in a model library; a service configuration module 420, configured to generate an inference flowchart corresponding to a target data analysis process in response to an editing operation and a configuration operation input by a user on a preset interface; the inference flowchart comprises a plurality of nodes connected in sequence according to the target data analysis process, and the nodes are model nodes or logic nodes; the editing operation comprises a node adding operation, a node deleting operation and / or a node connecting operation; the configuration operation comprises determining an algorithm model file corresponding to each model node from the model library and / or configuring an execution code corresponding to each logic node, the model node is configured to process input data according to the configured algorithm model file, and the logic node is configured to process input data according to the configured execution code; and a service publishing module 430, configured to generate an algorithm model service file corresponding to the target data analysis process according to the inference flowchart and send the algorithm model service file to the server in response to a publishing operation input by the user on the preset interface, so that the server deploys an algorithm model service according to the algorithm model service file, the algorithm model service is configured to process data to be analyzed according to the target data analysis process in response to a data analysis request sent by a second device, the data analysis request carries the data to be analyzed, and the data analysis result is sent to the second device.

[0137] Optionally, the management system 400 further comprises a running state monitoring module 440, the running state monitoring module 440 is configured to: send a running state monitoring instruction to the server, the running state monitoring instruction is configured to instruct the server to acquire a running state parameter corresponding to the algorithm model service, the running state parameter comprises a memory utilization rate, a processor utilization rate and a response time; the server generates a first alarm information in a case where any running state parameter is greater than or equal to a first threshold value and a duration is greater than or equal to a second threshold value, the first alarm information is configured to prompt an abnormal running state of the algorithm model service; and the server sends the first alarm information to the first device at a preset time interval.

[0138] Optionally, the management system 400 further comprises an analysis result evaluation module 450, configured to: acquire verification data, the verification data comprising a plurality of verification samples and a first analysis result of each verification sample based on the target data analysis process; send an analysis result evaluation instruction to the server, the analysis result evaluation instruction carrying the verification data, the analysis result evaluation instruction being used to instruct the server to process the plurality of verification samples according to the target data analysis process based on the algorithm model service, to obtain a second analysis result corresponding to each verification sample; and determine an evaluation index corresponding to the algorithm model service according to the first analysis result and the second analysis result of each verification sample, and generate a second alarm information in a case where any evaluation index is greater than or equal to a preset threshold, and send the second alarm information to the first device, the second alarm information being used to prompt an abnormal analysis result of the algorithm model service; wherein, in a case where the category of the algorithm model service is a binary classification model, the evaluation index comprises: accuracy, precision, recall, specificity, area under ROC curve, K-S test value and F1 measure; in a case where the category of the algorithm model service is a multi-classification model, the evaluation index comprises: accuracy, weighted precision, weighted recall and weighted F value; in a case where the category of the algorithm model service is a regression model, the evaluation index comprises: mean absolute error, mean square error, root mean square error and determination coefficient; and in a case where the category of the algorithm model service is a clustering model, the evaluation index comprises: silhouette coefficient-cosine similarity, silhouette coefficient-Euclidean distance, residual sum of squares and covariance.

[0139] Optionally, the management system 400 further comprises a service iterative updating module 460, configured to: in response to an iterative updating instruction input by a user, perform iterative training on an algorithm model file corresponding to one or more model nodes included in the algorithm model service according to the training code, to obtain a trained algorithm model file; update the algorithm model service file according to the trained algorithm model file, to obtain an updated algorithm model service file; and send the updated algorithm model service file to the server, so that the server deploys the algorithm model service according to the updated algorithm model service file.

[0140] It should be understood that the specific description of the above optional mode can refer to the foregoing method embodiments, and will not be described here. In addition, the description of the explanation and beneficial effects of any one of the above management systems 400 can refer to the corresponding method embodiments described above, and will not be described here.

[0141] The embodiment of the present application further provides a computer readable storage medium, which stores at least one computer instruction, and the at least one computer instruction is loaded and executed by a processor to realize the method of each of the above embodiments. The description of the explanation and beneficial effects of the related content in any of the above computer readable storage media can refer to the corresponding embodiment, and will not be repeated here.

[0142] The embodiment of the present application further provides a chip. The chip integrates a control circuit and one or more ports for realizing the function of the above management system 400. Optionally, the functions supported by the chip can refer to the above, and will not be repeated here.

[0143] Those skilled in the art can understand that all or part of the steps of the above embodiments can be completed by a program instructing related hardware, and the program can be stored in a computer readable storage medium. The storage medium mentioned above can be a read-only memory, a random access memory, etc. The processing unit or the processor mentioned above can be a central processing unit, a general processor, an application specific integrated circuit (ASIC), a microprocessor (digital signal processor, DSP), a field programmable gate array (FPGA) or other programmable logic device, a transistor logic device, a hardware component or any combination thereof.

[0144] The embodiments of the present application also provide a computer program product containing instructions, which, when executed on a computer, cause the computer to perform any of the methods described above. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on the computer, the processes or functions according to the embodiments of the present application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable apparatus. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center through a wired (for example, coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (for example, infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server, data center, etc. containing one or more available media sets. The available media can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, an SSD), etc.

[0145] It should be noted that the above-mentioned devices for storing computer instructions or computer programs provided by the embodiments of the present application, such as but not limited to the above-mentioned memory, computer-readable storage medium, and communication chip, etc., are all non-transitory. Those skilled in the art should be aware that in one or more of the above examples, the functions described by the embodiments of the present application can be implemented by hardware, software, firmware, or any combination thereof. When implemented by software, these functions can be stored in a computer-readable storage medium or transmitted as one or more instructions or codes on a computer-readable storage medium. The computer-readable storage medium includes a computer storage medium and a communication medium, wherein the communication medium includes any medium that facilitates the transfer of a computer program from one place to another. The storage medium can be any available medium accessible by a general-purpose or special-purpose computer.

[0146] Although the embodiments of the present application have been shown and described above, it should be understood that the above-mentioned embodiments are exemplary and should not be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements, and variations to the above-mentioned embodiments within the scope of the present application.

Claims

1. A method for full lifecycle management of aerospace engineering data algorithm models, characterized in that, Applied to a first device, the first device communicating with a server, the method includes: Multiple different types of algorithm model files are stored in the model library; In response to user input of editing and configuration operations on a preset interface, a reasoning flowchart corresponding to the target data analysis process is generated. The reasoning flowchart includes multiple configured nodes connected sequentially according to the target data analysis process, where each node is a model node or a logic node. The editing operations include adding nodes, deleting nodes, and / or connecting nodes. The configuration operations include determining the algorithm model file corresponding to each model node from the model library and / or configuring the execution code corresponding to each logic node. The model nodes are used to process the input data according to the configured algorithm model file, and the logic nodes are used to process the input data according to the configured execution code. In response to a user's publish operation input on a preset interface, an algorithm model service file corresponding to the target data analysis process is generated and sent to the server according to the inference flowchart, so that the server deploys the algorithm model service according to the algorithm model service file. The algorithm model service is used to respond to a data analysis request sent by a second device, the data analysis request carrying data to be analyzed, processing the data to be analyzed according to the target data analysis process to obtain data analysis results, and sending the data analysis results to the second device. The logical nodes include data processing nodes, routing nodes, and combiner nodes. Each data processing node includes at least one input terminal and at least one output terminal. The data processing node processes the data input from the input terminal according to configured execution code and outputs the processing result through the output terminal. Each combiner node includes at least one input terminal and multiple output terminals. The combiner node outputs the data input from the input terminal through the multiple output terminals according to configured execution code. Each routing node includes at least one input terminal and multiple output terminals. The routing node determines the target output terminal based on the data input from the input terminal using configured execution code and outputs the input data through the target output terminal, where the target output terminal is one of the multiple output terminals.

2. The method according to claim 1, characterized in that, After the server deploys the algorithm model service according to the algorithm model service file, the method further includes: A running status monitoring instruction is sent to the server, the running status monitoring instruction being used to instruct: the server to obtain the running status parameters corresponding to the algorithm model service, the running status parameters including memory utilization, processor utilization, and response time; if any running status parameter is greater than or equal to a first threshold and the duration is greater than or equal to a second threshold, the server generates a first alarm message, the first alarm message being used to indicate that the running status of the algorithm model service is abnormal; the server sends the first alarm message to the first device at a preset time interval.

3. The method according to claim 2, characterized in that, After the server deploys the algorithm model service according to the algorithm model service file, the method further includes: Acquire verification data, which includes multiple verification samples and a first analysis result obtained by processing each verification sample based on the target data analysis process; An analysis result evaluation instruction is sent to the server. The analysis result evaluation instruction carries the verification data and is used to instruct the server to process the multiple verification samples according to the target data analysis process based on the algorithm model service to obtain a second analysis result corresponding to each verification sample. The server determines the evaluation index corresponding to the algorithm model service based on the first and second analysis results corresponding to each verification sample. If any evaluation index is greater than or equal to a preset threshold, the server generates a second alarm message and sends the second alarm message to the first device. The second alarm message is used to indicate that the analysis result of the algorithm model service is abnormal. Specifically, when the algorithm model service is a binary classification model, the evaluation metrics include: accuracy, precision, recall, specificity, area under the ROC curve, KS test value, and F1 score; when the algorithm model service is a multi-class classification model, the evaluation metrics include: accuracy, weighted precision, weighted recall, and weighted F-score; when the algorithm model service is a regression model, the evaluation metrics include: mean absolute error, mean square error, root mean square error, and coefficient of determination; when the algorithm model service is a clustering model, the evaluation metrics include: silhouette coefficient-cosine similarity, silhouette coefficient-Euclidean distance, sum of squared residuals, and covariance.

4. The method according to claim 3, characterized in that, The method further includes: In response to the user's input iterative update command, the algorithm model file corresponding to one or more model nodes included in the algorithm model service is iteratively trained according to the training code to obtain the trained algorithm model file. The algorithm model service file is updated based on the trained algorithm model file to obtain the updated algorithm model service file; The updated algorithm model service file is sent to the server so that the server can deploy the algorithm model service based on the updated algorithm model service file.

5. The method according to claim 4, characterized in that, The method further includes: A data distribution difference monitoring instruction is sent to the server. The data distribution difference monitoring instruction carries training data for each algorithm model file included in the training algorithm model service. The data distribution difference monitoring instruction is used to indicate that: when the server receives a data analysis request sent by the second device, it determines the data distribution difference value between the data to be analyzed and the training data carried in the data analysis request; if the data distribution difference value is greater than or equal to a third threshold, the server generates a third alarm message; and the server sends the third alarm message to the second device. The third alarm message is used to indicate that there is a data distribution difference between the data to be analyzed and the training data.

6. The method according to claim 5, characterized in that, The algorithm model service is also used to respond to a data analysis request sent by the second device, the data analysis request carrying data to be analyzed, adding the data to be analyzed to a sequentially arranged task queue, processing multiple data to be analyzed in the task queue in sequence according to the target data analysis process, obtaining the data analysis result corresponding to each data to be analyzed, and sending the data analysis result corresponding to each data to be analyzed to the second device.

7. The method according to claim 6, characterized in that, When the server is deployed with multiple algorithm model services corresponding to the target data analysis process, the method further includes: Send a release mode setting instruction to the server, the release mode setting instruction being used to instruct the server to set the release mode of multiple algorithm model services corresponding to the target data analysis process; wherein, the release mode includes independent release, canary release, or comparative release; When the release mode is independent release, the release mode setting instruction carries the identifier of the target algorithm model service. The target algorithm model service is one of multiple algorithm model services. The server responds to the data analysis request sent by the second device. The data analysis request carries the data to be analyzed. The server processes the data to be analyzed through the target algorithm model service to obtain the data analysis result and sends the data analysis result to the second device. When the release mode is gray release, the server responds to the data analysis request sent by the second device. The data analysis request carries the data to be analyzed. The server processes the data to be analyzed through multiple algorithm model services to obtain the data analysis result corresponding to each algorithm model service, and sends the data analysis result corresponding to each algorithm model service to the second device. When the publishing mode is comparative publishing, the server responds to the data analysis request sent by the second device. The data analysis request carries the data to be analyzed. The server processes the data to be analyzed through multiple algorithm model services to obtain the data analysis result corresponding to each algorithm model service. The server determines the evaluation index corresponding to each algorithm model service. The server sends the data analysis result of the algorithm model service corresponding to the maximum value of the evaluation index to the second device.

8. A full lifecycle management system for aerospace engineering data algorithm models, characterized in that, The system is applied to a first device, which communicates with a server, and includes: The model management module is used to store multiple different categories of algorithm model files in the model library; The service configuration module is used to generate an inference flowchart corresponding to the target data analysis process in response to editing and configuration operations input by the user on a preset interface. The inference flowchart includes multiple configured nodes connected sequentially according to the target data analysis process, and the nodes are model nodes or logic nodes. The editing operations include adding nodes, deleting nodes, and / or connecting nodes. The configuration operations include determining the algorithm model file corresponding to each model node from the model library and / or configuring the execution code corresponding to each logic node. The model nodes are used to process the input data according to the configured algorithm model file, and the logic nodes are used to process the input data according to the configured execution code. The service publishing module is used to respond to the publishing operation input by the user on the preset interface, generate and send the algorithm model service file corresponding to the target data analysis process to the server according to the inference flowchart, so that the server can deploy the algorithm model service according to the algorithm model service file. The algorithm model service is used to respond to the data analysis request sent by the second device. The data analysis request carries the data to be analyzed. The data to be analyzed is processed according to the target data analysis process to obtain the data analysis result, and the data analysis result is sent to the second device. The logical nodes include data processing nodes, routing nodes, and combiner nodes. Each data processing node includes at least one input terminal and at least one output terminal. The data processing node processes the data input from the input terminal according to configured execution code and outputs the processing result through the output terminal. Each combiner node includes at least one input terminal and multiple output terminals. The combiner node outputs the data input from the input terminal through the multiple output terminals according to configured execution code. Each routing node includes at least one input terminal and multiple output terminals. The routing node determines the target output terminal based on the data input from the input terminal using configured execution code and outputs the input data through the target output terminal, where the target output terminal is one of the multiple output terminals.

9. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the aerospace engineering data algorithm model full lifecycle management method as described in any one of claims 1-7.

Images