Platform management-based application analysis method of electric power AI model

Abstract

The invention relates to an application analysis method of an electric power AI model based on platform management, which comprises the following steps: carrying out unified management on models uploaded by third-party manufacturers, and carrying out standard encapsulation of mirror images and monitoring of required hardware resources; formulating an optimal model application strategy; the optimal model application strategy is issued to a station end model analysis host; and carrying out practical evaluation on the optimal model application strategy issued to the station end, understanding practical application indexes of the model, and carrying out comparative analysis on the indexes before and after the model is updated. According to the method, a large number of models can be effectively managed, meanwhile, the multiple models can be subjected to same-dimension comparison under a unified index, and a model evaluation mechanism is perfected; advantage identification tags of different models are screened out for model coding application, the model application accuracy is improved, and the model application effect is improved. And the iteratively optimized model is automatically updated and deployed to a remote power system platform, the operation state of the model in a service system is monitored in real time, and the operation and maintenance efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an application analysis method of an electric power AI model based on platform management. Background Art

[0002] In recent years, with the advancement of computer technology, machine learning, especially deep learning, has made great progress and has become the hottest research field. It has been widely promoted and applied in many other fields, including the power industry. With the rapid development of the economy, the demand for electricity continues to grow, and the overall architecture of the power grid is becoming increasingly large and complex. Traditional power inspection and management methods are difficult to meet the actual needs of daily life. Therefore, more and more manufacturers tend to design platforms based on artificial intelligence technology to adapt to the complex needs of the power system.

[0003] However, many existing models based on different types of power system platforms have not been efficiently applied, and there are also redundant model deployments. At the same time, the application indicators of the models at different stations have not been effectively counted, there is a lack of reasonable evaluation basis, and there is a lack of quantifiable evaluation indicators and evaluation reports. It is impossible to objectively evaluate the actual application effects of different manufacturers' models on site, which is not conducive to the continuous iteration and optimization of each manufacturer's model performance. In addition, the actual deployment and update efficiency of the model is low, the distribution channels between the cloud and the station are not smooth, and the excellent models generated by the cloud training iteration cannot be quickly and automatically deployed and updated to the station to provide better analysis services. Summary of the invention

[0004] In order to solve the problems in the prior art of imperfect power AI model evaluation mechanism, difficulty in improving application effect, and low deployment and update efficiency, the purpose of the present invention is to provide an application analysis method of power AI model based on platform management.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: an application analysis method of an electric power AI model based on platform management, the method comprising the following steps in order:

[0006] (1) Models uploaded by third-party vendors are brought under unified management, and standardized image packaging and required hardware resources are monitored;

[0007] (2) Evaluate various performance indicators of each model, arrange the models through multi-model comparison of indicators, and formulate the optimal model application strategy;

[0008] (3) Send the optimal model application strategy to the station-side model analysis host to achieve remote automatic update of the model;

[0009] (4) Conduct a practical evaluation of the optimal model application strategy sent to the station, understand the actual application indicators of the model, and conduct a comparative analysis of the indicators before and after the model update.

[0010] The step (1) specifically comprises the following steps in order:

[0011] (1a) Formulate unified model application image production rules and parameter requirements;

[0012] (1b) The third-party manufacturer uploads the model according to the model application image production rules and provides basic information about the model, records the hardware model that the image is compatible with and the type of analysis that can be provided, supports the third-party manufacturer's image file fragment upload, disconnection resuming, and online update functions, and obtains the image model;

[0013] (1c) Standardize and encapsulate the image model;

[0014] (1d) Unify the model deployment and update methods and write scripts to automatically deploy and update the standardized packaged models;

[0015] (1e) Intensive management and real-time monitoring of deployed models: real-time monitoring of the resource utilization of the CPU, GPU, and memory of the model and the platform to which the model belongs; based on the resource utilization of the analysis service, formulate expansion strategies and set custom thresholds. When the utilization of a certain resource exceeds the custom threshold, support elastic expansion of the service and automatically add new processes to relieve analysis pressure.

[0016] The step (2) specifically comprises the following steps in order:

[0017] (2a) Evaluate the model indicators of third-party manufacturers through the model evaluation function;

[0018] (2b) According to the requirements of the model to be evaluated, the required sample set and hardware resources are set; the evaluation process includes model image loading, data set loading, image startup, container operation and return analysis of evaluation results; errors that occur during the evaluation process are recorded in real time and returned reminders; after the evaluation is completed, the returned results are analyzed and the indicator data is recorded;

[0019] (2c) Screen the advantages of different manufacturers' models in various indicators through multi-model comparison;

[0020] (2d) Combine and apply the best model services from each type of model, formulate model application strategies and related configurations, compare different combination application situations, and obtain the best model application strategy.

[0021] The step (3) specifically refers to: updating and issuing the model according to the optimal model application strategy, realizing the distribution scheduling and load balancing of different models, opening up the issuing channel between the cloud application analysis center and the remote application scenario, issuing the model application strategy to the station-side model analysis host, and realizing remote automatic updating of the artificial intelligence algorithm.

[0022] The step (4) specifically refers to: marking the missed reports and false reports of the actual operation results of the on-site model, automatically counting the missed reports and false reports of different models on-site, and truly understanding the application indicators of the model in the actual application scenario.

[0023] It can be seen from the above technical scheme that the beneficial effects of the present invention are: first, by formulating a unified model mirroring production specification, the present invention can effectively manage a large number of models, and at the same time can compare multiple models in the same dimension under unified indicators to improve the model evaluation mechanism; second, the present invention can screen out the advantage identification tags of different models for model encoding application, give play to the "long board effect" of different models, improve the accuracy of model application, and improve the model application effect; third, the present invention can automatically update and deploy the iteratively optimized model to the remote power system platform, monitor the operating status of the model in the business system in real time, and improve the operation and maintenance efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 is a flow chart of the method of the present invention;

[0025] Figure 2 A flowchart of the method for model image management;

[0026] Figure 3 Flowchart of methods orchestrated for model application;

[0027] Figure 4 Flowchart of the method for deploying and running the model. DETAILED DESCRIPTION

[0028] like Figure 1 As shown, an application analysis method of an electric power AI model based on platform management includes the following steps in sequence:

[0029] (1) Models uploaded by third-party vendors are brought under unified management, and standardized image packaging and required hardware resources are monitored;

[0030] (2) Evaluate various performance indicators of each model, arrange the models through multi-model comparison of indicators, and formulate the optimal model application strategy;

[0031] (3) Send the optimal model application strategy to the station-side model analysis host to achieve remote automatic update of the model;

[0032] (4) Conduct a practical evaluation of the optimal model application strategy sent to the station, understand the actual application indicators of the model, and conduct a comparative analysis of the indicators before and after the model update.

[0033] like Figure 2 As shown, the step (1) specifically includes the following steps in order:

[0034] (1a) Formulate unified model application image production rules and parameter requirements;

[0035] (1b) The third-party manufacturer uploads the model according to the model application image production rules and provides basic information about the model, records the hardware model that the image is compatible with and the type of analysis that can be provided, supports the third-party manufacturer's image file fragment upload, disconnection resuming, and online update functions, and obtains the image model;

[0036] (1c) Standardize and encapsulate the image model;

[0037] (1d) Unify the model deployment and update methods and write scripts to automatically deploy and update the standardized packaged models;

[0038] (1e) Intensive management and real-time monitoring of deployed models: real-time monitoring of the resource utilization of the CPU, GPU, and memory of the model and the platform to which the model belongs; based on the resource utilization of the analysis service, formulate expansion strategies and set custom thresholds. When the utilization of a certain resource exceeds the custom threshold, support elastic expansion of the service and automatically add new processes to relieve analysis pressure.

[0039] like Figure 3 As shown, the step (2) specifically includes the following steps in order:

[0040] (2a) Evaluate the model indicators of third-party manufacturers through the model evaluation function;

[0041] (2b) According to the requirements of the model to be evaluated, the required sample set and hardware resources are set; the evaluation process includes model image loading, data set loading, image startup, container operation and return analysis of evaluation results; errors that occur during the evaluation process are recorded in real time and returned reminders; after the evaluation is completed, the returned results are analyzed and the indicator data is recorded;

[0042] (2c) Screen the advantages of different manufacturers' models in various indicators through multi-model comparison;

[0043] (2d) Combine and apply the best model services from each type of model, formulate model application strategies and related configurations, compare different combination application situations, and obtain the best model application strategy.

[0044] like Figure 4As shown, step (3) specifically refers to: updating and issuing the model according to the optimal model application strategy, realizing the distribution scheduling and load balancing of different models, opening up the issuing channel between the cloud application analysis center and the remote application scenario, issuing the model application strategy to the station-side model analysis host, and realizing remote automatic updating of the artificial intelligence algorithm.

[0045] The step (4) specifically refers to: marking the missed reports and false reports of the actual operation results of the on-site model, automatically counting the missed reports and false reports of different models on-site, and truly understanding the application indicators of the model in the actual application scenario.

[0046] In summary, the present invention can effectively manage a large number of models by formulating a unified model application mirror production specification, and at the same time evaluate the model indicators of the model manufacturer through the model indicator evaluation function, compare the multi-model indicators in the same dimension, screen out the advantage identification tags of different models for model coding application, realize the complementary advantages of models from different manufacturers, and give play to the "long board effect" of different models, and improve the accuracy of model application by more than 3%. The present invention can automatically update and deploy the iteratively optimized model to the remote power system platform, open up the "last mile" of artificial intelligence research and application, and remotely realize the online and offline operations of business system applications, monitor the running status of the model in the business system in real time, and improve the operation and maintenance efficiency by 200%.

Claims

1. An application analysis method of an electric power AI model based on platform management, characterized by: The method comprises the following steps in order: (1) Models uploaded by third-party vendors are brought under unified management, and standardized image packaging and required hardware resources are monitored; (2) Evaluate various performance indicators of each model, arrange the models through multi-model comparison of indicators, and formulate the optimal model application strategy; (3) Send the optimal model application strategy to the station-side model analysis host to achieve remote automatic update of the model; (4) Conduct a practical evaluation of the optimal model application strategy sent to the station, understand the actual application indicators of the model, and conduct a comparative analysis of the indicators before and after the model update.

2. The application analysis method of the power AI model based on platform management according to claim 1 is characterized by: The step (1) specifically comprises the following steps in order: (1a) Formulate unified model application image production rules and parameter requirements; (1b) The third-party manufacturer uploads the model according to the model application image production rules and provides basic information about the model, records the hardware model that the image is compatible with and the type of analysis that can be provided, supports the third-party manufacturer's image file fragment upload, disconnection resuming, and online update functions, and obtains the image model; (1c) Standardize and encapsulate the image model; (1d) Unify the model deployment and update methods and write scripts to automatically deploy and update the standardized packaged models; (1e) Intensive management and real-time monitoring of deployed models: real-time monitoring of the resource utilization of the CPU, GPU, and memory of the model and the platform to which the model belongs; based on the resource utilization of the analysis service, formulate expansion strategies and set custom thresholds. When the utilization of a certain resource exceeds the custom threshold, support elastic expansion of the service and automatically add new processes to relieve analysis pressure.

3. The application analysis method of the power AI model based on platform management according to claim 1 is characterized by: The step (2) specifically comprises the following steps in order: (2a) Evaluate the model indicators of third-party manufacturers through the model evaluation function; (2b) According to the requirements of the model to be evaluated, the required sample set and hardware resources are set; the evaluation process includes model image loading, data set loading, image startup, container operation and return analysis of evaluation results; Record the errors that occur during the evaluation process in real time and return reminders; After the evaluation is completed, analyze the returned results and record the indicator data; (2c) Screen the advantages of different manufacturers' models in various indicators through multi-model comparison; (2d) Combine and apply the best model services from each type of model, formulate model application strategies and related configurations, compare different combination application situations, and obtain the best model application strategy.

4. The application analysis method of the power AI model based on platform management according to claim 1 is characterized by: The step (3) specifically refers to: updating and issuing the model according to the optimal model application strategy, realizing the distribution scheduling and load balancing of different models, opening up the issuing channel between the cloud application analysis center and the remote application scenario, issuing the model application strategy to the station-side model analysis host, and realizing remote automatic updating of the artificial intelligence algorithm.

5. The application analysis method of the power AI model based on platform management according to claim 1 is characterized by: The step (4) specifically refers to: marking the missed reports and false reports of the actual operation results of the on-site model, automatically counting the missed reports and false reports of different models on-site, and truly understanding the application indicators of the model in the actual application scenario.

Images