Refrigeration system operation method and apparatus, device, and storage medium

Abstract

The present application discloses a refrigeration system operation method and apparatus, a device, and a storage medium. The method comprises: obtaining an open source data set of a refrigeration system, wherein there is at least one target in a plurality of components included in the refrigeration system, and parameters configured for each first component comprise a controllable operation parameter; determining an initial model for performing power consumption prediction on the refrigeration system, and using an open toolbox to select a target algorithm, wherein the target algorithm is applied to the initial model for power consumption prediction; using the open source data set and the target algorithm to train the initial model to obtain a target model; and deploying the target model to a controller of the refrigeration system, so that the controller determines a power consumption optimization strategy of the refrigeration system by means of the target model and, on the basis of the parameter value corresponding to each controllable operation parameter defined by the determined power consumption optimization strategy, controls the operation of the corresponding first component in the refrigeration system so as to effectively reduce the energy consumption of the refrigeration system.

Description

Refrigeration system operation method, device, equipment and storage medium

[0001] This application claims priority to the Chinese patent application filed with the China Patent Office on June 25, 2023, with application number 2023107576139 and application name “Data processing method, device, equipment and storage medium”, all contents of which are incorporated by reference into this application. Technical Field

[0002] The present application relates to the field of Internet technology, specifically to the field of artificial intelligence technology, and in particular to an operating method, device, equipment and storage medium for a refrigeration system. Background Art

[0003] With the development of internet technology, the number of data centers continues to increase. Data centers require significant energy and cooling systems to operate properly. This energy consumption not only impacts the environment but also incurs significant costs. Therefore, reducing overall data center energy consumption has become a key focus in the industry. A data center's total energy consumption typically includes the energy consumption of IT (Internet) equipment and systems such as cooling and power distribution. Therefore, cooling system energy consumption directly impacts the overall energy consumption of a data center. Reducing cooling system energy consumption can optimize data center energy consumption. Consequently, reducing cooling system energy consumption has become a research hotspot.

[0004] Summary of the Invention

[0005] The embodiments of the present application provide a method, device, equipment, and storage medium for operating a refrigeration system, which can effectively reduce the energy consumption of the refrigeration system.

[0006] In one aspect, an embodiment of the present application provides a method for operating a refrigeration system, the method comprising:

[0007] Obtaining an open source dataset of a refrigeration system, the refrigeration system comprising a plurality of components, each of the components being configured with one or more parameters, the open source dataset comprising parameter values ​​corresponding to the parameters configured for each of the components; wherein at least one first component is present among the plurality of components, and the parameters configured for each of the first components include controllable operating parameters;

[0008] Determining an initial model for power consumption prediction of the refrigeration system, and selecting a target algorithm using an open toolbox, wherein the target algorithm is applied to the initial model for power consumption prediction;

[0009] Using the open source dataset and the target algorithm to train the initial model to obtain a target model;

[0010] deploying the target model to a controller of the refrigeration system, so that the controller determines a power consumption optimization strategy for the refrigeration system based on the target model, wherein the power consumption optimization strategy defines a parameter value corresponding to a controllable operating parameter of at least one of the first components;

[0011] Furthermore, the controller is enabled to control the operation of the corresponding first component in the refrigeration system according to the parameter values ​​corresponding to the various controllable operating parameters defined in the determined power consumption optimization strategy.

[0012] On the other hand, an embodiment of the present application provides an operating device for a refrigeration system, the device comprising:

[0013] an acquisition unit, configured to acquire an open source dataset of a refrigeration system, the refrigeration system comprising a plurality of components, each of the components being configured with one or more parameters, the open source dataset comprising parameter values ​​corresponding to the parameters configured for each of the components; wherein at least one first component is present among the plurality of components, and the parameters configured for each of the first components include controllable operating parameters;

[0014] a processing unit, configured to determine an initial model for predicting power consumption of the refrigeration system, and select a target algorithm using an open toolbox, wherein the target algorithm is applied to the initial model for predicting power consumption;

[0015] The processing unit is further configured to train the initial model using the open source dataset and the target algorithm to obtain a target model;

[0016] The processing unit is also used to deploy the target model to the controller of the refrigeration system, so that the controller determines the power consumption optimization strategy of the refrigeration system through the target model, and the power consumption optimization strategy defines the parameter value corresponding to the controllable operating parameter of at least one of the first components; and to control the operation of the corresponding first component in the refrigeration system according to the parameter value corresponding to each controllable operating parameter defined by the determined power consumption optimization strategy.

[0017] In another aspect, an embodiment of the present application provides a computer device, the computer device including an input interface and an output interface, and the computer device further including:

[0018] processors and computer storage media;

[0019] The processor is adapted to implement one or more instructions, the computer storage medium stores one or more instructions, and the one or more instructions are adapted to be loaded by the processor and executed as follows:

[0020] Obtaining an open source dataset of a refrigeration system, the refrigeration system comprising a plurality of components, each of the components being configured with one or more parameters, the open source dataset comprising parameter values ​​corresponding to the parameters configured for each of the components; wherein at least one first component is present among the plurality of components, and the parameters configured for each of the first components include controllable operating parameters;

[0021] Determining an initial model for power consumption prediction of the refrigeration system, and selecting a target algorithm using an open toolbox, wherein the target algorithm is applied to the initial model for power consumption prediction;

[0022] Using the open source dataset and the target algorithm to train the initial model to obtain a target model;

[0023] deploying the target model to a controller of the refrigeration system, so that the controller determines a power consumption optimization strategy for the refrigeration system based on the target model, wherein the power consumption optimization strategy defines a parameter value corresponding to a controllable operating parameter of at least one of the first components;

[0024] Furthermore, the controller is enabled to control the operation of the corresponding first component in the refrigeration system according to the parameter values ​​corresponding to the controllable operating parameters defined in the determined power consumption optimization strategy.

[0025] On the other hand, an embodiment of the present application provides a computer storage medium, which stores one or more instructions, and the one or more instructions are suitable for being loaded by a processor and executed by the above-mentioned operating method of the refrigeration system.

[0026] On the other hand, an embodiment of the present application provides a computer program product, which includes one or more instructions; when one or more instructions in the computer program are executed by a processor, the above-mentioned method for operating the refrigeration system is implemented.

[0027] The embodiment of the present application can determine an initial model for power consumption prediction of a refrigeration system, train the initial model to obtain a target model, and deploy the model in the controller of the refrigeration system, so that the controller can determine the power consumption optimization strategy of the refrigeration system through the target model, and automatically and intelligently tune and control the power consumption of the refrigeration system through the determined power consumption optimization strategy, thereby effectively reducing the energy consumption of the refrigeration system. Furthermore, by using an open toolbox to select a target algorithm for power consumption prediction for the initial model, the processing logic of the initial model can be white-boxed to a certain extent, so that the power consumption optimization strategy determined by the target model obtained by training the initial model has high interpretability and credibility, thereby enabling the power consumption of the refrigeration system to be more accurately tuned and controlled through the determined power consumption optimization strategy, thereby improving the energy consumption reduction effect. In addition, training the initial model through an open source dataset can break the closed algorithm and security implementation, so that the credibility of the trained target model can be effectively improved, thereby further improving the credibility of the power consumption optimization strategy obtained based on the target model, and further improving the energy consumption reduction effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] FIG1a is a schematic structural diagram of a refrigeration system provided in an embodiment of the present application;

[0029] FIG1b is a schematic diagram of an operation architecture provided in an embodiment of the present application;

[0030] FIG1c is a schematic diagram of the working principle of an operation architecture provided in an embodiment of the present application;

[0031] FIG2 is a schematic flow chart of an operating method of a refrigeration system provided in an embodiment of the present application;

[0032] FIG3 is a schematic diagram of a calculation principle of an energy saving value provided in an embodiment of the present application;

[0033] FIG4a is a schematic diagram of various nodes in a network topology diagram provided by an embodiment of the present application;

[0034] FIG4b is a partial structural diagram of a network topology diagram provided in an embodiment of the present application;

[0035] FIG5 is a flow chart of an operating method of a refrigeration system provided in another embodiment of the present application;

[0036] FIG6 is a schematic structural diagram of an operating device of a refrigeration system provided in an embodiment of the present application;

[0037] FIG7 is a schematic structural diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0038] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application.

[0039] In the embodiments of the present application, a refrigeration system refers to a system that removes heat from an object or space through refrigeration equipment to reduce the temperature of the corresponding object or space. It can be commonly used in places where a low temperature environment needs to be maintained, such as data center computer rooms, office buildings, etc. For ease of explanation, the refrigeration system in a data center is used as an example. Referring to Figure 1a: the refrigeration system includes components such as a refrigeration unit (including a refrigeration host), a cooling water pump, a chilled water pump, a cooling tower, a condenser, an evaporator, and a cold storage tank; among them, the core of the refrigeration system is the refrigeration unit, which absorbs heat by circulating refrigerant and then discharges the heat to achieve the cooling effect. Each component in the refrigeration system can have one or more parameters, among which the parameters that can support parameter value adjustment are controllable operating parameters (or adjustable operating parameters), and the components with controllable operating parameters can be called first components; for example, a cooling water pump can have parameters such as operating status, power, water output, number of operating units, and operating frequency. The parameter values ​​corresponding to parameters such as operating frequency, operating status, and number of operating units can be adjusted, so these parameters are all controllable operating parameters, and the cooling water pump is the first component. For example, a cooling tower may have parameters such as cooling water temperature difference, number of operating units, water output, and operating frequency; a refrigeration host may have parameters such as number of operating units, cooling side temperature difference, current percentage, freezing side outlet water temperature, power, and return water temperature. Then, the number of operating units and operating frequency of the cooling tower, and the freezing side outlet water temperature and number of operating units of the refrigeration host are all controllable operating parameters, and the cooling tower and the refrigeration host are both first components.

[0040] It should be noted that Figure 1a only illustrates the refrigeration system architecture and is not intended to be limiting. The actual architecture of a refrigeration system can vary based on business needs, and the design and use of an actual refrigeration system must take into account multiple factors, such as the required cooling capacity, ambient temperature, refrigeration unit efficiency, and maintenance. Furthermore, the efficiency and energy efficiency of refrigeration systems are also key concerns in the refrigeration industry. To improve the efficiency and reduce the energy consumption of refrigeration systems, new technologies and materials are being applied to their design and manufacture, such as renewable energy, high-efficiency refrigerants, and intelligent control.

[0041] For the above-mentioned refrigeration system, the embodiment of the present application proposes an operational architecture for energy-saving optimization of the refrigeration system based on AIOps technology. The operational architecture is designed through the best practices of multiple AI (Artificial Intelligence) energy-saving projects and R&D architectures. Among them, AIOps is the abbreviation of Artificial Intelligence and Operations. It is a technology that combines artificial intelligence (AI) and operations (Ops), aiming to improve the efficiency and quality of IT (Internet technology) operations through automation and intelligence. The so-called AI technology refers to the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, AI is a comprehensive technology in computer science; it mainly produces a new type of intelligent machine that can respond in a similar way to human intelligence by understanding the essence of intelligence and studying the design principles and implementation methods of various intelligent machines, so that the intelligent machine has multiple functions such as perception, reasoning and decision-making.

[0042] In other words, AIOps is an AI-driven intelligent operations methodology that evolved from DevOps thinking. Influenced by SRE (Site Reliability Engineer) practices, it focuses on overall system reliability, emphasizing one-stop platforms, low-code development, automated processes, and security. DevOps, a portmanteau of Development and Operations, is a culture, movement, or practice that prioritizes communication and collaboration between software developers (Dev) and IT (Internet Technology) Operations (Ops). By automating software delivery and architectural change processes, it enables faster, more frequent, and more reliable software builds, tests, and releases. AIOps leverages big data, machine learning, and other artificial intelligence technologies to automate IT operations processes. Its goal is to improve IT operations efficiency, reduce costs, minimize human error, and enhance service quality.

[0043] It should be noted that the embodiments of this application adhere to at least the following four basic principles when designing an operational architecture for energy-saving optimization of the refrigeration system based on AIOps:

[0044] ① Product thinking: AI energy saving should follow standard product design thinking, focus on design and manufacturing standards, and can be replicated in batches and covered in large-scale data center rooms.

[0045] ② Platform thinking: Support visual development platforms to connect development engineers, HVAC experts, and algorithm experts; support low-code development platforms to enable delivery engineers to implement instantiation configuration for specific projects; support centralized management platforms to enable operations experts to view service status, energy-saving effects, and emergency operations at any time.

[0046] ③ Security Thinking: Data centers are the digital foundation for maintaining the continuity of critical IT business operations, exceeding 99.9999% reliability requirements. Therefore, AI energy saving should follow a security-coverage-optimization design strategy. Security is the "first class" (best) citizen of the operational architecture. Then, it should be compatible with the heterogeneous systems of large-scale computer rooms. Finally, it should be oriented towards global optimization and sustainable optimization.

[0047] ④ Open thinking: The operational architecture can support open APIs (Application Programming Interfaces) and may also include features such as componentization, white-boxing, standardization, and process-based operations. That is, the platform is composed of standard components, and the components themselves have clear interface and indicator definitions. Components from multiple component suppliers can be replaced as long as they meet performance and security requirements. The platform follows a unified "Pipeline," and each component forms a standard input and output filter.

[0048] As shown in FIG1b , the operating principle of the AIOps-based refrigeration system energy-saving optimization operation architecture proposed in the embodiment of the present application is roughly as follows:

[0049] First, an open-source dataset can be used to model the refrigeration system and obtain modeling results. An open-source dataset refers to an open (i.e., publicly available) dataset that includes the collected values ​​of one or more parameters for each component in the refrigeration system. The modeling results can include a digital system corresponding to the refrigeration system and an initial model for power consumption prediction. Optionally, since the physical system (BA system) to which the refrigeration system belongs often operates under a few fixed operating modes, optimization changes can be added during the modeling phase to explore more combinations of units and operating frequency ranges, giving the AI ​​greater scope for optimization. Second, an open toolbox can be used to select an algorithm for power consumption prediction (called the target algorithm) based on the initial model. The open toolbox is an open library of algorithms that provides algorithm selection. Furthermore, a secure sandbox can be used to constrain the initial model. The initial model can then be trained to obtain a target model, which can then be simulated and tested. The AIOps management module can then analyze the simulation test results to determine whether to retrain the target model for further optimization. After obtaining the target model, the target model can be deployed to the controller (also known as the edge intelligent controller or edge controller, which is an operating system that can control the operation of the refrigeration system) through the model warehouse, so that the controller can automatically and intelligently tune the refrigeration system for energy saving through the target model.

[0050] Furthermore, the controller can also be service-monitored to obtain at least one of the operating status of the controller and the operating status of the target model in the controller, so as to perform optimization analysis based on the obtained operating status through the AIOps management module to determine whether to perform model training on the target model again. In addition, after the controller uses the target model to perform energy-saving tuning control on the refrigeration system, the AIOps management module can also evaluate the energy-saving effect of the target model according to a unified energy-saving effect evaluation strategy to determine whether to perform model training on the target model again to update the target model. If the target model is updated, the target model in the controller can be further upgraded online (OTA upgrade) using the updated target model through the model warehouse to optimize the target model in the controller.

[0051] Based on the above description, the working principle of the AIOps-based refrigeration system energy-saving optimization operation architecture proposed in the embodiment of this application can be roughly divided into the following stages: data processing, model building, simulation testing, tuning control and proactive operation and maintenance, etc. Among them:

[0052] (1) Data processing stage. This stage mainly includes point configuration, data collection, data cleaning, feature extraction, data governance and change optimization. Point configuration refers to the configuration of one or more points in the refrigeration system. The points are used to collect the parameter values ​​of the components in the refrigeration system. Data collection refers to the collection of the working condition data of each configured point. The working condition data of any point refers to the data collected when the corresponding point is working, which may include one or more parameter values. Data cleaning refers to the data deduplication, data format conversion and abnormal data processing of the collected working condition data. Feature extraction refers to the feature extraction of the results of data cleaning to obtain an open source data set. Data governance refers to the process of improving data quality. Change optimization refers to the process of collecting more reliable and valid data (i.e. parameter values) by changing the site of the refrigeration system. It should be noted that data governance and change optimization are optional processes, that is, these two processes can be omitted in the data processing stage. In addition, the result obtained through the data processing stage is a reliable open source data set. It is understandable that good data is the key to training a good model.

[0053] (2) Model construction phase, which mainly includes algorithm selection, security sandbox processing, and model training. Algorithm selection refers to the selection of the algorithm used for AI energy saving of the refrigeration system (i.e., the target algorithm mentioned above). The AI ​​energy saving algorithm can be developed from machine learning to neural networks (such as CNN (Convolutional Neural Network, Convolutional Neural Network), RNN (Recurrent Neural Networks, Recurrent Neural Networks, LSTM (Long Short-Term Memory) long short-term neural network, etc.), and gradually to reinforcement learning. With the rise of LLM (Large Natural Language Model), the Transformer model behind it (a model that uses attention mechanism to improve model training speed) as a foundation model has also begun to be paid attention to in the field of industrial intelligence, making the AI ​​energy saving algorithm also applicable to the Transformer model. The security sandbox processing includes constraints on the optimization process, data health constraints, and optimization result constraints, as well as setting reasonable value ranges for the parameter values ​​and states corresponding to all controllable operating parameters, real-time monitoring of data quality and equipment alarms, and performing over-limit checks on the parameter values ​​output by the model. Optionally, the security sandbox can set the range of parameter values ​​corresponding to controllable operating parameters by controlling the step size and taking a "small step, fast run" approach to reduce fluctuations in the cooling system. Model training refers to optimizing the model parameters of the initial model. The model training process can be completed on a local development machine or a cloud-based GPU (Graphic Processing Unit) server. Generally speaking, one data center can correspond to one AI energy-saving project, and for each AI energy-saving project, a separate model can be trained to optimize the cooling system in the corresponding data center for energy saving. New data accumulated during the automatic training cycle can also be used for transfer learning to ensure that the AI ​​model is optimized and reliably controlled within the boundaries defined by HVAC experts.

[0054] (3) Simulation test phase, this phase mainly includes the simulation test of the target model obtained after model training. Usually, the data used for model training can be divided into a training set and a test set. The training set is used to train the initial model, and the test set is used to test the prediction accuracy (abbreviated as accuracy) of the target model obtained after training. However, it is difficult to evaluate the security and energy saving of the model by testing only on static and limited data sets. If these models that have not been fully evaluated are put online, it will bring risks to the reliable operation of the data center. Based on this, the embodiment of the present application can obtain a simulation system through system modeling. The simulation system includes establishing a digital twin system corresponding to the physical system to which the refrigeration system belongs by integrating the mechanism model and data drive, and testing the accuracy of the target model through the simulation system. In addition, the simulation system can also respond to the tuning control of the controller to simulate the operating status of the refrigeration system for a period of time, so as to test the energy-saving performance of the target model according to the simulation results. By simulating the specific decision-making process of the AI ​​model on a computer, the iteration speed of the tuning algorithm can be greatly accelerated, and it is ensured that the AI ​​model is comprehensively evaluated before actual deployment.

[0055] (4) Optimization and control phase. This phase mainly includes the controller using the target model to optimize and control the energy saving of the refrigeration system. Specifically, the controller can use the target model to calculate a set of power consumption optimization strategies at regular intervals (for example, the default (default) is 10 minutes). The power consumption optimization strategy defines the parameter values ​​of the controllable operating parameters of at least one first component; and through the protocol interface (BACNet (communication protocol for intelligent buildings), Snmp (remote network monitoring), Modbus (application layer message transmission) and other protocol interfaces), the corresponding components in the refrigeration system are issued control commands, so that the corresponding components adjust the operating conditions according to the parameter values ​​corresponding to the controllable operating parameters defined by the power consumption optimization strategy, so as to achieve energy-saving optimization control of the refrigeration system. It should be noted that, for the requirements of data security and local control, the owners and operators of the data center may require that the AI ​​model support being deployed on a local server or AI algorithm card (abbreviated as AI card) and support offline operation.

[0056] (5) Active operation and maintenance phase. This phase mainly includes self-diagnosis of software or hardware faults in at least one of the controller and the target model by monitoring at least one of the operating status of the controller and the operating status of the target model in the controller, and active closed-loop maintenance of the diagnosed fault points. It should be understood that the cooling system is a key facility to ensure the sustainable operation of IT business. For the complex system of AI+BA, it is necessary to conduct potential failure mode analysis and formulate emergency operation plans in advance, and simulate equipment failures and system anomalies in the conversion test. In particular, it is necessary to support local clusters or hot standby to ensure high availability of AI energy saving, and the AI ​​model can be configured with a watchdog to regularly report the operating status (also called service status) to the active operation and maintenance system, so that the active operation and maintenance system can realize self-diagnosis of software or hardware faults and closed-loop maintenance process according to the reported operating status, and complete repair maintenance and service recovery within the SLA (Service Level Agreement).

[0057] Based on the relevant descriptions of the above stages, it can be seen that the operational architecture for energy-saving optimization of the refrigeration system based on AIOps technology proposed in the embodiment of the present application has opened up the entire life cycle of data, models, training, testing, deployment, operation and maintenance, as shown in Figure 1c. Among them, the strategy optimization mentioned in the algorithm call part of Figure 1c refers to the process of determining the power consumption optimization strategy of the refrigeration system; online debugging refers to the ability to open the algorithm model (i.e., the target model) and provide API use to facilitate debugging and verifying the performance of the target model, and making the target model available for use by other businesses; algorithm upgrades may include the aforementioned OTA upgrade of the target model in the controller. The edge deployment mentioned in the access deployment part of Figure 1c refers to deploying the algorithm and target model to the controller, and the controller is a small operating system. AI card reasoning refers to inserting an AI card in the controller that is used solely for algorithm program call model calculation to find the optimal power consumption optimization strategy.

[0058] Based on the above description, it can be seen that the operational architecture for energy-saving optimization of the refrigeration system based on AIOps technology proposed in the embodiment of the present application can bring new opportunities and advantages to the energy management of data centers. By utilizing open data and algorithm simulation models, not only can data be better managed and utilized, the value and efficiency of data can be improved, and the speed of AI iteration can be accelerated, but also data center industry engineers can use these open data and models to better understand the operating principles and performance characteristics of the refrigeration system, connect academia and industry, and thus more accurately optimize the energy utilization efficiency of the refrigeration system. At the same time, by breaking the closed algorithms and security implementations, collaborative work and knowledge sharing between various data centers can be achieved, thereby improving energy utilization efficiency and sustainable development levels, and achieving more efficient energy management and energy conservation and emission reduction.

[0059] In addition, as an innovative AI energy-saving architecture, this operation architecture can also provide a more efficient and reliable operation solution for the PUE (Power Usage Effectiveness) optimization project of the data center's cooling system, thereby promoting the sustainable development of the data center. PUE is an indicator used to evaluate the energy efficiency of a data center. It is equal to the ratio of all the energy consumed by the data center (i.e., total energy consumption) and the energy consumed by the equipment (IT equipment) load in the data center (i.e., IT equipment energy consumption), that is, PUE = total energy consumption of the data center / IT equipment energy consumption. Among them, the total energy consumption of the data center includes the energy consumption of IT equipment and the energy consumption of cooling, power distribution and other systems; it can be seen that the PUE value is greater than 1, and the closer the PUE value is to 1, the less energy is consumed by non-IT equipment, and thus the better the energy efficiency level of the data center.

[0060] It should be understood that the above is only an illustrative description of the general principles of the operational architecture for energy-saving optimization of refrigeration systems based on AIOps technology proposed in the embodiment of the present application, and does not limit this. For example, in actual applications, the operational architecture can also standardize metadata such as data points, data sets, algorithms, and models, and construct them into a network topology map, so that data can be better managed and utilized through the network topology map, thereby improving the value and efficiency of data. At the same time, the operational architecture can also use AIOps technology to automatically and intelligently improve the operational efficiency and quality of AI energy-saving projects, and then combine the network topology structure of metadata to improve key links such as data processing, model construction, simulation testing, tuning control, and proactive operation and maintenance, thereby achieving rapid implementation and stable operation, and providing a more efficient and reliable solution for the sustainable development of data centers. Through this automated and intelligent operation method, the interference of human factors can be reduced, the accuracy and efficiency of data center energy management can be improved, and the demand for operating costs and human resources can also be reduced.

[0061] Practice has shown that the AIOps-based refrigeration system energy-saving optimization operation architecture proposed in the embodiments of this application can bring about multiple beneficial effects, including but not limited to the following aspects:

[0062] ① Improve energy utilization efficiency: Intelligent management and optimization of the data center's cooling system through AIOps technology can effectively reduce unnecessary energy waste, improve energy utilization efficiency, and thus achieve energy-saving optimization of the data center.

[0063] ② Improve data center operational efficiency: Automated operations and maintenance through AIOps technology can quickly and accurately diagnose and resolve faults, improve data center operational efficiency, and reduce operating costs. Furthermore, operational architecture technology solutions enable data sharing and interconnection between different devices and systems, further improving data center operational efficiency.

[0064] ③ Improve data center security: Through real-time monitoring and early warning of data centers (such as controllers and target models located in controllers) through AIOps technology, security risks can be quickly identified and corresponding measures can be taken to improve the security and stability of data centers and reduce the risk of security accidents.

[0065] ④ Improve user experience: Through the intelligent management of AIOps technology, the service quality and response speed of the data center can be improved, and user satisfaction and experience can be improved.

[0066] In summary, the AIOps-based operational architecture and technical solution for energy-efficient data center cooling systems can deliver multiple benefits, including improved energy efficiency, data center operational efficiency, data center security, and enhanced user experience. These benefits will help drive data center development towards a greener and more sustainable direction, contributing to environmental protection and sustainable development.

[0067] It is worth noting that the methodology of the AIOps-based operational architecture for energy-saving optimization of refrigeration systems proposed in the embodiments of this application can also be applied to various other fields and industries, such as finance, healthcare, and industry. Taking the financial industry as an example, by building an operational architecture system, data resources from different financial institutions can be integrated and shared, thereby more accurately conducting risk assessments and credit ratings, and improving the risk control level and business efficiency of financial institutions. In the medical industry, the operational architecture system can be used to integrate the electronic medical data of medical institutions, thereby achieving optimal allocation of medical resources. In the industrial field, the operational architecture system can be used to manage and optimize the flow and conversion of data in the industrial production process, improving production efficiency and quality. Furthermore, in the specific implementation of applying this operational architecture to other fields and industries, if data related to user privacy is involved, when the above method embodiments are applied to specific products or technologies, these relevant data are collected with the user's permission or consent, and the collection, use and processing of the relevant data comply with the relevant laws, regulations and standards of the relevant region.

[0068] Based on the above-mentioned description of the operational architecture of the energy-saving optimization of the refrigeration system based on AIOps technology, an embodiment of the present application proposes a method for operating a refrigeration system. The method for operating the refrigeration system can be executed by a computer device, which can be a terminal or a server; or, the method for operating the refrigeration system can also be jointly executed by a terminal and a server, without limitation. Among them, the terminal can be a smart phone, a computer (such as a tablet computer, a laptop computer, a desktop computer, etc.), a smart wearable device (such as a smart watch, smart glasses), an intelligent voice interaction device, a smart home appliance (such as a smart TV), a vehicle terminal or an aircraft, etc.; the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and basic cloud computing services such as big data and artificial intelligence platforms, etc. Furthermore, the terminal and server can be located inside or outside the blockchain network, without limitation; further, the terminal and server can also upload any data stored internally to the blockchain network for storage to prevent the internally stored data from being tampered with, thereby improving data security.

[0069] For ease of explanation, the following description will be made by taking a computer device as an example to illustrate the method for operating the refrigeration system. Referring to FIG. 2 , the method for operating the refrigeration system may include the following steps S201-S204:

[0070] S201, obtain an open source dataset of a refrigeration system.

[0071] In a specific implementation, when executing step S201, the computer device can obtain an open source dataset for the refrigeration system from an open data platform via the Internet. Alternatively, based on the relevant description of the aforementioned operational architecture, when executing step S201, the computer device can configure points in the refrigeration system. A component in the refrigeration system can have one or more points, and each point can be used to collect a parameter value corresponding to a parameter of a component; collect operating condition data from each configured point, where the operating condition data of any point includes a parameter value corresponding to a parameter of a component; and perform data cleaning and feature extraction on the collected operating condition data to obtain an open source dataset for the refrigeration system.

[0072] It should be understood that the configuration of points corresponding to refrigeration systems of different architectures may vary, and the computer equipment can configure points in various components of the refrigeration system based on the opinions of on-site engineers to obtain multiple points. One component may be configured with one or more points, and each point is used to collect parameter values ​​corresponding to a parameter configured on the corresponding component. The open source dataset obtained in step S201 may include parameter values ​​corresponding to the parameters configured for each component in the refrigeration system. Among the multiple components included in the refrigeration system, there is at least one first component, and the parameters configured for each first component include controllable operating parameters. Any parameter that supports parameter value adjustment can be used as a controllable operating parameter.

[0073] For example, if the refrigeration system includes a chilled water pump, the points configured for the chilled water pump may include, but are not limited to: a point for collecting a parameter value corresponding to the chilled water pump power parameter (i.e., a power value); a point for collecting a parameter value corresponding to the chilled water pump operating frequency parameter (i.e., an operating frequency value); a point for collecting a parameter value corresponding to the chilled water pump operating status parameter (i.e., an operating status value); a point for collecting a parameter value corresponding to the number of chilled water pumps in operation, etc. In this case, the chilled water pump component is the first component, and the controllable operating parameters configured therein may include: the chilled water pump operating frequency, the chilled water pump operating status, the number of chilled water pumps in operation, etc.

[0074] For another example, if the refrigeration system includes a cooling tower, the points configured for the cooling tower may include, but are not limited to: points for collecting parameter values ​​corresponding to the operating frequency of the cooling tower (i.e., the operating frequency value), points for collecting parameter values ​​corresponding to the power of the cooling tower (i.e., the power value), points for collecting parameter values ​​corresponding to the operating status of the cooling tower (i.e., the operating status value), points for collecting parameter values ​​corresponding to the number of cooling towers in operation, etc. In this case, the cooling tower component is the first component, and the controllable operating parameters configured therein may include: the operating frequency of the cooling tower, the operating status of the cooling tower, the number of cooling towers in operation, etc.

[0075] For another example, if the refrigeration system includes a refrigeration host, the points configured for the refrigeration host may include but are not limited to: a point for collecting the parameter value corresponding to the parameter power of the refrigeration host (i.e., the power value), a point for collecting the parameter value corresponding to the parameter current percentage of the refrigeration host (the current percentage value), a point for collecting the parameter value corresponding to the parameter outlet water temperature of the refrigeration host (the outlet water temperature value of the refrigeration side), a point for collecting the parameter value corresponding to the parameter outlet water pressure of the refrigeration host (the outlet water pressure value of the refrigeration side), a point for collecting the parameter value corresponding to the parameter inlet water temperature of the refrigeration host (the inlet water temperature value of the refrigeration side), and a point for collecting the parameter value corresponding to the parameter outlet water pressure of the refrigeration host (the outlet water pressure value of the refrigeration side). The controllable operating parameters of the refrigeration unit may include: a point for collecting the parameter value corresponding to the parameter of the chilled side inlet water pressure (chilled side inlet water pressure value); a point for collecting the parameter value corresponding to the parameter of the cooling side outlet water temperature of the refrigeration unit (cooling side outlet water temperature value); a point for collecting the parameter value corresponding to the parameter of the cooling side outlet water pressure of the refrigeration unit (cooling side outlet water pressure value); a point for collecting the parameter value corresponding to the parameter of the cooling side inlet water temperature of the refrigeration unit (cooling side inlet water temperature value); a point for collecting the parameter value corresponding to the parameter of the cooling side inlet water pressure of the refrigeration unit (cooling side inlet water pressure value); a point for collecting the parameter value corresponding to the parameter of the cooling side temperature difference of the refrigeration unit (cooling side temperature difference value); etc. In this case, the refrigeration unit component is the first component, and the controllable operating parameters configured therein may include: the chilled side outlet water temperature of the refrigeration unit, the chilled side outlet water pressure of the refrigeration unit, the chilled side inlet water pressure of the refrigeration unit, the cooling side outlet water pressure of the refrigeration unit, the cooling side inlet water pressure of the refrigeration unit, etc.

[0076] S202 : determining an initial model for power consumption prediction of a refrigeration system, selecting a target algorithm using an open toolbox, and applying the target algorithm to the initial model for power consumption prediction.

[0077] In a specific implementation, a model can be uniformly set for each component in the refrigeration system; in this case, the initial model can be an independent model. Alternatively, at least one sub-model can be configured for each component in the refrigeration system; in this case, the initial model can be understood as a system model obtained by combining the sub-models corresponding to each component, that is, the initial model can include: a sub-model configured for each component in the refrigeration system. In this case, any two sub-models in the initial model can be independent of each other or have a coupling relationship; when the first component and the second component are connected, the sub-model corresponding to the first component can be coupled with the sub-model corresponding to the second component, and the first component and the second component refer to two different components in the refrigeration system. Furthermore, when any sub-model (let's say sub-model a) is coupled with another sub-model (let's say sub-model b), the output of any sub-model (i.e., sub-model a) serves as the input of the other sub-model (sub-model b).

[0078] For example, for the cooling tower component, multiple sub-models can be configured, such as the cooling tower cooling water temperature difference prediction model and the cooling tower power prediction model; for the refrigeration unit component, multiple sub-models can be configured, such as the chiller cooling side return water temperature prediction model and the chiller power prediction model. Since the cooling tower's output flows into the refrigeration unit, that is, the cooling tower and refrigeration unit are connected, the cooling tower cooling water temperature difference prediction model and the chiller power prediction model are coupled, that is, the output of the cooling tower cooling water temperature difference prediction model can serve as the input of the chiller power prediction model. Optionally, considering that the power of the chiller is related to the chiller's cooling side return water temperature, the chiller cooling side return water temperature prediction model can also be coupled with the chiller power prediction model, that is, the output of the chiller cooling side return water temperature prediction model can also serve as the input of the chiller power prediction model.

[0079] It should be understood that when the initial model includes multiple sub-models, the target algorithm selected by step S202 includes: the prediction algorithm adopted by each sub-model in the initial model, and the so-called prediction algorithm refers to an algorithm that predicts or classifies based on the input data of the model. The prediction algorithm adopted by any sub-model is used to predict the parameter value corresponding to the preset target parameter based on the parameter values ​​corresponding to the input parameters. The target parameters involved in the prediction algorithms adopted by different sub-models are different. For example, for the sub-model of the chiller power prediction model configured for the chiller, the prediction algorithm it adopts is used to predict the parameter value corresponding to the target parameter of the chiller power (i.e., the chiller power value) based on the parameter values ​​corresponding to multiple parameters such as the input chiller cooling side return water temperature and the cooling tower cooling water temperature difference.

[0080] It should also be noted that when the initial model includes multiple sub-models and the target algorithm includes the prediction algorithm used by each sub-model, each sub-model in the initial model uses its own prediction algorithm to collaboratively predict the power consumption of the refrigeration system. Specifically, there is at least one target sub-model among the multiple sub-models included in the initial model, and the target parameter involved in the prediction algorithm used by the target sub-model includes a system power consumption parameter, and the parameter value corresponding to the system power consumption parameter is used to represent the power consumption of the refrigeration system. When collaboratively predicting the power consumption of the refrigeration system, the p-th sub-model in the initial model can use the corresponding prediction algorithm to predict the parameter value of the target parameter involved in the prediction algorithm used by the p-th sub-model based on the input parameter value; if the p-th sub-model is coupled with another q-th sub-model, the predicted parameter value can be input into the q-th sub-model, so that the q-th sub-model can use the corresponding prediction algorithm to predict the parameter value of the target parameter involved in the prediction algorithm used by the q-th sub-model based on the input parameter value, and so on, until the target sub-model performs the prediction operation, thereby outputting the power consumption of the refrigeration system. It can be seen that when the power consumption is predicted through the initial model, each sub-model is called in turn to perform relevant predictions based on the coupling relationship between the sub-models, so that the power consumption of the refrigeration system is predicted through the target sub-model; in this way, the final predicted power consumption can be made more consistent with the actual operation of the refrigeration system, thereby improving the accuracy of the power consumption prediction.

[0081] In a specific implementation, when a computer device uses an open toolbox to select an algorithm for an initial model, it can select an appropriate algorithm based on the specific application scenario. Furthermore, the selection of an algorithm can also take into account multiple factors, such as the type of data, the distribution of data, the complexity of the algorithm, the accuracy of the algorithm, etc. For example, taking the target algorithm as an example, including the prediction algorithm used by the sub-model corresponding to each component, and assuming that the development toolbox can provide multiple algorithms such as rule-based algorithms, statistics-based algorithms, and neural network-based algorithms, then:

[0082] For the chilled water pump component, the chilled water pump power prediction model configured for it is used to predict the power of the chilled water pump. Since it is clear that the greater the operating frequency of the chilled water pump, the greater its power consumption, that is, it is known that the operating frequency and power consumption of the chilled water pump are strongly correlated, the prediction algorithm selected for the chilled water pump power prediction model can be a rule-based algorithm.

[0083] For the refrigeration unit component, the power prediction model configured for it is used to predict the refrigeration unit's power. However, since the refrigeration unit's power is related to the chiller's inlet and outlet water temperature, chiller's inlet and outlet pressure, cooling side's inlet and outlet water temperature, and cooling side's inlet and outlet pressure, it is difficult to use rules to determine the refrigeration unit's power based on these parameters. Therefore, the prediction algorithm selected for the refrigeration unit power prediction model can be a statistically based algorithm. If the output data of a sub-model is related to certain data, the prediction algorithm selected for this sub-model can be a neural network-based algorithm, which can automatically learn the relationship between the input data and output of the corresponding sub-model.

[0084] S203: Train the initial model using the open source dataset and the target algorithm to obtain the target model.

[0085] In a specific implementation, a computer device may first train an initial model using an open source dataset and a target algorithm to obtain an intermediate model. Specifically, if the initial model includes multiple sub-models, the intermediate model includes the intermediate sub-models obtained by training each sub-model. In this case, since the target algorithm includes the prediction algorithm used by each sub-model, and each prediction algorithm indicates which parameter values ​​the model's input data must include, the computer device can construct training data and label information for any sub-model based on the indication of the prediction algorithm used by the sub-model and the open source dataset. The training data includes the parameter values ​​indicated by the corresponding prediction algorithm; the label information of the training data includes the parameter values ​​corresponding to the target parameters involved in the prediction algorithm, obtained based on the open source dataset. Then, the training data and corresponding label information are used to train any sub-model in a supervised training manner. Specifically, any sub-model can be called to use the prediction algorithm to predict the parameter value corresponding to the target parameter based on the training data. Based on the difference between the predicted parameter value and the label information, the model parameters of any sub-model are optimized to train the sub-model and obtain the corresponding intermediate sub-model.

[0086] For example, for the sub-model of the chilled water pump power prediction model, the prediction algorithm it uses is used to predict the parameter value corresponding to the target parameter, the chilled water pump power, based on the parameter value corresponding to the chilled water pump operating frequency parameter. Therefore, the parameter value corresponding to the chilled water pump operating frequency parameter (i.e., the operating frequency value) can be obtained from the open source dataset as the training data for the chilled water pump power prediction model, and the parameter value corresponding to the chilled water pump power parameter (i.e., the power value) can be obtained from the open source dataset as the label information for the training data. Then, the chilled water pump power prediction model can be called to use the corresponding prediction algorithm to predict the chilled water pump power value based on the operating frequency value in the training data. Based on the difference between the predicted power value and the power value in the label information, the model parameters of the chilled water pump power prediction model are optimized to achieve training of the chilled water pump power prediction model.

[0087] After obtaining the intermediate model, the intermediate model can be directly used as the target model. Furthermore, to ensure that the target model has good performance, after obtaining the intermediate model, the computer device can also perform performance testing on the intermediate model through a simulation system to obtain test results. If the test results indicate that the intermediate model passes the performance test, the intermediate model is used as the target model; if the test results indicate that the intermediate model fails the performance test, the intermediate model can be trained to obtain the target model. It should be understood that the method for training the intermediate model is similar to the method for training the initial model and will not be repeated here. Among them, the simulation system mentioned here includes: a digital twin system corresponding to the refrigeration system established by integrating the mechanism model and data-driven learning. The so-called mechanism model, also known as the white box model, is a model obtained by modeling each component (equipment) in the refrigeration system based on the physical characteristics of each component (equipment) in the refrigeration system, combined with a big data-driven learning method; data-driven learning can be understood as the process of configuring the operating condition data of each component in the refrigeration system on each digital component in the model obtained by modeling. The simulation system constructed in this way can avoid large errors or even unreasonable situations under working conditions that do not appear in historical data.

[0088] The performance test of the intermediate model performed by the simulation system may include, but is not limited to, accuracy performance testing, energy-saving performance testing, etc. Based on this, when the intermediate model includes an intermediate sub-model obtained by model training of each sub-model, the specific implementation of the performance test of the intermediate model by the simulation system to obtain the test results may include at least one of the following implementations:

[0089] Implementation method one: For the kth intermediate sub-model in the intermediate model, obtain the test data of the kth intermediate sub-model by running the simulation system; and use the test data to test the accuracy performance of the kth intermediate sub-model to obtain the test results. Wherein, k∈[1, K], K is the number of intermediate sub-models included in the intermediate model. Through the accuracy performance test, it can be tested whether the kth intermediate sub-model has high accuracy, so that when it is determined that the accuracy of the kth intermediate sub-model is low, the kth intermediate sub-model can be further trained to make the accuracy of the kth sub-model in the target model finally obtained higher, thereby improving the accuracy performance of the model.

[0090] Implementation method two: within a preset time period, the operation of the simulation system is controlled through the intermediate model. Specifically, each component in the simulation system can be operated; within the preset time period, the power consumption optimization strategy of the simulation system is calculated periodically through the intermediate model; each time a set of power consumption optimization strategies is calculated, the operation of the corresponding first component in the simulation system is controlled according to the parameter values ​​corresponding to the controllable operating parameters defined by the currently calculated power consumption optimization strategy. Specifically, the parameter values ​​corresponding to the controllable operating parameters defined by the currently calculated power consumption optimization strategy can be used to update the parameter values ​​corresponding to the controllable operating parameters of the corresponding first component in the simulation system, so that the corresponding first component in the simulation system can operate according to the parameter values ​​corresponding to the controllable operating parameters defined by the currently calculated power consumption optimization strategy. In the embodiment of the present application, the power consumption optimization strategy of the simulation system is calculated periodically through the intermediate model, which can avoid the consumption of a large amount of processing resources due to the real-time calculation of the power consumption optimization strategy, thereby saving processing resources, and can avoid the situation where the computer equipment runs slowly due to the use of a large amount of processing resources. Among them, the method of calculating the power consumption optimization strategy of the simulation system through the intermediate model can be: generating multiple groups of parameter combinations according to the strategy optimization algorithm, each group of parameter combinations includes parameter values ​​corresponding to each controllable operating parameter; calling the intermediate model to predict the power consumption of the simulation system according to each group of parameter combinations, and selecting a group of parameter combinations from the multiple groups of parameter combinations according to the predicted power consumption of the simulation system under each group of parameter combinations as the power consumption optimization strategy of the simulation system.

[0091] After the preset time period ends, the target energy consumption of the data center during the preset time period is calculated. The data center generates multiple environmental data during the preset time period, where one environmental data point refers to a combination of a wet-bulb temperature and an equipment load interval. The target energy consumption includes an energy efficiency indicator value for the data center under each of the multiple environmental data points. The energy efficiency indicator value is the ratio between the total energy consumed by the data center and the energy consumed by the equipment load of the data center (i.e., the PUE value). The so-called equipment load can be understood as the power of IT equipment. Next, a reference energy consumption profile for the data center can be obtained. The reference energy consumption profile includes an energy efficiency indicator value for each of the multiple environmental data points without using an intermediate model. Then, the intermediate model can be tested for energy-saving performance based on the target energy consumption profile and the reference energy consumption profile to obtain test results. The energy-saving performance test can determine whether the intermediate model has high energy-saving performance. If the energy-saving performance of the intermediate model is determined to be low, the intermediate model can be further trained to ensure that the resulting target model has high energy-saving performance.

[0092] When the computer device tests the energy-saving performance of the intermediate model based on the target energy consumption situation and the reference energy consumption situation and obtains the test results, it can determine the similarity between the target energy consumption situation and the reference energy consumption situation. If the similarity is greater than the similarity threshold, a test result indicating that the intermediate model has failed the performance test can be generated. If the similarity is less than or equal to the similarity threshold, a test result indicating that the intermediate model has passed the performance test can be generated. Furthermore, the similarity between the target energy consumption situation and the reference energy consumption situation can be measured by the number of abnormal environmental data, and abnormal environmental data refers to environmental data that meets the following conditions: the difference between the energy efficiency indicator value corresponding to the target energy consumption situation and the energy efficiency indicator value corresponding to the reference energy consumption situation is less than a preset value; that is, if the difference between the energy efficiency indicator value corresponding to a certain environmental data in the target energy consumption situation and the energy efficiency indicator value corresponding to the environmental data in the reference energy consumption situation is less than a preset value, then the environmental data can be regarded as abnormal environmental data.

[0093] Alternatively, when the computer device tests the energy-saving performance of the intermediate model according to the target energy consumption and the reference energy consumption, and obtains the test results, the following steps s11-s12 may be executed:

[0094] s11, based on the target energy consumption and the reference energy consumption, the intermediate model is evaluated for energy saving to obtain the target energy saving value of the intermediate model.

[0095] Considering that energy conservation is a very important task in the operation of a data center, it can reduce the operating costs of the data center, reduce the impact on the environment, and extend the life of hardware equipment. Based on this, in the operation architecture system proposed in the embodiment of the present application, an energy-saving calculation formula is proposed, so that the computer equipment can calculate the target energy saving value of the target model through the energy-saving calculation formula. The energy-saving calculation formula can help data center managers to quantitatively evaluate the effect of energy-saving optimization. Specifically, it can calculate the specific effect of energy-saving optimization by comprehensively considering multiple factors such as the IT load power consumption, temperature, humidity, etc. of the data center. Through this energy-saving calculation formula, it will help data center managers to better understand whether the efforts of energy-saving optimization have achieved the expected results, and formulate more effective energy-saving strategies based on these data, so as to achieve more sustainable data center operations. Among them, the definition of the energy-saving calculation formula is as follows:

[0096] Among them, ES represents the target energy saving value of the refrigeration system, R i represents the combination of the i-th wet-bulb temperature and IT load interval (i.e., the i-th environmental data), where i ranges from 1 to n, and n is the number of environmental data. PUE is the ratio of all energy consumed by the data center to the energy consumed by the IT load (i.e., energy efficiency indicator), PUEa i represents the PUE corresponding to the i-th environmental data before energy-saving optimization (i.e., the energy efficiency indicator value corresponding to the i-th environmental data in the reference energy consumption situation), PUEb i represents the PUE of the i-th environmental data after energy-saving optimization (i.e., the energy efficiency indicator value corresponding to the i-th environmental data in the target energy consumption situation), IT i represents the mean IT load (i.e., the mean equipment load) corresponding to the i-th environmental data, L i = represents the operating time for energy saving comparison within the i-th environmental data (i.e., the duration of the i-th environmental data). For a comparison of cooling systems controlled by an AI system (i.e., the intermediate model) and a BA system, Figure 3 shows a schematic diagram of the energy savings calculated by the intermediate model for a single range of the same wet-bulb temperature and IT load (i.e., a single environmental data).

[0097] Based on the relevant description of the above energy-saving calculation formula, the computer device can specifically perform step s11 as follows: traverse multiple environmental data generated by the data center within a preset time period, determine the currently traversed environmental data as the current environmental data, and determine the average value of the device load corresponding to the current environmental data. Specifically, the average value of the device load corresponding to the current environmental data can be determined based on historical experience, or the maximum load value and minimum load value corresponding to the device load interval in the current environmental data can be determined, and the average value of the determined maximum load value and minimum load value can be calculated to obtain the average value of the device load corresponding to the current environmental data. Then, the energy-saving efficiency of the intermediate model under the current environmental data can be calculated based on the energy efficiency indicator value corresponding to the current environmental data in the target energy consumption scenario, the energy efficiency indicator value corresponding to the current environmental data in the reference energy consumption scenario, and the average value of the device load corresponding to the current environmental data. Specifically, the energy efficiency indicator value corresponding to the current environmental data in the target energy consumption scenario and the energy efficiency indicator value corresponding to the current environmental data in the reference energy consumption scenario can be subtracted, and the result of the subtraction can be multiplied by the average value of the device load corresponding to the current environmental data to obtain the energy-saving efficiency of the intermediate model under the current environmental data.

[0098] After calculating the energy-saving efficiency of the intermediate model under the current environmental data, the energy savings value of the intermediate model under the current environmental data can be calculated based on the calculated energy-saving efficiency and the duration of the current environmental data within a preset time period. Specifically, the calculated energy-saving efficiency and the duration of the current environmental data within the preset time period can be multiplied to obtain the energy savings value of the intermediate model under the current environmental data. After traversing multiple environmental data, the energy savings values ​​of the intermediate model under each environmental data can be integrated to obtain the target energy savings value of the intermediate model.

[0099] s12, based on the target energy saving value of the intermediate model and the preset energy saving conditions, the energy saving performance of the intermediate model is tested to obtain the test results.

[0100] Among them, the preset energy-saving condition can be used to indicate: the target energy-saving value of the intermediate model is greater than or equal to the preset energy-saving value; then in this case, the computer device tests the energy-saving performance of the intermediate model based on the target energy-saving value of the intermediate model and the preset energy-saving condition, and when obtaining the test result, the target energy-saving value of the intermediate model and the preset energy-saving value can be compared; if the target energy-saving value of the intermediate model is greater than or equal to the preset energy-saving value, a test result indicating that the intermediate model has passed the performance test is generated; if the target energy-saving value of the intermediate model is less than the preset energy-saving value, a test result indicating that the intermediate model has not passed the performance test is generated. It can be seen that the embodiment of the present application tests the energy-saving performance of the intermediate model by using the size relationship between the preset energy-saving value and the target energy-saving value, which can simplify the testing process and improve testing efficiency.

[0101] Alternatively, the predicted energy-saving condition can be used to indicate that the difference between the target energy-saving value and the historical energy-saving value of the intermediate model must be greater than the target value; the target value can be set based on an empirical value, and the historical energy-saving value refers to the energy-saving value generated by the refrigeration system's own energy-saving logic before the intermediate model is used. In this case, when the computer equipment tests the energy-saving performance of the intermediate model based on the target energy-saving value of the intermediate model and the preset energy-saving conditions and obtains the test result, it can calculate the difference between the target energy-saving value and the historical energy-saving value of the intermediate model. If the difference is greater than the target value, a test result is generated indicating that the intermediate model has passed the performance test. If the difference is less than the target value, a test result is generated indicating that the intermediate model has failed the performance test.

[0102] S204, deploying the target model to the controller of the refrigeration system, enabling the controller to determine the power consumption optimization strategy of the refrigeration system through the target model, wherein the power consumption optimization strategy defines parameter values ​​corresponding to controllable operating parameters of at least one first component, and enabling the controller to control the operation of the corresponding first component in the refrigeration system according to the parameter values ​​corresponding to each controllable operating parameter defined by the determined power consumption optimization strategy.

[0103] In a specific implementation, after the computer device deploys the target model to the controller of the refrigeration system, the controller can use the target model to determine the power consumption optimization strategy of the refrigeration system at a fixed time (for example, the default is 10 minutes), thereby controlling the operation of the corresponding first component in the refrigeration system according to the parameter values ​​corresponding to the controllable operating parameters defined by the determined power consumption optimization strategy. The controller controls the operation of the corresponding first component in the refrigeration system according to the parameter values ​​corresponding to the controllable operating parameters defined by the determined power consumption optimization strategy in a manner that: the parameter values ​​corresponding to the controllable operating parameters defined by the determined power consumption optimization strategy are adopted to update the parameter values ​​corresponding to the controllable operating parameters of the corresponding first component in the refrigeration system, so that the corresponding first component in the refrigeration system operates according to the parameter values ​​corresponding to the controllable operating parameters defined by the determined power consumption optimization strategy.

[0104] The controller determines the power consumption optimization strategy for the refrigeration system using the target model, including: generating multiple parameter combinations based on a strategy optimization algorithm, each parameter combination including parameter values ​​corresponding to each controllable operating parameter. The target model is then invoked to predict the power consumption of the refrigeration system based on each parameter combination, obtaining a power consumption prediction result. The power consumption prediction result includes the predicted power consumption of the refrigeration system under each parameter combination. Based on the power consumption prediction results, a parameter combination is selected from the multiple parameter combinations as the power consumption optimization strategy for the refrigeration system. Therefore, by using the power consumption prediction results as a basis for determining the power consumption optimization strategy, the strategy can be made reliable, thereby improving its reliability and accuracy. Specifically, the parameter combination corresponding to the minimum power consumption in the power consumption prediction results can be directly selected as the power consumption optimization strategy for the refrigeration system. Alternatively, valid power consumption can be selected from the power consumption prediction results. Valid power consumption refers to power consumption greater than a power consumption threshold. A parameter combination corresponding to a power consumption in the valid power consumption is randomly selected as the power consumption optimization strategy for the refrigeration system.

[0105] The embodiment of the present application can determine an initial model for power consumption prediction of a refrigeration system, train the initial model to obtain a target model, and deploy the model in the controller of the refrigeration system, so that the controller can determine the power consumption optimization strategy of the refrigeration system through the target model, and automatically and intelligently tune and control the power consumption of the refrigeration system through the determined power consumption optimization strategy, thereby effectively reducing the energy consumption of the refrigeration system. Furthermore, by using an open toolbox to select a target algorithm for power consumption prediction for the initial model, the processing logic of the initial model can be white-boxed to a certain extent, so that the power consumption optimization strategy determined by the target model obtained by training the initial model has high interpretability and credibility, thereby enabling the power consumption of the refrigeration system to be more accurately tuned and controlled through the determined power consumption optimization strategy, thereby improving the energy consumption reduction effect. In addition, training the initial model through an open source dataset can break the closed algorithm and security implementation, so that the credibility of the trained target model can be effectively improved, thereby further improving the credibility of the power consumption optimization strategy obtained based on the target model, and further improving the energy consumption reduction effect.

[0106] Optionally, the operational architecture proposed in the embodiment of the present application can also standardize metadata such as data points, data sets, algorithms, and models, and construct them into a network topology diagram, so that data can be better managed and utilized through the network topology diagram, thereby improving the value and efficiency of data. Referring to Figure 4a, in the network topology diagram constructed by the operational architecture proposed in the embodiment of the present application, the cloud data concept of AI is divided into four types of node metadata, namely point nodes, data nodes, model nodes, and algorithm nodes; wherein, the description (definition) of each node is as follows:

[0107] (1) Point nodes represent the source of data and the description of the collection point. In the process of data collection and processing, point nodes are used to indicate the specific location or data source of data collection. A point node can correspond to a point, and a point can serve as a data source. In practical applications, the points corresponding to point nodes can usually include sensors, instruments, metering equipment, network interfaces, system interfaces, etc. The points corresponding to each point node can collect specific data, such as physical quantities such as temperature, humidity, pressure, flow, speed, or device status information. In the process of data processing, point nodes can also be used to describe the data collection conditions and collection environment. For example, a point node corresponding to a temperature sensor may include information such as temperature range, measurement accuracy, sampling frequency, installation location, and ambient temperature. This information can help the data processing system better understand and analyze data and improve the accuracy and efficiency of data processing. In the data collection and processing system, the definition and description of point nodes are very important. Through clear point node definition and description, the data processing system can better manage and analyze data and improve data processing efficiency and data quality. At the same time, the definition and description of point nodes also provide more information and reference basis for data analysis and application, which helps to better understand and utilize data.

[0108] (2) Data nodes represent the integration and storage of data. In the process of data management and processing, data nodes are used to indicate the location of data integration and storage. Data integration refers to the integration of data from different data sources into a unified data storage and management system. Data storage refers to the storage of data on a specific storage medium, such as a database, file system, cloud storage, etc. A data node corresponds to a data set. Any data node can be connected to at least one point node or other data node through at least one directed edge, and at least one directed edge points to any data node; the data set corresponding to any data node can be constructed from the data corresponding to each node connected by at least one directed edge. In practical applications, data nodes usually involve three parts: data source, data storage space, and data processing algorithm. Data source refers to the device or system that generates data (i.e., the point mentioned above), such as sensors, instruments, computer systems, etc. Data storage space refers to the medium or system that stores data, such as a database, file system, etc. Data processing algorithm refers to the algorithm that processes and analyzes the data of other nodes that point to the data node through directed edges, such as data mining algorithm, machine learning algorithm, AI algorithm, etc. The definition and description of data nodes are very important for data management and processing. Clear data node definitions and descriptions help system administrators better manage and maintain data storage and processing systems, improving system reliability and efficiency. Furthermore, data node definitions and descriptions provide more information and references for data analysis and applications, facilitating a better understanding and utilization of data. In practical applications, the definition and description of data nodes must consider multiple factors, such as data source, data type, data volume, and data processing requirements. By comprehensively considering these factors, a data node solution tailored to specific application scenarios can be designed.

[0109] (3) Model nodes represent data training and twin model descriptions. In machine learning and AI applications, a model node corresponds to a sub-model. A model refers to a mathematical model used to describe the relationship between data, such as a neural network, a decision tree, a support vector machine, etc. The processing of the sub-model corresponding to the model node can include two parts: model training and model use. Model training refers to using an existing data set to train a model so that the model can predict or classify unknown data. Model training needs to consider multiple factors, such as data quality, data distribution, model complexity, model accuracy, etc. The model training process usually includes steps such as data preprocessing, feature extraction, model selection, and parameter adjustment. Model use refers to using the trained model to predict or classify new data. Model use needs to consider multiple factors, such as data quality, data distribution, model complexity, model accuracy, etc. The model use process usually includes steps such as data preprocessing, feature extraction, model prediction, and result analysis. The definition and description of model nodes are very important for machine learning and AI applications. Through clear model node definition and description, data scientists can better manage and maintain the model training and use process, and improve the accuracy and efficiency of the model. At the same time, the definition and description of model nodes also provide more information and reference for data analysis and application, helping to better understand and utilize the model. In practical applications, the definition and description of model nodes need to consider multiple factors, such as data type, application scenario, model complexity, and model accuracy. By comprehensively considering these factors, a model node solution suitable for a specific application scenario can be designed.

[0110] (4) Algorithm nodes represent model calls and prediction algorithm descriptions. In machine learning and AI applications, an algorithm node corresponds to the prediction algorithm used by a sub-model; it can be seen that algorithm nodes may involve several parts such as model use and prediction algorithms. Model call refers to calling a trained model and inputting new data into the model for prediction or classification. Model call needs to consider multiple factors, such as model complexity, model accuracy, and the design of the call interface. The model call process usually includes steps such as data preprocessing, feature extraction, model call, and result analysis. Prediction algorithm refers to the specific algorithm for predicting or classifying a trained model. Prediction algorithm needs to select appropriate algorithms according to specific application scenarios, such as rule-based algorithms, statistics-based algorithms, neural network-based algorithms, etc. The selection of prediction algorithm needs to consider multiple factors, such as data type, data distribution, algorithm complexity, algorithm accuracy, etc. The definition and description of algorithm nodes are very important for machine learning and AI applications. Strategy algorithm refers to the algorithm that uses multiple models and algorithm logic to find the optimal control execution parameters of the data center cooling system and output the corresponding control strategy. Clear algorithm node definitions and descriptions can help data scientists better manage and maintain model invocation, prediction algorithms, and strategy algorithms, improving algorithm accuracy and efficiency. Furthermore, the definition and description of algorithm nodes provide more information and references for data analysis and application, facilitating a better understanding and utilization of algorithms. In practical applications, the definition and description of algorithm nodes must consider multiple factors, such as data type, application scenario, algorithm complexity, and accuracy. By comprehensively considering these factors, an algorithm node solution tailored to a specific application scenario can be designed.

[0111] Furthermore, the above four nodes can be connected through directed edges. The abstract description of their connection relationship is as follows:

[0112] (1) The connection between multiple point nodes and a data node is described as data generation. That is, when multiple point nodes are connected to a data node through directed edges, the data processing algorithm involved in the data node can be used to process the parameter values ​​collected at the points corresponding to these multiple point nodes, thereby generating a data set corresponding to the data node.

[0113] (2) The connection between multiple data nodes is described as data processing. That is, when multiple data nodes are connected to another data node through directed edges, the data processing algorithm involved in the other data node can be used to process the data sets corresponding to the multiple data nodes, thereby obtaining the data set corresponding to the other data node.

[0114] (3) The connection between a single data node and a model node is described as model processing. That is, when a data node is connected to a model node via a directed edge, the data set corresponding to the data node can be used to perform model training or model application on the sub-model corresponding to the model node. Specifically, the data set corresponding to the data node is used as the input of the sub-model corresponding to the model node, so that the sub-model corresponding to the model node can perform prediction processing based on the input data; if it is model training, the corresponding sub-model is optimized based on the results of the prediction processing; if it is model application, the results of the prediction processing are directly output.

[0115] (4) The connection between multiple model nodes and an algorithm node is described as an algorithm call. That is, when multiple model nodes are connected to an algorithm node through directed edges, the sub-model corresponding to each model node in these multiple model nodes calls the prediction algorithm corresponding to the algorithm node to perform prediction processing. In addition, the connection between a single model node and a model node can be described as model coupling. That is, when a model node is connected to another model node through a directed edge, the output of the sub-model corresponding to the model node needs to be used as the input of the sub-model corresponding to the other model node. For example, as shown in Figure 4b, the temperature difference value on the cooling side of the refrigeration host output by the refrigerator temperature difference prediction model can be used as the input of the refrigerator power prediction model.

[0116] Ultimately, by connecting multiple algorithm nodes for computation, a network topology diagram can be formed for policy calculation. Algorithm nodes can collaborate to complete complex prediction and analysis tasks, further improving the value and efficiency of data. In data centers, this network topology can be used for computation and analysis to determine the optimal energy-saving conditions for the cooling system and output corresponding power optimization strategies, achieving energy-saving optimization of the data center's cooling system.

[0117] Based on the above description, in practical applications, the embodiment of the present application can configure a network topology for the refrigeration system, thereby improving the efficiency of model training, tuning control, and other links through the network topology. Specifically, the network topology may include T point nodes, X data nodes, Y model nodes, and Z algorithm nodes; T, X, Y, and Z are all positive integers; where:

[0118] A point node corresponds to a point, which is used to collect parameter values ​​of components in the refrigeration system;

[0119] One data node corresponds to one data set. For the xth data node, there are N nodes in the network topology graph, and each of the N nodes points to the xth data node through a directed edge. The data set corresponding to the xth data node is constructed based on the data corresponding to the N nodes. The N nodes include at least one of the following: at least one point node and at least one data node other than the xth data node; x∈[1,X], where N is a positive integer.

[0120] One model node corresponds to one sub-model. For the yth model node, there are M nodes in the network topology graph, and each of the M nodes points to the yth model node through a directed edge. The sub-model corresponding to the yth model node performs model processing according to the data corresponding to the M nodes. The M nodes include at least one of the following: at least one data node and at least one model node other than the yth model node; y∈[1, Y], M is a positive integer;

[0121] One algorithm node corresponds to one prediction algorithm. For the zth algorithm node, there are h model nodes in the network topology diagram, and each of the h model nodes points to the zth algorithm node through a directed edge. The prediction algorithm corresponding to the zth algorithm node is called by the model corresponding to each model node in the h model nodes, z∈[1, Z], h∈[1, X].

[0122] Based on this, an embodiment of the present application proposes another method for operating a refrigeration system; as shown in FIG5 , the method for operating a refrigeration system may include the following steps S501-S506:

[0123] S501. Obtain an open source dataset of a refrigeration system, where the refrigeration system includes multiple components, each component is configured with one or more parameters, and the open source dataset includes parameter values ​​corresponding to the parameters configured for each component; wherein there is at least one first component among the multiple components, and the parameters configured for each of the first components include controllable operating parameters.

[0124] S502 : determining an initial model for power consumption prediction of a refrigeration system, selecting a target algorithm using an open toolbox, and applying the target algorithm to the initial model for power consumption prediction.

[0125] The initial model includes: a sub-model configured for each of the components in the refrigeration system. In a specific implementation, when determining the initial model for power consumption prediction of the refrigeration system, the computer device can traverse each point in any component in the refrigeration system, determine the point node corresponding to the currently traversed point from the network topology diagram, and determine the model nodes indirectly connected to the point node through multiple directed edges. The sub-model corresponding to the model node is determined as a sub-model corresponding to the component. By continuing the traversal, the sub-models corresponding to the component can be determined.

[0126] Correspondingly, the target algorithm includes: the prediction algorithm used by each sub-model in the initial model. Similar to the model determination method mentioned above, when using the open toolbox to select the target algorithm, the computer device can locate the model node corresponding to any sub-model in the initial model from the network topology diagram; then, based on the topological structure of the network topology diagram, use the open toolbox to use the prediction algorithm corresponding to the algorithm node pointed to by the directed edge of the located model node as the prediction algorithm used by any sub-model. It can be seen that by using the network topology diagram to select the algorithm, the efficiency of algorithm selection can be improved.

[0127] S503: Use the open source dataset and the target algorithm to train the initial model to obtain the target model.

[0128] The training process for any sub-model in the initial model includes obtaining training data for each sub-model based on the parameter values ​​corresponding to each parameter of each component in the open source dataset and the topological structure of the refrigeration system network topology, thereby improving the efficiency of training data acquisition. After obtaining the training data for any sub-model, the sub-model can be trained using the training data using a supervised training method. The specific training method can be found in the description of the method embodiment shown in Figure 2 above and will not be repeated here.

[0129] The method for obtaining training data for any sub-model may include: determining a model node corresponding to any sub-model in a network topology graph, and determining each point node related to the model node based on the topological structure of the network topology graph, wherein each determined point node is indirectly connected to the model node through multiple directed edges, and each directed edge points to the model node. Parameter values ​​collected from the points corresponding to each determined point node are screened from an open source dataset, and each screened parameter value is input into the network topology graph for graph calculation to obtain training data for any sub-model. The graph calculation mentioned herein may include: performing data processing on the parameter values ​​corresponding to the corresponding point node according to the data processing algorithm involved in the data nodes connected to each determined point node to obtain a dataset corresponding to the corresponding data node. If the data node is directly connected to the model node corresponding to any sub-model through a directed edge, the dataset corresponding to the corresponding data node can be used as training data for any sub-model. If the data node is not directly connected to the model node corresponding to any sub-model, the dataset corresponding to the data node is further processed based on the topological structure of the network topology graph to obtain training data for any sub-model.

[0130] S504, storing the target model in the model warehouse, allowing the model warehouse to deploy the target model to the controller of the refrigeration system, allowing the controller to determine the power consumption optimization strategy of the refrigeration system through the target model, and allowing the controller to control the operation of the corresponding first component in the refrigeration system according to the parameter values ​​corresponding to each controllable operating parameter defined by the determined power consumption optimization strategy.

[0131] Furthermore, after the target model is deployed to the controller, after the controller uses the target model for a preset period of time, the time period during which the controller uses the target model can be used as the target time period, and the first energy consumption situation of the data center during the target time period can be counted; the data center can generate multiple environmental data during the target time period, and the first energy consumption situation includes: when the target model is used, the energy efficiency indicator value of the data center under each environmental data. And, the second energy consumption situation of the data center is obtained, and the second reference energy consumption situation includes: when the target model is not used, the energy efficiency indicator value of the data center under each environmental data. Then, the energy-saving performance of the target model can be tested based on the first energy consumption situation and the second energy consumption situation. The testing method is similar to the testing method of the intermediate model mentioned above, and will not be repeated here. If it is determined that the target model has not passed the energy-saving performance test, the target model can be trained again to update the target model.

[0132] Accordingly, whenever the model warehouse detects an update to the target model, it performs an online update on the target model deployed in the controller based on the updated target model to ensure that the target model in the controller is the latest model. This allows the model performance of the target model in the controller to be promptly improved, thereby improving the accuracy of the power consumption optimization strategy determined by the controller based on the target model and ensuring energy consumption reduction. Specifically, the model warehouse can send the model parameters of the updated target model to the controller, causing the controller to update the model parameters of the internally deployed target model to the model parameters sent by the model warehouse, thereby achieving online updates of the target model.

[0133] S505: Perform service monitoring on the controller to obtain a service monitoring result.

[0134] The service monitoring results include at least one of the following: the operating status of the controller and the operating status of the target model within the controller. If the target model includes multiple sub-models, the operating status of the target model includes the operating status of each sub-model. For the target model within the controller, a watchdog can be configured for each sub-model within the target model within the controller. The configured watchdog monitors the operating status of the corresponding sub-model in real time, thereby obtaining the operating status of each sub-model within the target model.

[0135] S506: Detect a fault point on the controller according to the service monitoring result, and repair the fault point when a fault point is detected.

[0136] In a specific implementation, after a fault point is detected, a prompt message can be output to the operation and maintenance personnel so that the operation and maintenance personnel can repair the fault point in a timely manner, thereby ensuring the performance of the controller and enabling it to operate normally, thereby enabling the controller to reduce the energy consumption of the refrigeration system. The fault point includes at least one of the following: a hardware facility that fails in the controller, and a model that fails in the controller. Specifically, if the service monitoring result includes the operating status of the controller, the operating status of the controller in the service monitoring result and the specified operating status of the controller can be checked for consistency; if they are consistent, it is determined that there is no fault point in the controller; if they are inconsistent, it is determined that there is a fault point in the controller. If the service monitoring result includes the operating status of the target model, the operating status of each sub-model in the target model in the service monitoring result and the specified model operating status can be checked for consistency; if the operating status of a sub-model is inconsistent with the specified model status, it is determined that a fault point is detected in the target model, and the existing sub-model can be used as a fault point; if the operating status of a sub-model is inconsistent with the specified model status, it is determined that no fault point is detected in the target model.

[0137] The embodiment of the present application can determine an initial model for power consumption prediction of a refrigeration system, train the initial model to obtain a target model, and deploy the model in the controller of the refrigeration system, so that the controller can determine the power consumption optimization strategy of the refrigeration system through the target model, and automatically and intelligently tune and control the power consumption of the refrigeration system through the determined power consumption optimization strategy, thereby effectively reducing the energy consumption of the refrigeration system. Furthermore, by using an open toolbox to select a target algorithm for power consumption prediction for the initial model, the processing logic of the initial model can be white-boxed to a certain extent, so that the power consumption optimization strategy determined by the target model obtained by training the initial model has high interpretability and credibility, thereby enabling the power consumption of the refrigeration system to be more accurately tuned and controlled through the determined power consumption optimization strategy, thereby improving the energy consumption reduction effect. In addition, training the initial model through an open source dataset can break the closed algorithm and security implementation, so that the credibility of the trained target model can be effectively improved, thereby further improving the credibility of the power consumption optimization strategy obtained based on the target model, and further improving the energy consumption reduction effect.

[0138] Based on the description of the above-mentioned refrigeration system operation method embodiment, the present application also discloses a refrigeration system operation device; the refrigeration system operation device can be a computer program (including one or more instructions) running on a computer device, and the refrigeration system operation device can execute each step of the method flow shown in Figure 2 or Figure 5. Referring to Figure 6, the refrigeration system operation device can operate the following units:

[0139] An acquisition unit 601 is configured to acquire an open source dataset of a refrigeration system, wherein the refrigeration system includes multiple components, each of which is configured with one or more parameters, and the open source dataset includes parameter values ​​corresponding to the parameters configured for each of the components; wherein at least one first component is included among the multiple components, and the parameters configured for each of the first components include controllable operating parameters;

[0140] The processing unit 602 is configured to determine an initial model for predicting power consumption of the refrigeration system, and select a target algorithm using an open toolbox, wherein the target algorithm is applied to the initial model for predicting power consumption;

[0141] The processing unit 602 is further configured to train the initial model using the open source dataset and the target algorithm to obtain a target model;

[0142] The processing unit 602 is also used to deploy the target model to the controller of the refrigeration system, so that the controller determines the power consumption optimization strategy of the refrigeration system through the target model, and the power consumption optimization strategy defines the parameter value corresponding to the controllable operating parameter of at least one of the first components; and enables the controller to control the operation of the corresponding first component in the refrigeration system according to the parameter value corresponding to each controllable operating parameter defined by the determined power consumption optimization strategy.

[0143] In another embodiment, when the processing unit 602 is used to train the initial model using the open source dataset and the target algorithm to obtain the target model, it can be specifically used to:

[0144] Using the open source dataset and the target algorithm to train the initial model to obtain an intermediate model;

[0145] Performing a performance test on the intermediate model through a simulation system to obtain a test result; wherein the simulation system includes: a digital twin system corresponding to the refrigeration system established by integrating a mechanism model and data drive;

[0146] If the test result indicates that the intermediate model passes the performance test, the intermediate model is used as the target model;

[0147] If the test result indicates that the intermediate model fails the performance test, the intermediate model is trained to obtain a target model.

[0148] In another embodiment, the intermediate model includes an intermediate sub-model obtained by training each sub-model; accordingly, when the processing unit 602 is used to perform a performance test on the intermediate model through the simulation system to obtain the test result, it can be specifically used to:

[0149] For the kth intermediate sub-model in the intermediate model, test data of the kth intermediate sub-model is obtained by running the simulation system; k∈[1, K], K is the number of intermediate sub-models included in the intermediate model;

[0150] The test data is used to test the accuracy performance of the kth intermediate sub-model to obtain a test result.

[0151] In another embodiment, the refrigeration system is located in a data center; accordingly, when the processing unit 602 is used to perform a performance test on the intermediate model using the simulation system to obtain the test results, it can be specifically used to:

[0152] During a preset time period, the operation of the simulation system is controlled by the intermediate model; after the preset time period ends, the target energy consumption of the data center during the preset time period is counted; wherein the data center generates a plurality of environmental data during the preset time period, and the target energy consumption includes: an energy efficiency indicator value of the data center under each of the plurality of environmental data;

[0153] Acquire a reference energy consumption condition of the data center, the reference energy consumption condition comprising: an energy efficiency indicator value of the data center under each of the plurality of environmental data when the intermediate model is not used;

[0154] According to the target energy consumption situation and the reference energy consumption situation, the energy-saving performance of the intermediate model is tested to obtain a test result.

[0155] In another embodiment, when the processing unit 602 is used to control the operation of the simulation system through the intermediate model within a preset time period, it can be specifically used to:

[0156] Running each component in the simulation system;

[0157] Within a preset time period, regularly calculating the power consumption optimization strategy of the simulation system through the intermediate model;

[0158] Each time a set of power consumption optimization strategies is calculated, the operation of the corresponding first component in the simulation system is controlled according to the parameter values ​​corresponding to the controllable operating parameters defined by the currently calculated power consumption optimization strategies.

[0159] In another embodiment, when the processing unit 602 is used to test the energy-saving performance of the intermediate model according to the target energy consumption and the reference energy consumption to obtain the test result, it can be specifically used to:

[0160] Performing an energy-saving evaluation on the intermediate model according to the target energy consumption and the reference energy consumption to obtain a target energy-saving value of the intermediate model;

[0161] Based on the target energy saving value of the intermediate model and the preset energy saving conditions, the energy saving performance of the intermediate model is tested to obtain a test result.

[0162] In another embodiment, the energy efficiency indicator value is a ratio between the total energy consumed by the data center and the energy consumed by the equipment load of the data center; accordingly, when the processing unit 602 is configured to perform energy-saving evaluation on the intermediate model based on the target energy consumption and the reference energy consumption to obtain the target energy saving value of the intermediate model, it can be specifically configured to:

[0163] Traversing multiple environmental data generated by the data center within the preset time period, determining the currently traversed environmental data as the current environmental data, and determining the average value of the equipment load corresponding to the current environmental data;

[0164] Calculating the energy-saving efficiency of the intermediate model under the current environmental data based on the energy efficiency indicator value corresponding to the current environmental data in the target energy consumption scenario, the energy efficiency indicator value corresponding to the current environmental data in the reference energy consumption scenario, and an average value of the device load corresponding to the current environmental data;

[0165] Calculating the energy saving value of the intermediate model under the current environmental data according to the calculated energy-saving efficiency and the duration of the current environmental data within the preset time period;

[0166] After all the plurality of environmental data are traversed, the energy consumption saving values ​​of the intermediate model under each environmental data are integrated to obtain the target energy consumption saving value of the intermediate model.

[0167] In another embodiment, one environmental data item refers to a combination of a wet-bulb temperature and an equipment load interval; accordingly, when determining the average value of the equipment load corresponding to the current environmental data, the processing unit 602 may be specifically configured to:

[0168] Determining a maximum load value and a minimum load value corresponding to the device load interval in the current environment data;

[0169] An average operation is performed on the determined maximum load value and minimum load value to obtain an average value of the device load corresponding to the current environment data.

[0170] In another embodiment, when the processing unit 602 is used to calculate the energy-saving efficiency of the intermediate model under the current environmental data based on the energy efficiency indicator value corresponding to the current environmental data in the target energy consumption situation, the energy efficiency indicator value corresponding to the current environmental data in the reference energy consumption situation, and the average of the device load corresponding to the current environmental data, it can be specifically used to:

[0171] performing a difference operation on the energy efficiency indicator value corresponding to the current environmental data in the target energy consumption situation and the energy efficiency indicator value corresponding to the current environmental data in the reference energy consumption situation;

[0172] The result of the difference operation is multiplied by the average value of the equipment load corresponding to the current environmental data to obtain the energy-saving efficiency of the intermediate model under the current environmental data.

[0173] In another embodiment, the preset energy-saving condition is used to indicate that the target energy-saving value of the intermediate model is greater than or equal to the preset energy-saving value; accordingly, when the processing unit 602 is used to test the energy-saving performance of the intermediate model based on the target energy-saving value of the intermediate model and the preset energy-saving condition and obtain the test result, it can be specifically used to:

[0174] If the target energy saving value of the intermediate model is greater than or equal to the preset energy saving value, generating a test result indicating that the intermediate model passes the performance test;

[0175] If the target energy saving value of the intermediate model is less than the preset energy saving value, a test result indicating that the intermediate model has failed the performance test is generated.

[0176] In another embodiment, the processing unit 602 may be used for a training process of any sub-model in the initial model, the training process comprising: obtaining training data of the any sub-model according to parameter values ​​corresponding to parameters of each component in the open source dataset and the topological structure of the network topology diagram of the refrigeration system;

[0177] According to a supervised training method, the training data is used to train any of the sub-models.

[0178] In another embodiment, when the processing unit 602 is used to select a target algorithm using an open toolbox, it may be specifically configured to:

[0179] For any sub-model in the initial model, locate the model node corresponding to the any sub-model from the network topology diagram;

[0180] According to the topological structure of the network topology diagram, an open toolbox is used to use the prediction algorithm corresponding to the algorithm node pointed to by the directed edge of the located model node as the prediction algorithm adopted by any sub-model.

[0181] In another embodiment, when the processing unit 602 is used to deploy the target model into the controller of the refrigeration system, it can be specifically used to:

[0182] Storing the target model in a model warehouse, so that the model warehouse deploys the target model to a controller of the refrigeration system;

[0183] Wherein, whenever the model warehouse detects that the target model is updated, the target model deployed in the controller is updated online according to the updated target model.

[0184] In another embodiment, after deploying the target model to the controller, the processing unit 602 may further be configured to:

[0185] Performing service monitoring on the controller to obtain a service monitoring result; the service monitoring result includes at least one of the following: an operating state of the controller, and an operating state of a target model in the controller;

[0186] The controller is detected for a fault point based on the service monitoring result, and when a fault point is detected, the fault point is repaired; the fault point includes at least one of the following: a hardware facility that fails in the controller, and a model that fails in the controller.

[0187] In another embodiment, the process of the controller determining the power consumption optimization strategy of the refrigeration system through the target model includes:

[0188] Generate multiple sets of parameter combinations according to the strategy optimization algorithm, each set of parameter combinations includes parameter values ​​corresponding to each controllable operating parameter;

[0189] Calling the target model to predict the power consumption of the refrigeration system according to each parameter combination to obtain a power consumption prediction result; the power consumption prediction result includes: the predicted power consumption of the refrigeration system under each parameter combination;

[0190] According to the power consumption prediction result, a group of parameter combinations is selected from the multiple groups of parameter combinations as a power consumption optimization strategy for the refrigeration system.

[0191] According to another embodiment of the present application, the various units in the operating device of the refrigeration system shown in Figure 6 can be individually or all combined into one or several other units to form, or one (or some) of the units can be further divided into multiple functionally smaller units to form, which can achieve the same operation without affecting the realization of the technical effects of the embodiments of the present application. The above-mentioned units are divided based on logical functions. In actual applications, the functions of one unit can also be implemented by multiple units, or the functions of multiple units can be implemented by one unit. In other embodiments of the present application, the operating device based on the refrigeration system can also include other units. In actual applications, these functions can also be implemented with the assistance of other units, and can be implemented by the collaboration of multiple units.

[0192] According to another embodiment of the present application, a computer program (including one or more instructions) capable of executing each step involved in the corresponding method shown in FIG. 2 or FIG. 5 can be run on a general-purpose computing device such as a computer, which includes processing elements and storage elements such as a central processing unit (CPU), a random access memory (RAM), and a read-only memory (ROM), to construct an operating device for a refrigeration system as shown in FIG. 6 and implement the operating method of the refrigeration system of the embodiment of the present application. The computer program can be recorded on, for example, a computer-readable storage medium, and loaded into the computing device via the computer-readable storage medium and run therein.

[0193] The embodiments of the present application can determine an initial model for power consumption prediction for a refrigeration system. By training the initial model to obtain a target model, and deploying the model in the controller of the refrigeration system, the controller can determine a power consumption optimization strategy for the refrigeration system based on the target model. The controller can then automatically and intelligently optimize and control the power consumption of the refrigeration system based on the determined power consumption optimization strategy, effectively reducing the energy consumption of the refrigeration system. Furthermore, by using an open toolbox to select a target algorithm for power consumption prediction for the initial model, the processing logic of the initial model can be white-boxed to a certain extent. This allows the power consumption optimization strategy determined by the target model obtained by training the initial model to have high interpretability and credibility, thereby enabling the power consumption of the refrigeration system to be more accurately optimized and controlled based on the determined power consumption optimization strategy, thereby improving the energy consumption reduction effect. In addition, training the initial model with an open source dataset can break the closed algorithm and security implementation, effectively improving the credibility of the trained target model, thereby further improving the credibility of the power consumption optimization strategy derived based on the target model, and further improving the energy consumption reduction effect.

[0194] Based on the description of the above method embodiment and device embodiment, the embodiment of the present application also provides a computer device. Referring to Figure 7, the computer device at least includes a processor 701, an input interface 702, an output interface 703 and a computer storage medium 704. Among them, the processor 701, the input interface 702, the output interface 703 and the computer storage medium 704 in the computer device can be connected via a bus or other means. The computer storage medium 704 can be stored in the memory of the computer device, and the computer storage medium 704 is used to store a computer program, and the computer program includes one or more instructions, and the processor 701 is used to execute one or more instructions in the computer program stored in the computer storage medium 704. The processor 701 (or CPU (Central Processing Unit, central processing unit)) is the computing core and control core of the computer device, which is suitable for implementing one or more instructions, and is specifically suitable for loading and executing one or more instructions to realize the corresponding method flow or corresponding function.

[0195] In one embodiment, the processor 701 described in the embodiment of the present application can be used to perform a series of energy-saving processing on the refrigeration system, specifically including: obtaining an open source data set of the refrigeration system, the refrigeration system including multiple components, each of the components configured with one or more parameters, and the open source data set including parameter values ​​corresponding to the parameters configured for each component; wherein there is at least one first component among the multiple components, and the parameters configured for each of the first components include controllable operating parameters; determining an initial model for power consumption prediction of the refrigeration system, and using an open toolbox to select a target algorithm, the target algorithm is applied to the initial model for power consumption prediction; using the open source data set and the target algorithm to train the initial model to obtain a target model; deploying the target model to the controller of the refrigeration system, so that the controller determines the power consumption optimization strategy of the refrigeration system through the target model, the power consumption optimization strategy defines the parameter value corresponding to the controllable operating parameter of at least one of the first components, and the controller controls the operation of the corresponding first component in the refrigeration system according to the parameter value corresponding to each controllable operating parameter defined by the determined power consumption optimization strategy, and so on.

[0196] The embodiment of the present application also provides a computer storage medium (Memory), which is a memory device in a computer device for storing computer programs and data. It is understandable that the computer storage medium here can include both built-in storage media in the computer device and, of course, extended storage media supported by the computer device. The computer storage medium provides a storage space that stores the operating system of the computer device. In addition, a computer program is also stored in the storage space, which includes one or more instructions suitable for being loaded and executed by the processor 701, and these instructions can be one or more program codes. It should be noted that the computer storage medium here can be a high-speed RAM memory, or a non-volatile memory (non-volatile memory), such as at least one disk storage; optionally, it can also be at least one computer storage medium located away from the aforementioned processor.

[0197] In one embodiment, a processor may load and execute one or more instructions stored in a computer storage medium to implement the corresponding steps in the method embodiments shown in FIG. 2 or FIG. 5 . In a specific implementation, the processor may load and execute the following steps:

[0198] Obtaining an open source dataset of a refrigeration system, the refrigeration system comprising a plurality of components, each of the components being configured with one or more parameters, the open source dataset comprising parameter values ​​corresponding to the parameters configured for each of the components; wherein at least one first component is present among the plurality of components, and the parameters configured for each of the first components include controllable operating parameters;

[0199] Determining an initial model for power consumption prediction of the refrigeration system, and selecting a target algorithm using an open toolbox, wherein the target algorithm is applied to the initial model for power consumption prediction;

[0200] Using the open source dataset and the target algorithm to train the initial model to obtain a target model;

[0201] deploying the target model to a controller of the refrigeration system, so that the controller determines a power consumption optimization strategy for the refrigeration system based on the target model, wherein the power consumption optimization strategy defines a parameter value corresponding to a controllable operating parameter of at least one of the first components;

[0202] Furthermore, the controller is enabled to control the operation of the corresponding first component in the refrigeration system according to the parameter values ​​corresponding to the controllable operating parameters defined in the determined power consumption optimization strategy.

[0203] In one embodiment, when the model is trained using the open source dataset and the target algorithm to obtain a target model, the one or more instructions may be loaded and specifically executed by the processor:

[0204] Using the open source dataset and the target algorithm to train the initial model to obtain an intermediate model;

[0205] Performing a performance test on the intermediate model through a simulation system to obtain a test result; wherein the simulation system includes: a digital twin system corresponding to the refrigeration system established by integrating a mechanism model and data drive;

[0206] If the test result indicates that the intermediate model passes the performance test, the intermediate model is used as the target model;

[0207] If the test result indicates that the intermediate model fails the performance test, the intermediate model is trained to obtain a target model.

[0208] In another embodiment, the intermediate model includes an intermediate sub-model obtained by training each sub-model; accordingly, when the performance test of the intermediate model is performed through a simulation system and the test results are obtained, the one or more instructions can be loaded and specifically executed by the processor:

[0209] For the kth intermediate sub-model in the intermediate model, test data of the kth intermediate sub-model is obtained by running the simulation system; k∈[1, K], K is the number of intermediate sub-models included in the intermediate model;

[0210] The test data is used to test the accuracy performance of the kth intermediate sub-model to obtain a test result.

[0211] In another embodiment, the refrigeration system is located in a data center; accordingly, when a performance test is performed on the intermediate model through a simulation system to obtain a test result, the one or more instructions may be loaded and specifically executed by the processor:

[0212] During a preset time period, the operation of the simulation system is controlled by the intermediate model; after the preset time period ends, the target energy consumption of the data center during the preset time period is counted; wherein the data center generates a plurality of environmental data during the preset time period, and the target energy consumption includes: an energy efficiency indicator value of the data center under each of the plurality of environmental data;

[0213] Acquire a reference energy consumption condition of the data center, the reference energy consumption condition comprising: an energy efficiency indicator value of the data center under each of the plurality of environmental data when the intermediate model is not used;

[0214] According to the target energy consumption situation and the reference energy consumption situation, the energy-saving performance of the intermediate model is tested to obtain a test result.

[0215] In another embodiment, during a preset time period, when the intermediate model is used to control the operation of the simulation system, the one or more instructions may be loaded and specifically executed by the processor:

[0216] Running each component in the simulation system;

[0217] Within a preset time period, regularly calculating the power consumption optimization strategy of the simulation system through the intermediate model;

[0218] Each time a set of power consumption optimization strategies is calculated, the operation of the corresponding first component in the simulation system is controlled according to the parameter values ​​corresponding to the controllable operating parameters defined by the currently calculated power consumption optimization strategies.

[0219] In another embodiment, when the energy-saving performance of the intermediate model is tested according to the target energy consumption and the reference energy consumption, and the test results are obtained, the one or more instructions may be loaded and specifically executed by the processor:

[0220] Performing an energy-saving evaluation on the intermediate model according to the target energy consumption and the reference energy consumption to obtain a target energy-saving value of the intermediate model;

[0221] Based on the target energy saving value of the intermediate model and the preset energy saving conditions, the energy saving performance of the intermediate model is tested to obtain a test result.

[0222] In another embodiment, the energy efficiency indicator value is a ratio between the total energy consumed by the data center and the energy consumed by the equipment load of the data center; accordingly, when performing an energy-saving assessment on the intermediate model based on the target energy consumption and the reference energy consumption to obtain the target energy saving value of the intermediate model, the one or more instructions may be loaded and specifically executed by the processor:

[0223] Traversing multiple environmental data generated by the data center within the preset time period, determining the currently traversed environmental data as the current environmental data, and determining the average value of the equipment load corresponding to the current environmental data;

[0224] Calculating the energy-saving efficiency of the intermediate model under the current environmental data based on the energy efficiency indicator value corresponding to the current environmental data in the target energy consumption scenario, the energy efficiency indicator value corresponding to the current environmental data in the reference energy consumption scenario, and an average value of the device load corresponding to the current environmental data;

[0225] Calculating the energy saving value of the intermediate model under the current environmental data according to the calculated energy-saving efficiency and the duration of the current environmental data within the preset time period;

[0226] After all the plurality of environmental data are traversed, the energy consumption saving values ​​of the intermediate model under each environmental data are integrated to obtain the target energy consumption saving value of the intermediate model.

[0227] In another embodiment, the environmental data refers to a combination of a wet-bulb temperature and a device load interval; accordingly, when determining the average value of the device load corresponding to the current environmental data, the one or more instructions may be loaded and specifically executed by the processor:

[0228] Determining a maximum load value and a minimum load value corresponding to the device load interval in the current environment data;

[0229] An average operation is performed on the determined maximum load value and minimum load value to obtain an average value of the device load corresponding to the current environment data.

[0230] In another embodiment, when calculating the energy-saving efficiency of the intermediate model under the current environmental data based on the energy efficiency indicator value corresponding to the current environmental data in the target energy consumption scenario, the energy efficiency indicator value corresponding to the current environmental data in the reference energy consumption scenario, and the average of the device load corresponding to the current environmental data, the one or more instructions may be loaded and specifically executed by the processor:

[0231] performing a difference operation on the energy efficiency indicator value corresponding to the current environmental data in the target energy consumption situation and the energy efficiency indicator value corresponding to the current environmental data in the reference energy consumption situation;

[0232] The result of the difference operation is multiplied by the average value of the equipment load corresponding to the current environmental data to obtain the energy-saving efficiency of the intermediate model under the current environmental data.

[0233] In another embodiment, the preset energy-saving condition is used to indicate that the target energy-saving value of the intermediate model is greater than or equal to the preset energy-saving value; accordingly, when the energy-saving performance of the intermediate model is tested based on the target energy-saving value of the intermediate model and the preset energy-saving condition, and the test result is obtained, the one or more instructions may be loaded and specifically executed by the processor:

[0234] If the target energy saving value of the intermediate model is greater than or equal to the preset energy saving value, generating a test result indicating that the intermediate model passes the performance test;

[0235] If the target energy saving value of the intermediate model is less than the preset energy saving value, a test result indicating that the intermediate model has failed the performance test is generated.

[0236] In another embodiment, the one or more instructions may be loaded and specifically executed by a processor:

[0237] For the training process of any sub-model in the target model, the training data of the sub-model is obtained according to the parameter values ​​corresponding to the parameters of each component in the open source data set and the topological structure of the network topology diagram of the refrigeration system;

[0238] According to a supervised training method, the training data is used to train any of the sub-models.

[0239] In another embodiment, when the open toolbox is used to select a target algorithm, the one or more instructions may be loaded and specifically executed by the processor:

[0240] For any sub-model in the initial model, locate the model node corresponding to the any sub-model from the network topology diagram;

[0241] According to the topological structure of the network topology diagram, an open toolbox is used to use the prediction algorithm corresponding to the algorithm node pointed to by the directed edge of the located model node as the prediction algorithm adopted by any sub-model.

[0242] In another embodiment, when the target model is deployed in the controller of the refrigeration system, the one or more instructions may be loaded and specifically executed by a processor:

[0243] Storing the target model in a model warehouse, so that the model warehouse deploys the target model to a controller of the refrigeration system;

[0244] Wherein, whenever the model warehouse detects that the target model is updated, the target model deployed in the controller is updated online according to the updated target model.

[0245] In another embodiment, after the target model is deployed to the controller, the one or more instructions may be loaded and specifically executed by the processor:

[0246] Performing service monitoring on the controller to obtain a service monitoring result; the service monitoring result includes at least one of the following: an operating state of the controller, and an operating state of a target model in the controller;

[0247] The controller is detected for a fault point based on the service monitoring result, and when a fault point is detected, the fault point is repaired; the fault point includes at least one of the following: a hardware facility that fails in the controller, and a model that fails in the controller.

[0248] In another embodiment, the process of the controller determining the power consumption optimization strategy of the refrigeration system through the target model includes:

[0249] Generate multiple sets of parameter combinations according to the strategy optimization algorithm, each set of parameter combinations includes parameter values ​​corresponding to each controllable operating parameter;

[0250] Calling the target model to predict the power consumption of the refrigeration system according to each parameter combination to obtain a power consumption prediction result; the power consumption prediction result includes: the predicted power consumption of the refrigeration system under each parameter combination;

[0251] According to the power consumption prediction result, a group of parameter combinations is selected from the multiple groups of parameter combinations as a power consumption optimization strategy for the refrigeration system.

[0252] The embodiment of the present application can determine an initial model for power consumption prediction of a refrigeration system, train the initial model to obtain a target model, and deploy the model in the controller of the refrigeration system, so that the controller can determine the power consumption optimization strategy of the refrigeration system through the target model, and automatically and intelligently tune and control the power consumption of the refrigeration system through the determined power consumption optimization strategy, thereby effectively reducing the energy consumption of the refrigeration system. Furthermore, by using an open toolbox to select a target algorithm for power consumption prediction for the initial model, the processing logic of the initial model can be white-boxed to a certain extent, so that the power consumption optimization strategy determined by the target model obtained by training the initial model has high interpretability and credibility, thereby enabling the power consumption of the refrigeration system to be more accurately tuned and controlled through the determined power consumption optimization strategy, thereby improving the energy consumption reduction effect. In addition, training the initial model through an open source dataset can break the closed algorithm and security implementation, so that the credibility of the trained target model can be effectively improved, thereby further improving the credibility of the power consumption optimization strategy obtained based on the target model, and further improving the energy consumption reduction effect.

[0253] It should be noted that, according to one aspect of the present application, a computer program product or computer program is also provided, which includes one or more instructions, and the one or more instructions are stored in a computer storage medium. The processor of the computer device reads the one or more instructions from the computer storage medium, and the processor executes the one or more instructions, so that the computer device performs the method provided in the various optional embodiments of the method embodiment shown in Figures 2 or 5 above. It should be understood that what is disclosed above is only a preferred embodiment of the present application, and it is certainly not used to limit the scope of rights of the present application. Therefore, equivalent changes made in accordance with the claims of the present application are still within the scope covered by the present application.

Claims

1. A method for operating a refrigeration system, characterized in that: include: Acquire an open source data set of a refrigeration system, wherein the refrigeration system includes multiple components, each of the components is configured with one or more parameters, and the open source data set includes parameter values ​​corresponding to the parameters configured for each of the components; wherein there is at least one first component among the multiple components, and the parameters configured for each of the first components include controllable operating parameters; Determining an initial model for predicting power consumption of the refrigeration system, and selecting a target algorithm using an open toolbox, wherein the target algorithm is applied to the initial model for predicting power consumption; The initial model is trained using the open source data set and the target algorithm to obtain a target model; Deploying the target model to a controller of the refrigeration system, so that the controller determines a power consumption optimization strategy of the refrigeration system through the target model, wherein the power consumption optimization strategy defines a parameter value corresponding to at least one controllable operating parameter of the first component; And, the controller is enabled to control the operation of the corresponding first component in the refrigeration system according to the determined parameter values ​​corresponding to each controllable operating parameter defined by the power consumption optimization strategy.

2. The method according to claim 1, characterized in that The initial model includes: a sub-model configured for each of the components in the refrigeration system, and the target algorithm includes: a prediction algorithm adopted by each sub-model in the initial model; Wherein, each sub-model in the initial model adopts its own prediction algorithm to collaboratively predict the power consumption of the refrigeration system.

3. The method according to claim 1 or 2, characterized in that The adopting the open source data set and the target algorithm to train the initial model to obtain the target model includes: The initial model is trained using the open source data set and the target algorithm to obtain an intermediate model; The intermediate model is subjected to a performance test through a simulation system to obtain a test result; wherein the simulation system comprises: a digital twin system corresponding to the refrigeration system and established by integrating a mechanism model and data drive; If the test result indicates that the intermediate model passes the performance test, the intermediate model is used as the target model; If the test result indicates that the intermediate model fails the performance test, the intermediate model is trained to obtain a target model.

4. The method according to any one of claims 1 to 3, characterized in that: The intermediate model includes an intermediate sub-model obtained by training each sub-model; the performance test of the intermediate model is performed by the simulation system to obtain the test result, including: For the kth intermediate sub-model in the intermediate model, test data of the kth intermediate sub-model is obtained by running the simulation system; k∈[1, K], K is the number of intermediate sub-models included in the intermediate model; The test data is used to test the accuracy performance of the kth intermediate sub-model to obtain a test result.

5. The method according to any one of claims 1 to 4, characterized in that: The refrigeration system is located in the data center; The performing of a performance test on the intermediate model through a simulation system to obtain a test result includes: Within a preset time period, the operation of the simulation system is controlled by the intermediate model; after the preset time period ends, the target energy consumption of the data center within the preset time period is counted; wherein the data center generates a plurality of environmental data within the preset time period, and the target energy consumption includes: the energy efficiency indication value of the data center under each of the plurality of environmental data; Acquire a reference energy consumption condition of the data center, the reference energy consumption condition comprising: an energy efficiency indicator value of the data center under each of the plurality of environmental data when the intermediate model is not used; According to the target energy consumption and the reference energy consumption, the energy-saving performance of the intermediate model is tested to obtain a test result.

6. The method according to any one of claims 1 to 5, characterized in that: The step of controlling the operation of the simulation system by using the intermediate model within a preset time period includes: Running each component in the simulation system; In a preset time period, regularly calculating the power consumption optimization strategy of the simulation system through the intermediate model; Each time a group of power consumption optimization strategies is calculated, the operation of the corresponding first component in the simulation system is controlled according to the parameter values ​​corresponding to the controllable operating parameters defined by the currently calculated power consumption optimization strategies.

7. The method according to any one of claims 1 to 6, characterized in that: The step of testing the energy-saving performance of the intermediate model according to the target energy consumption and the reference energy consumption to obtain a test result includes: According to the target energy consumption and the reference energy consumption, the intermediate model is evaluated for energy saving to obtain the target energy saving value of the intermediate model; Based on the target energy saving value of the intermediate model and the preset energy saving conditions, the energy saving performance of the intermediate model is tested to obtain a test result.

8. The method according to any one of claims 1 to 7, characterized in that: The energy efficiency indicator value is a ratio between the total energy consumed by the data center and the energy consumed by the equipment load of the data center; The step of performing energy-saving evaluation on the intermediate model according to the target energy consumption and the reference energy consumption to obtain the target energy saving value of the intermediate model includes: Traversing multiple environmental data generated by the data center within the preset time period, determining the currently traversed environmental data as the current environmental data, and determining the average value of the equipment load corresponding to the current environmental data; Calculate the energy-saving efficiency of the intermediate model under the current environmental data according to the energy efficiency indicator value corresponding to the current environmental data in the target energy consumption situation, the energy efficiency indicator value corresponding to the current environmental data in the reference energy consumption situation, and the average value of the equipment load corresponding to the current environmental data; Calculate the energy saving value of the intermediate model under the current environmental data according to the calculated energy saving efficiency and the duration of the current environmental data within the preset time period; After the plurality of environmental data are traversed, the energy consumption saving values ​​of the intermediate model under each environmental data are integrated to obtain the target energy consumption saving value of the intermediate model.

9. The method according to any one of claims 1 to 8, characterized in that An environmental data refers to a combination of a wet-bulb temperature and an equipment load interval; the determining the mean value of the equipment load corresponding to the current environmental data includes: Determine a maximum load value and a minimum load value corresponding to the device load interval in the current environment data; The determined maximum load value and minimum load value are averaged to obtain an average value of the equipment load corresponding to the current environment data.

10. The method according to any one of claims 1 to 9, characterized in that: The calculating the energy-saving efficiency of the intermediate model under the current environmental data according to the energy efficiency indication value corresponding to the current environmental data in the target energy consumption situation, the energy efficiency indication value corresponding to the current environmental data in the reference energy consumption situation, and the average value of the equipment load corresponding to the current environmental data includes: Performing a difference operation on the energy efficiency indication value corresponding to the current environmental data in the target energy consumption situation and the energy efficiency indication value corresponding to the current environmental data in the reference energy consumption situation; The result of the difference operation is multiplied by the average value of the equipment load corresponding to the current environmental data to obtain the energy-saving efficiency of the intermediate model under the current environmental data.

11. The method according to any one of claims 1 to 10, characterized in that: The preset energy-saving condition is used to indicate that: the target energy-saving value of the intermediate model is greater than or equal to the preset energy-saving value; The step of testing the energy-saving performance of the intermediate model based on the target energy-saving value of the intermediate model and the preset energy-saving condition to obtain the test result includes: If the target energy saving value of the intermediate model is greater than or equal to the preset energy saving value, generating a test result indicating that the intermediate model has passed the performance test; If the target energy saving value of the intermediate model is less than the preset energy saving value, a test result indicating that the intermediate model has failed the performance test is generated.

12. The method according to any one of claims 1 to 11, characterized in that: The refrigeration system is configured with a network topology diagram, which includes T point nodes, X data nodes, Y model nodes and Z algorithm nodes; T, X, Y and Z are all positive integers; wherein: One point node corresponds to one point, and the point is used to collect parameter values ​​of components in the refrigeration system; One data node corresponds to one data set. For the xth data node, there are N nodes in the network topology diagram, and each of the N nodes points to the xth data node through a directed edge. The data set corresponding to the xth data node is constructed according to the data corresponding to the N nodes. The N nodes include at least one of the following: at least one point node and at least one data node other than the xth data node; x∈[1,X], N is a positive integer; One model node corresponds to one sub-model. For the yth model node, there are M nodes in the network topology diagram, and each of the M nodes points to the yth model node through a directed edge. The sub-model corresponding to the yth model node performs model processing according to the data corresponding to the M nodes. The M nodes include at least one of the following: at least one data node and at least one model node other than the yth model node; y∈[1, Y], M is a positive integer; One algorithm node corresponds to one prediction algorithm. For the zth algorithm node, there are h model nodes in the network topology diagram, and each of the h model nodes points to the zth algorithm node through a directed edge. The prediction algorithm corresponding to the zth algorithm node is called by the model corresponding to each of the h model nodes, z∈[1, Z], h∈[1, X].

13. The method according to any one of claims 1 to 12, characterized in that: The training process of any sub-model in the initial model includes: Acquire training data of any of the sub-models according to parameter values ​​corresponding to the parameters of each of the components in the open source data set and the topological structure of the network topological graph of the refrigeration system; According to a supervised training method, the training data is used to train any of the sub-models.

14. The method according to any one of claims 1 to 13, characterized in that: The target algorithm selected using the open toolbox includes: For any sub-model in the initial model, locate a model node corresponding to the any sub-model from the network topology diagram; According to the topological structure of the network topology diagram, an open toolbox is used to use the prediction algorithm corresponding to the algorithm node pointed to by the directed edge of the located model node as the prediction algorithm adopted by any sub-model.

15. The method according to any one of claims 1 to 14, characterized in that: The deploying the target model to the controller of the refrigeration system includes: Storing the target model in a model warehouse, so that the model warehouse deploys the target model to a controller of the refrigeration system; Wherein, whenever the model warehouse detects that the target model is updated, the target model deployed in the controller is updated online according to the updated target model.

16. The method according to any one of claims 1 to 15, characterized in that: After deploying the target model to the controller, the method further includes: Perform service monitoring on the controller to obtain a service monitoring result; the service monitoring result includes at least one of the following: and the operating status of the target model in the controller; The controller is detected for a fault point according to the service monitoring result, and when a fault point is detected, the fault point is repaired; the fault point includes at least one of the following: a hardware facility that fails in the controller, and a model that fails in the controller.

17. The method according to any one of claims 1 to 16, characterized in that: The process of the controller determining the power consumption optimization strategy of the refrigeration system through the target model includes: Generate multiple groups of parameter combinations according to the strategy optimization algorithm, each group of parameter combinations includes parameter values ​​corresponding to each controllable operating parameter; The target model is called to predict the power consumption of the refrigeration system according to each parameter combination to obtain a power consumption prediction result; the power consumption prediction result includes: the predicted power consumption of the refrigeration system under each parameter combination; According to the power consumption prediction result, a group of parameter combinations is selected from the multiple groups of parameter combinations as a power consumption optimization strategy for the refrigeration system.

18. An operating device for a refrigeration system, characterized in that: include: An acquisition unit is used to acquire an open source data set of a refrigeration system, wherein the refrigeration system includes multiple components, each of the components is configured with one or more parameters, and the open source data set includes parameter values ​​corresponding to the parameters configured for each of the components; wherein there is at least one first component among the multiple components, and the parameters configured for each of the first components include controllable operating parameters; A processing unit, configured to determine an initial model for predicting power consumption of the refrigeration system, and select a target algorithm using an open toolbox, wherein the target algorithm is applied to the initial model for predicting power consumption; The processing unit is further used to train the initial model using the open source data set and the target algorithm to obtain a target model; The processing unit is also used to deploy the target model to the controller of the refrigeration system, so that the controller determines the power consumption optimization strategy of the refrigeration system through the target model, and the power consumption optimization strategy defines a parameter value corresponding to at least one controllable operating parameter of the first component; and to control the operation of the corresponding first component in the refrigeration system according to the parameter values ​​corresponding to each controllable operating parameter defined by the determined power consumption optimization strategy.

19. A computer device comprising an input interface and an output interface, characterized in that: Also includes: processor and computer storage media; The processor is suitable for implementing one or more instructions, the computer storage medium stores one or more instructions, and the one or more instructions are suitable for being loaded by the processor and executed by the operating method of the refrigeration system as described in any one of claims 1-17.

20. A computer storage medium, characterized in that The computer storage medium stores one or more instructions, and the one or more instructions are suitable for being loaded by a processor and executed by the operating method of the refrigeration system according to any one of claims 1-17.

21. A computer program product, characterized in that The computer program product includes one or more instructions, and when the one or more instructions are executed by a processor, the operating method of the refrigeration system according to any one of claims 1 to 17 is implemented.

Images