Heating and ventilation system multi-stage optimization and large model fusion promotion method and related equipment

Abstract

The invention discloses a heating and ventilation system multi-stage optimization and large model fusion promotion method and related equipment. The method comprises the following steps: S1, determining a system architecture and a topological relation; s2, collecting equipment operation data; s3, establishing an equipment performance model; s4, constructing an optimization demand matrix; s5, traversing the demand matrix, executing optimization, obtaining an optimization result matrix, and generating a first-stage optimization strategy; s6, single-day load and time-of-use electricity price consideration are introduced for optimization, and a two-stage optimization strategy is generated; and S7, multi-day scale optimization of cold / heat storage consideration is introduced, and a three-stage optimization strategy is generated. According to the scheme, the problems existing in an existing scheme can be systematically solved.

Description

Technical Field

[0001] The present invention relates to HVAC system technology, and in particular to HVAC system control and optimization technology. Background Art

[0002] As the core system for controlling energy consumption and environmental comfort in large buildings, the HVAC system's operational efficiency and control accuracy directly impact the building's energy performance and user experience. However, according to the "China Building Industry Automation White Paper," approximately 40% of newly built HVAC systems experience failure within a year due to automated control failures, resulting in long-term "malfunctioning" operations. Over 90% of HVAC systems are still manually controlled, characterized by chaotic strategies and inefficient control. Furthermore, the complex architecture and numerous related parameters of HVAC systems make optimization challenging. Traditional control and optimization methods often struggle to respond quickly and adapt to changing building usage, external environmental disturbances, and the coordinated operation of multiple devices.

[0003] In actual projects, the factors that affect the operating performance of HVAC systems exhibit multi-dimensional coupling characteristics in time and space, such as differences in cooling load demand in different time periods, time-of-use fluctuations in electricity prices, and the buffering characteristics of hot and cold energy storage systems, all of which place higher demands on control strategies. However, existing methods generally lack a holistic consideration of these factors, and most optimization strategies are still limited to local adjustments at a certain point in time or to a certain control object, making it difficult to achieve global energy efficiency optimization and strategy coordination. Therefore, there is an urgent need for a systematic control method that can integrate multiple operating stages and multiple optimization dimensions to guide parameter configuration and strategy formulation from a system perspective.

[0004] At the same time, with the continuous breakthroughs in big model technology, it has demonstrated significant advantages in knowledge generalization, semantic understanding, and task transfer, and is expected to enhance intelligence in energy management and system optimization. However, the direct application of big models to HVAC system optimization still faces a series of challenges: First, big models lack a native understanding of the physical structure, control logic, and operational boundaries of HVAC systems, making it difficult to establish system cognition through language information alone. Second, in tasks such as auxiliary query and strategy induction, execution-level obstacles such as ambiguous object recognition, inaccurate field matching, and incomplete interface parameter extraction exist, limiting their ability to serve as a leading optimization tool. Summary of the Invention

[0005] To address the challenges of existing HVAC system control and optimization solutions, this paper aims to provide an innovative multi-stage HVAC system operation optimization solution. By systematically introducing a phased, multi-scale strategy coordination mechanism, this approach connects the coupling relationships between equipment models, operational data, load response, and cost control, forming a closed-loop optimization path. Furthermore, by integrating the reasoning and generalization capabilities of large language models, this approach provides intelligent support for data modeling, semantic recognition, optimization strategy query, and dynamic adjustment, enabling a transition from "rule-driven" to "intelligently guided" HVAC system operation optimization.

[0006] To achieve the above objectives, the present invention provides a method for multi-stage optimization and large-model fusion improvement of a HVAC system, the method comprising:

[0007] S1: Determine the HVAC system architecture and topology relationship;

[0008] S2: Collect operating data of various equipment in the HVAC system;

[0009] S3: Based on the data collected in step S2, establish an equipment performance model for the equipment in the HVAC system;

[0010] S4: Determine the key input parameters and their optional ranges involved in the optimization task, and construct a discretized multi-dimensional optimization requirement matrix based on them;

[0011] S5: traverse the demand matrix constructed in step S4 to perform optimization and obtain the optimization result matrix;

[0012] S6: Based on the device-level optimization in S5, daily load distribution and time-of-use electricity prices are introduced to perform optimization within the daily scale. By constructing multi-period objective functions and constraints, a two-stage daily optimization strategy is formed.

[0013] S7: Based on the single-day optimization of S6, the consideration of cold / heat storage devices is introduced, and multi-day optimization is carried out. By constructing a comprehensive objective function and constraints considering cold / heat storage, and iteratively optimizing the heuristic algorithm of the large-scale solution, a three-stage multi-day optimization strategy is formed.

[0014] Furthermore, the step S3 establishes the equipment performance model through the following steps:

[0015] S301 Equipment classification, clarifying model categories: Classify equipment based on type and function, and clarify the model type that needs to be established for each type of equipment;

[0016] S302 Model Algorithm Selection: After the model type is determined, a specific modeling algorithm is selected based on the availability of equipment operation data and the complexity of model fitting;

[0017] S303 Model Construction: For the selected model, perform model parameter fitting or training to complete the mathematical expression construction of the equipment performance model, ensuring that the input and output variables cover the actual operating range;

[0018] S304 Model Accuracy and Effectiveness Evaluation Comparison: Evaluate the accuracy and applicability of various models by selecting model evaluation indicators;

[0019] S305 Equipment Performance Model Output: Output the final selected performance model structure and parameters into a standardized model format that can be called by optimization algorithms, control strategy modules or analysis tools.

[0020] Furthermore, in step S4, the optimization demand matrix is ​​constructed by the following steps:

[0021] S401: Identify the required input parameters, the number of which is N: Based on the scenarios and boundary conditions involved in system operation optimization, determine the input parameter dimensions that need to be included in modeling and optimization analysis;

[0022] S402 Determine each parameter range: For each parameter dimension determined in step S401, define its reasonable value range and discrete granularity;

[0023] S403 constructs an N-dimensional demand matrix: performs a multi-dimensional Cartesian product operation on the discrete value combination of each parameter in step S402 to construct an N-dimensional optimization demand matrix.

[0024] Furthermore, in step S5, the optimization result matrix is ​​generated by the following steps:

[0025] S501: traverse and obtain the demand parameters corresponding to each unit in the demand matrix;

[0026] S502: Optimize the calculation according to the device calculation order to obtain multiple control solutions;

[0027] S503: Using the control solution obtained in step S502 as an initial reference, an optimization is performed using a heuristic algorithm;

[0028] S504: Compare the algorithm output solution with the initial reference solution and select the better solution as the final solution;

[0029] S505: After completing the optimization for each unit in the matrix, fill it with the final solution to form an optimization result matrix with the same matrix structure.

[0030] Furthermore, in step S6, the optimization within a single day is completed through the following steps:

[0031] S601: Introducing or predicting daily load distribution and time-of-use electricity price distribution;

[0032] S602: Constructing an objective function based on the actual optimization goal;

[0033] S603: Setting key boundary conditions that affect the execution of the operation strategy;

[0034] S604: Within a limited time period, based on the optimization result matrix, a single-day hourly control plan is generated using dynamic programming or a similar algorithm.

[0035] Furthermore, in step S7, the multi-day optimization is completed by the following steps:

[0036] S701: Introducing or predicting daily load distribution and time-of-use electricity price distribution;

[0037] S702: Introducing parameters of the cold / heat storage device;

[0038] S703: Determine a multi-day comprehensive objective function;

[0039] S704: Based on the key boundary conditions affecting the execution of the operation strategy set in step S603, adjustments are made to additionally consider the switching constraints of the cold / heat storage device and the continuity of the remaining amount;

[0040] S705: Randomly generate large-scale control plans on a multi-day scale and optimize them using heuristic algorithms;

[0041] S706: Select an optimal or approximately optimal scheduling strategy based on the objective function value, constraint satisfaction, and operational feasibility.

[0042] In order to achieve the above object, the present invention further provides a computer-readable storage medium having a program stored thereon, which implements the above steps when executed by a processor.

[0043] In order to achieve the above-mentioned object, the present invention also provides a processor, which is used to run a program, and when the program is running, the steps of the above-mentioned HVAC system multi-stage optimization and large model fusion improvement method are executed.

[0044] In order to achieve the above-mentioned objectives, the present invention also provides a terminal device, which includes a processor, a memory, and a program stored in the memory and runnable on the processor. The program code is loaded and executed by the processor to implement the steps of the above-mentioned HVAC system multi-stage optimization and large model fusion improvement method.

[0045] In order to achieve the above objectives, the present invention also provides a computer program product, which, when executed on a data processing device, is suitable for executing the steps of the above-mentioned HVAC system multi-stage optimization and large model fusion improvement method.

[0046] The multi-stage optimization and large-model fusion improvement method for HVAC systems provided by the present invention has the following beneficial effects compared with the existing technology:

[0047] 1) The solution of the present invention significantly reduces the need for manual intervention through complete optimization process planning and automated execution, improves the continuity and stability of optimization tasks, and is suitable for long-term operation and complex project scenarios;

[0048] 2) The solution of the present invention supports multi-level linkage of equipment, system, and energy, as well as joint optimization on time scales such as hours, days, and multiple days;

[0049] 3) Through a modular optimization system and cost-controllable design, the solution of the present invention allows users to configure optimization paths of different stages and depths according to specific needs, ensuring a balance between optimization benefits and implementation costs, and improving practical feasibility;

[0050] 4) The solution of the present invention leverages the semantic understanding and logical reasoning capabilities of the large model to migrate experience across projects, automatically generate strategies, and adjust parameters, with strong generalization capabilities and deployment efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The present invention is further described below with reference to the accompanying drawings and specific embodiments.

[0052] Figure 1 The overall process of the multi-stage optimization and large-model fusion improvement method of the HVAC system in this invention;

[0053] Figure 2 This is a flow chart of sub-step S3 in the present invention;

[0054] Figure 3 This is a flow chart of sub-step S4 in the present invention;

[0055] Figure 4 This is a flow chart of sub-step S5 in the present invention;

[0056] Figure 5 This is a flow chart of sub-step S6 in the present invention;

[0057] Figure 6 This is a flow chart of sub-step S7 in the present invention;

[0058] Figure 7 Flowchart for the implementation of sub-step S3 in the example of the present invention;

[0059] Figure 8 Flowchart for the implementation of sub-step S4 in the present invention;

[0060] Figure 9 Flowchart for the implementation of sub-step S5 in the present invention;

[0061] Figure 10 Flowchart for the implementation of sub-step S6 in the example of the present invention;

[0062] Figure 11 Flowchart for the implementation of sub-step S7 in the present invention;

[0063] Figure 12 This is an example diagram of the large model architecture in the example of the present invention. DETAILED DESCRIPTION

[0064] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below with reference to specific illustrations.

[0065] Through a thorough study of existing HVAC system performance optimization and improvement solutions, the present invention provides a systematic solution to the problems existing in the existing HVAC system performance optimization and improvement solutions.

[0066] In response to the problems in the existing technology such as the lack of standardized optimization processes, high reliance on manual intervention in each link, and difficulty in dynamic linkage, the present invention establishes a multi-stage optimization framework with a clear structure and complete logic. Through systematic planning of optimization goals, execution paths and control parameters, it realizes the process management and automated execution of the whole process optimization tasks, reduces the intensity of manual intervention, and improves the continuity and stability of the optimization work.

[0067] In response to the problems that the existing HVAC system optimization scale is limited, it only stays at the equipment level, and it is difficult to achieve coordinated optimization at the system level and time series level, this invention introduces a linkage mechanism with multiple time scales and multiple control objects, so that the optimization strategy can cover multiple levels such as equipment operation, load response, energy storage regulation, and electricity price matching, thereby improving the coordination and energy efficiency of the overall operation.

[0068] In response to the problems that existing technologies do not modularize the optimization process, lack phased planning capabilities, and have difficulty in flexibly controlling the optimization depth and cost, the present invention proposes a configurable and combinable phased optimization method. It flexibly configures the optimization objectives and module content according to the system scale, operating status and user needs, and simultaneously evaluates the corresponding execution costs, providing decision makers with a more cost-effective basis for selecting optimization solutions.

[0069] To address the problems in traditional optimization methods such as reliance on fixed workflows and manual experience, low intelligence, and poor cross-project adaptability, the present invention further integrates the reasoning and generalization capabilities of large language models, introduces intelligent agent collaboration mechanisms in strategy generation, parameter identification, historical solution migration, and other links, improves the intelligent response capabilities and generalization promotion efficiency of the optimization system, and realizes rapid replication and deployment among multiple projects.

[0070] Based on this, the present invention develops an innovative multi-stage optimization and large-scale model fusion enhancement method for HVAC systems. This solution systematically introduces a phased, multi-scale strategy coordination mechanism to connect the coupling relationships between equipment models, operational data, load response, and cost control, forming a closed-loop optimization path. Furthermore, it further integrates the reasoning and generalization capabilities of large language models to provide intelligent support for data modeling, semantic recognition, optimization strategy query, and dynamic adjustment, achieving a transition from "rule-driven" to "intelligently guided" HVAC system operation optimization.

[0071] See also Figure 1 , which shows the overall flow chart of the multi-stage optimization and large-model fusion improvement method of the HVAC system given by the present invention.

[0072] Based on the diagram, the overall execution process of the multi-stage optimization and large-model fusion improvement method of the HVAC system provided by the present invention is mainly composed of the following seven main steps:

[0073] S1: Determine the system architecture and topology relationship;

[0074] S2: Collect equipment operation data;

[0075] S3: Establish equipment performance model;

[0076] S4 builds an optimization demand matrix;

[0077] S5 traverses the demand matrix to perform optimization and obtain the optimization result matrix, generating a first-stage optimization strategy;

[0078] S6 introduces single-day load and time-of-use electricity price considerations for optimization and generates a two-stage optimization strategy;

[0079] S7 introduces multi-day optimization considering cold / heat storage and generates a three-stage optimization strategy.

[0080] The following describes the specific implementation of each step in the multi-stage optimization and large-model fusion improvement method for the HVAC system, as well as the equipment that may be involved.

[0081] In this improvement method, step S1 determines the system architecture and topology relationship:

[0082] This step abstracts and extracts the components of the HVAC system and their connection relationships to clarify the architecture and physical topology of important equipment in the system.

[0083] Architectural information includes the number and model numbers of key equipment in the system, including heat and cold sources, terminals, transmission and distribution systems, and energy storage devices. Physical topology includes device connection logic and energy flow paths. After extracting this information, a structured topology map is constructed to support subsequent modeling, optimization, and control processes.

[0084] In this improvement method, step S2 collects equipment operation data:

[0085] This step aims to obtain basic HVAC system operating data to support subsequent performance modeling and strategy optimization. Data collection covers key equipment such as chillers, pumps, cooling towers, valves, and terminals, capturing their operating status parameters (such as power, flow, temperature, and opening) as well as system-level information (such as load changes, energy storage status, external weather conditions, and electricity prices).

[0086] In this improvement method, step S3 establishes an equipment performance model:

[0087] This step is to establish a performance model for the main equipment in the system. The performance model is mainly used to characterize the mathematical relationship between key parameters in the equipment and is used for the calculation and analysis of final evaluation indicators such as energy consumption and power.

[0088] See also Figure 2 In this step, the equipment performance model is established through the following five sub-steps. The basic process is as follows:

[0089] S301 Equipment classification, clarifying model types: Divide according to equipment type and function, clarify the type of model that needs to be established for each device, that is, different devices need to establish the mathematical relationship between the key parameters.

[0090] S302 Model Algorithm Selection: After determining the model type, select a specific modeling algorithm based on the availability of equipment operating data and the complexity of model fitting. Commonly used algorithms include polynomial regression, neural networks (ANNs), decision trees, random forests, and XGBoost. Depending on the actual situation, empirical formulas derived from research may also be used.

[0091] S303 Model Construction: For the selected model, perform model parameter fitting or training to complete the construction of the mathematical expression of the equipment performance model, ensuring that the input and output variables cover the actual operating range.

[0092] S304: Comparison of model accuracy and effectiveness evaluation: The accuracy and applicability of various models are evaluated by selecting model evaluation indicators. Common indicators include the coefficient of determination (R 2 ), mean square error (MSE), mean absolute error (MAE), etc. Set the model accuracy threshold. If the selected model accuracy threshold does not meet the standard, readjust the model or return to step S302 to reselect another algorithm.

[0093] S305 Equipment Performance Model Output: Output the final selected performance model structure and parameters into a standardized model format that can be used by optimization algorithms, control strategy modules, or analysis tools. The output can be in the form of a function expression, model object, or rule set, and should be compatible and scalable.

[0094] As an alternative, the above-mentioned equipment performance model establishment steps can also be integrated into a large model to realize automatic establishment of the equipment performance model.

[0095] By integrating the big model into the equipment performance model establishment process, the parts that originally required manual intervention are now controlled by the big model. The process is as follows:

[0096] S3011 Equipment information / data input: The basic information and operating data of the acquisition equipment need to be input into the large model, including equipment identification, type parameters, structural parameters and its operating sequence data.

[0097] S3021 large model automatically matches equipment and models: After obtaining structured input, the large model automatically matches equipment types and model types based on the existing domain knowledge base and semantic mapping capabilities. For example, a water pump needs to establish a head-flow model, an efficiency-flow model, etc., and a chiller needs to establish at least a compressor model. Evaporator or condenser models can also be established separately according to demand and accuracy requirements.

[0098] S3031 Autonomous model algorithm selection: The large model will autonomously select appropriate modeling algorithms, such as polynomial regression, decision tree, neural network, etc., based on the scale, volatility and variable relationships of equipment operation data.

[0099] S3041 Autonomous model construction: The large model generates corresponding modeling parameter settings, and automatically calls the fitting tools in the tool function or independently writes code to execute in the code sandbox, which is mainly used to isolate the code running environment.

[0100] S3051 Model Accuracy and Effectiveness Evaluation and Comparison: The large model automatically performs model performance evaluation and cross-comparison based on established evaluation indicators, such as automatically generating model accuracy reports, stability analysis, and generalization ability analysis. If the preset threshold requirements are not met, the model will be readjusted or returned to S3031 to reselect other algorithms.

[0101] S3061 Equipment performance model output: consistent with the process and method of S305.

[0102] In this improvement method, step S4 constructs the optimization demand matrix:

[0103] This step aims to identify the key input parameters involved in the optimization task and their selectable ranges, and construct a discretized multidimensional optimization requirement matrix based on these parameters. The optimization requirement matrix enumerates the external load requirements and environmental boundary conditions that the system may face under typical operating conditions, providing a comprehensive input space for subsequent optimization strategy generation.

[0104] See also Figure 3 In this step, the optimization demand matrix is ​​constructed through the following three sub-steps. The basic process is as follows:

[0105] S401 clearly inputs the required parameters, the number is N:

[0106] In this step, the input parameter dimensions to be included in the modeling and optimization analysis are determined based on the scenarios and boundary conditions involved in the system operation optimization. These parameters typically include system load requirements, outdoor environmental conditions, and other key influencing factors, such as cooling capacity and wet-bulb temperature.

[0107] S401 determines the range of each parameter:

[0108] This step defines a reasonable range of values ​​and discretization granularity for each parameter dimension determined in substep S401. Parameter ranges can be set based on historical operating data statistics, design operating ranges, boundary extremes, or engineering experience to ensure representativeness and computability of the matrix construction. Discretization of each parameter can employ methods such as equal-interval partitioning, quantile distribution, and cluster representative values ​​to control the scale and resolution of the parameter space.

[0109] S403 constructs an N-dimensional demand matrix through multi-dimensional Cartesian product operations:

[0110] This step performs a multi-dimensional Cartesian product operation on the discrete value combinations of each parameter in sub-step S402 to construct an N-dimensional optimization requirement matrix. Each cell of this matrix represents an independent operating condition combination, and each subsequent combination will serve as the input condition for an independent optimization task.

[0111] As an alternative, the above-mentioned optimization demand matrix construction solution can also be integrated with a large model to identify user intent and connect with historical data to complete the generation of the demand matrix. The corresponding implementation process is as follows:

[0112] S4011 Natural language description optimization requirements:

[0113] The current optimization goals and operational scenario requirements can be expressed through natural language input, such as "optimizing the cooling efficiency of the air conditioning system in hot weather." This description does not need to be structured and can be freely expressed based on business understanding.

[0114] S4012 large model semantic recognition, automatic generation of required parameters, structured output quantity N:

[0115] The large model, combining its semantic understanding capabilities with its domain knowledge base, parses the input content, automatically identifies key optimization dimensions and input parameters, and generates a structured parameter list. The model extracts factors related to the optimization objective, such as cooling load, ambient temperature, and humidity, and outputs clear parameter names, units, and the number of parameters N, which serve as the input basis for subsequent processes.

[0116] S4031 determines the range of each parameter:

[0117] For each identified parameter, the large model automatically combines historical operating data, equipment performance boundaries, engineering experience, or set rules to generate a feasible value range and discrete granularity for that parameter. If necessary, the large model can integrate with external databases and call on statistical information to generate a more representative set of values, enabling intelligent tailoring and streamlining of the parameter space.

[0118] S4041 constructs an N-dimensional demand matrix through multi-dimensional Cartesian product operation: the implementation process of this step is consistent with the process and method of sub-step S403.

[0119] In this improvement method, step S5 traverses the demand matrix to perform optimization and obtain the optimization result matrix:

[0120] This step performs device-level control optimization for each discrete parameter combination in the optimization requirements matrix. Based on the pre-defined performance model and optimization objectives, the system independently solves for each input parameter set and generates the corresponding optimal control strategy or device configuration. The optimization results are then added to a result matrix that matches the structure of the requirements matrix, ultimately forming a complete optimization result matrix and establishing the first-stage device-level optimization strategy.

[0121] See also Figure 4 In this step, the following five sub-steps are combined to form a first-stage device-level optimization strategy. The basic process is as follows:

[0122] S501 traverses and obtains the demand parameters corresponding to each unit in the demand matrix:

[0123] The system traverses all cells in the optimization demand matrix in turn, extracts the corresponding demand parameter groups, including cooling / heating loads, outdoor environmental conditions, etc., and uses them as the input basis for the current optimization operation.

[0124] S502 optimizes the calculation sequence of the equipment and obtains multiple control solutions:

[0125] Based on the established equipment performance model, the system performs combined calculations according to the logical order of equipment operation (such as chillers, refrigeration pumps, cooling towers), and generates several groups of candidate solutions that meet the system operation constraints (such as operating range, power limit, load matching requirements, etc.). Each group of solutions includes the number of equipment starts and stops and the control parameter settings.

[0126] S503 uses the obtained solution as an initial reference and uses a heuristic algorithm for optimization:

[0127] Based on the candidate solutions, a heuristic algorithm is introduced to iteratively optimize the solutions with the goal of minimizing energy consumption to obtain a better output solution.

[0128] S504 compares the algorithm output solution with the initial reference solution and selects the better solution as the final solution:

[0129] The system compares the optimized output solution with the lowest energy consumption solution among the original candidate solutions, and selects the solution with better energy consumption performance as the optimal control solution corresponding to the current parameter group.

[0130] After each unit in the S505 matrix is ​​optimized, it is filled with the final solution to form an optimization result matrix with the same matrix structure:

[0131] Fill each group of final optimization solutions into the positions with consistent structure in turn, complete the one-to-one mapping corresponding to the demand matrix, and thus construct a complete optimization result matrix.

[0132] As an alternative, the aforementioned first-stage device-level optimization strategy can be integrated with a large model to dynamically control the execution of the heuristic algorithm. This model also adds more flexible judgments and semantic explanations to the solution selection process. The corresponding implementation process is as follows:

[0133] Among them, steps S5011 and S5021 are consistent with steps S501 and S502, and the rest of the process is as follows:

[0134] S5031 uses a large model to expand and generate solutions based on the obtained solutions, and then inputs them into a heuristic algorithm for optimization. Based on the candidate solutions, the large model's language generation and reasoning capabilities are introduced to automatically expand multiple groups of control strategies with similar structures as enhanced input. By dynamically adjusting parameter boundaries and combination strategies, it helps construct a richer initial solution space. The expanded solutions are then input into the heuristic algorithm for iterative optimization to further improve the quality of the local optimal solution.

[0135] The S5041 large-scale model uses flexible judgment to select solutions based on hard indicators, and provides explanations for the selected solutions. After completing the heuristic optimization, the large-scale model participates in the screening and evaluation of the resulting solutions, comprehensively judging the advantages and disadvantages of the solutions based on both hard indicators and flexible dimensions such as system stability and balance. The final output solution is accompanied by semantic explanations, explaining the reasons for the selection and applicable scenarios, providing intelligent assistance for understanding and adjustment.

[0136] S5051 forms an optimization result matrix with the same matrix structure. The large model can generate visual charts and perform semantic queries: The difference between this step and sub-step S505 is that the large model can generate additional visual charts based on the matrix and perform semantic queries to quickly locate solutions.

[0137] In this improvement method, step S6 introduces single-day load and time-of-use electricity price considerations for optimization: Based on the equipment level optimization in step S5, this step introduces single-day load distribution and time-of-use electricity price to perform optimization within the single-day scale. By constructing multi-period objective functions and constraints, a two-stage single-day scale optimization strategy is formed.

[0138] See also Figure 5 In this step, the following four sub-steps are combined to form a two-stage single-day optimization strategy. The basic process is as follows:

[0139] S601: Introducing or predicting daily load distribution and time-of-use electricity price distribution: This step uses historical data analysis or load forecasting models to obtain cooling and heating load demand curves for each time period within the target operating day. Combined with time-of-use electricity price data from the power market, a 24-hour or finer-grained electricity price time series curve is constructed.

[0140] S602 Determine the objective function: Construct an objective function based on the actual optimization goal. The goal may include but is not limited to indicators such as minimizing the total energy consumption in a single day, minimizing the electricity bill, and maximizing the energy efficiency per unit load.

[0141] S603 Determine Constraints: Set key boundary conditions that affect the execution of the operation strategy, such as the minimum start and stop time of the equipment, the upper limit of the operation frequency, the start and stop cost, the upper and lower limits of the cooling output, etc., to ensure that the generated hourly strategy can be actually deployed, meet the control requirements of the continuous period, and comply with the requirements for safe operation of the equipment.

[0142] S604 uses the optimization matrix as the basis for a single-day, hourly control plan within a limited timeframe. Based on the constructed optimization matrix, S604 selects optimization strategies for different load levels and environmental boundaries as candidate solutions. Using dynamic programming or other time-series optimization algorithms, S604 combines the load-price distribution to construct a daily decision path. Ultimately, it outputs an hourly control strategy for the entire 24-hour period, achieving optimal energy consumption and economic efficiency throughout the entire day.

[0143] As an alternative, the aforementioned two-stage, single-day optimization strategy can be integrated with a large model. This model can replace manual work in establishing load distribution, constructing objective functions, and building constraints. Furthermore, the model can autonomously control the algorithm, generate multiple alternative paths, and select the final solution after flexible judgment. The corresponding implementation process is as follows:

[0144] S6011 The large model autonomously constructs a load forecasting model and establishes a distribution, introducing the distribution of time-of-use electricity prices: The large model calls external data and automatically calls the matching load forecasting method to construct the hourly distribution of cold / hot loads within a single day; the rest is consistent with sub-step S601.

[0145] S6011 automatically generates multi-objective functions based on user needs: it identifies user intent, generates an adaptive objective function with adjustable weights through semantic analysis, and outputs it through a reasonable mathematical expression.

[0146] S6011 automatically generates and determines constraints based on user needs: based on system equipment characteristics, operating rules and historical policy libraries, the model automatically identifies the equipment constraints, system boundary conditions and scheduling logic that need to be applied, and generates a structured mathematical constraint model.

[0147] S6041 Dynamic programming or similar algorithm generates a daily hourly control plan: The implementation process of this step is consistent with sub-step S604.

[0148] S6051 multi-path alternatives and flexible selection output the final solution: Based on the original optimization, the large model independently outputs multiple alternative paths through equivalent path storage, Top-K pruning, etc., and adds flexible selection judgment to form the final solution.

[0149] In step S7 of this improved method, multi-day optimization is introduced with consideration of cold / heat storage. This step aims to introduce consideration of cold / heat storage devices on the basis of the single-day optimization in S6 and perform multi-day optimization. By constructing a comprehensive objective function and constraints considering cold / heat storage, and iteratively optimizing the heuristic algorithm of the large-scale solution, a three-stage multi-day optimization strategy is formed.

[0150] See also Figure 6In this step, the following six sub-steps are combined to form a three-stage multi-day optimization strategy. The basic process is as follows:

[0151] The implementation process of step S701 is the same as that of S601.

[0152] S702 introduces the parameters of the cold / heat storage device;

[0153] S703 determines the multi-day comprehensive objective function: based on step S602, it is necessary to adjust to an objective function with a longer time span, and additional consideration needs to be given to the energy consumption of cold / heat storage and switching losses, etc.

[0154] S704 Determine Constraints: This step is based on step S603 and is adjusted to additionally consider the switching constraints of the cold / heat storage device, the continuity of the remaining amount, etc.

[0155] S705 randomly generates large-scale control solutions on a multi-day scale and uses heuristic algorithms for optimization: for the solution space expanded to the multi-day time dimension, heuristic algorithms (such as genetic algorithms, simulated annealing, ant colony optimization, etc.) are used to find the global optimal solution in the control parameter space through iteration and local search strategies based on large-scale random initial solutions.

[0156] S706 outputs the final plan: This selects an optimal or near-optimal scheduling strategy based on the objective function value, constraint satisfaction, and operational feasibility. This plan specifies the equipment operating status, energy storage charging and discharging paths, and load distribution plan for each daily period, and outputs deployable control instructions or strategy recommendations.

[0157] As an alternative, the aforementioned three-stage, multi-day optimization strategy can be integrated with a large model. This model can replace manual work in establishing load distribution, constructing objective functions, and building constraints. Furthermore, the model can autonomously control the algorithm, generate multiple alternative paths, and select the final solution after flexible judgment. The corresponding implementation process is as follows:

[0158] The S7011 large model independently builds a load forecasting model and establishes a distribution, introducing the distribution of time-of-use electricity prices: the large model calls external data and automatically calls the matching load forecasting method to build the hourly distribution of cold / hot loads within a single day.

[0159] S7021 introduces parameters of cold / heat storage device;

[0160] S7031 automatically generates multi-objective functions based on user needs: it identifies user intent, generates an adaptive objective function with adjustable weights through semantic analysis, and outputs it through a reasonable mathematical expression.

[0161] S7041 automatically generates and determines constraints based on user needs: Based on system equipment characteristics, operating rules and historical policy libraries, the model automatically identifies the equipment constraints, system boundary conditions and scheduling logic that need to be applied, and generates a structured mathematical constraint model.

[0162] S7051 Dynamic programming or similar algorithms generate single-day hourly control plans: On a multi-day scale, large-scale control plans are generated from large model control and heuristic algorithms are selected for optimization.

[0163] S7061 multi-path alternatives and flexible selection output final solution: Based on the original optimization, the large model independently outputs multiple alternative paths through equivalent path storage, Top-K pruning, etc., and adds flexible selection judgment to form the final solution.

[0164] Compared with the existing technology, the method provided by the present invention generally lacks standardized optimization processes and relies on manual settings and temporary scheduling. By constructing a structured and phased optimization process system, it realizes the automated management of the entire process from goal setting to result output.

[0165] Compared with traditional methods that mostly focus on static optimization at the device level, the method provided by the present invention realizes timing-system joint control by introducing a collaborative optimization mechanism of multiple time scales (such as hours, days, and weeks) and multiple control levels (equipment, system, energy, etc.).

[0166] Compared with traditional methods, the method provided by the present invention does not modularize the optimization tasks and lacks flexible combination capabilities. By introducing a configurable optimization module system, it supports on-demand combination of different optimization objectives and execution depths, significantly improving the adjustability and adaptability of the solution.

[0167] Compared with traditional methods that rely on rule engines and manual experience judgment, the method provided by the present invention introduces an intelligent agent system assisted by a large model and defines tasks in different process steps to achieve rapid cross-project adaptation and generalized execution.

[0168] The multi-stage HVAC system optimization and large-model fusion improvement method provided in this invention can be implemented as a corresponding software program, forming a corresponding HVAC system multi-stage optimization and large-model fusion improvement system. When running, this software program will execute the aforementioned HVAC system multi-stage optimization and large-model fusion improvement method process and store it in a corresponding storage medium for access and execution by a processor.

[0169] The following further illustrates the HVAC system multi-stage optimization and large-model fusion improvement method provided by the present invention through specific application examples.

[0170] In this example, the multi-stage optimization of the HVAC system and the large-scale model fusion improvement method provided by the present invention are used to implement the multi-stage optimization of the HVAC system as follows:

[0171] First, the system architecture and topology relationship are determined in step S1, and the equipment operation data is collected in step S2. There is no significant difference in the implementation of the operations of step S1 and step S2, and they can be executed according to the description of the technical solution.

[0172] Step S3, this example takes the establishment of cooling tower equipment performance model as an example, Figure 6 As shown, the performance model required for the cooling tower is first determined to include the flow-head (GH) and flow-efficiency (Gh) curves. Data corresponding to flow, head, and efficiency are collected from business trip data or equipment operation data, and a data table is created. Algorithms are selected for fitting, and the fitting effect is judged by indicators. If the effect is good, a mathematical expression or model weight coefficient is output and solidified into the final performance model. The large model can assist in data screening to ensure that the data range covers the actual usage range. Alternatively, the algorithm can be independently selected and multiple rounds of trials can be performed until the ideal fitting effect is achieved.

[0173] Step S4, as Figure 7 As shown, in this example, the scope of the optimization solution is first determined, and the external demand parameters to be input into the optimization process should be output. The large model can assist in identifying user intentions and thus independently determine the demand parameters. e and outdoor wet-bulb temperature t w For example, decide or let the large model output according to the historical data range and upper and lower limits to obtain the coverage range of the two parameters [G1, G2], [t1, t2] and discrete granularity m and k (that is, divided into m and k segments respectively), and obtain the corresponding optimization demand matrix through Cartesian product operation.

[0174] Step S5: In this example, the optimization demand matrix is ​​first traversed to extract a set of total cooling capacity G e and outdoor wet-bulb temperature t w Parameters are input into the equipment performance model. Taking the basic combination of chiller, refrigeration pump, cooling pump and cooling tower as an example, according to Figure 8The device analysis and calculation process presented first calculates the control scheme and energy consumption (power) for the chiller, then for the refrigeration pump and cooling pump, and finally for the cooling tower. Because multiple schemes may meet local optimality under the same demand parameters, i.e., the optimal energy consumption for a single device, multiple schemes may be output. The quantitative indicators involved in the scheme are abstracted into an array and used as the initial scheme. With energy consumption as the objective function, this is input into a heuristic algorithm for optimization, resulting in a series of optimized schemes. The initial scheme is then compared with the optimized scheme to find the optimal one. This optimal scheme is then added to the corresponding position in the optimization result matrix consistent with the structure of the optimization demand matrix. Upon completion, the next cycle is repeated to calculate the scheme for the next set of demand parameters. In this example, the large model can be reasonably expanded when inputting the initial scheme and incorporate flexible judgments and explanations when selecting the scheme, thereby enhancing the flexibility of the optimization scheme.

[0175] Step S6, as Figure 9 As shown, this example first requires inputting the load distribution, electricity price distribution, and optimization result matrix, where the load distribution and electricity price distribution are hourly time series distributions. Because the subsequent optimization in this step uses time series algorithms such as dynamic programming, the objective function needs to be matched. The optimization functions that can be used are as follows:

[0176] f(t)=min[f(t-1)+c t +ω1f trans (p t-1 ,p t )]

[0177] Where f(t) is the objective function at time t, f(t-1) is the value of the objective function at time t-1, and c t is the energy consumption of HVAC system equipment at time t, w1 is the adjustable weight coefficient, f trans (p t-1 ,p t ) is the penalty term for switching between schemes from t-1 to t, because in actual use, frequent switching schemes is not conducive to stable operation of the system. In practical applications, other factors such as stability can be added to the objective function as needed. In order to ensure normal operation and daily load usage specifications, constraints are added, including ensuring that the hourly cooling capacity needs to be greater than the load demand, and that changes in cooling demand below a certain threshold cannot implement equipment start-up and shutdown changes, etc., which can be increased or decreased according to demand. After completing the objective function and constraints, according to the hourly load and environmental conditions, the adaptation scheme is taken out from the optimization result matrix, the constraints are verified, the objective function is calculated, and the time period in the range of [1, T] is traversed (T is the effective time period that needs to be controlled), and the overall control scheme for a single day and hour can be obtained. In each step, the figure has shown the fusion operations that can be performed by the large model.

[0178] Step S7, as Figure 10 As shown, in this example, the input in this step adds cold / heat storage parameters compared to step S6, including storage capacity, efficiency, and inherent energy consumption. Subsequently, the objective function and constraints of multi-day scale and cold / heat storage are established in sequence. The objective function example is:

[0179]

[0180] Where T is the total number of time periods over multiple days, such as 2 days can be set to 48, f(t) is the objective function at time t, c t is the energy consumption of HVAC system equipment at time t, c q,t is the energy consumption of the cold / heat storage equipment at time t, w1 is the adjustable weight coefficient, f trans (p t-1 ,p t ) is the penalty term for switching between the schemes from t-1 to t, that is, the objective function considers the total energy consumption and switching penalty on a multi-day scale.

[0181] Examples of constraints are:

[0182] e t+1 =e t +q t,ch -q t,dis

[0183]

[0184] G t,e ≤q t +q t,ch -q t,dis

[0185] where e t+1 and e t are the reserves of cold / heat storage equipment at time t+1 and t, respectively, q t,ch is the deposit or recharge amount, q t,dis is the release amount; e max is the maximum reserve; G t,e is the cooling capacity required at time t, q t is the direct cooling capacity of other equipment at time t. After completing the objective function and constraints, a large-scale control solution is randomly generated within certain pre-set constraints, either hourly or within a custom time range. Each solution includes the control parameters for each time unit in the range. This solution is then fed into a heuristic algorithm (genetic algorithm or particle swarm algorithm) for iterative optimization, outputting the final solution. The diagram illustrates the fusion operations that can be performed on the large model at each step.

[0186] The big model used in this example does not refer to the big model itself, but rather to the intelligent agent instance built around the big model. Figure 11 As shown, the large model is bound to a tool function cluster that can call relevant data and code sandboxes. The former is used to assist with judgment and output in some steps, while the code sandbox allows the code written by the large model to run in an isolated environment for algorithm modification and indicator verification. The output layer of the large model adds structured output judgment. When structured output is required, constraints are used to achieve structured output in a fixed format.

[0187] Based on the above-mentioned HVAC system multi-stage optimization and large-model fusion improvement solution, an embodiment of the present invention also provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the steps of the above-mentioned HVAC system multi-stage optimization and large-model fusion improvement method are implemented.

[0188] An embodiment of the present invention further provides a processor, which is used to run a program, wherein when the program is running, the steps of the above-mentioned HVAC system multi-stage optimization and large model fusion improvement method are executed.

[0189] An embodiment of the present invention also provides a terminal device, which includes a processor, a memory, and a program stored in the memory and runnable on the processor. The program code is loaded and executed by the processor to implement the steps of the above-mentioned HVAC system multi-stage optimization and large model fusion improvement method.

[0190] The present invention also provides a computer program product, which, when executed on a data processing device, is suitable for executing the steps of the above-mentioned HVAC system multi-stage optimization and large model fusion improvement method.

[0191] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0192] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and modules described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0193] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0194] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products of the embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as the combination of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0195] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0196] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0197] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0198] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0199] Computer-readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission medium that can be used to store information that can be accessed by a computing device.

[0200] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0201] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0202] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-stage optimization and large-model fusion improvement method for HVAC systems, characterized by: The method comprises: S1: Determine the HVAC system architecture and topology relationship; S2: Collect operating data of various equipment in the HVAC system; S3: Based on the data collected in step S2, establish an equipment performance model for the equipment in the HVAC system; S4: Determine the key input parameters and their optional ranges involved in the optimization task, and construct a discretized multi-dimensional optimization requirement matrix based on them; S5: traverse the demand matrix constructed in step S4 to perform optimization and obtain the optimization result matrix; S6: Based on the device-level optimization in S5, daily load distribution and time-of-use electricity prices are introduced to perform optimization within the daily scale. By constructing multi-period objective functions and constraints, a two-stage daily optimization strategy is formed. S7: Based on the single-day optimization of S6, the consideration of cold / heat storage devices is introduced, and multi-day optimization is carried out. By constructing a comprehensive objective function and constraints considering cold / heat storage, and iteratively optimizing the heuristic algorithm of the large-scale solution, a three-stage multi-day optimization strategy is formed.

2. The HVAC system multi-stage optimization and large model fusion improvement method according to claim 1 is characterized in that: The step S3 establishes the equipment performance model through the following steps: S301 Equipment classification, clarifying model categories: Classify equipment based on type and function, and clarify the model type that needs to be established for each type of equipment; S302 Model Algorithm Selection: After the model type is determined, a specific modeling algorithm is selected based on the availability of equipment operation data and the complexity of model fitting; S303 Model Construction: For the selected model, perform model parameter fitting or training to complete the mathematical expression construction of the equipment performance model, ensuring that the input and output variables cover the actual operating range; S304 Model Accuracy and Effectiveness Evaluation Comparison: Evaluate the accuracy and applicability of various models by selecting model evaluation indicators; S305 Equipment Performance Model Output: Output the final selected performance model structure and parameters into a standardized model format that can be called by optimization algorithms, control strategy modules or analysis tools.

3. The HVAC system multi-stage optimization and large model fusion improvement method according to claim 1 is characterized in that: In step S4, the optimization demand matrix is ​​constructed through the following steps: S401: Identify the required input parameters, the number of which is N: Based on the scenarios and boundary conditions involved in system operation optimization, determine the input parameter dimensions that need to be included in modeling and optimization analysis; S402 Determine each parameter range: For each parameter dimension determined in step S401, define its reasonable value range and discrete granularity; S403 constructs an N-dimensional demand matrix: performs a multi-dimensional Cartesian product operation on the discrete value combination of each parameter in step S402 to construct an N-dimensional optimization demand matrix.

4. The HVAC system multi-stage optimization and large model fusion improvement method according to claim 1 is characterized in that: In step S5, the optimization result matrix is ​​generated by the following steps: S501: traverse and obtain the demand parameters corresponding to each unit in the demand matrix; S502: Optimize the calculation according to the device calculation order to obtain multiple control solutions; S503: Using the control solution obtained in step S502 as an initial reference, an optimization is performed using a heuristic algorithm; S504: Compare the algorithm output solution with the initial reference solution and select the better solution as the final solution; S505: After completing the optimization for each unit in the matrix, fill it with the final solution to form an optimization result matrix with the same matrix structure.

5. The HVAC system multi-stage optimization and large model fusion improvement method according to claim 1 is characterized in that: In step S6, the optimization within a single day is completed by the following steps: S601: Introducing or predicting daily load distribution and time-of-use electricity price distribution; S602: Constructing an objective function based on the actual optimization goal; S603: Setting key boundary conditions that affect the execution of the operation strategy; S604: Within a limited time period, based on the optimization result matrix, a single-day hourly control plan is generated using dynamic programming or a similar algorithm.

6. The HVAC system multi-stage optimization and large model fusion improvement method according to claim 1 is characterized in that: In step S7, the multi-day optimization is completed by the following steps: S701: Introducing or predicting daily load distribution and time-of-use electricity price distribution; S702: Introducing parameters of the cold / heat storage device; S703: Determine a multi-day comprehensive objective function; S704: Based on the key boundary conditions affecting the execution of the operation strategy set in step S603, adjustments are made to additionally consider the switching constraints of the cold / heat storage device and the continuity of the remaining amount; S705: Randomly generate large-scale control plans on a multi-day scale and optimize them using heuristic algorithms; S706: Select an optimal or approximately optimal scheduling strategy based on the objective function value, constraint satisfaction, and operational feasibility.

7. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps of the multi-stage optimization and large-model fusion improvement method of the HVAC system as described in any one of claims 1 to 6 are implemented.

8. A processor for running a program, characterized in that: When the program is running, the steps of the HVAC system multi-stage optimization and large model fusion improvement method described in any one of claims 1 to 6 are executed.

9. A terminal device comprising a processor, a memory, and a program stored in the memory and executable on the processor, characterized in that: The program code is loaded and executed by the processor to implement the steps of the HVAC system multi-stage optimization and large model fusion improvement method described in any one of claims 1-6.

10. A computer program product, characterized in that When executed on a data processing device, it is suitable for executing the steps of the HVAC system multi-stage optimization and large model fusion improvement method described in any one of claims 1 to 6.

Images