CN115468278A - Double-layer regulation and control management frame capable of reducing energy consumption of cooling system - Google Patents

Abstract

The invention relates to a cooling system double-layer regulation and control management framework capable of reducing energy consumption of the cooling system, which comprises a controlled entity domain, a connection model domain, a regulation and control algorithm domain, an optimization layer of an air-cooled rear-end water system and an optimization layer of an air-cooled front-end wind system; the controlled entity domain collects various variable data in real time through sensor networks distributed indoors, outdoors and on cooling entities in the data center; the energy consumption evaluation model connected with the model domain utilizes the data collected by the controlled entity domain to carry out real-time energy consumption evaluation on various cooling entities with the aid of the input parameter constraint model; the control algorithm domain optimizes the control parameters of the cooling entity. The invention relates to the field of energy consumption calculation models. The invention has the advantages of quickly reducing the energy consumption required by cooling the entity and having stronger applicability.

Description

A two-layer regulation and management framework for cooling system that can reduce its energy consumption

technical field

The invention relates to the field of energy consumption calculation models, in particular to a two-layer regulation and management framework of a cooling system capable of reducing energy consumption.

Background technique

As an important infrastructure in the digital society, data centers widely serve consumption, production, and social governance application scenarios such as the Internet, industrial Internet, and e-government affairs. The energy consumption required for their operations and services is increasing rapidly year by year. The cooling system is the main energy-consuming facility in the data center. Its function is to absorb the heat generated by the network and computing power equipment in the computer room during work, thereby ensuring the reliable operation of the network and computing power equipment.

As the core infrastructure supporting digital transformation, the data center carries most of the intensive computing tasks in the next-generation information technology scenarios. While building the digital cornerstone of social development, the scale of energy consumption is very large, and it has become one of the high-energy-consuming technology industries. one.

The huge energy consumption of the data center cannot be ignored, and the electricity consumption is huge, and currently about 70% of its electricity supply still comes from coal power. The State Grid Energy Research Institute predicts that by 2030, the electricity consumption of data centers will exceed 400 billion kWh, accounting for 3.7% of the electricity consumption of the whole society.

More and more ICT equipment will generate a lot of heat under high load, which will seriously affect its stable operation. In order to ensure the highly reliable operation of ICT equipment, the cooling system has become an indispensable part of DC, but its energy consumption accounts for about 40% of the power consumption of the data center, and it is one of the core subsystems of the data center's energy consumption.

Therefore, the energy consumption optimization of the cooling system is the most prominent issue in the refined management of data center energy consumption.

Contents of the invention

According to the deficiencies in the prior art, the purpose of the present invention is to provide a two-layer regulation and management framework for the cooling system that can reduce its energy consumption, which has the effect of rapidly reducing the energy consumption required by the cooling entity and has strong applicability.

Above-mentioned technical purpose of the present invention is achieved through the following technical solutions:

A two-layer control and management framework for cooling systems that can reduce energy consumption, including the controlled entity domain, the connection model domain, the control algorithm domain, the optimization layer of the air-cooled back-end water system, and the optimization layer of the air-cooled front-end air system;

The controlled entity domain collects various variable data in real time through the sensor network distributed in the indoor, outdoor and cooling entities of the data center;

The energy consumption assessment model connected to the model domain is assisted by the input parameter constraint model, and uses the data collected by the controlled entity domain to perform real-time energy consumption assessment for various cooling entities;

The control algorithm domain realizes the optimization of the control parameters of the cooling entity.

In a preferred example, the present invention can be further configured as follows: the controlled entity domain is the bottom layer domain of the two-layer regulation and management framework, and each cooling entity is connected with the model and algorithm of the two-layer regulation and management framework.

In a preferred example, the present invention can be further configured as follows: the connection model domain is a middle-level domain between the controlled entity domain and the control algorithm domain, and provides the entity energy consumption and parameter values for the parameter optimization process of the control algorithm domain Evaluate the computing power and realize the two-way data flow between the controlled entity domain and the control algorithm domain.

In a preferred example, the present invention can be further configured as follows: the connection model domain includes the energy consumption evaluation model of each cooling entity and the cooling system state quantity calculation evaluation model.

In a preferred example, the present invention can be further configured as follows: the control algorithm domain is the top-level domain of the two-layer control management framework, and on the premise of ensuring the cooling effect, search for the optimal combination of energy consumption for the control parameters of each cooling entity in the water system;

The optimal energy consumption can be reflected in the reduction of CLF value in terms of indicators;

The CLF formula is:

E _CS represents the energy consumption of the cooling system in the data center, E _{ICT represents} the energy consumption of information and communication equipment in the data center, and E _CS usually includes the energy consumption of the back-end water system E _MRSS and the energy consumption of the front-end wind system E _TCSS ;

The water system energy consumption E _MRSS includes the energy consumption E _CT of the cooling tower, the energy consumption E _CP of the cooling pump, the energy consumption E _C of the refrigeration unit, the energy consumption E _RP of the refrigeration pump, and the energy consumption E _PAC of the precision air conditioner.

In a preferred example, the present invention can be further configured as follows: the control algorithm domain includes two types of energy-saving optimization algorithms: a model-driven energy-saving optimization genetic algorithm and a model-free energy-saving optimization reinforcement learning algorithm.

In summary, the present invention includes at least one of the following beneficial technical effects:

1. Constructed three types of models, namely cooling system input quantity constraint model, cooling entity energy consumption evaluation model, and control quantity constraint model, and proposed two types of energy consumption optimization algorithms suitable for different conditions of data completeness, in order to solve the problem of data The energy consumption of the central cooling system provides a complete solution for optimal regulation and management.

2. Deploying various models and optimization algorithms of the two-tier regulation and management framework in the cooling system effectively reduces the energy consumption of the cooling entity and has strong practicability.

Description of drawings

FIG. 1 is a structure diagram of a two-layer regulation and management framework of a data center cooling system in this embodiment;

Fig. 2 is the heat flow diagram in the dual-cycle air-cooled refrigeration scheme in this embodiment

Fig. 3 is the flow chart of the genetic algorithm for energy consumption optimization driven by the model in this embodiment;

FIG. 4 is a flow chart of the model-free energy consumption optimization reinforcement learning algorithm in this embodiment;

Fig. 5 is a schematic diagram of the influence relationship between various entity energy consumption models in this embodiment.

detailed description

The present invention will be described in further detail below in conjunction with the accompanying drawings.

Example:

Referring to Figures 1 to 5, a two-layer regulation and management framework (referred to as "two-layer framework") of a cooling system that can reduce its energy consumption. The double-layer frame optimizes and controls each cooling entity in the front-end air system and the rear-end water system of the double-circulation air-cooling cooling system, and reduces the total energy consumption of each entity in the cooling system on the premise of ensuring the cooling effect. According to the cooling entity, the framework can be divided into an optimization layer for the air-cooled front-end wind system and an optimization layer for the air-cooled back-end water system. Among them, the optimization object of the former is the precision air conditioner of the front-end air system, which consumes electric energy to absorb the heat generated by information equipment; the optimization object of the latter is the refrigeration unit, cooling tower, cooling pump, and refrigeration pump of the back-end water system , these four types of cooling entities consume electric energy and discharge the heat absorbed by the front-end air system to the outdoor environment of the data center in a double-cycle manner.

The optimal control of various cooling entities by the double-layer framework is jointly realized by three functional domains, which are the controlled entity domain, the connection model domain, and the control algorithm domain.

The controlled entity domain firstly collects various variable data in real time through the sensor network distributed in the indoor, outdoor and cooling entities of the data center.

The energy consumption evaluation model connected to the model domain, with the assistance of the input parameter constraint model, can use the data collected by the controlled entity domain to conduct real-time energy consumption evaluation for various cooling entities.

The control algorithm domain realizes the optimization of the control parameters of the cooling entity. The optimization algorithm can use the energy consumption evaluation model connected to the model domain to evaluate the energy consumption calculated by different numerical combinations of the control parameters of the cooling entity, so as to search for the optimal combination of control parameter values for energy consumption; it can also use the end-to-end method to directly According to the collected data of the controlled entity domain, the end-to-end results of the optimal energy consumption control parameter combination are given.

In the control algorithm field, two types of optimization control algorithms are proposed for the back-end water system. The model-driven energy consumption optimization genetic algorithm is applicable under the condition of establishing various energy consumption evaluation models, and it is more applicable when the ability to establish energy consumption evaluation models is lacking. A model-free energy-optimized reinforcement learning algorithm.

For the wind system, a room temperature evaluation method is proposed, which is used to constrain the room temperature in the optimization process of the genetic algorithm to achieve the cooling effect of the wind system. By establishing a time-series deep learning model, the room temperature, control parameters (control parameters of each cooling entity), state parameters (temperature, humidity), and IT load at k moments before this moment are used as input variables to predict the changes in room temperature. , so that the influence of each group of optimization control parameters on the room temperature can be known during the optimization of the genetic algorithm.

Aiming at the water system of the data center cooling system, an optimal control algorithm for energy consumption of the cooling entity is proposed. Aiming at the water system, a global optimal control scheduling algorithm for cooling entities including two parts, an energy consumption evaluation model and a genetic algorithm, is proposed.

A model-driven energy optimization genetic algorithm is used to configure decision parameters for each cooling entity in the cooling system. The process of parameter configuration can be modeled as an optimization problem with constraints, that is, the energy-saving optimization of the control parameters of the cooling entity must be based on the premise that the cooling effect meets the normal operation of ICT, which is specifically reflected in the need to ensure that the room temperature is less than a preset value. By iterating two optimization steps, the combination of control parameters in the optimal state is obtained based on the genetic algorithm, so that the cooling system consumes lower power while ensuring the cooling capacity. In this algorithm, the configurable variables of various cooling entities are discretized and mapped as individual chromosome codes, and the sum of the outputs of the energy consumption evaluation models of each cooling entity is used as the fitness degree. Through several iterations of the genetic algorithm, it can be obtained Recommended configuration parameters for each cooling entity; model-free energy consumption optimization reinforcement learning algorithm uses neural network and DDPG algorithm to determine the decision parameters for each cooling entity in the cooling system without relying on external energy consumption modeling configure.

The control algorithm domain is the top-level domain of the two-layer control management framework, which searches for the optimal combination of energy consumption for the control parameters of each cooling entity in the water system under the premise of ensuring the cooling effect;

The optimal energy consumption can be reflected in the reduction of CLF value in terms of indicators;

The CLF formula is:

E _CS represents the energy consumption of the cooling system in the data center, E _ICT represents the energy consumption of information and communication equipment in the data center, and E _CS usually includes the energy consumption of the back-end water system E _MRSS and the energy consumption of the front-end wind system E _TCSS ;

The water system energy consumption E _MRSS includes the energy consumption E _CT of the cooling tower, the energy consumption E _CP of the cooling pump, the energy consumption E _C of the refrigeration unit, the energy consumption E _RP of the refrigeration pump, and the energy consumption E _PAC of the precision air conditioner.

The two-tier control management framework includes the monitoring and control of indoor and outdoor environmental variables, ICT equipment operating status, and multiple types of parameter variables in each cooling entity. These variables can be divided into controllable variables (A), observable variables (O), variable evaluation values (E) and system constant parameters (C) according to their roles in the framework. Watch Table 1:

When the data center meets the conditions for establishing an energy consumption evaluation model such as sufficient prior knowledge, the energy-saving optimization of its cooling system can use the model-driven energy optimization genetic algorithm.

This type of algorithm uses genetic algorithm as a parameter optimization method, models the total energy consumption of the cooling system and the room temperature inequality constraints as a Lagrangian dual problem, and uses the Lagrangian equation to express the above two objectives for iterative optimization. Each round of iteration is divided into two steps: the first step is to use the genetic algorithm to optimize the above objectives, and the second step is to automatically update the Lagrangian multipliers. The above two steps are carried out alternately in sequence, and the control parameters that meet the two requirements of energy saving and room temperature can be obtained when iterating until convergence. In the first step, the genetic algorithm first performs chromosome coding on the cooling entity control variables to be optimized. Here, the encoding method is the discretization of continuous eigenvalues. In the initial stage of the algorithm, a value is randomly generated for each controllable variable within its own safety range, and then this real value is mapped to a discrete interval with a certain step size. The above mapping is performed on all the controllable variables to be decided, and after splicing, the binary code of a single individual in the genetic algorithm can be obtained. In the optimization process of the genetic algorithm, for each combination of variable values searched in the numerical space, the algorithm uses the energy consumption evaluation model to calculate the energy consumption evaluation value of each cooling entity under the current combination; uses the room temperature evaluation model to calculate this group The room temperature changes that may be caused by the combination of parameter values. After substituting the evaluation results of the above two types of models into the Lagrangian equation, the optimization objective of the genetic algorithm can be obtained. According to the genetic algorithm optimization parameters obtained in the first step, the model-driven energy optimization genetic algorithm dynamically adjusts the Lagrangian multipliers according to the results of the room temperature evaluation model in the second step. The stop condition of the above two-step loop iteration is that the change value of the Lagrangian multiplier in the two successive iterations is within a certain threshold. At this point, the control variable combination code that is energy-saving and room temperature meets the requirements is obtained. Finally, this discrete code is reverse-mapped back to a continuous value, which is the final output result of the genetic algorithm for model-driven energy optimization. Numerical results of energy optimization for controllable variables of cooling entities. To sum up, this type of optimization algorithm has high requirements for the historical data storage of cooling entities in the data center, so as to ensure that various models in the connected model domain have high evaluation accuracy

When the data completeness condition of a data center is not enough to realize accurate energy consumption assessment in the connected model domain, the energy consumption optimization of its cooling system can use the model-free energy consumption optimization reinforcement learning algorithm proposed in this paper.

This type of algorithm no longer depends on the input quantity constraint model and energy consumption evaluation model in the connection model domain, but directly uses various parameters in the controlled entity domain as input to obtain control parameter optimization results in an end-to-end manner. In the overall optimization process, the algorithm first uses a neural network to estimate the state value related to the energy consumption of the cooling system, and then outputs the control quantity strategy through another neural network. optimization.

Experimental data:

The data center in the experiment has multiple sets of cooling systems with the same structure to provide cooling for computer rooms in different areas of different parks. The cooling system supporting the temperature adjustment of the computer room includes the refrigeration cycle and cooling cycle of the back-end water system, and the cold and hot air cycle of the front-end air system. The cooling system consists of five groups of cooling entities, each group includes a cooling tower (CT), a cooling pump (CP), a refrigerator, a primary refrigeration pump (PRP) and a secondary refrigeration pump (SRP). The cold water provided by SPR will serve as the cold source of the PAC system.

The embodiments of this specific implementation mode are all preferred embodiments of the present invention, and do not limit the scope of protection of the present invention accordingly. Therefore: all equivalent changes made according to the structure, shape and principle of the present invention should be covered by the present invention. within the protection scope of the present invention.

Claims (6)

1. A cooling system double-layer control and management framework that can reduce its energy consumption is characterized in that it includes the controlled entity domain, the connection model domain, the control algorithm domain, the optimization layer of the air-cooled back-end water system and the air-cooled front-end wind The optimization layer of the system; The controlled entity domain collects various variable data in real time through the sensor network distributed in the indoor, outdoor and cooling entities of the data center; The energy consumption assessment model connected to the model domain is assisted by the input parameter constraint model, and uses the data collected by the controlled entity domain to perform real-time energy consumption assessment for various cooling entities; The control algorithm domain realizes the optimization of the control parameters of the cooling entity. 2. a kind of cooling system double-layer regulation and control management framework that can reduce its energy consumption according to claim 1, it is characterized in that: described controlled entity domain is the bottom domain of double-layer regulation and control management framework, each cooling entity and The model and algorithm of the two-tier regulation and management framework are connected. 3. a kind of cooling system double-layer regulation and control management frame that can reduce its energy consumption according to claim 1, it is characterized in that: the connection model domain is the middle-level domain between the controlled entity domain and the control algorithm domain, for The parameter optimization process in the control algorithm domain provides the evaluation and calculation capabilities of entity energy consumption and parameter values, and realizes the bidirectional data flow between the controlled entity domain and the control algorithm domain. 4. A two-layer control and management framework for a cooling system capable of reducing energy consumption according to claim 3, wherein the connection model domain includes an energy consumption evaluation model of each cooling entity and a cooling system state quantity calculation and evaluation model. 5. A kind of cooling system double-layer regulation and control management framework capable of reducing its energy consumption according to claim 1, characterized in that: the regulation algorithm domain is the top-level domain of the double-layer regulation and management framework, which is water under the premise of ensuring the cooling effect The control parameters of each cooling entity in the system search for the optimal combination of energy consumption; The optimal energy consumption can be reflected in the reduction of CLF value in terms of indicators; The CLF formula is:

E _CS represents the energy consumption of the cooling system in the data center, E _ICT represents the energy consumption of information and communication equipment in the data center, and E _CS usually includes the energy consumption of the back-end water system E _MRSS and the energy consumption of the front-end wind system E _TCSS ; The water system energy consumption E _MRSS includes the energy consumption E _CT of the cooling tower, the energy consumption E _CP of the cooling pump, the energy consumption E _C of the refrigeration unit, the energy consumption E _RP of the refrigeration pump, and the energy consumption E _PAC of the precision air conditioner. 6. A kind of cooling system double-layer control and management framework capable of reducing its energy consumption according to claim 5, characterized in that: the control algorithm domain includes two types of energy-saving optimization algorithms: model-driven energy consumption optimization genetic algorithm and wireless Model-dependent energy-optimized reinforcement learning algorithms.

Images