dc pue optimization model generation service

NAIEV200R021C30

DC PUE Optimization ModelGeneration Service

Issue 01

Date 2020-12-30

HUAWEI TECHNOLOGIES CO., LTD.

Copyright © Huawei Technologies Co., Ltd. 2021. All rights reserved.

No part of this document may be reproduced or transmitted in any form or by any means without priorwritten consent of Huawei Technologies Co., Ltd. Trademarks and Permissions

and other Huawei trademarks are trademarks of Huawei Technologies Co., Ltd.All other trademarks and trade names mentioned in this document are the property of their respectiveholders. NoticeThe purchased products, services and features are stipulated by the contract made between Huawei andthe customer. All or part of the products, services and features described in this document may not bewithin the purchase scope or the usage scope. Unless otherwise specified in the contract, all statements,information, and recommendations in this document are provided "AS IS" without warranties, guaranteesor representations of any kind, either express or implied.

The information in this document is subject to change without notice. Every effort has been made in thepreparation of this document to ensure accuracy of the contents, but all statements, information, andrecommendations in this document do not constitute a warranty of any kind, express or implied.

Issue 01 (2020-12-30) Copyright © Huawei Technologies Co., Ltd. i

Contents

1 Documentation Guide............................................................................................................ 1

2 Product Overview.................................................................................................................... 22.1 What is the DC PUE Optimization Model Generation Service................................................................................ 22.2 Application Scenarios............................................................................................................................................................. 22.3 Product Features...................................................................................................................................................................... 32.4 Product Values..........................................................................................................................................................................32.5 Restrictions................................................................................................................................................................................ 52.6 Basic Concepts.......................................................................................................................................................................... 52.7 Service Dependencies.............................................................................................................................................................62.8 Billing Description................................................................................................................................................................... 62.9 Accessing the DC PUE Optimization Model Generation Service............................................................................ 72.10 Change History...................................................................................................................................................................... 8

3 Quick Start................................................................................................................................ 93.1 Using the PUE Optimization Model Generation Service to Quickly Generate a DC PUE OptimizationModel.................................................................................................................................................................................................. 93.1.1 Prerequisites...........................................................................................................................................................................93.1.2 Subscribing to the DC PUE Optimization Model Generation Service................................................................ 93.1.3 Accessing the DC PUE Optimization Model Generation Service.......................................................................103.1.4 Procedure............................................................................................................................................................................. 113.1.5 Creating a Service..............................................................................................................................................................113.1.6 Generating a Model..........................................................................................................................................................123.1.7 Evaluating a Model........................................................................................................................................................... 223.1.8 Verifying a Model.............................................................................................................................................................. 233.1.9 Publishing an Application Package............................................................................................................................. 243.1.10 Packaging Multiple Models......................................................................................................................................... 273.1.11 Other Operations............................................................................................................................................................ 293.2 Change History...................................................................................................................................................................... 29

4 FAQs..........................................................................................................................................304.1 What Factors Can Affect the PUE................................................................................................................................... 304.2 What Are the Measures to Lower the PUE.................................................................................................................. 304.3 How Can We Optimize Cooling Systems......................................................................................................................304.4 What Is Transfer Learning?................................................................................................................................................31

NAIEDC PUE Optimization Model Generation Service Contents

Issue 01 (2020-12-30) Copyright © Huawei Technologies Co., Ltd. ii

4.5 How Do I Obtain NMS Information............................................................................................................................... 324.6 How Do I Obtain Tenant Information........................................................................................................................... 324.7 Change History...................................................................................................................................................................... 33

5 Glossary................................................................................................................................... 34

NAIEDC PUE Optimization Model Generation Service Contents

Issue 01 (2020-12-30) Copyright © Huawei Technologies Co., Ltd. iii

1 Documentation Guide

Documents including Introduction, Quick Start, FAQs, and Glossary are given tohelp customers learn and use the DC PUE optimization model generation servicein order to customize site-oriented models.

Table 1-1 Documentation guide

Document Description

Introduction This document describes the positioning, application scenarios,functions, benefits, and restrictions of the DC PUE optimizationmodel generation service.

Quick Start This document describes how to use the DC PUE optimizationmodel generation service to quickly generate DC PUEoptimization models, helping users quickly get familiar with anduse the DC PUE optimization model generation service.

FAQs This document provides answers to frequently asked questions(FAQs) for users of the DC PUE optimization model generationservice.

Glossary This document describes the product terms related to the DCPUE optimization model generation service.

NAIEDC PUE Optimization Model Generation Service 1 Documentation Guide

Issue 01 (2020-12-30) Copyright © Huawei Technologies Co., Ltd. 1

2 Product Overview

2.1 What is the DC PUE Optimization ModelGeneration Service

The data center (DC) power usage effectiveness (PUE) optimization modelgeneration service is oriented to DC energy power saving scenarios. It combinesartificial intelligence (AI) with traditional heating, ventilation, and air conditioning(HVAC) knowledges experience and collects data of the equipment room outerdoors environment, IT loads, cooling stations equipment status, and end airconditioner status for analysis and modeling. The group control system can beautomatically adjusted by the model to reduce energy power consumptionwithout sacrificing service quality.

HVAC experts only need to set parameters based on the DC cooling mode andcorresponding HVAC techniques. The model generation service will automaticallyperform modeling, generating a DC PUE optimization model through training, anddeploy the model for customers. After the scheduling mode of the DC PUEoptimization model and interaction mode with the network management system(NMS) are configured for customers, real-time data exchange and policy deliverybetween the PUE optimization model and NMS can be implemented, and thereducing DC energy power consumption will reduce.

2.2 Application ScenariosPower consumption costs make up a large proportion of large DC operationscosts. In addition to the IT loads, other systems (such as cooling systems) shouldreduce power consumption costs. Currently, there are two delivery scenarios.

Initial Generation of the DC PUE Optimization Model

If a DC PUE optimization model is introduced to a DC for the first time, HVACexperts need to import engineering parameters and historical data collections ofthe cooling system. The service will generates a PUE optimization model based onthe parameters and the data.

NAIEDC PUE Optimization Model Generation Service 2 Product Overview


Update an existing DC PUE optimization model.If a DC already has a PUE optimization model, but the cooling system haschanged (for example, device aging or upgrade) or the service IT loads havechanged greatly, HVAC experts need to adjust the cooling system parameters andimport new environment data to update the PUE optimization model, avoiding thedeterioration of the model prediction effect.

2.3 Product Features

Cooling Mode and HVAC Technique ConfigurationHVAC experts can set parameters based on the DC cooling scenarios and HVACimplementation techniques.

Data importHVAC experts can import data from a local path or an object storage service(OBS) bucket.

Service modeling topology modificationThe service modeling topology can be automatically generated based on the dataimported by HVAC experts. In addition, the service topology can be modified.

Automatic modelingAlgorithms are automatically selected for model training. In this way, the DC PUEoptimization model can be generated automatically.

2.4 Product Values

Multi-Scenario AdaptationModels adapting to different cooling modes (such as mechanical cooling and freecooling) and different cooling device engineering techniques (such as constant-frequency or variable-frequency water chillers, open or closed cooling towers, andin-row or in-room air conditioners) can be generated based on Huawei's extensiveexperience in DC energy engineering, helping HVAC experts handle multiple DCscenarios.



Figure 2-1 Data Center

Comprehensive Control PolicyThe models can infer a policy to control all DC cooling devices (such as chillers,cooling water pumps, cooling towers, chilled water pumps, and heat exchangers)based on the capability of optimizing the DC cooling device control policy, helpingHVAC experts flexibly control the cooling system.

Figure 2-2 DC infrastructure

Oriented to HVAC Experts, Solving AI Modeling Problems Without CodingHVAC experts only need to provide data of the DC chilled water system (such asdata of IT loads, cooling stations, and air conditioners), and technique parametersof DC cooling devices (such as the header system/single pipe system and the



system where the chiller and heat exchanger are connected in parallel/in series).An AI model that matches the DC can be automatically obtained.

Figure 2-3 Process of AI modeling by an HVAC expert

2.5 RestrictionsThis service is applicable to a DC that meets the following conditions:

(1) Requires AI technologies to reduce the PUE.

(2) Has inference framework components.

(3) Lacks the AI model development capability.

For example, a DC that uses the chilled water system for temperature adjustmentwants to use AI technologies to perform energy saving optimization on the chilledwater system. The prediction framework has been deployed, but the PUEoptimization model that matches the DC environment is unavailable. In this case,the service can be used to reduce the PUE.

2.6 Basic Concepts

PUEPUE is used to measure the energy efficiency of DCs. If the PUE value is 2, itindicates that when an IT device consumes electricity of 1 watt, additionalelectricity of 1 watt needs to be consumed to cool and distribute the IT device. Ifthe PUE value is close to 1, almost all power is consumed for IT device running.



2.7 Service Dependencies

ModelArts Service

The NAIE platform uses the ModelArts service provided by the Huawei publiccloud system to implement data preprocessing and large-scale distributed modeltraining.

IAM Service

The NAIE platform uses the Identity and Access Management (IAM) serviceprovided by the Huawei public cloud system to implement unified identityauthentication and permission management.

API Gateway

The NAIE platform must interconnect with the unified API gateway provided bythe Huawei public cloud system. The API gateway provides a unified entrance forusers to invoke NAIE cloud service APIs. APIs provided by the NAIE cloud servicefor tenants must be registered with the API gateway before being released.

Relationship with the OBS

The NAIE platform uses the Object Storage Service (OBS) to store data and modelbackup and snapshots, achieving secure, reliable, and low-cost storage.

Relationship with the CCE

The NAIE platform uses the Cloud Container Engine (CCE) to deploy models asonline services, satisfying requirements for high concurrency and elastic scaling.

2.8 Billing Description

Billing Items

The DC PUE optimization model generation service is charged based on theselected instance specifications and usage duration. The billing items include themodel generation service and cloud-based inference, as shown in Table 2-1.

Table 2-1 Billing items

Billing Item Description

Modelgenerationservice

The DC PUE optimization model generation service is chargedbased on the CPU and GPU specifications and usage duration. Ifthe service is not used, no fee is charged.



Billing Item Description

Cloud-basedinference

The cloud-based inference service is charged based on the CPUand GPU specifications and usage duration. If the service is notused, no fee is charged.Once a model is deployed in cloud-based inference, is started,and the instance is in the Running status, fees are charged.Stop unnecessary instances in a timely manner to avoidunnecessary fees.

Billing Mode

Pay-per-use mode is used. Fees are charged based on the specifications and usageduration of Running instances.

● Billing formula: Unit price x Number of instances x Usage duration. The fee isdeducted by cent.

● With pay-per-use pricing, if the estimated price is a decimal numeral, it will beaccurate to two decimal places with the third digit rounded off. For example,if the estimated price is less than 0.01 after being rounded off, 0.01 isdisplayed.

● The DC PUE optimization model generation service uses the OBS.

Changing Billing Mode

Subscribing to the DC PUE optimization model generation service does not incurfees, but running instances incur fees. Therefore, service change configuration isnot involved. You can select and run instances with the required specifications.

Renewal

Users can recharge their accounts in time as required to ensure that the DC PUEoptimization model generation service can be used properly.

Expiration and Overdue Payment

If you do not renew your subscription on time, the cloud platform provides a graceperiod and a retention period. For details, see Grace Period and RetentionPeriod.

If the account is not recharged after the retention period expires, the resources arecleared.

2.9 Accessing the DC PUE Optimization ModelGeneration Service

Step 1 Enter https://console-intl.huaweicloud.com/naie/ in the address box of abrowser on a user PC and press Enter to access the NAIE service official website.



https://support.huaweicloud.com/intl/en-us/usermanual-billing/en-us_topic_0083039587.html

https://support.huaweicloud.com/intl/en-us/usermanual-billing/en-us_topic_0083039587.html

https://console-intl.huaweicloud.com/naie/

Step 2 Click Sign In in the upper right corner to access the login page.

Step 3 Select IAM User Login and enter the tenant name, user name, and password.

You can also log in using an account. Change the password after the firstsuccessful login and change the password periodically.

Step 4 Click Log In to access the NAIE service official website.

Step 5 Choose AI Services > Model and Training Service > Model Generation Service >DC PUE Optimization Model Generation Service. The introduction page of theDC PUE optimization model generation service is displayed.

Step 6 Click Enter Service. The DC PUE optimization model generation service page isdisplayed.

----End

2.10 Change HistoryDate Change Description

2020-06-30 Added section "Billing Description."

2019-12-30 Optimized the outline of Product Overview and rewrote theentire document.

2019-04-30 Released this document officially for the first time.



3 Quick Start

3.1 Using the PUE Optimization Model GenerationService to Quickly Generate a DC PUE OptimizationModel

This document describes how to use the DC PUE optimization model generationservice to quickly generate a DC PUE optimization model, helping users quicklyget familiar with and use this service.

3.1.1 Prerequisites● You have registered a HUAWEI CLOUD account.

● The administrator tenant and IAM user of the NAIE platform have beenregistered.

● You have subscribed to the DC PUE optimization model generation service ofthe NAIE.

3.1.2 Subscribing to the DC PUE Optimization ModelGeneration Service


When you access the NAIE service official website for the first time, the AccessAuthorization page is displayed. Click Authorize.

Step 2 Click Sign In in the upper right corner of the page. The login page is displayed.

Step 3 Enter the tenant name and password, and click Log In to access the NAIE serviceofficial website.

Change the password after the first successful login and change the passwordperiodically.

NAIEDC PUE Optimization Model Generation Service 3 Quick Start



Step 4 Choose AI Services > Model and Training Service > AI Model GenerationService > DC PUE Optimization Model Generation Service. The introductionpage of the DC PUE optimization model generation service is displayed.

Step 5 Click Buy Now. The page shown in Figure 3-1 is displayed.

You can click Learn about billing details to better understand the resources,specifications, and price information provided by the DC PUE optimization modelgeneration service. In addition, when you use a specific resource, the servicedisplays an eye-catching charging prompt on the page.

The parameters are described as follows:

● Region: HUAWEI CLOUD region that provides services.● MapReduce Service: Set this parameter as required.

Figure 3-1 Subscribing to the DC PUE optimization model generation service

Step 6 Click Use Immediately. The service subscription is complete.

----End

3.1.3 Accessing the DC PUE Optimization Model GenerationService


Step 2 Click Sign In in the upper right corner to access the login page.

Step 3 Select IAM User Login and enter the tenant name, user name, and password.

You can also log in using an account. Change the password after the firstsuccessful login and change the password periodically.

Step 4 Click Log In to access the NAIE service official website.

Step 5 Choose AI Services > Model and Training Service > Model Generation Service >DC PUE Optimization Model Generation Service. The introduction page of theDC PUE optimization model generation service is displayed.

Step 6 Click Enter Service. The DC PUE optimization model generation service page isdisplayed.

----End




3.1.4 ProcedureFigure 3-2 shows the procedure of the DC PUE optimization model generationservice.

Figure 3-2 Procedure

3.1.5 Creating a ServiceStep 1 On the Service List homepage of the DC PUE optimization model generation

service, click Add Service in the upper right corner.



The Add Service dialog box is displayed, as shown in Figure 3-3.

Figure 3-3 Adding a service

Step 2 Click OK. The service is added successfully.

----End

3.1.6 Generating a Model

HVAC System Configuration

Step 1 Click corresponding to the new service in the Operation column. The HVACSystem Configuration page is displayed, as shown in Figure 3-4.

The system provides the following configuration items: Optimization Scenario,Cooling Mode, and Pipe Type. For example, the Pipe Type configuration itemincludes the following options: Header and Single Pipe. Currently, HVAC devicesdo not support these configurations.



Figure 3-4 HVAC system configuration

Step 2 Expand the advanced option Cooling Station Configuration and select or add acooling station configuration, as shown in Figure 3-5. By default, this parameter isleft empty.● Click Select and select any configured cooling station configuration, as shown

in Figure 3-6. The cooling station configuration is displayed by NMSinformation and cooling station information.

● Click Add to add a cooling station configuration, as shown in Figure 3-7. Youneed to enter the basic information and NMS information about the coolingstation. Cloudsop Product UUID is mandatory.

The selected or added cooling station information is displayed as the coolingstation configuration, shown in Figure 3-8. If you select the configurationprovided by the system, the information cannot be modified. If the configuration isadded by the user, the configuration can be modified.

Note that only one piece of cooling station information can be configured. Youneed to delete the existing cooling station information before selecting or addingnew cooling station information.

Figure 3-5 Cooling station configurations



Figure 3-6 Select cooling station configurations

Figure 3-7 Adding cooling station configurations

Figure 3-8 Cooling station configurations

Step 3 If Optimization Scenario is set to Parameter optimization for air conditionersand cooling stations, you need to configure the micro-module cluster, as shownin Figure 3-9.



● Data Source is set to None by default. Retain the default value.

● Sample data is a preset file and does not need to be uploaded.

● If Local is selected, CSV files can be uploaded locally.

Figure 3-9 Micro-module cluster configurations

Step 4 Click Next. The Selecting Data page is displayed.

----End

Data Selection

Step 1 Set dataset information, as shown in Figure 3-10.

● Dataset Source: The default value is Sample data. The sample dataset hasbeen preset and does not need to be uploaded. You can directly import data.Retain the default value. If Data catalog is selected, you need to subscribe toand download a dataset on the data catalog platform and import the datasetto the DC PUE optimization model generation service platform. If Local isselected, upload files locally.

● Select Data File: Select a local data file. For details about the data file formatrequirements, see the dataset downloaded on the Download Sample Dataset

page or click on the right in the Select Data File area. You can directlyupload or download the corresponding sample dataset packagetrainTemplate_fc.zip or trainTemplate_ec.zip for trial use.

Figure 3-10 Dataset information

Step 2 In the Data Verification Configuration area, click Add to add verificationconfigurations, as shown in Figure 3-11. After a verification configuration is

added, click to save the configuration.

You can configure the parameter value range before feature combination to filterdata in feature engineering. You can also edit, cancel, save, and delete theverification configuration.

If you need to configure multiple data records, click Add. Before configuring thenext data record, ensure that the previous data record has been saved.



Figure 3-11 Data verification configurations

Step 3 Expand Feature Manage to configure the mapping between features andparameters, as shown in Figure 3-12.

Retain the default settings. You can modify the default policy configurations asrequired. The default policy configurations cannot be deleted. You can edit thedefault configurations, add a configuration, and modify or delete configurations.

Figure 3-12 Configuring the feature source

Step 4 Click Import Data. The system automatically uploads and checks the data.

If the feature name of the dataset matches the parameter topology, a data checkreport shown in Figure 3-13 is displayed. The table lists the number of invalidvalues, percentage of invalid values, maximum value, and minimum value of eachfeature value.

If the import fails, check whether the dataset or data verification configurationsare correct. If no, correct them and import the data again.

Figure 3-13 Importing a dataset

Step 5 (Optional) If you need to upload and import data again after the import issuccessful, click Re-import. On the returned dataset and data verificationconfiguration page, upload and import data again.



Step 6 Click Next. The Training Configuration page is displayed.

----End

Training Configurations

Step 1 Set the control parameter optimization range, as shown in Figure 3-14.

The Name column lists the names of all control parameters of the cooling system.You can set the value range and adjustment precision of each control parameter.The system automatically records the latest data settings and updates the valuerange of the corresponding control parameter in the parameter topology.

Figure 3-14 Configuring the control parameter optimization range

Step 2 Configure optimization constraints, as shown in Figure 3-15.

You can add, edit, delete, and save constraint configurations. Each constraintconfiguration contains at least one comparison operator (==, !=, >, <, >=, and <=).The saved optimization constraint configurations are used to optimize the trustedspace for model verification. You can click Optimization RestrictionReasonableness Check to check the percentage of datasets that pass thetopology restriction.

Figure 3-15 Configuring optimization constraints

Step 3 Click the drop-down arrow next to Parameter Topology Configuration. Theparameter topology configuration information is displayed, as shown in Figure3-16.

The parameters in different areas are described as follows:● All parameters in area 3 are control parameters, which can be adjusted based

on the site requirements.Hover the pointer over a parameter icon to view the alias, type, maximumvalue, minimum value, and adjustment precision of the parameter. You canright-click the icon and choose Modify from the shortcut menu. In the dialog



box that is displayed, modify the parameter information, as shown in Figure3-17. After the maximum value, minimum value, and adjustment precisionare modified, the modification is automatically synchronized to the controlparameter optimization range configuration area shown in Figure 3-14.

● All parameters in area 2 are environment parameters, which cannot becontrolled.Hover the pointer over an icon to view the parameter alias and type. You canright-click the icon and choose Modify from the shortcut menu. In the dialogbox that is displayed, modify the parameter information, as shown in Figure3-18.

● All parameters in area 1 are correlative parameters.Hover the pointer over a correlative parameter. Then lines are displayedbetween the correlative parameter and the environment parameters andcontrol parameters. The lines indicate that the correlative parameters areaffected by the environment parameters and control parameters in the ring.Hover the pointer over an icon to view the parameter alias and type. You canright-click the icon and choose Modify from the shortcut menu. In the dialogbox that is displayed, modify the parameter information, as shown in Figure3-19.

Figure 3-16 Parameter topology



Figure 3-17 Control parameter information

Figure 3-18 Environment parameter information



Figure 3-19 Correlative parameter information

Step 4 Perform the following operations as required if you are a service expert.● Right-click the icon of an unnecessary parameter and choose Delete from the

shortcut menu to delete the parameter. Then, click Save for the parameterdeletion to take effect.

● Click Add. A dialog box is displayed, as shown in Figure 3-20. Select a definedparameter name from the parameter name drop-down list box and set otherparameters based on the site requirements.

● Right-click a line and choose Delete from the shortcut menu to delete thecorrelation relationship between the two parameters.

● Draw a line between two icons to establish a correlation relationship.



Figure 3-20 Adding a parameter

Step 5 Choose Parameter optimization for air conditioners and cooling stations inOptimization Scenario on the HVAC System Configuration page. You can viewParameter Association under Parameter Topology Configuration, as shown inFigure 3-21.

You can add, modify, delete, or save associated parameter configurations. Eachparameter association must contain a cooling station parameter and an airconditioner parameter. Cooling station parameters are transferred to airconditioners for use. The saved associated parameters are used for air conditioneroptimization, improving the model inference optimization.

Figure 3-21 Parameter association

Step 6 Click Generate Model.

Step 7 In the displayed Selecting the Computing Node Specifications dialog box, selectthe computing node specifications and click Generate Model.

The system returns to the Service List page. The service process indicates theprogress of the current model training. If the service status is FINISHED in theStatus column, the model has been generated.



CA UTION

If the service status is RUNNING, the foreground of the DC PUE optimizationmodel generation service platform keeps sending messages to the background toquery the current service status. Even if the platform access times out, theinterface for querying service status keeps sending query messages to thebackground and the messages do not time out. When the service status changesto FINISHED, the interface automatically stops querying the service status.

----End

3.1.7 Evaluating a ModelModel evaluation is to obtain some data that is not used for training fromhistorical data packages and evaluate the model.

Step 1 After a model is generated, click in the Operation column of a correspondingservice on the Service List page.

Model evaluation starts and a model evaluation report is generated, as shown inFigure 3-22.

The red curve represents the actual energy consumption of the cooling station.The black curve represents the predicted energy consumption of the coolingstation, which is deduced from the AI model. The more the red curve overlaps theblack curve, the smaller the deviation is and the more accurate the AI model is. Ifthe evaluation report shows that the accuracy is less than 95%, you need to adjustthe parameter topology again. For example, you can add, delete, and modify afitting relationship between a correlative parameter and a control variable orenvironment variable, and add, delete, and modify a correlative parameter. Afterthe modification is complete, generate a new model.

Move the cursor to on the right of Model availability evaluation in the modelevaluation report dialog box. You can view the model availability evaluationresults and model training suggestions.



Figure 3-22 Model evaluation report

Step 2 You can drag the left and right boxes in the gray area below the time axis, asshown in Figure 3-22.

Drag the red box on the left to the right and the red box on the right to the left tonarrow the range. You can view the curve fitting information of a day in the chart.

Step 3 Click Back to close the model evaluation report.

----End

3.1.8 Verifying a ModelYou need to upload a new dataset for model verification.

Step 1 After a model is generated, click in the Operation column of acorresponding service on the Service List page.

Step 2 Upload a file, as shown in Figure 3-23.

You can click Download Sample Dataset to download the corresponding sampledataset and view its format.● Dataset Source: Use the default value Local. Sample data indicates preset

sample datasets. If you select this dataset source, you can click Run and verifythe model directly without uploading datasets.

● Select File: Click the icon in the red box to upload a local data file or thedownloaded sample dataset predictTemplate_fc.zip or predictTemplate_ec.zip.



Figure 3-23 Uploading a dataset

Step 3 After the dataset is uploaded and Status on the left of the page changes toUPLOADED, click Run in the right upper corner of the page.

Step 4 In the displayed Selecting the Computing Node Specifications dialog box, selectthe computing node specifications and click Run. The system starts to verify themodel.

After the model verification is complete, a model evaluation report is generated.The report is displayed on the Optimization Result tab page, which contains theExpected Energy Saving Effect and Energy Saving Control Policy areas, asshown in Figure 3-24.

● Expected Energy Saving Effect: View the bar chart of the predicted energysaving effects. You can drag the scroll bar below the chart to narrow downthe time range and view the data of a certain day or time.

● Energy Saving Control Policy: View the adjustment value of the AI energysaving control policy in different time segments compared with the originalcontrol policy. You can adjust the value of a control parameter of the datacenter based on the data in the Adjusted Value row.

Figure 3-24 Optimization result

Step 5 Click Back to return to the Service List page of the DC PUE optimization modelgeneration service.

----End

3.1.9 Publishing an Application PackageA user can publish an application package only after the user is registered as adeveloper.



Step 1 Register as a developer.

1. Enter https://console-intl.huaweicloud.com/naie/ in the address box of abrowser and press Enter to access the NAIE service official website.

2. In the avatar drop-down list in the upper right corner, click Workspace. Thepersonal workspace page is displayed.

3. Click Register as a Developer, as shown in Figure 3-25.

1. Figure 3-25 Registering as a developer

In the Register as a Developer dialog box, select Individual Developer andcomplete the developer registration as prompted.

Step 2 Return to the Service List page of the DC PUE optimization model generation

service and click in the Operation column corresponding to the targetservice.

Step 3 In the displayed Package Publish dialog box, click Generate, as shown in Figure3-26.

The default application package name is DCPUE-Cooling mode-Cooling stationDN-Timestamp-Random letter, which is configurable. By default, the initial versionof an application package is 1.0.0. When the application package is repacked, theversion number is automatically incremented by one.

In the Advanced area, you can select the use case version, which corresponds todifferent inference platform frameworks. Retain the default value.

CA UTION

The application package name and version must be unique. Otherwise, thepacking fails.




Figure 3-26 Packing an application package

It takes about 2 minutes to generate an application package.

Step 4 After the application package is generated, click Publish to push the applicationpackage to the AI marketplace, as shown in Figure 3-27.

You can also perform the following operations:

Click Regenerate, reconfigure the application package information in thedisplayed Package Publish dialog box, and click Generate to generate a newapplication package.

Figure 3-27 Message indicating that the application package has been generated

Step 5 After the application package is successfully pushed, the page shown in Figure3-28 is displayed. You can click the AI market link to go to the developer space,



publish the pushed application package, and wait for the publication applicationto be approved. During the waiting period, you can click the AI market link to goto the developer space and view the application package approval progress.

You can also perform the following operations:

Click Regenerate, reconfigure the application package information in thedisplayed Package Publish dialog box, and click Generate to generate a newapplication package.

Figure 3-28 Message indicating that the application package has been published

Step 6 Click Back to close the application package publication window.

----End

3.1.10 Packaging Multiple ModelsStep 1 Click Multi Model Package on the top of the DC PUE optimization model

generation service page.

On the Multi Model Package Management page, you can pack the models ofmultiple services that have been trained. (The services must belong to differentcooling scenarios.)

Step 2 Click New. The multi-model packaging dialog box is displayed. The applicationpackage name and version have been preset. You can also manually change them.Select a service name from the Packaging Service drop-down list box, as shownin Figure 3-29.



Figure 3-29 Multi-model packaging

Step 3 Click Generate and wait until the package is generated.

Step 4 After the package is generated, click to view the services associated withthe package, as shown in Figure 3-30. In addition, you can re-generate, release,and delete the generated application package.

Figure 3-30 Viewing associated services

----End



3.1.11 Other Operations

Step 1 Click in the upper right corner to view all reference documentsof the DC PUE optimization model generation service.

Step 2 Click corresponding to a service in the Operation column and select Edit orDelete from the drop-down list box to edit or delete the service. For a runningservice, you can click Stop to stop the model generation for the current service.You can click Log to view created feature engineering, model generation, modelevaluation, and model verification tasks as well as corresponding logs.

----End

3.2 Change HistoryDate Description

2020-12-30 Revised section "Publishing an Application Package."

2020-08-17 Optimized the HVAC configurations and updated the"Generating a Model" section.

2020-07-16 Deleted the function of publishing a model package as aninference service and section "Publishing an Inference Service."

2020-06-16 Added multi-model packaging. For details, see PackagingMultiple Models.Added cooling station configurations during model generation.For details, see Generating a Model.

2020-03-30 Added the following sections:● Publishing an Application Package● Publishing an Inference Service

2019-12-30 Provided service subscription and added section "Subscribing tothe DC PUE Optimization Model Generation Service."Added descriptions about range configurations for controlparameter optimization and parameter associationconfigurations in Generating a Model.

2019-10-30 Optimized the service page and functions and updatedcorresponding descriptions in Generating a Model.

2019-04-30 Released this document officially for the first time.



4 FAQs

4.1 What Factors Can Affect the PUEThe following factors can affect the PUE:

● Power consumption of non-IT systems, including cooling systems, such as thecompressors, pumps, fans, and air conditioners.

● Power loss of the power supply and distribution systems.

4.2 What Are the Measures to Lower the PUEThe following measures can lower the PUE:

● Improve the working efficiency of the cooling systems to reduce the costs ofheat exchange.

● Optimize the power supply and distribution systems to improve the powerconversion efficiency.

4.3 How Can We Optimize Cooling SystemsThe cooling systems can be optimized in the following ways:

● Accurately calculate the cooling capacity to match the IT heat sources.● Improve the working efficiency of each device, for example, making full use of

the optimal working ranges of compressors and pumps.● Improve the working efficiency of each system, for example, properly

adjusting the ratio of compressors to pump.● Set the best working parameters for the systems, for example, setting the

ratio of cooling towers to water chillers, setting parameter linkage betweencooling towers and water chillers, and setting parameter linkage betweenwater chillers and end systems.

NAIEDC PUE Optimization Model Generation Service 4 FAQs


4.4 What Is Transfer Learning?

Basic ConceptsTransfer learning is a field of machine learning. It focuses on applying models (orknowledge) that resolve existing problems to other similar problems. For example,the ability of identifying trucks can be improved using the vehicle identificationmodel (or knowledge). The model capability for resolving problems of new datacenters (or the target domain) can be improved using the energy consumptionmodel or knowledge learned from existing data centers (or the source domain).

BenefitsTransfer learning requires a small amount of training data in the target domain, asshown in the following figure. You can transfer knowledge or models in the sourcedomain to enhance the regression effect for non-training data, improve the modelgeneralization capability, and implement cross-domain model deployment.

Typical Methods● Feature transfer

Knowledge to be transferred from the source to target domain should be assimilar as possible to improve transferability. This eliminates the need forconsidering specific domain characteristics impeding the transfer. Onlycommon characteristics between the two domains need to be considered.Direct comparison of knowledge learned by the automatic decoder is notideal. Proactive measures should be taken to make knowledge from thesource and target domains more similar and identify the differences toimprove the transfer result.

● Model transferIt is assumed that the source domain and the target domain share modelparameters. The target domain uses the models trained with a large amountof data in the source domain for prediction. For example, if an imagerecognition system has been trained using millions of images and a new



image problem occurs, you do not need to prepare hundreds of thousands ofimages for training. Instead, you only need to transfer a trained model to thenew field and provide only tens of thousands of images to reach highprecision. The advantage is that the similarity between models can be fullyutilized.

4.5 How Do I Obtain NMS InformationPrerequisites: You have the O&M permission.

The procedure is as follows:

1. Log in to the offline NAIE-I.

2. Click System Configurations in the upper right corner of the page.

3. Click the NMS Configurations tab page and check the values of Framework ID(EMS UUID) and NMS Name.

4.6 How Do I Obtain Tenant InformationYou can obtain tenant information from the customer in compliant methods.



To view tenant information, perform the following steps:

1. Log in to the DC PUE optimization model generation service page using anaccount.

2. Press F12 to open the console and click the Network tab page.3. Find and click interface me. The interface invoking record window is displayed

on the right.4. Click the preview tab to view the values of the following parameters:

– id: Application ID.– domainId: Tenant ID.

4.7 Change HistoryDate Change Description

2020-09-30 Added the following chapters:● How Do I Obtain NMS Information● How Do I Obtain Tenant Information

2020-03-30 Added section "What Is Transfer Learning?"

2019-04-30 This is the first official release.



5 Glossary

PPUE

Power usage effectiveness (PUE) is used to measure the energy efficiency of datacenters. If the PUE value is 2, it indicates that when an IT device consumeselectricity of 1 watt, additional electricity of 1 watt needs to be consumed to cooland power the IT device. If the PUE value is close to 1, almost all power isconsumed for device running.

NAIEDC PUE Optimization Model Generation Service 5 Glossary
