CN116527416A - Intelligent AI energy-saving control system and method applied to data center - Google Patents

Intelligent AI energy-saving control system and method applied to data center Download PDF

Info

Publication number
CN116527416A
CN116527416A CN202310799650.6A CN202310799650A CN116527416A CN 116527416 A CN116527416 A CN 116527416A CN 202310799650 A CN202310799650 A CN 202310799650A CN 116527416 A CN116527416 A CN 116527416A
Authority
CN
China
Prior art keywords
experience
energy
saving control
content
air conditioner
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202310799650.6A
Other languages
Chinese (zh)
Other versions
CN116527416B (en
Inventor
夏玉学
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Liwan Technology Co ltd
Original Assignee
Shenzhen Liwan Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shenzhen Liwan Technology Co ltd filed Critical Shenzhen Liwan Technology Co ltd
Priority to CN202310799650.6A priority Critical patent/CN116527416B/en
Publication of CN116527416A publication Critical patent/CN116527416A/en
Application granted granted Critical
Publication of CN116527416B publication Critical patent/CN116527416B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L12/00—Data switching networks
    • H04L12/02—Details
    • H04L12/10—Current supply arrangements
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/092—Reinforcement learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/04—Inference or reasoning models
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16Y—INFORMATION AND COMMUNICATION TECHNOLOGY SPECIALLY ADAPTED FOR THE INTERNET OF THINGS [IoT]
    • G16Y10/00—Economic sectors
    • G16Y10/75—Information technology; Communication
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16Y—INFORMATION AND COMMUNICATION TECHNOLOGY SPECIALLY ADAPTED FOR THE INTERNET OF THINGS [IoT]
    • G16Y10/00—Economic sectors
    • G16Y10/80—Homes; Buildings
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16Y—INFORMATION AND COMMUNICATION TECHNOLOGY SPECIALLY ADAPTED FOR THE INTERNET OF THINGS [IoT]
    • G16Y20/00—Information sensed or collected by the things
    • G16Y20/10—Information sensed or collected by the things relating to the environment, e.g. temperature; relating to location
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16Y—INFORMATION AND COMMUNICATION TECHNOLOGY SPECIALLY ADAPTED FOR THE INTERNET OF THINGS [IoT]
    • G16Y40/00—IoT characterised by the purpose of the information processing
    • G16Y40/10—Detection; Monitoring
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16Y—INFORMATION AND COMMUNICATION TECHNOLOGY SPECIALLY ADAPTED FOR THE INTERNET OF THINGS [IoT]
    • G16Y40/00—IoT characterised by the purpose of the information processing
    • G16Y40/20—Analytics; Diagnosis
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16Y—INFORMATION AND COMMUNICATION TECHNOLOGY SPECIALLY ADAPTED FOR THE INTERNET OF THINGS [IoT]
    • G16Y40/00—IoT characterised by the purpose of the information processing
    • G16Y40/30—Control
    • G16Y40/35—Management of things, i.e. controlling in accordance with a policy or in order to achieve specified objectives
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/12—Protocols specially adapted for proprietary or special-purpose networking environments, e.g. medical networks, sensor networks, networks in vehicles or remote metering networks
    • H04L67/125—Protocols specially adapted for proprietary or special-purpose networking environments, e.g. medical networks, sensor networks, networks in vehicles or remote metering networks involving control of end-device applications over a network
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/2866—Architectures; Arrangements
    • H04L67/30—Profiles
    • H04L67/303—Terminal profiles
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04Q—SELECTING
    • H04Q9/00—Arrangements in telecontrol or telemetry systems for selectively calling a substation from a main station, in which substation desired apparatus is selected for applying a control signal thereto or for obtaining measured values therefrom
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04Q—SELECTING
    • H04Q2209/00—Arrangements in telecontrol or telemetry systems
    • H04Q2209/70—Arrangements in the main station, i.e. central controller
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04Q—SELECTING
    • H04Q2209/00—Arrangements in telecontrol or telemetry systems
    • H04Q2209/80—Arrangements in the sub-station, i.e. sensing device

Abstract

The invention provides an intelligent AI energy-saving control system and method applied to a data center, wherein the system comprises: the AI reasoning server is arranged in the data center; the edge control hosts are arranged on each machine room precise air conditioner in the data center one by one and are respectively in communication connection with the AI reasoning server and the machine room precise air conditioner arranged in the data center one by one; the temperature and humidity sensors are arranged in a machine room cold channel in the data center one by one and are respectively in communication connection with the AI reasoning server one by one; and the machine room cooling channel is communicated with a cooling output end of the machine room precise air conditioner. The invention reduces the PUE and saves the power consumption to a great extent.

Description

Intelligent AI energy-saving control system and method applied to data center
Technical Field
The invention relates to the technical field of energy-saving control of data centers, in particular to an intelligent AI energy-saving control system and method applied to a data center.
Background
At present, the data center is operated continuously, and energy conservation control is needed to be carried out on the data center in order to save energy consumption of the data center. Normally, the energy-saving control of the data center is realized by manually adjusting operation parameters of a machine room precise air conditioner in the data center by operation and maintenance personnel, so that the labor cost is high, in addition, the energy of the operation and maintenance personnel is limited, the conditions of untimely energy-saving control and unreasonable energy-saving control can exist, and the energy-saving control efficiency of the data center is reduced.
The communication of the precise air conditioner in the computer room in the data center is usually carried out by adopting a serial port communication mode, and common protocols comprise RS-232, RS-422 and RS-485, and serial port communication lines are usually completed by using 3 lines: the serial port two-end equipment for serial port communication needs to adopt the same baud rate, data bit, stop bit and parity check, serial port communication is low in serial communication efficiency, one-to-many communication is difficult to support, and therefore if one computer room precision air conditioner passive ring system in a data center collects through serial ports, other systems cannot collect the equipment in a serial port communication mode directly, otherwise signal interference can be caused, both sides cannot normally communicate, when a plurality of systems are required to communicate with equipment supporting serial ports, only one system can directly dock with the equipment, other systems dock with data through a software interface, the data center directly docks with a computer room precision air conditioner usually a ring system, other systems only indirectly collect and control infrastructure equipment in a rotating ring docking mode, a large amount of software docking development customization work is caused, and more importantly, the direct docking of the software system is unstable, and real-time performance is also poor.
In addition, re-wiring is difficult in data centers, and in particular some air conditioning units, for example, may not have 220 volt mains plug, and are also typically provided for temporary operation and cannot be used as a long term power supply.
Thus, for the above-mentioned problems, a solution is needed.
Disclosure of Invention
The invention aims to provide an intelligent AI energy-saving control system applied to a data center, which reduces PUE, saves power consumption to a great extent, can enable the system to directly collect the operation information of a machine room precise air conditioner through an edge control host, and enables a movable ring system to continuously collect the operation information of the machine room precise air conditioner as usual without butting the movable ring system to obtain the operation information of the machine room precise air conditioner, thereby reducing software butting development and customization work and avoiding the problems of unstable direct butting of a software system and poor real-time property. The network interface and the power supply interface are not required to be independently arranged for the edge control host in the data center, so that the data center is ensured to be free from rewiring, and the convenience is improved. The operation parameters of the machine room precise air conditioner in the data center are not required to be adjusted manually by operation and maintenance personnel, the labor cost is reduced, the situations of untimely energy-saving control and unreasonable energy-saving control caused by limited energy of the operation and maintenance personnel are avoided, and the energy-saving control efficiency of the data center is improved.
The embodiment of the invention provides an intelligent AI energy-saving control system applied to a data center, which comprises the following components:
the AI reasoning server is arranged in the data center;
the edge control hosts are arranged on each machine room precise air conditioner in the data center one by one and are respectively in communication connection with the AI reasoning server and the machine room precise air conditioner arranged in the data center one by one;
the temperature and humidity sensors are arranged in a machine room cold channel in the data center one by one and are respectively in communication connection with the AI reasoning server one by one; the machine room cooling channel is communicated with a cooling output end of the machine room precise air conditioner;
the AI reasoning server executes the following operations in one execution period:
based on communication, acquiring operation information of the precise air conditioner of the machine room and real-time temperature and humidity information in a cold channel of the machine room;
inputting the operation information and the real-time temperature and humidity information into a pre-trained IT machine room heat load prediction AI model to obtain a heat load trend of the data center for a future heat load preset time;
inputting the heat load trend into a pre-trained computer room precise air conditioner AI energy-saving control model to obtain air conditioner adjusting parameters of each computer room precise air conditioner;
And issuing the air conditioner adjusting parameters to the corresponding precise air conditioner in the machine room.
Preferably, the AI reasoning server performs the following operations in the next execution period:
continuously acquiring the actual heat load detected by the machine room precise air conditioner;
comparing the thermal load difference between the actual thermal load and the thermal load trend;
the heat load difference is supplemented and input into the AI energy-saving control model of the precise air conditioner of the machine room;
and the computer room precise air conditioner AI energy-saving control model performs energy-saving control compensation on the computer room precise air conditioner based on communication.
Preferably, the pre-training process of the IT machine room thermal load prediction AI model includes:
acquiring a large amount of heat load prediction experience;
training the initial AI model by taking a large amount of heat load prediction experience as a training sample until convergence;
and taking the converged initial AI model as the IT machine room heat load prediction AI model.
Preferably, the pre-training process of the computer room precise air conditioner AI energy-saving control model comprises the following steps:
based on online reinforcement learning, pre-training the AI energy-saving control model of the precise air conditioner of the machine room;
based on online reinforcement learning, the machine room precise air conditioner AI energy-saving control model is pre-trained, and the machine room precise air conditioner AI energy-saving control model comprises:
Acquiring a large number of historical temperature regulation records of the precise air conditioner of the machine room; the historical attemperation record includes: a first tempering target and tempering process;
determining the refrigeration performance of the machine room precise air conditioner based on the first temperature regulation target and the temperature regulation process;
acquiring a preset temperature regulation target set; the set of tempering targets includes: a plurality of second tempering targets;
based on the refrigeration performance, determining an optimal air conditioner control scheme of the machine room precise air conditioner under any second temperature regulation target;
pairing the second temperature regulation targets with the optimal air conditioner control scheme one by one to obtain a plurality of pairing groups;
training the neural network model by taking the pairing groups as training samples until convergence;
and taking the neural network model as the computer room precise air conditioner AI energy-saving control model to finish pre-training.
Preferably, the pre-training process of the AI energy-saving control model of the precise air conditioner of the machine room comprises the following steps:
acquiring equipment information of a precise air conditioner of a machine room;
acquiring a condition template based on equipment information and preset energy-saving control experience, and determining energy-saving control experience acquisition conditions;
acquiring energy-saving control experience conforming to energy-saving control experience acquisition conditions from a preset energy-saving control experience acquisition scene;
Training the neural network model by taking the energy-saving control experience as a training sample until convergence;
and taking the converged neural network model as an AI energy-saving control model of the precise air conditioner of the machine room.
Preferably, the energy-saving control experience meeting the energy-saving control experience acquisition condition is acquired from a preset energy-saving control experience acquisition scene, which comprises the following steps:
searching experience source content meeting the energy-saving control experience acquisition conditions from the energy-saving control experience acquisition scene;
analyzing the content type distribution of the experience source content; the content type distribution includes: at least one group of one-to-one first content type and first content proportion;
matching the content type distribution with a preset standard content type distribution to obtain a matching degree;
when the matching degree is greater than or equal to a preset matching degree threshold value, determining the content position of a second content proportion preset before in the content of the experience source, and extracting first local content in a first content range preset before and after the content position from the content of the experience source;
matching a first content sentence in the first local content with a standard content sentence in a preset standard content sentence library;
when the matching is met, taking the first content statement which is met by the matching as a first target content statement, and acquiring a preset content direction, a preset second content range and a preset second content type which correspond to the standard content statement which is met by the matching;
Extracting second local content of a second content type in a second content range in the content direction of the first target content statement from experience source content;
determining an experience evaluation value based on the second local content and a preset experience evaluation index library;
when the experience evaluation value is greater than or equal to a preset experience evaluation value threshold value, extracting all third local contents in the opposite direction of the content direction of the first target content sentence from experience source contents;
and sharing the semantic library based on the third local content and preset experience, and determining the energy-saving control experience.
Preferably, determining the experience evaluation value based on the second local content and a preset experience evaluation index library includes:
extracting a plurality of groups of experience evaluation indexes, coincidence-score tables and index weights which are in one-to-one correspondence from an experience evaluation index library;
determining the coincidence degree of the second local content to any one of the tested evaluation indexes;
determining a target score based on the fitness and the corresponding fitness-score table;
giving index weight corresponding to the target score to obtain an evaluation score;
and accumulating and calculating the evaluation score to obtain the experience evaluation value.
Preferably, the determining the energy-saving control experience based on the third local content and the preset experience sharing semantic library includes:
Extracting a plurality of groups of experience sharing semantics and experience descriptions which are in one-to-one correspondence from an experience sharing semantic library;
based on semantic analysis technology, carrying out semantic analysis on second content sentences in the third local content to obtain content semantics;
matching the content semantics with any one of the tested shared semantics;
when the matching is met, taking a second content sentence corresponding to the matched content semantics as a second target content sentence, and taking an experience description corresponding to the matched experience sharing semantics as a target experience description;
acquiring statement position relations among second target content statements in the third local content;
determining a standard logic keyword sequence based on the sentence position relation and a preset standard logic keyword sequence library;
extracting logic keywords in the second target content sentences, and arranging according to sentence position relations to obtain a logic keyword sequence;
matching the logic keyword sequence with a standard logic keyword sequence;
when the target experience description is matched and met, based on a preset first logic relation corresponding to a standard logic keyword sequence matched and met, carrying out logic integration on the target experience description to obtain energy-saving control experience;
Otherwise, extracting a second logic relation between target experience descriptions from the experience sharing semantic library;
and based on the second logic relation, performing logic integration on the target experience description to obtain the energy-saving control experience.
Preferably, before searching the experience source content meeting the energy-saving control experience acquisition condition from the energy-saving control experience acquisition scene, the method further comprises:
the method comprises the steps of obtaining an experience index of an experience sharing party in an energy-saving control experience obtaining scene, wherein a specific obtaining formula is as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,for the experience index of the experience sharing party,is the first of experience sharing partiesThe experience value of the individual participants,is the first of experience sharing partiesThe person weight of the individual participating person,a total number of participants in the experience sharing party;
when the experience index is greater than or equal to a preset experience index threshold, acquiring the sharing history of the experience sharing party;
evaluating the sharing history based on a preset sharing history evaluation template to obtain a sharing history evaluation value;
based on experience indexes and sharing history evaluation values, the credibility of the energy-saving control experience acquisition scene is acquired, and a specific acquisition formula is as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,the credibility of the scene is obtained for the energy-saving control experience, As an intermediate variable, the number of the variables,in order to share the history evaluation value,is a constant which is set in advance,for a preset minimum empirical index threshold,a preset maximum experience index threshold value;
and searching experience source content meeting the energy-saving control experience acquisition conditions from the energy-saving control experience acquisition scene when the reliability is greater than or equal to a preset reliability threshold.
The intelligent AI energy-saving control method applied to the data center provided by the embodiment of the invention comprises the following steps:
the AI reasoning server acquires the operation information of the machine room precise air conditioner and the real-time temperature and humidity information of the detected data center based on communication;
the AI reasoning server inputs the operation information and the real-time temperature and humidity information into a pre-trained IT machine room heat load prediction AI model to obtain a heat load trend of the data center for a preset time of future heat load;
the AI reasoning server inputs the heat load trend to a pre-trained machine room precise air conditioner AI energy-saving control model;
and the AI reasoning server performs energy-saving control on the precise air conditioner of the machine room based on communication by using the precise air conditioner AI energy-saving control model of the machine room.
Preferably, the intelligent AI energy-saving control method further comprises the following steps:
The AI reasoning server continuously acquires the actual heat load detected by the machine room precise air conditioner;
the AI reasoning server compares the thermal load difference between the actual thermal load and the thermal load trend;
the AI reasoning server inputs the heat load difference supplement to the AI energy-saving control model of the precise air conditioner of the machine room to obtain air conditioning adjustment parameters of each precise air conditioner of the machine room;
and the AI reasoning server issues the air conditioner adjusting parameters to the corresponding precise air conditioner of the machine room.
Preferably, the pre-training process of the AI energy-saving control model of the precise air conditioner of the machine room comprises the following steps:
acquiring equipment information of a precise air conditioner of a machine room;
acquiring a condition template based on equipment information and preset energy-saving control experience, and determining energy-saving control experience acquisition conditions;
acquiring energy-saving control experience conforming to energy-saving control experience acquisition conditions from a preset energy-saving control experience acquisition scene;
training the neural network model by taking the energy-saving control experience as a training sample until convergence;
and taking the converged neural network model as an AI energy-saving control model of the precise air conditioner of the machine room.
Preferably, the energy-saving control experience meeting the energy-saving control experience acquisition condition is acquired from a preset energy-saving control experience acquisition scene, which comprises the following steps:
Searching experience source content meeting the energy-saving control experience acquisition conditions from the energy-saving control experience acquisition scene;
analyzing the content type distribution of the experience source content; the content type distribution includes: at least one group of one-to-one first content type and first content proportion;
matching the content type distribution with a preset standard content type distribution to obtain a matching degree;
when the matching degree is greater than or equal to a preset matching degree threshold value, determining the content position of a second content proportion preset before in the content of the experience source, and extracting first local content in a first content range preset before and after the content position from the content of the experience source;
matching a first content sentence in the first local content with a standard content sentence in a preset standard content sentence library;
when the matching is met, taking the first content statement which is met by the matching as a first target content statement, and acquiring a preset content direction, a preset second content range and a preset second content type which correspond to the standard content statement which is met by the matching;
extracting second local content of a second content type in a second content range in the content direction of the first target content statement from experience source content;
Determining an experience evaluation value based on the second local content and a preset experience evaluation index library;
when the experience evaluation value is greater than or equal to a preset experience evaluation value threshold value, extracting all third local contents in the opposite direction of the content direction of the first target content sentence from experience source contents;
and sharing the semantic library based on the third local content and preset experience, and determining the energy-saving control experience.
Preferably, determining the experience evaluation value based on the second local content and a preset experience evaluation index library includes:
extracting a plurality of groups of experience evaluation indexes, coincidence-score tables and index weights which are in one-to-one correspondence from an experience evaluation index library;
determining the coincidence degree of the second local content to any one of the tested evaluation indexes;
determining a target score based on the fitness and the corresponding fitness-score table;
giving index weight corresponding to the target score to obtain an evaluation score;
and accumulating and calculating the evaluation score to obtain the experience evaluation value.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims thereof as well as the appended drawings.
The technical scheme of the invention is further described in detail through the drawings and the embodiments.
Drawings
The accompanying drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate the invention and together with the embodiments of the invention, serve to explain the invention. In the drawings:
FIG. 1 is a schematic diagram of an intelligent AI energy-saving control system applied to a data center in an embodiment of the invention;
FIG. 2 is a flowchart illustrating operations performed by the AI reasoning server in accordance with an embodiment of the invention;
FIG. 3 is a schematic diagram of an iEdge implementation in an embodiment of the present invention;
fig. 4 is a network diagram of POE implementation in an embodiment of the present invention.
Detailed Description
The preferred embodiments of the present invention will be described below with reference to the accompanying drawings, it being understood that the preferred embodiments described herein are for illustration and explanation of the present invention only, and are not intended to limit the present invention.
The embodiment of the invention provides an intelligent AI energy-saving control system applied to a data center, as shown in FIG. 1, comprising:
the AI reasoning server is arranged in the data center;
the edge control hosts are arranged on each machine room precise air conditioner in the data center one by one and are respectively in communication connection with the AI reasoning server and the machine room precise air conditioner arranged in the data center one by one;
The temperature and humidity sensors are arranged in a machine room cold channel in the data center one by one and are respectively in communication connection with the AI reasoning server one by one; the machine room cooling channel is communicated with a cooling output end of the machine room precise air conditioner;
as shown in fig. 2, the AI reasoning server performs the following operations in one execution period:
step S1: based on communication, acquiring operation information of the precise air conditioner of the machine room and real-time temperature and humidity information in a cold channel of the machine room; wherein, the operation information includes: air conditioning status data;
step S2: inputting the operation information and the real-time temperature and humidity information into a pre-trained IT machine room heat load prediction AI model to obtain a heat load trend of the data center for a future heat load preset time; the IT machine room thermal load prediction AI model is pre-trained based on a DNN algorithm model; the heat load trend is, for example: whether hot spots exist, the hot spot change trend and the like;
step S3: inputting the heat load trend into a pre-trained computer room precise air conditioner AI energy-saving control model to obtain air conditioner adjusting parameters of each computer room precise air conditioner; the AI energy-saving control model is pre-trained based on a fuzzy/DNN algorithm model; the air conditioning parameters are, for example: setting return air temperature, setting return air humidity and the like;
Step S4: and issuing the air conditioner adjusting parameters to the corresponding precise air conditioner in the machine room.
The working principle and the beneficial effects of the technical scheme are as follows:
the edge control host (iEdge) provides at least one serial port, is in serial port communication with the machine room precise air conditioner and is used for collecting operation information of the machine room precise air conditioner, and the other serial port is in serial port communication with the movable ring system, so that the movable ring system is not influenced and operation information of the machine room precise air conditioner is also collected. By the aid of the method, the system can directly collect the operation information of the precise air conditioner of the machine room through the edge control host, the movable ring system can continuously collect the operation information of the precise air conditioner of the machine room as usual, the movable ring system does not need to be in butt joint to obtain the operation information of the precise air conditioner of the machine room, software butt joint development and customization work is reduced, the problems that the direct butt joint of the software system is unstable and poor in instantaneity are avoided, and the influence on the original system 2 of a user is 0. The edge control host computer and the AI reasoning server are in communication connection and networking based on the POE switch, the POE switch is connected with each edge control host computer through a POE network cable, the network and the power supply are also carried out on each edge control host computer while communication is realized, a network interface and a power supply interface are not required to be arranged for the edge control host computers in a data center independently, the fact that the data center is not required to be re-wired is guaranteed to the greatest extent, convenience is improved, the problems of power supply and networking of iEdae are solved by the application of the POE technology, engineering is extremely simple, and the POE switch and the AI reasoning server can be networking through the network cable or optical fiber. The AI reasoning server directly inputs the collected operation information to the machine room precise air conditioner AI energy-saving control model, the machine room precise air conditioner AI energy-saving control model can directly take over to carry out energy-saving control on the machine room precise air conditioner, operation and maintenance personnel are not required to manually adjust operation parameters of the machine room precise air conditioner in the data center, labor cost is reduced, the situations that energy-saving control is not timely and unreasonable due to limited energy of the operation and maintenance personnel are avoided, and the efficiency of energy-saving control of the data center is improved. Real-time data (temperature and humidity real-time data and air conditioner operation real-time data) are input into an IT machine room heat load prediction AI model, a heat load trend of the IT machine room heat load for 1-5 minutes is predicted, the predicted heat load trend is input into an AI energy-saving control model of a machine room precise air conditioner, and adjustment parameters (aim to adjust air conditioner cold output in advance) of each machine room precise air conditioner are calculated.
The data center infrastructure equipment communication usually adopts a serial port communication mode, common protocols comprise RS-232, RS-422 and RS-485, and serial port communication lines are usually completed by using 3 lines: the method comprises the steps that (1) a ground wire, (2) a data wire is sent, and (3) the data wire is received, wherein equipment at two ends of serial port communication needs to adopt the same baud rate, data bit, stop bit and parity check, serial port communication is difficult to support 1-to-many communication except that serial communication efficiency is low, if one infrastructure equipment is collected by a dynamic ring system through serial ports, other systems cannot collect the equipment in a serial port communication mode directly, otherwise signal interference is caused, and both sides cannot normally communicate; when a plurality of systems need to communicate with equipment supporting serial ports, only one system can be directly in butt joint with the equipment, other systems are in butt joint with the system in direct butt joint through a software interface to exchange data, at present, a data center is directly in butt joint with a general dynamic ring system of infrastructure equipment, other systems can only indirectly acquire and control the infrastructure equipment in a mode of being in butt joint with the dynamic ring, a large number of software butt joint development and customization work are caused, and more importantly, the direct butt joint of the software system is unstable and the real-time performance is also very poor. The device supporting the serial port through the iEdge device can communicate with a plurality of upper computers, and comprises an existing dynamic ring system, as shown in fig. 3, the implementation principle is as follows:
(1) And (3) a wiring mode: the iEdge provides at least 2 serial ports, 1 is used for collection and is physically connected with the serial ports (ground wire, transmitting data wire and receiving data wire) of the managed infrastructure equipment, and the other 1 serial port is connected with the dynamic ring to collect the serial ports, which is the same as the physical connection.
(2) And (3) realizing software: the iEdge reads the serial port instruction issued by the dynamic ring system, and reads the acquisition/control instruction issued by the AI reasoning server (serial port protocol format, sending/receiving through network protocol) through the TCP/IP port, and the other main thread issues the current instruction to the managed infrastructure equipment through the serial port, acquires the return information, and sends the return information to the requester (dynamic ring system, or AI reasoning server, or other systems). And ensuring the stable operation of the iEdge working thread through daemon approach.
In the data center of stock, the rewiring transformation is difficult, and particularly an air conditioning room may not have a 220V mains plug, and is also usually provided for temporary operation and cannot be used as long-term power supply; in the scheme design, the POE technology is utilized to realize networking and power supply of the iEdge through the network cable, so that engineering construction difficulty of energy-saving transformation of a machine room and dependence on site conditions of the machine room are greatly reduced.
The specific scheme is shown in fig. 4:
(1) POE exchanger (can integrate to AI reasoning server, also can arrange POE exchanger alone) place in IT computer lab district, get the electricity and come from frame PDU, POE exchanger passes through POE net twine and connects every iEdge in the air conditioning room, and survey at iEdge and separate out 12V/24V/48V direct current power supply line, and hundred megabit/giga net twine through the POE separator, butt joint iEdge's power supply mouth and hundred megabit/giga net gape.
(2) Even if the air conditioner is not provided with a network interface and a power supply interface, the iEdge deployed beside each air conditioner can solve the problems of network and power supply, in addition, the iEdge can be installed in each air conditioner to support magnetic installation or track installation, so that the reliability and easy maintenance of physical links and the clean appearance are ensured.
(3) The network between the POE switch and the AI reasoning server can be networked, or can be networked by network cable or optical fiber remote (and the AI reasoning server can be arranged in different machine rooms, so long as the network is communicated).
In one embodiment, the AI reasoning server performs the following operations at the next execution cycle:
continuously acquiring the actual heat load detected by the machine room precise air conditioner;
comparing the thermal load difference between the actual thermal load and the thermal load trend;
The heat load difference is supplemented and input into the AI energy-saving control model of the precise air conditioner of the machine room;
and the computer room precise air conditioner AI energy-saving control model performs energy-saving control compensation on the computer room precise air conditioner based on communication.
In the next execution period, the AI performs corresponding control compensation according to the heat load difference between the actual heat load and the heat load trend of the data center detected by the machine room precise air conditioner, and realizes a Online Onplicy Reinforcement Learning framework.
Preferably, the pre-training process of the IT machine room thermal load prediction AI model includes:
acquiring a large amount of heat load prediction experience; the heat load prediction experience can be a great number of heat load records of a data center in history, experience of expert personnel for heat load prediction and the like;
training the initial AI model by taking a large amount of heat load prediction experience as a training sample until convergence; the initial AI model may be a neural network model or the like;
and taking the converged initial AI model as the IT machine room heat load prediction AI model. After training to converge, the initial AI model may be used for thermal load prediction.
In one embodiment, the pre-training process of the computer room precise air conditioner AI energy-saving control model comprises the following steps:
Based on online reinforcement learning, pre-training the AI energy-saving control model of the precise air conditioner of the machine room;
based on online reinforcement learning, the machine room precise air conditioner AI energy-saving control model is pre-trained, and the machine room precise air conditioner AI energy-saving control model comprises:
acquiring a large number of historical temperature regulation records of the precise air conditioner of the machine room; the historical attemperation record includes: a first tempering target and tempering process; the first temperature adjustment target is specifically, for example: adjusting the room temperature to 15 ℃; the temperature adjusting process specifically includes, for example: the room temperature is adjusted from 20 ℃ to 15 ℃ for a long time;
determining the refrigeration performance of the machine room precise air conditioner based on the first temperature regulation target and the temperature regulation process; the refrigerating performance is measured by the time spent by the machine room precise air conditioner for adjusting the room temperature to the temperature adjusting target;
acquiring a preset temperature regulation target set; the set of tempering targets includes: a plurality of second tempering targets; the second temperature regulation target is different in temperature;
based on the refrigeration performance, determining an optimal air conditioner control scheme of the machine room precise air conditioner under any second temperature regulation target; the optimal air conditioner control scheme is a control scheme for adjusting different room temperatures to a second temperature adjustment target;
Pairing the second temperature regulation targets with the optimal air conditioner control scheme one by one to obtain a plurality of pairing groups;
training the neural network model by taking the pairing groups as training samples until convergence;
and taking the neural network model as the computer room precise air conditioner AI energy-saving control model to finish pre-training.
In one embodiment, the pre-training process of the computer room precise air conditioner AI energy-saving control model comprises the following steps:
acquiring equipment information of a precise air conditioner of a machine room; wherein the device information includes: host model, historical working condition, historical maintenance condition and the like of the machine room precise air conditioner;
acquiring a condition template based on equipment information and preset energy-saving control experience, and determining energy-saving control experience acquisition conditions; the energy-saving control experience acquisition condition template is a template for converting equipment information into energy-saving control experience acquisition conditions, and specifically, for example: the equipment information is the host model xx, the energy-saving control experience acquisition condition template is "the equipment with the host model xx is suitable for the energy-saving control experience";
acquiring energy-saving control experience conforming to energy-saving control experience acquisition conditions from a preset energy-saving control experience acquisition scene; the energy-saving control experience acquisition scene can be a forum, a bar, a WeChat/QQ group and the like for energy-saving control sharing by data center operation and maintenance personnel; the energy-saving control experience is the experience of manually adjusting the operation parameters of the machine room precise air conditioner in the data center by a large amount of operation and maintenance personnel in history, and particularly, the experience can be operation parameter adjustment logic and the like;
Training the neural network model by taking the energy-saving control experience as a training sample until convergence; after the neural network model is trained to be converged based on energy-saving control experience, the neural network model can learn how to manually adjust the operation parameters of the machine room precise air conditioner in the data center, and when the operation information of the machine room precise air conditioner is input, the neural network model can automatically control the energy saving of the machine room precise air conditioner;
and taking the converged neural network model as an AI energy-saving control model of the precise air conditioner of the machine room.
The working principle and the beneficial effects of the technical scheme are as follows:
generally, if some rules for performing energy-saving control on the precise air conditioner of the machine room in the data center are directly set to control the precise air conditioner of the machine room, the operation of the precise air conditioner of the machine room cannot meet the manual normal operation scope, so that quality assurance and the like of manufacturers on the precise air conditioner of the machine room are affected. According to the embodiment of the invention, the neural network model is trained by utilizing the energy-saving control experience of energy-saving control of the machine room precise air conditioner in the manual history, and the trained machine room precise air conditioner AI energy-saving control model can keep consistent with the logic of manual normal operation when the machine room precise air conditioner is energy-saving controlled, so that the quality assurance of the machine room precise air conditioner by manufacturers and the like are prevented from being influenced. And secondly, energy-saving control experience acquisition conditions are introduced, and model training is performed by pertinently acquiring energy-saving control experiences applicable to the data center, so that the applicability of an energy-saving reliable value AI model to the current data center is improved. In one embodiment, acquiring the energy saving control experience conforming to the energy saving control experience acquisition condition from the preset energy saving control experience acquisition scene includes:
Searching experience source content meeting the energy-saving control experience acquisition conditions from the energy-saving control experience acquisition scene; the experience source content accords with the energy-saving control experience acquisition condition, and is an energy-saving control sharing forum, a bar, experience sharing posting in WeChat/QQ group, chat speaking and the like for operation and maintenance personnel;
analyzing the content type distribution of the experience source content; the content type distribution includes: at least one group of one-to-one first content type and first content proportion; wherein the first content type comprises: text, charts, live images, etc.; the first content proportion is the ratio of the content of the first content type to the content of the whole experience source;
matching the content type distribution with a preset standard content type distribution to obtain a matching degree; wherein, the standard content types are distributed as a complete energy-saving control experience to share the content types and the corresponding content proportions, for example: sharing and introducing an energy-saving control scheme, attaching a graph for comparing energy-saving effects (for example, comparing the power consumption of a data center before the energy-saving control scheme is executed with the power consumption of the data center after the energy-saving control scheme is executed) and the like, wherein the standard content types are distributed into 70% of characters and 30% of the graph;
When the matching degree is greater than or equal to a preset matching degree threshold value, determining the content position of a second content proportion preset before in the content of the experience source, and extracting first local content in a first content range preset before and after the content position from the content of the experience source; the preset second content ratio is a ratio of the content of experience sources, for example: when sharing the energy-saving control experience, the operation and maintenance personnel generally introduces an energy-saving control scheme, and attaches a graph with energy-saving effect comparison to the latter half as a data support, so that the second content proportion is 70%, and the corresponding content position is the position of the first 70% of the experience source content;
matching a first content sentence in the first local content with a standard content sentence in a preset standard content sentence library; the standard content sentences are sentences for introducing and leading out a graph for comparing the energy saving effect, and specifically, for example: the energy-saving effect comparison table of the shared energy-saving control scheme is seen below;
when the matching is met, taking the first content statement which is met by the matching as a first target content statement, and acquiring a preset content direction, a preset second content range and a preset second content type which correspond to the standard content statement which is met by the matching; the content direction reflects in which direction of the first target content sentence the graph introducing and leading out the energy saving effect contrast is, for example: the standard content statement is "look at the energy-saving effect comparison table of the energy-saving control scheme shared by me below", then the content direction is back; the second content type is a content type of a graph reflecting introduction and leading out energy saving effect comparison, for example: the standard content statement is "look at the energy-saving effect comparison table of the energy-saving control scheme shared by me, and the second content type is a table;
Extracting second local content of a second content type in a second content range in the content direction of the first target content statement from experience source content;
determining an experience evaluation value based on the second local content and a preset experience evaluation index library; the larger the experience evaluation value is, the better the energy-saving effect of the energy-saving control experience in the experience source content is;
when the experience evaluation value is greater than or equal to a preset experience evaluation value threshold value, extracting all third local contents in the opposite direction of the content direction of the first target content sentence from experience source contents; if the operation and maintenance personnel provide energy-saving effect comparison after sharing, an energy-saving control scheme is introduced before, so that a fourth local content is extracted in the direction opposite to the content direction of the first target content statement;
and sharing the semantic library based on the third local content and preset experience, and determining the energy-saving control experience.
The working principle and the beneficial effects of the technical scheme are as follows:
generally, the energy-saving control experience is uploaded after the operation and maintenance personnel are required to carry out experience description arrangement, so that the labor cost is high, in addition, the number of operation and maintenance personnel of an internal department is limited, and the uploaded energy-saving control experience may not be comprehensive. Normally, the operation and maintenance personnel actively share the energy-saving control experience with the internal personnel in the time of historic work, for example: the energy-saving control experience is shared in an energy-saving control experience acquisition scene by means of internal forum posting, communication and communication in a WeChat nail group and the like. The embodiment of the invention designs the method for acquiring the energy-saving control experience from the energy-saving control experience acquisition scene, improves the comprehensiveness of the energy-saving control experience acquisition, can perform the energy-saving control experience acquisition scene sharing among different enterprises, further improves the comprehensiveness of the energy-saving control experience acquisition, and can indirectly improve the working capacity of the computer room precise air conditioner AI energy-saving control model trained by the energy-saving control experience. And secondly, when the energy-saving control experience is extracted from the energy-saving control experience acquisition scene, firstly, introducing content type distribution and standard content type distribution, and rapidly determining experience source content shared by complete energy-saving control experience, thereby reducing extraction resources for extracting the energy-saving control experience from the energy-saving control experience acquisition scene and improving extraction efficiency. And then, introducing standard content sentences, preset content directions, preset second content ranges and preset second content types, rapidly determining the demonstration support information of the operation and maintenance personnel sharing the energy-saving control experience on the experience shared by the operation and maintenance personnel, introducing an experience evaluation index library, evaluating the experience quality degree, searching for the introduction content of the energy-saving control scheme when the experience evaluation value is greater than or equal to a preset experience evaluation value threshold, and determining the energy-saving control experience by combining the experience sharing semantic library. The accuracy and the acquisition quality of the energy-saving control experience are greatly prompted.
In one embodiment, determining the empirical evaluation value based on the second local content and a preset empirical evaluation index base includes:
extracting a plurality of groups of experience evaluation indexes, coincidence-score tables and index weights which are in one-to-one correspondence from an experience evaluation index library; the experience evaluation index is an index for evaluating the demonstration support information of the experience shared by the operation and maintenance personnel for sharing the energy-saving control experience, and specifically, for example: the difference of the real-time power consumption before and after the execution of the energy-saving control scheme is more than 2kw, and the more the difference of the real-time power consumption before and after the execution of the energy-saving control scheme in the second local content exceeds 2kw, the higher the coincidence degree is; the coincidence-score table has scores corresponding to different coincidence degrees, and the higher the coincidence degree is, the higher the score is; the higher the index weight is, the more the target score under the index of representing experience evaluation can represent the excellent degree of the energy-saving control experience shared by operation and maintenance personnel;
determining the coincidence degree of the second local content to any one of the tested evaluation indexes;
determining a target score based on the fitness and the corresponding fitness-score table;
giving index weight corresponding to the target score to obtain an evaluation score; when the method is endowed, multiplying the index weight with the target score to obtain an evaluation score;
And accumulating and calculating the evaluation score to obtain the experience evaluation value.
The working principle and the beneficial effects of the technical scheme are as follows:
experience evaluation indexes, a coincidence-score table and index weights are introduced, so that the comprehensiveness and the accuracy of experience evaluation are improved.
In one embodiment, the determined energy saving control experience based on the third local content and the preset experience sharing semantic library comprises:
extracting a plurality of groups of experience sharing semantics and experience descriptions which are in one-to-one correspondence from an experience sharing semantic library; the experience sharing semantics are semantics of performing energy-saving control experience sharing, and experience description is energy-saving control experience of experience sharing semantic reaction, specifically, for example: experience sharing semantics are that the host operating frequency with lower task priority is immediately reduced, and corresponding experience shows that the host operating frequency with lower task priority is reduced;
based on semantic analysis technology, carrying out semantic analysis on second content sentences in the third local content to obtain content semantics;
matching the content semantics with any one of the tested shared semantics;
when the matching is met, taking a second content sentence corresponding to the matched content semantics as a second target content sentence, and taking an experience description corresponding to the matched experience sharing semantics as a target experience description;
Acquiring statement position relations among second target content statements in the third local content; the sentence position relationship is the relative position relationship of the second target content sentence in the third local content;
determining a standard logic keyword sequence based on the sentence position relation and a preset standard logic keyword sequence library; the standard logic keyword sequence library has standard logic keyword sequences corresponding to different sentence position relations, specifically, for example: the sentence position relationship is A, B, C, and the standard logic keyword sequences are first, then, last and the like;
extracting logic keywords in the second target content sentences, and arranging according to sentence position relations to obtain a logic keyword sequence; wherein, the "logic keyword" is a keyword reflecting logic before and after the operation of the energy-saving control scheme, for example: "first", "then", etc.;
matching the logic keyword sequence with a standard logic keyword sequence;
when the target experience description is matched and met, based on a preset first logic relation corresponding to a standard logic keyword sequence matched and met, carrying out logic integration on the target experience description to obtain energy-saving control experience; the first logic relationship is logic before and after the operation of the energy-saving control scheme of the standard logic keyword sequence reaction, for example: the standard logic keyword sequence is 'first, next and last', and then the first logic relationship is first … …, next … … and last … …; when integrating the target experience explanation, carrying out permutation integration based on the logic keywords in the target experience explanation and combining the first logic relation;
Otherwise, extracting a second logic relation between target experience descriptions from the experience sharing semantic library; when the experience is not matched, the description operation staff does not have front and back logic (for example, sequence) between the described sentences when sharing the energy-saving control experience, and the experience sharing semantic library has a second logic relationship between target experience descriptions, specifically, for example: the target experience shows that the second logic relationship is independent when the air conditioner in the data center is closed and the exhaust fan in the data center is opened, and the two logic relationships are not required to be distinguished;
and based on the second logic relation, performing logic integration on the target experience description to obtain the energy-saving control experience.
The working principle and the beneficial effects of the technical scheme are as follows:
normally, when sharing the energy-saving control scheme, an operation and maintenance person sharing the energy-saving control experience can insert more content irrelevant to the energy-saving control experience, for example: the method comprises the steps of turning off part of air conditioners in a data center, wherein the air conditioners consume too much energy, and the air conditioners consume too much energy in the middle and are irrelevant. Therefore, the embodiment of the invention introduces experience sharing semantics to quickly determine the second target content statement related to the energy-saving control experience. The determination efficiency of the energy-saving control experience determination is improved. In addition, the energy-saving control experience shared by operation and maintenance personnel is often a logic process, and the target experience description needs to be logically integrated based on the experience description of the operation and maintenance personnel to serve as the energy-saving control experience.
In one embodiment, before searching the experience source content meeting the energy-saving control experience acquisition condition from the energy-saving control experience acquisition scene, the method further comprises:
the method comprises the steps of obtaining an experience index of an experience sharing party in an energy-saving control experience obtaining scene, wherein a specific obtaining formula is as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,for the experience index of the experience sharing party,is the first of experience sharing partiesThe experience value of the individual participants,is the first of experience sharing partiesThe person weight of the individual participating person,a total number of participants in the experience sharing party; normally, when sharing the energy-saving control experience, it is possible that a plurality of operation and maintenance personnel participate in the sharing process together (share that they have commonly performed on a certain data centerHistory of energy-saving control), therefore, the experience sharing party may have a plurality of participants, the greater the experience value of the participant is related to the working years, job levels, etc., the higher the specific working years, job levels, etc., the experience value may be preset manually, the higher the person weight is related to the number of historical posts/utterances of the participant, and the greater the number of historical posts/utterances is, the higher the person weight is, and the person weight may be preset manually;
When the experience index is greater than or equal to a preset experience index threshold, acquiring the sharing history of the experience sharing party; the sharing history is other people's evaluation of the energy-saving control experience historically shared by the experience sharing party, and the like;
evaluating the sharing history based on a preset sharing history evaluation template to obtain a sharing history evaluation value; the sharing history evaluation template is a template for evaluating the overall situation of the experience sharing party in the history of energy-saving control according to the sharing history, specifically, for example: the worse the experience sharing party historically shares the other people's evaluation of the energy-saving control experience (for example, the effect is not achieved after the reference is implemented, etc.), the lower the historical evaluation value is;
based on experience indexes and sharing history evaluation values, the credibility of the energy-saving control experience acquisition scene is acquired, and a specific acquisition formula is as follows:
；
the intermediate variables are set to indicate that the more valuable the experience of the energy-saving control is shared by the experience sharing party, the higher the reliability is when the experience value is smaller, and the greater the obtained sharing history evaluation value is; when the experience value is larger, the obtained sharing history evaluation value is smaller, which means that the experience sharing party shares the energy-saving control experience with worse evaluation, the less the experience is supposed to be, and the lower the credibility is;
Wherein, the liquid crystal display device comprises a liquid crystal display device,the credibility of the scene is obtained for the energy-saving control experience,as an intermediate variable, the number of the variables,in order to share the history evaluation value,is a constant which is set in advance,for a preset minimum empirical index threshold,a preset maximum experience index threshold value;
and searching experience source content meeting the energy-saving control experience acquisition conditions from the energy-saving control experience acquisition scene when the reliability is greater than or equal to a preset reliability threshold.
The working principle and the beneficial effects of the technical scheme are as follows:
before searching experience source content meeting the energy-saving control experience acquisition conditions from the energy-saving control experience acquisition scene, the embodiment of the invention calculates the credibility of the energy-saving control experience acquisition scene, and acquires the information from the energy-saving control experience acquisition scene when the credibility is greater than or equal to a preset credibility threshold value, thereby improving the accuracy and the quality of acquiring the experience source content. Normally, in forums, bar posts, weChat/QQ/nail groups and the like for sharing the energy-saving control experience of a data center by operation and maintenance personnel, the energy-saving control experience with lower quality (mainly poor energy-saving effect after reference implementation) is inevitably shared, along with the input and use of the system, the credibility of an energy-saving control experience acquisition scene is continuously updated, whether the experience source content meeting the energy-saving control experience acquisition condition is required to be searched from the energy-saving control experience acquisition scene is judged based on the credibility, and the method is particularly applicable.
The embodiment of the invention provides an intelligent AI energy-saving control method applied to a data center, which comprises the following steps:
the AI reasoning server acquires the operation information of the machine room precise air conditioner and the real-time temperature and humidity information of the detected data center based on communication;
the AI reasoning server inputs the operation information and the real-time temperature and humidity information into a pre-trained IT machine room heat load prediction AI model to obtain a heat load trend of the data center for a preset time of future heat load;
the AI reasoning server inputs the heat load trend to a pre-trained machine room precise air conditioner AI energy-saving control model to obtain air conditioner adjusting parameters of each machine room precise air conditioner;
and the AI reasoning server issues the air conditioner adjusting parameters to the corresponding precise air conditioner of the machine room.
In one embodiment, the intelligent AI energy-saving control method further includes:
the AI reasoning server continuously acquires the actual heat load detected by the machine room precise air conditioner;
the AI reasoning server compares the thermal load difference between the actual thermal load and the thermal load trend;
the AI reasoning server supplements and inputs the heat load difference into the AI energy-saving control model of the precise air conditioner of the machine room;
And the AI reasoning server performs energy-saving control compensation on the precise air conditioner of the machine room based on communication by the AI energy-saving control model of the precise air conditioner of the machine room.
In one embodiment, the pre-training process of the computer room precise air conditioner AI energy-saving control model comprises the following steps:
acquiring equipment information of a precise air conditioner of a machine room;
acquiring a condition template based on equipment information and preset energy-saving control experience, and determining energy-saving control experience acquisition conditions;
acquiring energy-saving control experience conforming to energy-saving control experience acquisition conditions from a preset energy-saving control experience acquisition scene;
training the neural network model by taking the energy-saving control experience as a training sample until convergence;
and taking the converged neural network model as an AI energy-saving control model of the precise air conditioner of the machine room.
In one embodiment, acquiring the energy saving control experience conforming to the energy saving control experience acquisition condition from the preset energy saving control experience acquisition scene includes:
searching experience source content meeting the energy-saving control experience acquisition conditions from the energy-saving control experience acquisition scene;
analyzing the content type distribution of the experience source content; the content type distribution includes: at least one group of one-to-one first content type and first content proportion;
Matching the content type distribution with a preset standard content type distribution to obtain a matching degree;
when the matching degree is greater than or equal to a preset matching degree threshold value, determining the content position of a second content proportion preset before in the content of the experience source, and extracting first local content in a first content range preset before and after the content position from the content of the experience source;
matching a first content sentence in the first local content with a standard content sentence in a preset standard content sentence library;
when the matching is met, taking the first content statement which is met by the matching as a first target content statement, and acquiring a preset content direction, a preset second content range and a preset second content type which correspond to the standard content statement which is met by the matching;
extracting second local content of a second content type in a second content range in the content direction of the first target content statement from experience source content;
determining an experience evaluation value based on the second local content and a preset experience evaluation index library;
when the experience evaluation value is greater than or equal to a preset experience evaluation value threshold value, extracting all third local contents in the opposite direction of the content direction of the first target content sentence from experience source contents;
And sharing the semantic library based on the third local content and preset experience, and determining the energy-saving control experience.
In one embodiment, determining the empirical evaluation value based on the second local content and a preset empirical evaluation index base includes:
extracting a plurality of groups of experience evaluation indexes, coincidence-score tables and index weights which are in one-to-one correspondence from an experience evaluation index library;
determining the coincidence degree of the second local content to any one of the tested evaluation indexes;
determining a target score based on the fitness and the corresponding fitness-score table;
giving index weight corresponding to the target score to obtain an evaluation score;
and accumulating and calculating the evaluation score to obtain the experience evaluation value.
It will be apparent to those skilled in the art that various modifications and variations can be made to the present invention without departing from the spirit or scope of the invention. Thus, it is intended that the present invention also include such modifications and alterations insofar as they come within the scope of the appended claims or the equivalents thereof.

Claims (10)

1. Intelligent AI energy-saving control system for data center, characterized by comprising:
the AI reasoning server is arranged in the data center;
the edge control hosts are arranged on each machine room precise air conditioner in the data center one by one and are respectively in communication connection with the AI reasoning server and the machine room precise air conditioner arranged in the data center one by one;
The temperature and humidity sensors are arranged in a machine room cold channel in the data center one by one and are respectively in communication connection with the AI reasoning server one by one; the machine room cooling channel is communicated with a cooling output end of the machine room precise air conditioner;
the AI reasoning server executes the following operations in one execution period:
based on communication, acquiring operation information of the precise air conditioner of the machine room and real-time temperature and humidity information in a cold channel of the machine room;
inputting the operation information and the real-time temperature and humidity information into a pre-trained IT machine room heat load prediction AI model to obtain a heat load trend of the data center for a future heat load preset time;
inputting the heat load trend into a pre-trained computer room precise air conditioner AI energy-saving control model to obtain air conditioner adjusting parameters of each computer room precise air conditioner;
and issuing the air conditioner adjusting parameters to the corresponding precise air conditioner in the machine room.
2. The intelligent AI energy-saving control system for a data center of claim 1, wherein the AI reasoning server performs the following operations in the next execution cycle:
continuously acquiring the actual heat load detected by the machine room precise air conditioner;
Comparing the thermal load difference between the actual thermal load and the thermal load trend;
the heat load difference is supplemented and input into the AI energy-saving control model of the precise air conditioner of the machine room;
and the computer room precise air conditioner AI energy-saving control model performs energy-saving control compensation on the computer room precise air conditioner based on communication.
3. The intelligent AI energy-saving control system for a data center of claim 1, wherein the pre-training process of the IT machine room thermal load prediction AI model comprises:
acquiring a large amount of heat load prediction experience;
training the initial AI model by taking a large amount of heat load prediction experience as a training sample until convergence;
and taking the converged initial AI model as the IT machine room heat load prediction AI model.
4. The intelligent AI energy-saving control system for a data center of claim 1, wherein the pre-training process of the machine room precise air conditioner AI energy-saving control model comprises the following steps:
based on online reinforcement learning, pre-training the AI energy-saving control model of the precise air conditioner of the machine room;
based on online reinforcement learning, the machine room precise air conditioner AI energy-saving control model is pre-trained, and the machine room precise air conditioner AI energy-saving control model comprises:
Acquiring a temperature regulation record of the precise air conditioner of the machine room in a previous execution period; the temperature adjustment record includes: temperature regulation expected information and temperature regulation actual information;
training the neural network model by using the expected temperature regulation information and the actual temperature regulation information until convergence;
and taking the converged neural network model as the energy-saving control model of the precise air conditioner AI of the machine room to finish pre-training.
5. The intelligent AI energy-saving control system for a data center of claim 1, wherein the pre-training process of the machine room precise air conditioner AI energy-saving control model comprises the following steps:
acquiring equipment information of the precise air conditioner of the machine room;
acquiring a condition template based on the equipment information and a preset energy-saving control experience, and determining an energy-saving control experience acquisition condition;
acquiring energy-saving control experience conforming to the energy-saving control experience acquisition conditions from a preset energy-saving control experience acquisition scene; the energy-saving control experience comprises the following steps: determining experience of air conditioner adjusting parameters in the next execution period of the machine room precise air conditioner based on temperature adjustment expectation and temperature adjustment reality of the machine room precise air conditioner;
training the neural network model by taking the energy-saving control experience as a training sample until convergence;
And taking the converged neural network model as the computer room precise air conditioner AI energy-saving control model.
6. The intelligent AI energy-saving control system for a data center of claim 5, wherein acquiring energy-saving control experiences conforming to the energy-saving control experience acquisition conditions from a preset energy-saving control experience acquisition scene comprises:
searching experience source content meeting the energy-saving control experience acquisition conditions from the energy-saving control experience acquisition scene;
analyzing the content type distribution of the experience source content; the content type distribution includes: at least one group of one-to-one first content type and first content proportion;
matching the content type distribution with a preset standard content type distribution to obtain a matching degree;
when the matching degree is greater than or equal to a preset matching degree threshold value, determining a content position of a preset second content proportion in the experience source content, and extracting first local content in a preset first content range before and after the content position from the experience source content;
matching a first content sentence in the first local content with a standard content sentence in a preset standard content sentence library;
When the content is matched and met, taking the first content statement matched and met as a first target content statement, and acquiring a preset content direction, a preset second content range and a preset second content type corresponding to the standard content statement matched and met;
extracting second local content of the second content type within the second content range in the content direction of the first target content sentence from the experience source content;
determining an experience evaluation value based on the second local content and a preset experience evaluation index library;
when the experience evaluation value is greater than or equal to a preset experience evaluation value threshold value, extracting all third local contents in the reverse direction of the content direction of the first target content sentence from the experience source content;
and determining the energy-saving control experience based on the third local content and a preset experience sharing semantic library.
7. The intelligent AI energy-saving control system for a data center of claim 6, wherein determining an empirical evaluation value based on the second local content and a preset empirical evaluation index base comprises:
extracting a plurality of groups of experience evaluation indexes, coincidence-score tables and index weights which are in one-to-one correspondence from the experience evaluation index library;
Determining the coincidence degree of the second local content to any one of the experience evaluation indexes;
determining a target score based on the correspondence and the corresponding correspondence-score table;
giving the index weight corresponding to the target score to obtain an evaluation score;
and accumulating and calculating the evaluation score to obtain the experience evaluation value.
8. The intelligent AI energy-saving control system for a data center of claim 6, wherein determining the energy-saving control experience based on the third local content and a pre-determined experience-sharing semantic library comprises:
extracting a plurality of groups of experience sharing semantics and experience descriptions which are in one-to-one correspondence from the experience sharing semantics library;
based on semantic analysis technology, carrying out semantic analysis on the second content statement in the third local content to obtain content semantics;
matching the content semantics with any experience sharing semantics;
when the experience sharing semantic is matched and matched, taking the second content statement corresponding to the matched and matched content semantic as a second target content statement, and taking the experience description corresponding to the matched and matched experience sharing semantic as a target experience description;
acquiring statement position relations among the second target content statements in the third local content;
Determining a standard logic keyword sequence based on the sentence position relation and a preset standard logic keyword sequence library;
extracting logic keywords in the second target content sentences, and arranging according to the sentence position relation to obtain a logic keyword sequence;
matching the logic keyword sequence with the standard logic keyword sequence;
when the target experience description is matched and met, carrying out logic integration on the target experience description based on a preset first logic relation corresponding to the standard logic keyword sequence matched and met to obtain the energy-saving control experience;
otherwise, extracting a second logic relation between the target experience descriptions from the experience sharing semantic library;
and based on the second logic relation, logically integrating the target experience specification to obtain the energy-saving control experience.
9. The intelligent AI energy-saving control system for a data center of claim 6, further comprising, prior to searching for experience source content from the energy-saving control experience acquisition scene that meets the energy-saving control experience acquisition condition:
the experience index of the experience sharing party in the energy-saving control experience acquisition scene is acquired, and a specific acquisition formula is as follows: Wherein (1)>For the experience index of the experience sharing party, < +.>For +.>Experience value of individual participants,/->The +.f. in the experience sharing party>Personnel weight of individual participants, +.>A total number of participants in the experience sharing party;
when the experience index is greater than or equal to a preset experience index threshold, acquiring the sharing history of the experience sharing party;
evaluating the sharing history based on a preset sharing history evaluation template to obtain a sharing history evaluation value;
based on the experience index and the sharing history evaluation value, the credibility of the energy-saving control experience acquisition scene is acquired, and a specific acquisition formula is as follows:，/>wherein (1)>Obtaining the credibility of a scene for the energy-saving control experience,/->Is an intermediate variable +.>For the sharing of the history evaluation value,for a preset constant, ++>For a preset minimum empirical index threshold, < +.>For a preset maximum empirical index thresholdA value; and searching experience source content meeting the energy-saving control experience acquisition conditions from the energy-saving control experience acquisition scene when the credibility is greater than or equal to a preset credibility threshold.
10. The intelligent AI energy-saving control method applied to the data center is characterized by comprising the following steps:
The AI reasoning server acquires the operation information of the precise air conditioner of the machine room and the real-time temperature and humidity information of the detected data center based on communication;
the AI reasoning server inputs the operation information and the real-time temperature and humidity information into a pre-trained IT machine room heat load prediction AI model to obtain a heat load trend of the data center for a preset time of future heat load;
the AI reasoning server inputs the heat load trend to a pre-trained machine room precise air conditioner AI energy-saving control model to obtain air conditioner adjusting parameters of each machine room precise air conditioner;
and the AI reasoning server issues the air conditioner adjusting parameters to the corresponding precise air conditioner of the machine room.
CN202310799650.6A 2023-07-03 2023-07-03 Intelligent AI energy-saving control system and method applied to data center Active CN116527416B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310799650.6A CN116527416B (en) 2023-07-03 2023-07-03 Intelligent AI energy-saving control system and method applied to data center

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310799650.6A CN116527416B (en) 2023-07-03 2023-07-03 Intelligent AI energy-saving control system and method applied to data center

Publications (2)

Publication Number Publication Date
CN116527416A true CN116527416A (en) 2023-08-01
CN116527416B CN116527416B (en) 2023-09-01

Family

ID=87403261

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310799650.6A Active CN116527416B (en) 2023-07-03 2023-07-03 Intelligent AI energy-saving control system and method applied to data center

Country Status (1)

Country Link
CN (1) CN116527416B (en)

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2013154556A1 (en) * 2012-04-11 2013-10-17 Empire Technology Development Llc Data center access and management settings transfer
CN109800066A (en) * 2018-12-13 2019-05-24 中国科学院信息工程研究所 A kind of data center's energy-saving scheduling method and system
CN111237989A (en) * 2020-02-04 2020-06-05 青岛海信网络科技股份有限公司 Building ventilation air conditioner control method and device based on load prediction
CN114001443A (en) * 2021-09-27 2022-02-01 中盈优创资讯科技有限公司 Machine room energy saving method and device based on AI
CN114916209A (en) * 2022-06-06 2022-08-16 深圳市云海网能科技有限公司 System and method for realizing micro-module energy-saving control based on AI technology
CN115405996A (en) * 2022-08-30 2022-11-29 北京智能建筑科技有限公司 Method and device for controlling low-carbon operation of heat supply system and nonvolatile storage medium
US20230132431A1 (en) * 2021-11-01 2023-05-04 Samsung Electronics Co., Ltd. Method for controlling surface heat generation electronic device and storage medium therefor
CN116193819A (en) * 2023-01-19 2023-05-30 中国长江三峡集团有限公司 Energy-saving control method, system and device for data center machine room and electronic equipment

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2013154556A1 (en) * 2012-04-11 2013-10-17 Empire Technology Development Llc Data center access and management settings transfer
CN109800066A (en) * 2018-12-13 2019-05-24 中国科学院信息工程研究所 A kind of data center's energy-saving scheduling method and system
CN111237989A (en) * 2020-02-04 2020-06-05 青岛海信网络科技股份有限公司 Building ventilation air conditioner control method and device based on load prediction
CN114001443A (en) * 2021-09-27 2022-02-01 中盈优创资讯科技有限公司 Machine room energy saving method and device based on AI
US20230132431A1 (en) * 2021-11-01 2023-05-04 Samsung Electronics Co., Ltd. Method for controlling surface heat generation electronic device and storage medium therefor
CN114916209A (en) * 2022-06-06 2022-08-16 深圳市云海网能科技有限公司 System and method for realizing micro-module energy-saving control based on AI technology
CN115405996A (en) * 2022-08-30 2022-11-29 北京智能建筑科技有限公司 Method and device for controlling low-carbon operation of heat supply system and nonvolatile storage medium
CN116193819A (en) * 2023-01-19 2023-05-30 中国长江三峡集团有限公司 Energy-saving control method, system and device for data center machine room and electronic equipment

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
张琦;谢慕君;贾其臣;曹开发;: "热负荷预测中最有影响力参数测定", 长春工业大学学报, no. 02 *

Also Published As

Publication number Publication date
CN116527416B (en) 2023-09-01

Similar Documents

Publication Publication Date Title
US20190302713A1 (en) Method and device for automatically diagnosing and controlling apparatus in intelligent building
Lee et al. Energy savings by energy management systems: A review
CN101344779B (en) Network type remote monitoring management system for air-conditioner device
CN110836514B (en) Control method and device of air conditioning unit
CN107942960A (en) A kind of intelligentized information processing system
CN110030674A (en) A kind of wisdom classroom air-conditioner control system and control method
CN112365030A (en) Building energy consumption management method and system, electronic equipment and computer storage medium
CN110220288A (en) Central air-conditioning system intelligent optimized control method and device based on big data cloud platform
CN109140723A (en) A kind of distribution building HVAC monitoring system and method
CN107449100A (en) Central air-conditioning flexibility peak regulation long-range control method based on distributed structure/architecture
CN111780332B (en) Household metering method and device for central air conditioner
CN112113314A (en) Real-time temperature data acquisition system and temperature adjusting method based on learning model
CN112329338B (en) Cold source system control method and device based on fish swarm optimization BP neural network
CN115220351B (en) Intelligent energy-saving optimization control method for building air conditioning system based on cloud side end
CN111158245A (en) Building energy-saving monitoring and management method based on Internet of things
CN113465121A (en) Remote monitoring system based on air conditioner of Internet of things and control method thereof
CN115272675A (en) Energy management system and method based on multi-sensor information fusion
CN116527416B (en) Intelligent AI energy-saving control system and method applied to data center
CN109253514A (en) A kind of intelligent residence cooling system
CN112484255B (en) Energy-saving heating ventilation air conditioning system and building automatic control method
Park et al. Energy-saving decision making framework for HVAC with usage logs
CN115496628A (en) Smart hotel management system
CN110823291A (en) Method and system for monitoring indoor temperature and humidity environment of building based on K-means clustering algorithm
CN113701232B (en) Heat supply system building-level regulation and control method and system based on temperature diversity analysis
CN114548445A (en) Building operation and maintenance management system based on real-time feedback

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant