A Control Method for Central Air Conditioning Water Systems Based on Large Language Models and Digital Twins

By combining large language models with digital twin technology, intelligent control of building air conditioning water systems has been achieved, solving the problems of insufficient flexibility and accuracy in traditional methods, improving the system's operational efficiency and safety, and reducing reliance on professional technicians.

CN118391792BActive Publication Date: 2026-11-17ZHEJIANG UNIV
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202410482021.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-22
Publication Date
2026-11-17
Estimated Expiration
2044-04-22

AI Technical Summary

Technical Problem

Traditional building air conditioning water system control methods lack flexibility and precision when dealing with complex environmental dynamics and changing usage demands, resulting in low energy efficiency, high operational risks, and high dependence on professional technicians.

Method used

By combining large language models with digital twin technology, the system receives natural language commands, analyzes user needs, identifies devices and their control variables, and verifies control strategies in a virtual environment, thereby achieving precise and flexible control.

Benefits of technology

It improves the intelligence level and operational efficiency of building air conditioning water systems, lowers the professional knowledge threshold, ensures the accurate execution of control commands and system safety, and enhances adaptability to complex environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118391792B_ABST
    Figure CN118391792B_ABST
Patent Text Reader

Abstract

This invention discloses a central air conditioning water system control method based on a large language model and digital twin, comprising: receiving control commands issued by users and preprocessing successfully matched control commands; inputting the processed commands to the large language model, determining whether the processed commands specify the type and number of the device to be controlled and whether the device to be controlled is usable; determining the names of control variables and their corresponding setpoints, querying the operating status of each device according to the verification level, changing the setpoints of the control variables according to the control strategy, and simultaneously performing simulation. The large language model then analyzes the control effect and feeds it back to the user, and transmits the control strategy that meets the user's needs to the field edge controller for execution. This invention marks an important step towards more intelligent and information-based building automation system management, demonstrating the enormous potential and application value of intelligent control methods in the field of building technology.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of automatic control of building heating, ventilation and air conditioning systems, and particularly relates to a control method for central air conditioning water systems based on large language models and digital twins. Background Technology

[0002] In the field of automatic control of building heating, ventilation, and air conditioning (HVAC) systems, efficient management of air conditioning water systems is key to achieving energy savings and improving indoor environmental comfort. Traditional control strategies mainly rely on preset logic and manual intervention. This approach is rather crude and often proves inflexible and ineffective in dealing with complex building environments and changing usage demands. Especially in today's increasingly information-driven and intelligent world, the operation and maintenance management of building systems faces the challenge of transitioning from traditional models to intelligent and automated systems.

[0003] With the rapid development of artificial intelligence technology, the emergence of large language models and digital twin technology has provided new solutions to the aforementioned problems. Large language models, through deep learning, can understand and process natural language instructions, providing cognitive and comprehension abilities similar to humans. Digital twin technology, on the other hand, can accurately simulate objects and processes in the physical world within a virtual environment, providing a powerful tool for predicting and verifying control strategies. The combination of these two technologies indicates enormous potential for intelligent control of building air conditioning water systems, enabling more precise and flexible control strategies, optimizing energy use, and improving user experience. Nevertheless, effectively integrating the natural language processing capabilities of large language models with the high simulation accuracy of digital twin models to achieve real-time, dynamic, and precise control of building air conditioning water systems remains a major challenge. Traditional methods have limitations in understanding complex user needs, real-time data processing, and flexible adjustment of control strategies, resulting in low energy efficiency, high operational risks, and high dependence on maintenance personnel. Summary of the Invention

[0004] The purpose of this invention is to solve the problems existing in the prior art and to provide a central air conditioning water system control method based on large language model and digital twin.

[0005] To achieve the above-mentioned objectives, the present invention specifically adopts the following technical solution:

[0006] A control method for a central air conditioning water system based on a large language model and digital twin, comprising the following steps:

[0007] S1. The input module receives the control command issued by the user, matches the control command with the pre-built keyword library, and determines whether the control command contains the keywords in the keyword library: if so, the control command is preprocessed to obtain the processed command; otherwise, the user is prompted that the control command entered does not belong to the building energy field and needs to be re-entered.

[0008] S2. Input the processing command into the large language model and determine whether the processing command simultaneously specifies the type and number of the device to be controlled. If not, prompt the user on the interactive interface to supplement the device information and re-enter the control command. If both are specified, retrieve the equipment list from the building air conditioning water system in the digital twin model system. Match the device to be controlled in the digital twin model system with the device type and number in the processing command. If the device to be controlled cannot be used, prompt the user on the interactive interface that the device to be controlled cannot be used, please reconfirm the device to be controlled, and re-enter the control command.

[0009] S3. The large language model determines whether the processing instruction matches the control command type of the device to be controlled: If they do not match, the user is prompted to reconfirm the control method and re-enter the control instruction on the interactive interface; if they match, the automatic control variable list of the device to be controlled is obtained from the building air conditioning water system; if the processing instruction and the automatic control variable list match, the control variable name is determined from the automatic control variable list, and the control variable set value corresponding to the control command type is output; if the processing instruction and the automatic control variable list do not match, the user is prompted to reconfirm the control variables of the device to be controlled and re-enter the control instruction on the interactive interface.

[0010] S4. Obtain the name of the device to be controlled from the building air conditioning water system and obtain the topology connection relationship of the device to be controlled; sort the devices in the topology connection relationship according to the pre-built device inspection level to obtain the sorted device list; query the sorted device list in turn, for the current device in the sorted device list, call the building air conditioning water system to first obtain the current device's operating status. If the current device's operating status meets the requirements for executing processing instructions and control variable settings, then query the next device until the device to be controlled is found. Otherwise, call the building air conditioning water system to adjust the current device's operating status until the current device's operating status meets the requirements for executing processing instructions and control variable settings.

[0011] S5. Obtain the control strategy of the building air conditioning water system corresponding to the processing command, input the control strategy into the building air conditioning water system, change the set value of the control variable, obtain the performance data of each device of the building air conditioning water system before the change, and perform digital twin model simulation on the building air conditioning water system after the change to obtain the performance data of each device of the building air conditioning water system after the change. Input the performance data before the change and the performance data after the change into the large language model to obtain the analysis results of the control effect and feed them back to the user.

[0012] S6. After receiving feedback from the user regarding the analysis results, input the user's feedback results into the large language model. The large language model determines whether the control strategy meets the user's needs. If it does, the modified control variable settings are saved to the historical information database, and the control strategy that meets the user's needs is passed to the field edge controller of the building air conditioning water system. The field edge controller then executes the control strategy that meets the user's needs. Otherwise, the control command is re-entered.

[0013] Based on the above scheme, each step can be implemented in the following preferred manner.

[0014] Preferably, in step S1, the input module is capable of receiving various forms of control command input and has the ability to convert speech into text; the preprocessing includes correcting grammatical errors in the control commands, extracting key information points from the control commands, and eliminating ambiguity in control commands with multiple interpretations by using contextual information or requesting more information from the user.

[0015] Preferably, in step S3, if the control command type is an open command, the control variable setting value is 1; if the control command type is a close command, the control variable setting value is 0; if the control command type is a numerical adjustment command, the large language model uses the setting value in the processing instruction as the control variable setting value; if there is no setting value in the processing instruction, the retrieval model searches for the historical setting value of the automatic control variable in the historical information database based on the current operating status of the building's air conditioning water system, and uses the historical setting value with the closest value as the control variable setting value.

[0016] Preferably, in step S4, if the control command type is an open command or a value adjustment command, the equipment order corresponding to the inspection level is: all types of valves, all types of cooling water pumps, all types of chilled water pumps, all types of cooling towers, and all types of chiller units; if the control command type is a close command, the equipment order corresponding to the inspection level is: all types of chiller units, all types of cooling towers, all types of cooling water pumps, all types of chilled water pumps, and all types of valves.

[0017] Preferably, the building air conditioning water system also includes equipment component modules that can dynamically update and maintain equipment information.

[0018] As a preferred option, the historical information database is used to collect and store historical control commands, analysis results corresponding to control strategies, and user feedback results.

[0019] Preferably, the building air conditioning water system supports remote control operation, and users or maintenance personnel can remotely monitor and control the building air conditioning water system through network connection.

[0020] As a preferred option, when acquiring control strategies, the control strategies can be adjusted according to the real-time operating status of each device in the building's air conditioning water system, or users can select control strategies based on the control effect, and the control strategies can be automatically assessed for risk.

[0021] As a preferred option, a self-learning mechanism allows the building air conditioning water system to automatically optimize control strategies and adjust the understanding and generation logic of the large language model based on data in the historical information database.

[0022] Preferably, the edge controller is deployed on the building site, responsible for executing control strategies and directly managing and controlling the building air conditioning water system equipment on site; the edge controller supports real-time data acquisition and analysis, and can instantly feed back the operating status of each piece of equipment in the building air conditioning water system to the digital twin platform, providing a basis for the generation and optimization of control strategies; through a secure communication protocol established between the digital twin platform and the edge controller, the secure transmission and execution of processing instructions are ensured, preventing data tampering and unauthorized access.

[0023] Compared with the prior art, the present invention has the following advantages:

[0024] First, by introducing large language models and digital twin technology, this invention significantly improves the intelligence and efficiency of building air conditioning water system control. Users do not need in-depth professional knowledge to interact directly with the system through natural language, greatly simplifying the operation process and lowering the professional threshold for user operation. This method not only improves the user experience but also reduces reliance on professional technicians.

[0025] Secondly, this invention ensures accurate execution of control commands and safe system operation by precisely parsing user instructions and monitoring the building's water system status in real time. This advanced control logic and status monitoring mechanism overcomes the limitations of traditional control systems in equipment identification, status judgment, and control strategy execution, providing more accurate and reliable control results.

[0026] Furthermore, through simulation analysis and optimization using digital twin models, this invention can predict the implementation effects of control strategies, providing a scientific basis for adjusting and optimizing these strategies. This not only improves the system's control accuracy but also enhances its adaptability to complex environmental changes. Finally, this invention provides an innovative intelligent control solution for building air conditioning water systems in the field of modern smart buildings. It not only meets the current needs of intelligent building management and maintenance but also has broad application prospects and promotional value, opening up new paths for the intelligent control and management of building energy systems. These features make this invention a significant technological advancement in improving energy efficiency, reducing operation and maintenance costs, and enhancing user experience. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating the implementation of the present invention;

[0028] Figure 2 This is a topology diagram of the water system of a factory's HVAC equipment in this embodiment. Detailed Implementation

[0029] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below. Technical features in the various embodiments of the present invention can be combined accordingly without mutual conflict.

[0030] This invention aims to introduce an innovative intelligent control method for building air conditioning water systems. This method combines the advanced understanding capabilities of large language models with the simulation and prediction capabilities of digital twin technology. It can not only accurately analyze the control needs of maintenance personnel and automatically identify relevant equipment and their control variables, but also pre-verify the effectiveness and security of control strategies in a virtual environment. Furthermore, by supporting multimodal data input, such as text and voice, this method further optimizes the user-system interaction process and improves the flexibility and accuracy of the control logic. Finally, through the collaborative work of a cloud platform and edge controllers, efficient execution of control strategies is achieved, significantly reducing operational risks, improving system operating efficiency and security, and promoting the intelligent and automated development of building air conditioning water system management, thus opening up new avenues for technological innovation and application.

[0031] This invention aims to achieve intelligent control of building air conditioning water systems by integrating large language models and digital twin technology, thereby improving the system's automation level and operational efficiency while ensuring the accuracy and security of the control strategy. In a preferred embodiment of this invention, a central air conditioning water system control method based on large language models and digital twins is provided, such as... Figure 1 As shown, the above method includes the following steps S1 to S6, and the specific process is as follows:

[0032] S1. The input module receives control commands from the user, matches the control commands with a pre-built keyword library, and determines whether the control commands contain keywords from the keyword library. If so, the control commands are pre-processed to obtain processed commands; otherwise, the user is prompted that the input control commands do not belong to the building energy field and that the control commands need to be re-entered.

[0033] It should be noted that in step S1 of the present invention, the input module is capable of receiving various forms of control command input and has the ability to convert speech into text.

[0034] In this embodiment, the input module receives control commands issued by the user in natural language. These commands can relate to various aspects of the corresponding building's energy system. These commands can take many forms, such as text input, voice input, and other forms of input. Specifically, the user can submit commands via text or voice. For text input, the user can directly input text using an input device such as a keyboard. For users who wish to ask questions via voice, the input module will use a speech-to-text function to convert the voice information into text. It is also designed to process multimodal information and has corresponding conversion capabilities to handle other forms of input besides text and voice, further ensuring that different forms of input can be accurately understood and processed, such as gestures or image input (if applicable), and can convert this information into a format that can be understood and processed.

[0035] In this embodiment, the speech-to-text module of the input module uses the Wav2Vec 2.0 model, which learns representations directly from the original speech signal using unsupervised pre-training. This process can be represented as follows:

[0036] Encoded Representation=Wav2Vec2.0(Raw Audio)

[0037] Wherein, Encoded Representation represents the encoded representation, Raw Audio represents the original speech signal, and Wav2Vec2.0(·) represents the Wav2Vec 2.0 model.

[0038] In this embodiment, if the user input includes images and videos, the input module's processing model uses the ViLT model, which understands multimodal content by fusing images (or video frames) and text. The formula for the ViLT model to process images and text can be simplified as follows:

[0039] Unified Representation=ViLT(Image Embeddings,Text Embeddings)

[0040] Here, Unified Representation refers to synthetic representation, Image Embeddings refers to image embedding, and Text Embeddings refers to text embedding.

[0041] It should be noted that in step S1 of this invention, after receiving the user's control command, the input module matches the control command with a pre-built keyword library, using a keyword-based retrieval method to analyze the control command issued by the user through natural language. The input module sets a series of keywords related to building energy systems and forms a keyword library. When receiving the user's input control command, it first performs a rapid scan to identify these keywords. If the user's question does not contain these preset keywords, the question will be marked as currently unsolvable because it does not meet the query parameters of the building energy system, and the user will be prompted that the input control command does not belong to the building energy field. For example, when a user asks "1+1=?", it will be considered that the question does not belong to the building energy field. This keyword matching mechanism ensures that the response is focused on highly relevant questions, thereby improving the overall processing efficiency and the relevance of the response. In this way, the method of this invention can more effectively manage user inquiries and ensure that queries to the building energy information system receive timely and accurate responses. In this process, a keyword matching model is used, that is, selection and retrieval based on key variables and variables in the system, such as text-embedding-3-small. For example, if a user needs to issue the control command "X1", the keyword matching model can load and read the text, segment and vectorize the text, match the n most similar words to the question vector in the text vector, and add the matched text as context and question together to X1.

[0042] It should be noted that in step S1 of the present invention, the above preprocessing includes correcting syntax errors in the control instructions, extracting key information points from the control instructions, and eliminating ambiguity in the control instructions for those with multiple interpretations by using contextual information or requesting more information from the user.

[0043] In this embodiment, after receiving the user's control command, the input module performs a series of preprocessing operations to improve the efficiency and accuracy of control command processing. These operations include, but are not limited to, grammar correction, keyword extraction, and disambiguation, to enhance the efficiency and accuracy of question parsing. Specifically, grammar correction automatically corrects grammatical errors in the control command to reduce misunderstandings caused by inaccurate language expression. Keyword extraction extracts key information points from the control command to help the system more quickly locate the user's query intent. Disambiguation involves the input module eliminating ambiguity for expressions that may have multiple interpretations, either through contextual information or by requesting more information from the user, to ensure the control command is correctly understood.

[0044] In this embodiment, the grammar correction and context understanding sections utilize language models and the context-aware model BERT to correct grammatical errors in the text and understand the context of the user's query. These models can infer the most appropriate word replacements or rearrangements based on the context to correct grammatical errors. Furthermore, these models learn deep-level language features on large amounts of text data through pre-training, enabling them to extract contextual information relevant to the query, thereby improving the accuracy of query parsing.

[0045] Through the detailed implementation steps described above, the input module can effectively receive and process control commands issued by users in natural language. Regardless of the form in which the user issues the control commands, the input module can accurately understand and respond, thereby providing users with accurate and efficient services.

[0046] S2. Input the processing command into the large language model and determine whether the processing command simultaneously specifies the type and number of the device to be controlled. If not, prompt the user on the interactive interface to supplement the device information and return to S1 to re-enter the control command. If both are specified, retrieve the equipment list from the building air conditioning water system in the digital twin model system. Match the device to be controlled in the digital twin model system with the type and number of the device to be controlled in the equipment list according to the processing command. If the device to be controlled cannot be used, prompt the user on the interactive interface that the device to be controlled cannot be used, ask them to reconfirm the device to be controlled, and return to S1 to re-enter the control command.

[0047] It should be noted that in step S2 and subsequent steps of using a large language model in this invention, the method of this invention is not limited to using a specific large language model. It can adapt to and integrate multiple large language models, including but not limited to open-source and commercial models, to improve the adaptability and accuracy of the method. In step S2 and subsequent steps of using a large language model in this embodiment, the CHATGLM model is used as the large language model, and the transmission process of questions and responses is realized by calling the API of the corresponding model. CHATGLM (Chat General Language Model) is a general language model that, through training on a large amount of text data, can understand and generate natural language, primarily in Chinese and English usage environments. The CHATGLM model can perform deep analysis of user questions, identifying key variables and information needs in the questions. Specifically, by analyzing the semantic and grammatical structure of user questions, the CHATGLM model can determine the topic and intent of the questions, and thus identify key variables related to the questions. At the same time, the CHATGLM model can also understand the user's information needs based on the context of the questions, thereby providing more accurate and relevant answers.

[0048] It should be noted that in step S2 of this invention, after receiving the control request submitted by the user in natural language, the system first performs initial parsing to determine whether the processing instruction explicitly indicates the type and device number of the device to be controlled. If the processing instruction lacks the above necessary information, the system interacts with the user, requesting them to supplement the missing information through interactive dialogue and guiding the user to provide the control instruction again, before returning to S1.

[0049] It should be noted that in step S2 of this invention, before executing the control command, the system verifies the real-time status of the device to be controlled, including its existence and availability. This step ensures the effective execution of the control command, avoids control failure due to the unavailability of the target device, and thus improves the safety and reliability of the operation. Specifically, the program uniformly calls the API interface connected to the digital twin model system to obtain the current equipment list of the building's air conditioning water system. For example, by adding and executing the script in a specific project, result = platform.get_all_model() will return information: a set of names, such as ["Chiller_1", "Chiller_2"] etc. The equipment list is a list of all existing and usable equipment in the current building air conditioning water system, which the large language model uses to select the device to be controlled in the digital twin model system based on the equipment information (type and number of the device to be controlled) specified in the processing command. During this process, the system takes into account the potential ambiguity of the equipment names provided by the user. For example, the user might abbreviate "centrifugal chiller unit_1" to "No. 1 chiller unit." Through advanced natural language processing technology, the system can flexibly identify and match such ambiguous equipment names. When determining the device to be controlled, the system also checks the availability of the selected device to ensure that it exists in the current water system and is usable. If the device to be controlled does not exist in the current building air conditioning water system or is unusable, the system interacts with the user again to clearly point out the problem: the device to be controlled does not exist or is unusable. The user is then asked to reconfirm the device to be controlled and provide the control command again, returning to S1.

[0050] In this embodiment, when the control commands provided by the user are incomplete or contain errors, the system will guide the user to supplement or correct the information through interactive dialogue. This mechanism not only optimizes the user experience but also improves the success rate of control command execution. For example, if the control device specified in the user's control command does not exist, the system will provide feedback to the user: "The device you want to control does not exist or is unusable. Please reconfirm the device you wish to control and provide the control command again." By automatically adding and supplementing the corresponding explanatory text in each interaction round, the system assists the large language model in more accurately judging the user's intent and needs, improving the accuracy of command parsing and the quality of response. For example, in this step, the example of a "promote" passed to the large language model is:

[0051] 1. "Can you analyze from the following text which specific device the user wants to control (e.g., chiller unit 1)? If the corresponding device and number exist, answer "YES", otherwise answer "NO". For example, if the user's question is: 'Start centrifugal chiller unit 1', then answer "YSE". The user's question is: {Question raised by the user}."

[0052] 2. "Please analyze from the following text which device the user wants to control? Please follow the format: 'Controlled device:', such as: 'Controlled device: Chiller No. 1'. The user's question is: {Question raised by the user}."

[0053] 3. "This is the system's equipment list" + Equipment List + "These are the devices the user needs to control:" + Device to be controlled + "Please select the devices the user needs to control from the system's equipment list. If no matching device is found, answer 'NO'. The numbers _1, 2, 3 after the device name represent the device number, such as centrifuge unit_1 referring to centrifuge unit number one.\nNote! Do not change the device names in the following operations! The devices must be selected from the equipment list, not fabricated! Please select the relevant devices according to the user's needs and output them directly, such as: 'centrifuge water chiller unit_1', or 'NO'.

[0054] It should also be noted that the equipment in the aforementioned building air conditioning water system generally includes various types of chillers, chilled water pumps, cooling water pumps, cooling towers, and plate heat exchangers. The specific equipment used depends on the actual composition of the building air conditioning water system in the project and is not limited in this invention.

[0055] As can be seen, by calling and reading the digital twin system, a list of currently available devices can be dynamically generated and provided. Based on real-time data from the digital twin model, accurate device options are provided for user commands. When a device is unavailable or does not exist, the system will proactively interact with the user to ensure the accuracy of commands and the effectiveness of their execution.

[0056] S3. The large language model determines whether the processing instruction matches the control command type of the device to be controlled: If they do not match, the user is prompted to reconfirm the control method on the interactive interface, and returns to S1 to re-enter the control instruction; if they match, the automatic control variable list of the device to be controlled is obtained from the building air conditioning water system; if the processing instruction and the automatic control variable list match, the control variable name is determined from the automatic control variable list, and the control variable set value corresponding to the control command type is output; if the processing instruction and the automatic control variable list do not match, the user is prompted to reconfirm the control variables of the device to be controlled on the interactive interface, and returns to S1 to re-enter the control instruction.

[0057] It should be noted that in step S3 of this invention, it is necessary to confirm the name of the control variable corresponding to the processing instruction and its set value. Upon receiving the processing instruction, the program immediately performs a preliminary analysis. The purpose of this stage is to determine whether the processing instruction is a clear, explicit, and executable control command, and, in conjunction with the characteristics of the target control device (such as a building air conditioning water system), to confirm its acceptable command type. Control command types for building air conditioning water systems are divided into three categories: on / off commands, off commands, and numerical adjustment commands. The large language model analyzes whether the processing instruction matches the control command type of the target device. If the processing instruction does not belong to any of the corresponding control methods, the system interacts with the user, indicating that the target device does not have a corresponding control method, requesting reconfirmation of the control method, and returning to S1 to provide a new control instruction, ensuring that the control command type of the target device is clear and explicit. Specifically, when interacting with the user, if the processing instruction fails to clearly match any of the above control command types, the system will consider the instruction ambiguous. In this case, the system will proactively interact with the user, guiding the user to reconfirm the control method and return to S1 to provide a new control instruction. This may involve returning to the initial stage of control instruction reception to ensure that an accurate and explicit control command is obtained.

[0058] Once the control command type of the device to be controlled by the processing instruction is successfully matched, the name of the control variable for the processing instruction is determined by combining the list of automatic control variables of the device to be controlled, and then the set value of the control variable is judged and determined. Specifically, the program uniformly calls the API interface connected to the digital twin model system to obtain the list of all modifiable automatic control variables in the device to be controlled. This is the corresponding functional module of the digital twin model system, which is used by the large language model to determine the name of the control variable for the processing instruction in combination with the automatic control variables. In specific implementation, for example, adding and executing `result = models.get_variables_by_type("Chiller_1","Automatic")` in a specific project script will return information: a set of names, examples ["Remote / Local", "On / Off", "Chiller Outlet Temperature Setpoint"], etc. At the same time, the large language model can understand the abstract intent in the user's instructions, allowing users' questions to have a certain degree of ambiguity.

[0059] If the processing instruction and the list of automatic control variables match, the control variable names are confirmed. The program prints out the list of automatic control variables for the current target control device to help confirm the control variables involved in the processing instruction. During this process, the large language model intelligently identifies and determines the most matching control variable names based on the vague descriptions provided by the user and the device's list of automatic control variables.

[0060] After the control variable name is determined, the control variable setpoint will be determined. In this invention, the control variable setpoint is determined based on the control command type. Specifically, if the control command type is an "on" command, the control variable setpoint is 1; if the control command type is a "off" command, the control variable setpoint is 0; if the control command type is a numerical adjustment command, the large language model will use the setpoint value in the processing instruction as the control variable setpoint. If the processing instruction does not contain a setpoint value, the retrieval model will search for historical setpoint values ​​of the automatic control variable in the historical information database based on the current operating status of the building's air conditioning water system, and use the historical setpoint value with the closest value as the control variable setpoint.

[0061] In this embodiment, for numerical adjustment commands, further analysis will be conducted to determine whether the set value of the control variable is explicitly given in the processing instruction. If it is, the set value in the processing instruction will be used as the set value of the control variable. Otherwise, if the user does not explicitly give a set value, the retrieval model will be used in conjunction with the current operating status of the building's air conditioning water system to search for similar cases in the historical information database to infer the most suitable set value, which will then be used as the set value of the control variable. The historical information database contains historical operating data of the building's air conditioning water system, from which historical set value records of the automatic control variable can be obtained. Furthermore, when finally determining the set value of the control variable, the retrieval model will also comprehensively consider the commonly used range of the automatic control variable corresponding to the target control device (the device to be controlled), thereby selecting the most suitable set value and applying it.

[0062] In this embodiment, the retrieval model described above is the BGE-M3 model, the first text retrieval model with multi-functionality, multi-language, and multi-granularity characteristics. It efficiently supports multi-language, long text, and mixed retrieval, ensuring the accuracy and relevance of the answers. To improve the performance of a single retrieval mode, it includes self-knowledge distillation functionality, selects optimized batch processing strategies, and supports large batch sizes, which can be easily used when fine-tuning vectors for long text or large language models.

[0063] It should be noted that this invention uses an advanced large language model with Chinese as the primary language, and implements the transmission of questions and responses by calling the corresponding model's API. By automatically adding and supplementing relevant explanatory text in each interaction round, it assists the large language model in more accurately judging the user's intent and needs, thereby improving the accuracy of command parsing and the quality of response. For example, the `promot` passed to the large language model in this step is as follows:

[0064] 1. "Can you analyze the user's command type from the following text? Is it a device power-on / power-off command or a device setting adjustment command? If it's device power-on, answer '1'; if it's device power-on, answer '2'; if it's device setting adjustment, answer '3'; if it's neither, answer 'NO'. For example: 'User question: Power on chiller #1', then answer '1'. User question: {User question}"

[0065] 2. "Device: {Device to be controlled}, the device's automatic control variables are: {All automatic control variables of the device}. Can you analyze the user's command type from the following text and determine which automatic control variable of the device needs to be adjusted? Please do not change the device name in the following operations! The device must be selected from the device list, not fabricated! Please select the relevant device according to the user's needs and output it directly.\nUser's question: {User's question}"

[0066] 3. "Equipment: {To be controlled equipment}, the automatic control variables that need to be set for the equipment are: {Automatic control variables that need to be set for the equipment}. Can you analyze from the following text what value the user's command requires for the automatic control variables to be set? If a corresponding value exists, please answer directly with the corresponding number, such as "1"; if not, answer "NO".\nUser's question: {User's question}."

[0067] As can be seen, when the control commands provided by the user are incomplete or contain errors, the system will guide the user to supplement or correct the information through interactive dialogue. This mechanism not only optimizes the user experience but also improves the success rate of control command execution. For example, if the control device specified in the user's control command does not exist, the system will provide feedback to the user: "The device you want to control does not have the corresponding automatic control variable. Please reconfirm the control method of the device you wish to control and provide the control command again." Based on the control requirements of building air conditioning water systems, the method of this invention can identify and execute control commands such as the opening, closing, and numerical adjustment of various devices, and effectively communicate with the user when the commands are unclear, ensuring the clear and accurate execution of control commands.

[0068] S4. Obtain the name of the device to be controlled from the building air conditioning water system, and obtain the topology connection relationship of the device to be controlled; sort the devices in the topology connection relationship according to the pre-built device inspection level, and obtain the sorted device list; query the sorted device list in turn. For the current device in the sorted device list, call the building air conditioning water system to first obtain the operating status of the current device. If the operating status of the current device meets the requirements of executing processing instructions and control variable settings, then query the next device until the device to be controlled is found. Otherwise, call the building air conditioning water system to adjust the operating status of the current device until the operating status of the current device meets the requirements of executing processing instructions and control variable settings.

[0069] It should be noted that step S4 of this invention is to check and confirm that the current status of the building's air conditioning water system (water system) meets the control conditions. The specific implementation process of step S4 is briefly described below.

[0070] First, it is necessary to confirm the topological connections between the device to be controlled and its related water system equipment and valves. This includes all equipment and valves involved in the chilled water and cooling water circulation, including the device's own valves, to ensure the relevance and effectiveness of control commands. To this end, the system is designed with a program input mechanism that allows maintenance personnel to input the name of the device to be controlled. The system automatically queries the related equipment and valves in the current water system that are topologically connected to the device and returns a list of these topologically connected equipment and valves, providing necessary information for subsequent status checks. Specifically, the program calls the API interface connected to the digital twin model system. The topological connections between the device to be controlled and its related water system equipment and valves are the corresponding functional modules of the digital twin platform. In specific implementation, for example, by adding and executing `result = models.get_related_equipments("pump1")` in a specific project script, it will return a list of equipment names including all equipment and valves involved in the chilled water and cooling water circulation.

[0071] Secondly, based on the specific requirements of the control commands, the system prioritizes the equipment to be controlled and its related water system equipment and valves. Following the necessity of equipment and valves in the water circulation circuit, the system sequentially checks the operating status of connected equipment and valves with higher priority than the equipment to be controlled, ensuring the reasonable execution order of the control strategy and that the current water system status meets the control conditions. This process employs different standards depending on the control method to ensure that each step meets the conditions for executing the control commands.

[0072] Specifically, the system equipment and valves are sorted by inspection level according to the different control command types. In step S4, if the control command type is an open command or a value adjustment command, the equipment sorting corresponding to the inspection level is: all types of valves, all types of cooling water pumps, all types of chilled water pumps, all types of cooling towers, and all types of chiller units. If the control command type is a close command, the equipment sorting corresponding to the inspection level is: all types of chiller units, all types of cooling towers, all types of cooling water pumps, all types of chilled water pumps, and all types of valves. For example, before executing an open command or a value adjustment command, valves have a higher priority than water pumps. Therefore, it is necessary to ensure that the corresponding series of valves are open before opening the corresponding water pump, thus allowing the water pump to start or adjust its frequency. Conversely, before executing a close command, water pumps have a higher priority than valves. Therefore, it is necessary to ensure that the corresponding water pump is closed before closing the corresponding valve.

[0073] In addition, for numerical adjustment commands, it is also necessary to ensure that the device to be controlled is turned on. Therefore, during the level sorting process, the device to be controlled will be added to the list of control devices to be checked in advance before the check is performed.

[0074] During the inspection process, the system automatically executes a pre-defined program to check the status of devices in the equipment list. This primarily involves calling the corresponding interfaces of the digital twin platform. First, it obtains the current on / off status of the devices. If any problems are found with the operating status of a device or valve, and the control conditions are not met, the system will proactively open / close the device or valve according to preset control commands, performing automated adjustments to ensure that the water system's operating status meets the execution conditions of the control commands. For example, for an open command checking the on / off status of water pump_1, the command is: `result = models.get_var_value_by_modelname_and_varname("valve_1","on / off status"). If the query finds that valve_1 is closed, then valve_1 should be opened using the command `models.set_parameter_value("valve_1","start / stop control command","1.0")`.

[0075] During the inspection process, when issuing start-up commands and adjusting values ​​for cooling towers, chillers, and large heat exchange equipment in systems with complex series and parallel connections, the minimum standard should be to ensure the normal circulation of chilled water and cooling water.

[0076] During the inspection, if two or more chiller units and primary chilled water pumps or cooling water pumps are connected and operating through a shared manifold, the valves on the inlet or outlet pipes of each chiller unit should be interlocked with the corresponding operating chilled water pumps and cooling water pumps; if two or more cooling towers are connected and operating through a shared manifold, the valves on the branch pipes of each cooling tower should be interlocked with the corresponding operating cooling water pumps.

[0077] It is evident that during the parsing and execution of control commands, the large language model can automatically adjust the priority of control variables according to the specific content of the user command, ensuring that control commands are executed in the optimal order, thereby improving the system's operating efficiency and energy efficiency.

[0078] S5. Obtain the control strategy of the building air conditioning water system corresponding to the processing command, input the control strategy into the building air conditioning water system, change the set value of the control variable, obtain the performance data of each device in the building air conditioning water system before the change, and perform digital twin model simulation on the building air conditioning water system after the change to obtain the performance data of each device in the building air conditioning water system after the change. Input the performance data before the change and the performance data after the change into the large language model to obtain the analysis results of the control effect and feed them back to the user.

[0079] It should be noted that in step S5 of this invention, the digital twin system automatically performs a risk assessment before executing control commands. This assessment evaluates the potential risks and impacts of command execution, including their effects on system stability and equipment safety, ensuring the security of the control commands. Furthermore, the digital twin system supports multi-level control strategy verification. By simulating the control effects under different operating conditions, it automatically selects the optimal control strategy to maximize the control effect.

[0080] It should be noted that step S5 of the present invention consists of the following processes: automatic generation and execution of control strategy, execution of simulation of building air conditioning water system, acquisition of performance data before and after execution of control strategy, analysis of control effect and feedback of analysis results to user. The specific implementation process of step S5 is briefly described below.

[0081] It should be noted that step S5 of this invention first involves the automatic generation and execution of the control strategy. After successfully collecting information on the device to be controlled and its corresponding control variables and setpoints, the system enters the automatic generation stage of the control strategy. At this point, the system utilizes advanced algorithms integrated with the digital twin model to automatically construct a control strategy that matches the user's control requirements based on the acquired input information (the device to be controlled determined in S2 and the control variables and setpoints determined in S3) and executes changes to the twin system. This process not only considers the operating parameters of the device but also integrates the current operating state of the system to ensure that the generated control strategy is both effective and practical.

[0082] Subsequently, the control strategy is directly applied to the digital twin model, enabling real-time modification and control of the twin system. The core of this step lies in translating the theoretical control strategy into practically executable operations. Through precise mapping of the digital twin model, it ensures that the designed control strategy can be pre-verified and executed in the virtual environment. This not only significantly reduces the risks that may be encountered during actual execution but also improves the efficiency and accuracy of the overall control process. While the control strategy is being executed, the system monitors the execution process in real time, ensuring that each operation is completed accurately and without error according to the predetermined strategy.

[0083] It should be noted that the digital twin system supports real-time data synchronization and can adjust control strategies based on the real-time status information of each device in the building's air conditioning water system. It also supports multi-level control strategy verification. When acquiring control strategies, the system can adjust the control strategies based on the real-time operating status of each device in the building's air conditioning water system. It can also simulate the control effects under different operating conditions, allowing users to select control strategies based on the control effects and automatically conduct risk assessments on the control strategies to ensure the maximization of control effects.

[0084] As can be seen, the automatic generation and execution process of the above control strategy fully demonstrates the application value of digital twin technology in intelligent control systems. Through the close integration of virtual and reality, the control process becomes more flexible and efficient, and can be dynamically adjusted based on real-time feedback. The main approach involves calling the corresponding interfaces of the digital twin platform and using the command `models.set_parameter_value("device", "automatic variable", "set value")` to control various devices and valves within the digital twin platform.

[0085] It should be noted that step S5 of this invention involves performing a digital twin model simulation of the building's air conditioning water system. Once the control strategy is applied in the digital twin model, the method of this invention immediately initiates the simulation program. This process utilizes high-precision computational simulation to predict and analyze the actual impact of the control strategy on the building water system under control operations, including equipment response, changes in system state, and various possible scenarios, and waits for the simulation calculation to complete. Through this simulation, the system can predict the effects of the control strategy without risk, and identify and resolve potential problems in advance.

[0086] During the simulation, performance data needs to be collected to provide empirical evidence for evaluating the effectiveness of the control strategy. Based on this data, the system will predict the performance of the water system and its control equipment after the implementation of control operations, including energy efficiency, stability, and potential risks. After the simulation is completed, the system will comprehensively evaluate the simulation results to determine whether the control strategy needs to be adjusted or optimized.

[0087] It is evident that the simulation process of digital twin models is a crucial step in intelligent control systems. It makes the verification and optimization of control strategies more scientific and precise, greatly improving the success rate of control scheme implementation and the overall performance of the system.

[0088] In addition, the digital twin system integrates a fault detection and diagnosis module, which can automatically detect and diagnose possible faults during the execution of control strategies, provide timely fault recovery suggestions, reduce system downtime, and thus ensure that the control commands issued by users will not cause actual system failures or even paralysis.

[0089] It should be noted that in step S5 of this invention, the next step is to obtain key performance variable information (performance data) of the water system and control equipment before and after the implementation of the control strategy. After the control strategy is implemented and the simulation is completed, the performance data acquisition stage begins. During this process, the interface of the digital twin model is called to record in detail the key variable data of the building water system and related control equipment before and after the implementation of the control strategy. These variables include parameters such as temperature, pressure, and flow rate, which are directly related to the system performance and the effectiveness evaluation of the control strategy. The program needs to call the values ​​of various variables in the digital twin model before and after the simulation. The specific variables are fixed as a corresponding list. Using the developed interface models.get_var_value_by_modelname_and_varname("equipment","variable"), the corresponding values ​​of the variables can be obtained. The result returned during the recording process is "equipment + variable: value".

[0090] By analyzing the changing trends of these key variables, the system can accurately assess the actual impact of control strategies on water system performance. This analysis not only helps in understanding the effects of each control operation but also provides data support for subsequent strategy adjustments. Furthermore, the acquisition and analysis of this key variable information also provides valuable data resources for the long-term operation and maintenance and performance optimization of the system.

[0091] It is evident that the process of acquiring and analyzing performance data demonstrates the powerful capabilities of digital twin models in real-time data processing and analysis, providing solid data support for intelligent control and ensuring the scientific nature and effectiveness of control decisions.

[0092] It should be noted that in step S5 of this invention, the large language model ultimately analyzes the control effect by combining the user's control commands and the performance data of the water system and control equipment before and after the execution of the control strategy, and feeds back the analysis results to the user. Specifically, after collecting the performance data, the system performs an in-depth analysis of the control effect. During the analysis, the system evaluates whether the control strategy has achieved the expected goal, identifies the causes of any deviations, and proposes possible improvement measures. This step utilizes the large language model to comprehensively consider the user's control commands and the acquired performance data to fully evaluate the execution effect of the control strategy. In this embodiment, the large language model will generate feedback information that is easy for the user to understand, clearly indicating the execution effect of the control strategy, existing problems, and suggested optimization directions. This timely and detailed feedback not only enhances the user's confidence in operating the system but also provides guidance and assistance for the user in actual operation. The analysis results will be displayed on the control interface and played back in voice format.

[0093] Furthermore, when using a large language model to obtain analysis results, the example of the prompt passed to the large language model in this process is: "This is the system state information before control {system state information before control}, this is the system state information after control {system state information after control}. Please summarize the changes to the user by combining the information before and after the changes."

[0094] As can be seen, the process of analyzing the control effects highlights the intelligent control system's ability to drive decision-making through data. Through precise analysis and feedback generation, even maintenance personnel without a strong background in HVAC and automatic control can easily understand the control effects of control commands. This method further enhances the transparency of control strategies and user satisfaction, ensuring the efficiency and effectiveness of the intelligent control process.

[0095] S6. After receiving feedback from the user regarding the analysis results, input the user's feedback results into the large language model. The large language model determines whether the control strategy meets the user's needs. If it does, the modified control variable settings are saved to the historical information database, and the control strategy that meets the user's needs is passed to the field edge controller of the building air conditioning water system. The field edge controller then executes the control strategy that meets the user's needs. Otherwise, the control command is re-entered.

[0096] It should be noted that step S6 of the present invention is the process of receiving, analyzing and processing user feedback, as well as the process of executing control strategies on-site. The specific implementation process of step S6 will be briefly described below.

[0097] It should be noted that in step S6 of this invention, the user feedback is first received and analyzed, and then processed. If the user provides positive feedback such as satisfaction or permission to execute, this control operation (processing instructions and the corresponding system operation status and control strategy before and after control) is recorded and stored in the historical information database as a valuable reference for future decisions and operations. If the user provides negative feedback such as dissatisfaction or disallowance to execute, the system interacts with the user, provides interactive guidance, helps the user make necessary adjustments based on the control effect, asks the user to reconfirm the control method based on the previous instructions and corresponding control effects, and re-provides control instructions, returning to S1 to optimize the control effect.

[0098] Additionally, when inputting user feedback into the large language model, the example of the prompt passed to the large language model is: "Please judge the user's feedback information: {user feedback}. Does the feedback indicate that the user's needs have been met and can be executed? If it has been met, answer "YES", otherwise answer "NO". For example: 'User question: It's okay, execute is allowed', then answer "YES".

[0099] This process demonstrates not only the system's high flexibility in user interaction and feedback processing, but also its commitment to continuous improvement and optimization of control strategies. By effectively handling user feedback, the method of this invention enhances user engagement and promotes iterative updates and performance improvements in the intelligent control system.

[0100] It should be noted that when inputting user feedback into the large language model, the user feedback can be in any modality, either text or speech. That is, the large language model can accept various forms of user feedback input. If it is necessary to convert the speech input into text form during this process, the Wav2Vec 2.0 model is still used.

[0101] It should be noted that in step S6 of this invention, the cloud-based digital twin platform finally transmits the corresponding control strategy to the edge controllers at the building water system site, and these controllers execute the control strategy on-site. Specifically, the optimized and user-confirmed control strategy is transmitted to the edge controllers located at the building site via the cloud-based digital twin platform. The edge controllers then execute corresponding control operations in the actual building water system according to the received control strategy, achieving real-time control and adjustment of the physical environment.

[0102] Furthermore, the on-site execution of the control strategy fully utilizes edge computing technology, ensuring the efficiency and timeliness of control operations. By tightly integrating the cloud platform and edge controller, the method of this invention achieves technological innovation and application breakthroughs in the field of intelligent building control, providing a new, efficient, accurate, and responsive model for building automation system management.

[0103] It should be noted that in step S6 of this invention, the edge controller is deployed on the building site, responsible for executing control strategies and directly managing and controlling the building's air conditioning water system equipment. The edge controller supports real-time data acquisition and analysis, and can instantly feed back the operating status of each device in the building's air conditioning water system to the digital twin platform, providing real-time operating status information for control decisions and a basis for the generation and optimization of control strategies. Through a secure communication protocol established between the digital twin platform and the edge controller, the secure transmission and execution of processing instructions are ensured, effectively preventing data tampering and unauthorized access, and guaranteeing system security.

[0104] Additionally, it should be noted that in step S6 of this invention, the historical information database is used to collect and store historical control commands, analysis results corresponding to control strategies, and user feedback results. This data is used to guide the generation of future control strategies, improving the accuracy of control strategies and user satisfaction. Furthermore, this invention, through a self-learning mechanism, allows the building air conditioning water system to automatically optimize control strategies and adjust the understanding and generation logic of the large language model based on data in the historical information database, thereby continuously improving the system's performance and accuracy.

[0105] It is evident that the field execution steps of the edge controller mark a crucial transition from virtual verification to practical application of the control strategy. This not only ensures that the implementation effect of the control strategy is consistent with expectations, but also significantly improves the actual operational effectiveness and user satisfaction of the system.

[0106] It should be noted that the building air conditioning water system of the present invention can support remote control operation, enabling users or maintenance personnel to connect to the digital twin platform via the network, thereby realizing remote monitoring and control of the building air conditioning water system.

[0107] It should be noted that the building air conditioning water system also includes equipment component modules, which can dynamically update and maintain equipment information, including the addition, deletion, replacement and status change of equipment, to ensure the consistency and real-time performance of equipment information between the digital twin model system and the actual building air conditioning water system.

[0108] It should be noted that the method of the present invention includes an interactive interface through which users can directly input multimodal control commands, view the generation and execution of control strategies, and receive analysis results on the control effects.

[0109] It's important to note that digital twins are simulation processes that fully utilize physical models, sensor measurements, and operational history data to integrate multiple disciplines, physical quantities, scales, and probabilities. They map the actual system in a virtual space, reflecting the entire lifecycle of the corresponding physical entity. Digital twins possess excellent characteristics such as real-time synchronization, accurate mapping, and high fidelity. They have already been implemented and achieved significant success in several related fields, for example, in industry, significantly driving changes in product design, production, operation, and maintenance. Water systems are large-scale and have complex topologies, far exceeding human real-time simulation and computational capabilities. Digital twin technology can achieve "transparency" of the entire system through simulation calculations, providing a virtual sandbox for energy-efficient system operation. Unlike traditional simulation, digital twins establish a precise two-way connection between the physical and digital worlds. While reflecting the dynamic process of the physical entity's entire lifecycle, they also support precise simulation and control of interconnected networks from multiple perspectives. For practical engineering scenarios, digital twins significantly expand the scope and meaning of simulation, forming a solution that fully leverages the value of data in real-world applications. A digital twin system is a virtual device system built in the cloud based on actual field equipment; it is a cloud-based mapping of the physical world. In this invention, the digital twin system refers to the software platform that provides this mapping function. The water system is a subsystem of the digital twin system, and because of the different types of mapped equipment, it can be divided into water systems and others.

[0110] To better demonstrate the specific implementation and technical effects of the present invention, the central air conditioning water system control method based on large language model and digital twin shown in steps S1 to S6 of the above preferred implementation is applied to a specific example.

[0111] Example

[0112] like Figure 2 As shown, the water system equipment of a factory's HVAC system includes 5 open variable frequency fan cooling towers, 5 chilled water pumps, 5 cooling water pumps, 5 centrifugal chiller units, as well as tees and valves.

[0113] The user's voice control command was: "Start chiller unit number one".

[0114] Step 1. Accept user commands and convert speech to text.

[0115] Step 2. The device list obtained from the system is: ['chilled water pump_1','chilled water pump_2','chilled water pump_3','chilled water pump_4','chilled water pump_5','cooling water pump_1','cooling water pump_2','cooling water pump_3','cooling water pump_4','cooling water pump_5','cooling load_1','open variable frequency fan cooling tower_1','open variable frequency fan cooling tower_2','open variable frequency fan cooling tower_3','open variable frequency fan cooling tower_4','open variable frequency fan cooling tower_4','centrifugal chiller_1','centrifugal chiller_2','centrifugal chiller_3','centrifugal chiller_4','centrifugal chiller_5']. Analysis shows that the device to be controlled is: centrifugal chiller_1.

[0116] Step 3. The automatic control variables of the centrifugal chiller unit _1 obtained by calling the system are: ['Start / Stop Control Command', 'Temperature Setpoint ℃', 'Remote / Local']. The analysis shows that the control type is an on command, the automatic control variable is: Start / Stop Control Command, and the setpoint is 1.0.

[0117] Step 4. Ensure that the system conditions for controlling the centrifugal chiller unit 1 are met by inspecting all equipment and valves related to the chilled water and cooling water circulation.

[0118] Step 5. Generate control commands and start the centrifugal chiller unit_1 in the system to perform digital twin model simulation of the building water system.

[0119] Step 6. Call the building water system to obtain performance data of the water system and control equipment before and after the execution of the control strategy. Specifically: Centrifugal chiller_1: Chilled water supply and return temperature difference: 3℃, Cooling water supply and return temperature difference: 5℃, Chilled water flow rate: 0 m³ / h, Cooling water flow rate: 0 m³ / h, Chiller_1_COP: 0, etc. The obtained system and equipment information after the execution of the control command includes: Chilled water supply and return temperature difference: 5℃, Cooling water supply and return temperature difference: 4℃, Chilled water flow rate: 74.48 m³ / h, Cooling water main pipe flow rate: 104.82 m³ / h, Chiller_1_COP: 2.1, etc.

[0120] Step 7. The large language model combines the user's control commands and the obtained performance data to analyze the control effect of the control strategy execution. The response received was: "The start / stop control command for the centrifugal chiller unit_1 has been set to 1 as per your requirements, and the simulation results show the following changes in the system: Based on the information you provided, the system's chilled water circulation efficiency has improved, and the heat transfer performance has been enhanced. The chilled water has started flowing, the system has started operating, and it provides cooling. The cooling water supply and return temperature difference has decreased, indicating a reduction in the heat load of the cooling water system or an increase in the cooling water flow rate, which improves cooling efficiency. The cooling water has also changed from a stationary state to a flowing state, indicating that the system is in normal operation and the cooling water is carrying away heat. The chiller COP has increased from 0 to 2.1, indicating that the chiller has started working effectively and has a certain energy efficiency ratio. In summary, after the control strategy is implemented, the centrifugal chiller unit_1 is operating normally. Overall, after the control, the chilled water and cooling water flow rates ensure the system's heat exchange capacity, and the temperature difference reflects the effectiveness of heat exchange. The COP has increased from 0 to 2.1, indicating that the chiller has started working effectively." The user feedback was: "Meets expectations, execution is permitted."

[0121] Step 8. Receive and analyze user feedback. If the feedback is positive, allow the command to be executed. The cloud-based digital twin platform transmits the corresponding control strategy to the edge controllers at the building water system site, and these controllers execute the control strategy on-site to complete the control.

[0122] In summary, this invention provides a central air conditioning water system control method based on large language models and digital twins, comprising:

[0123] 1) Receiving user control commands. This process is the starting point for building an intelligent control system, aiming to capture and parse user needs expressed in natural language through an efficient input module. This process is designed based on understanding the complexity and diversity of user needs, including multimodal information such as direct text input and voice input, ensuring the system can accurately understand and process input in different formats. The input module's preprocessing functions, such as syntax correction, keyword extraction, and disambiguation, further improve the efficiency and accuracy of problem parsing, providing a clear and accurate description of user needs for subsequent steps.

[0124] 2) Determining the target device for the control command. This process involves accurately identifying the target device and operation type in the user's intent. First, the system ensures the clarity and executability of the command by checking if the user command explicitly specifies the device type and number. If information is insufficient, the system proactively interacts with the user, requesting additional information, re-issuing the command, and returning to the first step, demonstrating the system's adaptability and user-friendliness. Second, the system invokes the digital twin model system to obtain a list of devices currently permitted for control in the central air conditioning water system, and selects the target device based on the user command, showcasing the powerful capabilities of digital twin technology in simulating and matching devices in the physical world.

[0125] 3) Confirm the control variables and setpoints of the command. This process is the core of intelligent control, aiming to transform the user's abstract instructions into specific control commands. First, this process, based on the components of the device to be controlled, determines and ensures that the user's control command is clear, unambiguous, and executable. If information is insufficient, the system proactively interacts with the user, requesting additional information, a revised command, and returns to step one. Second, by analyzing the self-control variables of the device components the user intends to control, the system can accurately identify the names of the self-control variables in the user's intent and determine the setpoints of the control variables based on the type of control command (e.g., on, off, or numerical adjustment). If the command does not belong to any corresponding control method, the system interacts with the user, indicating that the device component to be controlled does not have a corresponding control method, requesting the user to re-issue the command, and returns to step one. This process is designed to consider the diversity and ambiguity of user commands. For example, in numerical adjustment control, if the user does not explicitly specify the setpoint of the control variable, the system can match the specific setpoint by combining historical information. This process demonstrates the system's intelligence and flexibility in handling fuzzy information.

[0126] 4) Verify and confirm that the current water system status meets the control conditions. This process ensures the safety and effectiveness of control command execution. By confirming the topological connections of the equipment to be controlled and checking the operating status of related equipment and valves, the system verifies whether the current state of the water system meets the preset control conditions before executing any control operation. The logic of this process reflects the emphasis on system stability and operational safety, ensuring that control commands are executed under appropriate system conditions.

[0127] 5) Automatically generate and execute control strategies. This process demonstrates the automation capabilities of the intelligent control system. Based on the control equipment and control variables determined in the previous steps, the system can automatically generate the corresponding control strategy for the digital twin building water system and execute the corresponding system changes. This process showcases the powerful capabilities of digital twin technology in achieving real-time synchronization between virtual and real environments and the automated deployment of control strategies.

[0128] 6) Perform digital twin model simulation. This process is a crucial step in control strategy optimization, using high-precision computational simulation to predict and analyze performance changes in the building water system under control operations. This process is designed to allow the system to assess the potential impact and effects of control strategies before actual application, thereby optimizing control commands and strategies to ensure optimal control effectiveness.

[0129] 7) Obtain key variable information for control strategy execution. This process, by invoking the digital twin model, enables the system to acquire key variable information for the water system and control equipment before and after the control strategy execution, providing data support for subsequent control effect analysis. This process demonstrates the application of digital twin technology in monitoring and recording system state changes, providing an accurate data foundation for the evaluation and adjustment of control strategies.

[0130] 8) Control Effect Analysis. The large language model comprehensively considers user control commands and acquired key variable information to analyze the effect of control strategy execution and feeds the analysis results back to the user. This process demonstrates the system's ability to evaluate and optimize the execution effect after completing the control command execution, ensuring that the user can obtain the expected control effect and make adjustments when necessary.

[0131] 9) Receive and analyze user feedback. The large language model completes the final evaluation of the control strategy by analyzing user feedback on the control effect. Positive feedback will lead to the saving of control records, providing experience for future control operations; while negative feedback will trigger further interaction with the user, requiring reconfirmation of the control method and returning to step one, demonstrating the system's adaptability and user-centricity in continuously optimizing the control effect.

[0132] 10) The cloud-based digital twin platform interacts with the edge controller. This process marks the transition of control strategies from virtual simulation to real-world implementation. By transmitting optimized control strategies to the edge controllers in the field via the cloud platform, this process demonstrates the ability of digital twin technology to integrate cloud computing and edge computing resources to achieve efficient and reliable control strategy deployment.

[0133] In summary, the core innovation of this invention lies in the integrated application of large language models and digital twin technology, providing a new, efficient, accurate, and low-risk method for the intelligent control of building air conditioning water systems. This method cleverly combines large language models and digital twin technology to address the problems of insufficient informatization in the construction industry and excessive reliance on manual operation and maintenance. By introducing natural language processing technology, this method can directly respond to control requests from maintenance personnel in natural language form, utilizing the advanced understanding and cognitive capabilities of large language models to achieve precise control and intelligent management of building air conditioning water systems. This method can not only accurately parse control commands from maintenance personnel and automatically identify relevant equipment, their control variables, and setpoints, but also perform pre-verification in the digital twin model to ensure the accuracy and completeness of the control scheme. By first verifying the control scheme in a virtual environment and then adjusting the edge controller commands through a cloud platform for implementation, this invention significantly reduces the risks of actual operation and avoids potential errors, thereby improving the efficiency and security of system operation. Furthermore, this method supports multimodal data input, including text and voice, enhancing user interaction with the system and optimizing the understanding of control logic and user needs. The implementation of this method not only reduces the costs that may be caused by misoperation, but also promotes the transformation of system operation and maintenance management towards intelligence and automation, opening up new avenues for technological innovation and application.

[0134] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, all technical solutions obtained through equivalent substitution or transformation fall within the protection scope of the present invention.

Claims

1. A control method for a central air conditioning water system based on a large language model and digital twin, characterized in that, Includes the following steps: S1. The input module receives the control command issued by the user, matches the control command with the pre-built keyword library, and determines whether the control command contains the keywords in the keyword library: if so, the control command is preprocessed to obtain the processed command; otherwise, the user is prompted that the control command entered does not belong to the building energy field and needs to be re-entered. S2. Input the processing command into the large language model and determine whether the processing command simultaneously specifies the type and number of the device to be controlled. If not, prompt the user on the interactive interface to supplement the device information and re-enter the control command. If both are specified, retrieve the equipment list from the building air conditioning water system in the digital twin model system. Match the device to be controlled in the digital twin model system with the device type and number in the processing command. If the device to be controlled cannot be used, prompt the user on the interactive interface that the device to be controlled cannot be used, please reconfirm the device to be controlled, and re-enter the control command. S3. The large language model determines whether the processing instruction matches the control command type of the device to be controlled: If they do not match, the user is prompted to reconfirm the control method and re-enter the control instruction on the interactive interface; if they match, the automatic control variable list of the device to be controlled is obtained from the building air conditioning water system; if the processing instruction and the automatic control variable list match, the control variable name is determined from the automatic control variable list, and the control variable set value corresponding to the control command type is output; if the processing instruction and the automatic control variable list do not match, the user is prompted to reconfirm the control variables of the device to be controlled and re-enter the control instruction on the interactive interface. S4. Obtain the name of the device to be controlled from the building air conditioning water system and obtain the topology connection relationship of the device to be controlled; sort the devices in the topology connection relationship according to the pre-built device inspection level to obtain the sorted device list; query the sorted device list in turn, for the current device in the sorted device list, call the building air conditioning water system to first obtain the current device's operating status. If the current device's operating status meets the requirements for executing processing instructions and control variable settings, then query the next device until the device to be controlled is found. Otherwise, call the building air conditioning water system to adjust the current device's operating status until the current device's operating status meets the requirements for executing processing instructions and control variable settings. S5. Obtain the control strategy of the building air conditioning water system corresponding to the processing command, input the control strategy into the building air conditioning water system, change the set value of the control variable, obtain the performance data of each device of the building air conditioning water system before the change, and perform digital twin model simulation on the building air conditioning water system after the change to obtain the performance data of each device of the building air conditioning water system after the change. Input the performance data before the change and the performance data after the change into the large language model to obtain the analysis results of the control effect and feed them back to the user. S6. After receiving feedback from the user regarding the analysis results, input the user's feedback results into the large language model. The large language model determines whether the control strategy meets the user's needs. If it does, the modified control variable settings are saved to the historical information database, and the control strategy that meets the user's needs is passed to the field edge controller of the building air conditioning water system. The field edge controller executes the control strategy that meets the user's needs. Otherwise, the control command is re-entered. In step S1, the input module is capable of receiving various forms of control command input and has the ability to convert speech into text; the preprocessing includes correcting grammatical errors in the control commands, extracting key information points from the control commands, and eliminating ambiguity in control commands with multiple interpretations by using contextual information or requesting more information from the user. In step S3, if the control command type is an open command, the control variable setting value is 1; if the control command type is a close command, the control variable setting value is 0; if the control command type is a numerical adjustment command, the large language model will use the setting value in the processing instruction as the control variable setting value; if there is no setting value in the processing instruction, the retrieval model will search for the historical setting value of the automatic control variable in the historical information database based on the current operating status of the building air conditioning water system, and use the historical setting value with the closest value as the control variable setting value. Through a self-learning mechanism, the building air conditioning water system is allowed to automatically optimize control strategies and adjust the understanding and generation logic of the large language model based on data in the historical information database; The edge controller is deployed on the building site and is responsible for executing control strategies and directly managing and controlling the building air conditioning water system equipment on site. The edge controller supports real-time data acquisition and analysis, and can instantly feed back the operating status of each device in the building's air conditioning water system to the digital twin platform, providing a basis for the generation and optimization of control strategies. Through the secure communication protocol established between the digital twin platform and the edge controller, the secure transmission and execution of processing instructions are ensured, preventing data tampering and unauthorized access. In step S4, if the control command type is an open command or a value adjustment command, the equipment order corresponding to the inspection level is: all types of valves, all types of cooling water pumps, all types of chilled water pumps, all types of cooling towers, and all types of chiller units; if the control command type is a close command, the equipment order corresponding to the inspection level is: all types of chiller units, all types of cooling towers, all types of cooling water pumps, all types of chilled water pumps, and all types of valves.

2. The central air conditioning water system control method based on large language model and digital twin as described in claim 1, characterized in that, The building air conditioning water system also includes equipment modules that can dynamically update and maintain equipment information.

3. The central air conditioning water system control method based on large language model and digital twin as described in claim 1, characterized in that, The historical information database is used to collect and store historical control commands, analysis results of control strategies, and user feedback.

4. The central air conditioning water system control method based on large language model and digital twin as described in claim 1, characterized in that, The building air conditioning water system supports remote control operation, allowing users or maintenance personnel to remotely monitor and control the system via network connection.

5. The central air conditioning water system control method based on large language model and digital twin as described in claim 1, characterized in that, When acquiring control strategies, the control strategies are adjusted according to the real-time operating status of each device in the building's air conditioning water system, or the user selects a control strategy based on the control effect, and the system automatically performs a risk assessment of the control strategy.

Citation Information

Patent Citations

  • Method and device for processing operating instructions of air conditioner control system, equipment and medium

    CN110044012A

  • Intelligent voice infrared equipment control system and method

    CN111276145A

  • Central control system control method and device based on digital twinning, equipment and medium

    CN117518957A

  • Air conditioning method, device and equipment and storage medium

    CN117889531A