Cooling load prediction and energy-saving optimization system and method based on large model

Abstract

The invention relates to the technical field of intelligent control and efficient energy saving of heating ventilation air conditioners, and particularly discloses a cooling load prediction and energy-saving optimization system and method based on a large model. The system comprises a data acquisition and preprocessing module, a feature extraction and construction module, a large model prediction engine, an energy-saving optimization decision module and a visual display and monitoring module. According to the method, multi-source characteristics such as weather, time and historical load are fused, a multi-model dynamic weighted fusion mechanism is adopted to realize high-precision advanced prediction of the cooling load, and the starting time, load distribution and water supply temperature of a water chilling unit are optimized based on a prediction result, so that intelligent operation of energy-saving refrigeration air conditioning equipment is realized. The method solves the problems of lag and high energy consumption of a traditional control method, improves the self-adaptive capability and operation stability of the system, is suitable for various building scenes such as business and industry, and has remarkable energy-saving and consumption-reducing effects.

Description

Technical Field

[0001] This invention relates to the field of intelligent control and energy-saving technology for heating, ventilation and air conditioning (HVAC), specifically to a cooling load prediction and energy-saving optimization system and method based on a large model. It is particularly suitable for improving the operating efficiency of energy-saving refrigeration and air conditioning equipment and represents an innovative application of energy-saving technology in building environment control. Background Technology

[0002] As building energy consumption continues to increase as a proportion of total social energy consumption, HVAC systems, as a major source of building energy consumption, have made energy-saving optimization an important research direction in building energy conservation. Chillers, as the core refrigeration equipment in central air conditioning systems, directly affect the energy consumption of the entire system through their operating efficiency.

[0003] Traditional cooling load forecasting and control technologies have significant limitations: In terms of forecasting, they typically rely on empirical formulas or simple statistical models, which struggle to accurately depict the complex nonlinear relationships between multiple factors such as temperature, humidity, solar radiation, and building usage patterns, resulting in limited forecast accuracy, especially with significant errors during sudden weather changes or changes in usage patterns. In terms of control, feedback-based control strategies are commonly used, meaning adjustments are made only after load changes actually occur. This model exhibits significant response lag, making it difficult to achieve timely and accurate control during chiller startup or when loads fluctuate drastically, potentially leading to energy waste or impacting indoor comfort. Furthermore, traditional optimization strategies are often based on fixed rules or static parameters, lacking the ability to dynamically and globally optimize based on real-time and forecast information, making it difficult to adapt to the changing needs of different building types and diverse operating conditions.

[0004] In recent years, large-scale model technologies, represented by deep learning and ensemble learning, have demonstrated powerful pattern recognition and complex relationship learning capabilities, providing new technical pathways for solving the aforementioned problems. However, how to deeply integrate these advanced data-driven algorithms with the physical characteristics, operational constraints, and engineering practices of HVAC systems to construct an integrated intelligent system capable of high-precision prediction, real-time dynamic optimization, and stable reliability remains a pressing technical challenge in this field. Therefore, developing an intelligent prediction and optimization system and method that can significantly reduce the energy consumption of energy-efficient air conditioning equipment and improve the system's adaptive capabilities has significant engineering value and market prospects. Summary of the Invention

[0005] The purpose of this invention is to provide a cooling load prediction and energy-saving optimization system and method based on a large model to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides a cooling load prediction and energy-saving optimization system based on a large model, comprising:

[0008] The data acquisition and preprocessing module is used to collect meteorological data, building operation schedule information, historical cooling load data and chiller unit operating parameters in real time, and to clean, handle outliers, fill missing values ​​and standardize the collected raw data to generate preprocessed data.

[0009] The feature extraction and construction module is connected to the data acquisition and preprocessing module. It receives the preprocessed data and extracts time features, meteorological features, and lag features from it. The time features include at least hour, weekday, month, holiday identifier, weekday identifier, and business hours identifier. The meteorological features include at least temperature, humidity, perceived temperature, air enthalpy, solar radiation intensity, and wind speed. The lag features include at least historical cooling load values, rolling average, peak values, and trends. This module constructs the extracted features into a multidimensional feature vector.

[0010] A large model prediction engine is connected to the feature extraction and construction module and receives the multi-dimensional feature vector. The large model prediction engine integrates at least one deep learning model and an ensemble learning model and is configured with a model fusion unit. The model fusion unit dynamically allocates weights based on the historical prediction accuracy of each model, performs weighted fusion of the output prediction values ​​of each model, and generates the final cold load prediction value and prediction confidence interval.

[0011] The energy-saving optimization decision module is connected to the large model prediction engine. It receives the final cooling load prediction value and calculates the optimal start-up time of the chiller unit, the optimal load distribution scheme among multiple units, and the optimal supply temperature of chilled water based on the prediction value through optimization algorithms, forming a comprehensive operation strategy.

[0012] The visualization and monitoring module is connected to the large model prediction engine and the energy-saving optimization decision module. It is used to display the comparison between the cold load prediction curve and the actual load curve in real time, the detailed information of the comprehensive operation strategy, the energy-saving effect statistics and the equipment operation status, and to provide early warning information.

[0013] Preferably, the large model prediction engine integrates a neural network, a random forest, and an XGBoost gradient boosting decision tree, wherein the neural network is selected from a multilayer perceptron or a long short-term memory network.

[0014] Preferably, the formula for dynamic weighted fusion performed by the model fusion unit is:

[0015] Among them, Q final Q is the final predicted value.nn Q rf Q xgb These are the predicted values ​​of the neural network, random forest, and XGBoost models, respectively. The weights w1, w2, and w3 are dynamically calculated based on the prediction accuracy of each model on the recent validation set and satisfy w1+w2+w3=1.

[0016] Preferably, the perceived temperature T in the meteorological characteristics feel The air enthalpy h is calculated using a function that includes temperature T, relative humidity RH, and wind speed Vw; the air enthalpy h is calculated using a function that includes temperature T and humidity W.

[0017] Preferably, the startup time optimization unit in the energy-saving optimization decision module minimizes the time from startup time T. start To the predicted peak load time T peak The energy consumption integral between the two points is used as the objective function to satisfy the condition at demand time T. need Given the constraint that the cooling capacity is not lower than the predicted cooling load, the optimal start-up time is determined.

[0018] Preferably, the load allocation optimization unit in the energy-saving optimization decision module establishes an integer programming model with the goal of minimizing the total power consumption of all operating chiller units. The constraints include the load rate operating range of each unit, the total cooling capacity provided by all units not being less than the predicted total demand, and the start-up and shutdown logic restrictions of the units.

[0019] Preferably, the water supply temperature optimization unit in the energy-saving optimization decision module aims to maximize the coefficient of performance (COP) of the chiller unit and solves for the optimal water supply temperature under the constraint of meeting the cooling capacity requirements of the system.

[0020] Secondly, the present invention also provides a method for predicting cooling load and optimizing energy saving based on a large model, applied to the above-mentioned system, the method comprising the following steps:

[0021] The data acquisition and preprocessing module collects real-time meteorological data, building operation schedules, historical cooling load data, and equipment operating parameters, and performs data cleaning and standardization.

[0022] The feature extraction and construction module is used to construct a multidimensional feature vector containing time features, meteorological features, and lag features from the preprocessed data.

[0023] The multidimensional feature vector is input into the large model prediction engine, which performs parallel prediction through multiple integrated models, and the model fusion unit dynamically weights and fuses the results of each model to output the predicted value and confidence interval of the cold load for a specified future period.

[0024] The predicted cooling load is input into the energy-saving optimization decision module, which performs startup time optimization, load allocation optimization, and water supply temperature optimization respectively, generating a comprehensive operation strategy containing specific parameters.

[0025] The visualization and monitoring module displays the predicted values ​​and the comprehensive operation strategy, and sends the strategy to the chiller unit control system for execution. At the same time, it continuously collects actual operation data for model iteration and updates.

[0026] Preferably, after generating the comprehensive operation strategy, a continuous optimization step is also included: periodically retraining each model in the large model prediction engine using newly collected actual operation data, and updating the dynamic weights in the model fusion unit based on the performance of each model on the new validation set after retraining.

[0027] Preferably, the perceived temperature in the meteorological characteristics is calculated using the following formula:

[0028] in, RH is the dry bulb temperature, Vw is the relative humidity, and RH is the wind speed.

[0029] Compared with the prior art, the beneficial effects of the present invention are:

[0030] This invention integrates multi-layer neural networks, random forests, and XGBoost large models, combined with a dynamically weighted multi-model fusion strategy. This allows the system to deeply mine the complex nonlinear relationships between load and various factors such as weather, time, and building usage patterns in historical data, thereby achieving high-precision and proactive prediction of cooling load. Based on these predictions, the system can proactively and dynamically optimize key operating parameters such as chiller start-up time, load distribution, and water supply temperature, transforming traditional passive response into proactive preventative control. This avoids energy waste at the source and significantly improves the operating efficiency of energy-saving air conditioning systems.

[0031] The high-precision load prediction of this invention enables the system to plan the equipment operating status in advance, smoothing out frequent start-ups and shutdowns and large-scale loading / unloading operations caused by sudden load changes; the optimized operating strategy ensures that the chiller unit operates stably within the high-efficiency range, reduces mechanical and thermal stress on the equipment, effectively reduces the risk of failure, and thus significantly improves the operational reliability and service life of the entire refrigeration and air conditioning equipment.

[0032] The feature engineering and model architecture of this invention are modular and scalable, and can quickly adapt to different types of buildings and their specific operating modes by incorporating new features and data. At the same time, the model supports continuous learning and updating, and can track changes in building usage habits and equipment performance degradation, maintaining optimized efficiency in the long term. Ultimately, while ensuring indoor environmental comfort or process requirements, the system achieves a significant reduction in building air conditioning system energy consumption through global optimization, which has important energy-saving, environmental protection and economic value. Attached Figure Description

[0033] Figure 1 This is a schematic block diagram of the cooling load prediction and energy-saving optimization system based on a large model according to the present invention;

[0034] Figure 2 This is a flowchart of the cooling load prediction and energy-saving optimization method based on a large model according to the present invention.

[0035] Figure 3 This is a schematic diagram of the human-computer interaction interface of the visualization and monitoring module of the present invention;

[0036] Figure 4 This is a schematic diagram of the energy consumption statistics page in the human-computer interaction interface of the present invention. Detailed Implementation

[0037] To further explain the technical solution of the present invention, a detailed description is provided below through specific embodiments.

[0038] Please see Figure 1 , Figure 3 and Figure 4 This invention provides a cooling load prediction and energy-saving optimization system based on a large model. This system is an integrated intelligent control system combining hardware and software, including the following modules:

[0039] Data Acquisition and Preprocessing Module: As the system's data input, this module is responsible for acquiring raw data from various data sources in real-time or near real-time. Data sources include: meteorological data APIs connected to the internet (for acquiring temperature, humidity, wind speed, solar radiation intensity, etc.), temperature and humidity sensors deployed within the building, electricity / energy meters, chiller controllers, and building management systems (BMS). This module has a built-in data cleaning engine that imputes missing values, detects and removes outliers based on statistical methods (such as the 3σ principle), and finally performs standardization processing (such as Z-score standardization) to transform data of different dimensions into a distribution with a mean of 0 and a standard deviation of 1, providing a high-quality, standardized data foundation for subsequent model processing.

[0040] Feature extraction and construction module: Connected to the data acquisition and preprocessing module, its task is to transform the cleaned data into machine-understandable, information-rich high-dimensional feature vectors. This module automatically constructs three main categories of features:

[0041] Time characteristics include basic timestamps (hours, days of the week, months), business logic identifiers (whether it is a weekday, holiday, or business hours), and periodic encoding introduced to better capture periodic patterns (encoding 24 hours a day and 7 days a week using sine and cosine functions).

[0042] Meteorological characteristics: This includes not only raw meteorological observations but also derived characteristics closely related to human comfort and air conditioning load, such as perceived temperature (considering temperature, humidity, and wind speed) and air enthalpy (reflecting the total heat of the air, a key parameter for air conditioning load calculation). Specifically, perceived temperature T... feel The air enthalpy h is calculated using a function that includes temperature T, relative humidity RH, and wind speed Vw; the air enthalpy h is calculated using a function that includes temperature T and humidity W.

[0043] Lag and statistical characteristics: Extracted from historical cold load sequences, such as the load of the previous hour, the load of the previous 24 hours (capturing the daily cycle pattern), the rolling average, peak value, trend and standard deviation of the past 24 hours, etc., to characterize the load inertia, peak characteristics and volatility.

[0044] Finally, the module concatenates the above features into a fixed-dimensional feature vector, which is then used as the model input.

[0045] Large-scale prediction engine: This framework constructs a multi-model fusion prediction framework. At least three types of prediction models run in parallel within the framework.

[0046] A deep learning model, such as a multilayer perceptron (MLP) or a long short-term memory network (LSTM) that is good at processing temporal data.

[0047] An ensemble learning model, such as Random Forest (RF) or XGBoost based on decision trees.

[0048] Each model receives the same feature vector, makes predictions independently, and outputs its own predicted cold load value. The model fusion unit of the large model prediction engine uses a dynamic weighted average strategy to fuse the results: each model is assigned a weight, which is dynamically calculated and updated based on the model's prediction accuracy on recent historical data. The higher the accuracy, the larger the weight. The final predicted value is the weighted sum of the predictions from all models, and the prediction confidence interval is also output. This fusion strategy effectively integrates the advantages of different models, improving the robustness and accuracy of the prediction.

[0049] Energy-saving optimization decision module: Connected to the forecasting engine, this module translates the predicted future cooling load curve into specific, executable equipment control commands. This module comprises three sub-units:

[0050] Start-up time optimization unit: For the start-up process of chiller units in the morning or after intermittent operation, with the goal of minimizing energy consumption during the start-up phase, the optimal start-up time of the chiller unit is calculated under the premise of meeting the cooling demand, so as to avoid the waste caused by starting too early or the impact on comfort by starting too late.

[0051] Load Allocation Optimization Unit: For multiple chiller units operating in parallel, an optimization model is established with the goal of minimizing the total system operating power. Under constraints such as meeting the total cooling capacity demand, the allowable range of load rate for each unit, and the limit on the number of start-ups and shutdowns of the units, the optimal load rate or the combination of operating units for each unit is calculated.

[0052] Water supply temperature optimization unit: Under the premise of ensuring the cooling capacity of the system, find the chilled water supply temperature that maximizes the coefficient of performance (COP) of the chiller unit.

[0053] This module ultimately outputs a comprehensive set of optimized operating strategies, including specific startup times, load commands for each unit, and water supply temperature setpoints.

[0054] Visualization and monitoring module: serving as the human-computer interaction interface ( Figure 3 It provides a panoramic view for operation and management personnel. It displays real-time comparisons of predicted and actual cooling load values, details of optimization strategies, and post-implementation energy-saving statistics in the form of dashboards, graphs, and reports. Figure 4 This module also tracks the operating status of key equipment. Furthermore, it has an early warning function, proactively sending alarm messages and operational suggestions when prediction deviations are too large, energy-saving effects fail to meet expectations, or equipment malfunctions.

[0055] Please see Figure 1 and Figure 2 Corresponding to the above system, the present invention also provides a method for cooling load prediction and energy-saving optimization based on a large model, comprising the following steps:

[0056] Step S100, Data Acquisition and Preprocessing: Real-time acquisition of multi-source data such as meteorological data, building schedules, historical load data, and equipment parameters; cleaning, filling, and standardization to form a time-series database.

[0057] Step S200, Feature Engineering: Automatically extract and construct time features, meteorological features (including derived features such as perceived temperature and enthalpy) and lag statistical features from the preprocessed data, and combine them into the model input feature vector.

[0058] Step S300, Multi-model fusion prediction: Input the feature vectors into multiple pre-trained large models (such as MLP / LSTM, RF, XGBoost) simultaneously; each model infers independently to obtain preliminary prediction results; dynamically calculate and update the fusion weights based on the recent online prediction performance of each model; perform a weighted average on the preliminary prediction results to obtain the final cold load prediction value and uncertainty assessment.

[0059] Step S400: Energy-saving optimization decision: Based on the future load forecast curve obtained in step S300, execute sequentially or in parallel:

[0060] S410: Startup time optimization, aiming to find the optimal startup time with the goal of minimizing energy consumption during startup;

[0061] S420: Load allocation optimization, which aims to find the optimal load rate of each unit with the goal of minimizing the total power consumption of the system;

[0062] S430: Water supply temperature optimization, aiming to find the optimal water supply temperature with the goal of maximizing the overall energy efficiency of the system;

[0063] By integrating the results from S410 to S430, a comprehensive optimization strategy containing specific control parameters is generated.

[0064] Step S500: Strategy Execution and Closed-Loop Optimization. The optimization strategy is distributed to the chiller group control system for execution; real-time monitoring of actual system operating data is performed, and prediction errors and actual energy-saving effects are calculated; new operating data is fed back to the model training process, and the prediction model is periodically subjected to incremental learning or retraining to achieve continuous model evolution and continuous improvement of system performance.

[0065] Example 1: Application in Commercial Complexes

[0066] This embodiment uses a large commercial complex as an application scenario. The building has a large floor area and is equipped with multiple centrifugal chiller units, resulting in significant energy consumption of the air conditioning system.

[0067] System Deployment:

[0068] 1. Hardware Deployment: The core server is deployed in the computer room and connects to existing chiller controllers, smart meters, building management systems (BMS), and sensors via standard industrial communication protocols. Simultaneously, a high-precision meteorological data source is integrated.

[0069] 2. Software Deployment: Install and run the software suite of this system on the server, configure the data interface and algorithm model, and provide a visual monitoring interface via the network.

[0070] 3. Model initialization: Collect historical operational data of the building, and complete feature engineering, model training and initialization.

[0071] Operation process:

[0072] The system initiates a forecasting and optimization process daily at a set time. In the early morning, the system automatically acquires future weather forecasts and building schedules, constructing feature vectors. The forecasting engine then calls upon the model for inference and fusion, generating a full-day cooling load forecast curve.

[0073] The optimization decision-making module calculates optimization strategies based on the prediction curve, considering startup time, unit load combination, and supply water temperature. Once confirmed, these strategies are automatically or manually sent to the chiller group control system for execution. The system monitors operating status and prediction deviations around the clock.

[0074] After long-term operation and observation, the system has demonstrated the following effects:

[0075] Prediction accuracy: The high accuracy of cooling load prediction provides a reliable basis for optimized control.

[0076] Energy saving effect: By optimizing the startup strategy, the ineffective running time of the unit is effectively reduced; by optimizing load distribution and operating parameters, the overall operating energy efficiency of the chiller unit is improved. The system achieves significant energy saving benefits and greatly reduces the operating cost of the air conditioning system.

[0077] Stability: The system's optimized operation strategy makes the unit run more smoothly, significantly reducing the frequency of equipment start-ups and shutdowns and the amplitude of load fluctuations, which helps to extend the service life of the equipment.

[0078] Example 2: Industrial Building Applications

[0079] This embodiment uses a precision manufacturing workshop as an application scenario. This scenario has extremely high requirements for the precision of environmental temperature and humidity control, and the internal process heat dissipation load varies greatly.

[0080] System customization adjustments:

[0081] 1. Feature Enhancement: In addition to the standard features, features related to internal heat sources, such as production scheduling information and the operating status of key process equipment, are introduced.

[0082] 2. Zonal Prediction: Zonal load modeling and prediction are performed for different temperature control zones within the workshop to adapt to the differentiated process environment requirements.

[0083] 3. Optimization target adjustment: In the optimization decision-making, the accuracy of environmental control is set as a priority constraint.

[0084] After deploying this system, the following results were achieved:

[0085] Prediction accuracy: The system can still maintain a high level of accuracy in predicting cooling loads even when faced with complex internal and external disturbances.

[0086] Quality control: The overall temperature and humidity control accuracy in the workshop has been effectively guaranteed and improved, meeting the high standards of process production requirements.

[0087] Energy saving and overall benefits: While ensuring precise environmental control, the total energy consumption of the air conditioning system was significantly reduced. At the same time, more stable and precise environmental control also made a positive contribution to improving production process stability and product yield.

[0088] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A cooling load prediction and energy-saving optimization system based on a large model, characterized in that, include: The data acquisition and preprocessing module is used to collect and preprocess meteorological data, building operation schedules, historical cooling loads, and equipment operating parameters. The feature extraction and construction module is connected to the data acquisition and preprocessing module, and is used to extract time features, meteorological features and lag features from the preprocessed data and construct them into a multi-dimensional feature vector. The large model prediction engine, connected to the feature extraction and construction module, integrates multiple prediction models and is configured with a model fusion unit, which is used to perform fusion prediction based on the multi-dimensional feature vector and output cold load prediction results. An energy-saving optimization decision module, connected to the large model prediction engine, includes a startup time optimization unit, a load distribution optimization unit, and a water supply temperature optimization unit, used to generate an optimized operation strategy for the chiller unit based on the prediction results. The visualization and monitoring module is connected to the large model prediction engine and the energy-saving optimization decision module, and is used to display prediction results, optimization strategies and system operating status.

2. The cooling load prediction and energy-saving optimization system based on a large model according to claim 1, characterized in that, The large model prediction engine integrates neural networks, random forests, and XGBoost gradient boosting decision trees. The neural network is selected from multilayer perceptrons or long short-term memory networks.

3. The cooling load prediction and energy-saving optimization system based on a large model according to claim 2, characterized in that, The formula for dynamic weighted fusion performed by the model fusion unit is as follows: Among them, Q final Q is the final predicted value. nn Q rf Q xgb These are the predicted values ​​of the neural network, random forest, and XGBoost models, respectively. The weights w1, w2, and w3 are dynamically calculated based on the prediction accuracy of each model on the recent validation set and satisfy w1+w2+w3=1.

4. The cooling load prediction and energy-saving optimization system based on a large model according to claim 1, characterized in that, The perceived temperature T in the meteorological characteristics feel The air enthalpy h is calculated using a function that includes temperature T, relative humidity RH, and wind speed Vw; the air enthalpy h is calculated using a function that includes temperature T and humidity W.

5. The cooling load prediction and energy-saving optimization system based on a large model according to claim 1, characterized in that, The startup time optimization unit in the energy-saving optimization decision module aims to minimize the time from startup time T. start To the predicted peak load time T peak The energy consumption integral between the two points is used as the objective function to satisfy the condition at demand time T. need Given the constraint that the cooling capacity is not lower than the predicted cooling load, the optimal start-up time is determined.

6. The cooling load prediction and energy-saving optimization system based on a large model according to claim 1, characterized in that, The load allocation optimization unit in the energy-saving optimization decision module establishes an integer programming model with the goal of minimizing the total power consumption of all operating chiller units. The constraints include the load rate operating range of each unit, the total cooling capacity provided by all units not being less than the predicted total demand, and the start-up and shutdown logic restrictions of the units.

7. The cooling load prediction and energy-saving optimization system based on a large model according to claim 1, characterized in that, The water supply temperature optimization unit in the energy-saving optimization decision module aims to maximize the coefficient of performance (COP) of the chiller unit and solves for the optimal water supply temperature under the constraint of meeting the cooling capacity requirements of the system.

8. A method for predicting cooling load and optimizing energy saving based on a large model, applied to the system according to any one of claims 1-7, characterized in that, The method includes the following steps: The data acquisition and preprocessing module collects real-time meteorological data, building operation schedules, historical cooling load data, and equipment operating parameters, and performs data cleaning and standardization. The feature extraction and construction module is used to construct a multidimensional feature vector containing time features, meteorological features, and lag features from the preprocessed data. The multidimensional feature vector is input into the large model prediction engine, which performs parallel prediction through multiple integrated models, and the model fusion unit dynamically weights and fuses the results of each model to output the predicted value and confidence interval of the cold load for a specified future period. The predicted cooling load is input into the energy-saving optimization decision module, which performs startup time optimization, load allocation optimization, and water supply temperature optimization respectively, generating a comprehensive operation strategy containing specific parameters. The visualization and monitoring module displays the predicted values ​​and the comprehensive operation strategy, and sends the strategy to the chiller unit control system for execution. At the same time, it continuously collects actual operation data for model iteration and updates.

9. The cooling load prediction and energy-saving optimization method based on a large model according to claim 8, characterized in that, After generating the comprehensive operation strategy, a continuous optimization step is also included: periodically retraining each model in the large model prediction engine using newly collected actual operation data, and updating the dynamic weights in the model fusion unit based on the performance of each model on the new validation set after retraining.

10. The cooling load prediction and energy-saving optimization method based on a large model according to claim 8, characterized in that, The perceived temperature in the meteorological characteristics is calculated using the following formula: in, RH is the dry bulb temperature, Vw is the relative humidity, and RH is the wind speed.

Images