Air conditioner load prediction and feed-forward energy-saving control method and device integrating airport building mechanism and multi-mode large model

Abstract

The invention provides an air conditioner load prediction and feed-forward energy-saving control method and device integrating an airport building mechanism and a multi-mode large model, and is applied to the technical field of data processing. According to the airport air conditioning system energy saving and accurate regulation and control, demand side prediction and supply side regulation and control serve as a core framework. The method comprises the following steps: firstly, acquiring multi-source original data of airport building characteristics, environment, personnel flow and air conditioner working conditions, and generating standardized cooling capacity demand characteristic data through integration and time sequence rule mining; a dual-prediction model is constructed, fusion analysis is carried out, and the precise cooling capacity demand of each time-space dimension is output; based on the established energy consumption standard model, determining the equipment cooperation constraint, and further optimizing and calculating the operation scheme of the equipment such as the water chilling unit; and finally, air conditioning system early warning information giving consideration to comfort and energy conservation is generated in combination with the double-layer architecture, and intelligent management is achieved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and device for predicting air conditioning load and feedforward energy-saving control that integrates airport building mechanisms and multimodal large models. Background Technology

[0002] During airport operations, the air conditioning system, as a core infrastructure, plays a crucial role in maintaining suitable temperature and humidity within the terminal building and ensuring the comfort of passengers and staff. However, airport buildings have unique characteristics that significantly distinguish them from ordinary buildings: on the one hand, terminal buildings are mostly tall, open-plan structures with complex heat transfer characteristics in their envelope (such as large glass curtain walls), resulting in multiple paths for cooling loss; on the other hand, airport passenger and vehicle traffic fluctuates significantly due to flight schedules, with substantial differences in passenger density between peak and off-peak hours, leading to dramatic and irregular fluctuations in air conditioning cooling demand across time and space. Furthermore, the air conditioning system comprises multiple equipment units, including chillers, fan coil units, and cooling towers, with strong coupling between these units, further increasing the complexity of system control.

[0003] The existing airport air conditioning system control methods generally have the following problems:

[0004] Insufficient accuracy in load forecasting: Traditional forecasting methods often rely on a single data dimension or a general time series model, failing to fully integrate airport building mechanisms (such as spatial structure and thermal resistance of the building envelope) with airport-specific operational characteristics (such as peak flight times and passenger trajectories). This makes it difficult to capture the spatiotemporal fluctuations and lag characteristics of cooling demand, resulting in significant discrepancies between forecast results and actual demand, and failing to provide a reliable basis for precise regulation.

[0005] Lack of coordination in equipment control: Existing control modes mostly adopt "single-point control" or "fixed parameter operation", without establishing a collaborative optimization mechanism for equipment groups, ignoring the coupling relationship between various equipment and the energy efficiency threshold constraints, often resulting in "supply exceeding demand" or "supply shortage", which not only wastes energy but may also affect indoor comfort.

[0006] Imbalance between energy saving and comfort: There is a lack of a systematic regulatory framework that takes into account both energy saving goals and comfort requirements. Traditional strategies often focus on optimizing a single goal (such as simply pursuing energy reduction or excessively ensuring comfort) without establishing a dynamic balance mechanism. This makes it difficult to meet the needs of both simultaneously in complex scenarios such as peak flight seasons and seasonal changes.

[0007] The early warning response mechanism is lagging behind: existing early warnings are mostly based on passive triggering by equipment failure or parameter exceeding limits, without combining demand-side forecast data and supply-side regulation effects for advance prediction. They lack the ability to proactively identify and warn of deviations in the coordination between "prediction and regulation", and cannot avoid the risks of excessive energy consumption or decreased comfort in a timely manner.

[0008] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure and does not constitute information on prior art known to those skilled in the art. Summary of the Invention

[0009] According to one aspect of this application, a method for predicting air conditioning load and feedforward energy-saving control by integrating airport building mechanisms and a multimodal large-scale model is provided, comprising: acquiring multi-source raw airport data including airport building characteristics, environmental parameters, passenger flow, and air conditioning system operating conditions; processing the multi-source raw airport data by integrating multi-dimensional historical data of outdoor meteorology, indoor environment, passenger trajectories, and equipment operation, and using recurrent neural network technology to mine time-series data patterns to generate standardized cooling demand-related feature data; processing the standardized cooling demand-related feature data to construct a multi-regional room temperature neural network large-lag prediction model and a central air conditioning system dynamic energy consumption trend prediction model; and generating various data through multimodal large-scale model fusion analysis. The system generates precise cooling demand forecasts across time and space. These forecasts are then processed to construct a standard air conditioning energy consumption model encompassing peak and off-peak periods, regional load levels, and equipment energy efficiency thresholds. Based on a strategy of prioritizing evaluation criteria followed by dynamic scheduling, collaborative optimization control constraints for the equipment group are generated. These constraints are then processed, and combined with cooling demand forecast data, collaborative optimization calculations are performed on the air conditioning equipment group, including chillers, fan coil units, and cooling towers, to generate target equipment operation plans. Finally, the target equipment operation plans and related data are processed, and a two-tiered architecture combining precise demand-side forecasting and intelligent supply-side control is used to generate early warning information for airport air conditioning systems that balances comfort and energy-saving goals.

[0010] Another aspect of this application discloses an air conditioning load prediction and feedforward energy-saving control device that integrates airport building mechanisms and a multimodal large-scale model. The device includes: an acquisition module for acquiring multi-source raw airport data containing airport building characteristics, environmental parameters, passenger flow, and air conditioning system operating conditions; a processing module for processing the multi-source raw airport data, integrating multi-dimensional historical data on outdoor weather, indoor environment, passenger trajectories, and equipment operation, and using recurrent neural network technology to mine time-series data patterns to generate standardized cooling demand-related feature data; further processing the standardized cooling demand-related feature data to construct a multi-regional room temperature neural network large-lag prediction model and a central air conditioning system dynamic energy consumption trend prediction model, which are then fused using a multimodal large-scale model. The system analyzes and generates accurate cooling demand forecasts across various spatiotemporal dimensions. These forecasts are then processed to construct a standard air conditioning energy consumption model encompassing peak and off-peak periods, regional load levels, and equipment energy efficiency thresholds. Based on a strategy of prioritizing evaluation criteria before dynamic scheduling, collaborative optimization control constraints for the equipment group are generated. These constraints are further processed, and combined with cooling demand forecast data, collaborative optimization calculations are performed on the air conditioning equipment group, including chillers, fan coil units, and cooling towers, to generate target equipment operation plans. Finally, the target equipment operation plans and related data are processed, and a two-tiered architecture combining accurate demand-side forecasting and intelligent supply-side control is used to generate early warning information for the airport air conditioning system that balances comfort and energy-saving goals.

[0011] According to another aspect of this application, an electronic device includes: a first processor; and a memory for storing executable instructions of the first processor; wherein the first processor is configured to execute the above-described method for air conditioning load prediction and feedforward energy-saving control that integrates airport building mechanisms and multimodal large models by executing the executable instructions.

[0012] According to another aspect of this application, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a second processor, implements the above-described method for predicting air conditioning loads and feeding-forward energy-saving control by integrating airport building mechanisms and multimodal large models.

[0013] This application provides a method and device for air conditioning load forecasting and feedforward energy-saving control that integrates airport building mechanisms and a multimodal large-scale model. Focusing on energy saving and precise regulation of airport air conditioning systems, it constructs a core architecture of "demand-side forecasting + supply-side regulation" and proposes an air conditioning load forecasting and feedforward energy-saving control scheme that integrates airport building mechanisms and a multimodal large-scale model. First, it acquires multi-source raw data on airport building characteristics, environmental parameters, personnel flow, and air conditioning operating conditions. Through spatiotemporal alignment, quality grading, and recurrent neural network technology, standardized cooling demand characteristic data is generated. Then, a multi-regional room temperature large-lag prediction model and a central air conditioning dynamic energy consumption trend prediction model are constructed, and precise cooling demand in each spatiotemporal dimension is output through multimodal fusion. Based on this, an energy consumption standard model including peak-valley division, load grading, and energy efficiency thresholds is established to generate equipment collaborative constraints. Subsequently, the equipment group, including chiller units, is collaboratively optimized to formulate a target operation plan. Finally, combined with a two-layer architecture, early warning information that balances comfort and energy saving is generated. By integrating airport building mechanisms with multimodal models, the shortcomings of inaccurate load forecasting and lack of equipment coordination in traditional control are addressed, enabling intelligent management of airport air conditioning systems and significantly improving control accuracy and energy efficiency.

[0014] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0015] Figure 1 The flowchart illustrates a method for predicting air conditioning load and feeding forward energy-saving control that integrates airport building mechanisms and a multimodal large model, according to an embodiment of this application.

[0016] Figure 2 The diagram shows a structural schematic of an air conditioning load prediction and feedforward energy-saving control device that integrates airport building mechanisms and multimodal large models, according to an embodiment of this application. Detailed Implementation

[0017] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0018] In one implementation, Figure 1 The schematic diagram illustrates a flow chart of an air conditioning load prediction and feedforward energy-saving control method that integrates airport building mechanisms and multimodal large models according to an embodiment of this application.

[0019] S101, acquire multi-source raw airport data including airport building characteristics, environmental parameters, personnel flow, and air conditioning system operating conditions.

[0020] In one implementation, the scope of data collection is clearly defined, covering core architectural parameters of the terminal building: including the total building area, the physical boundaries and area ratios of each functional area (departure hall, waiting area, baggage sorting area, etc.), the floor height, span, and enclosure structure type (glass curtain wall, concrete wall, etc.) of high-ceilinged spaces, and key indicators affecting heat transfer such as building thermal insulation coefficient and door and window sealing performance parameters. Example: By retrieving airport construction project archives, it was found that the waiting hall in Terminal A has a floor height of 18 meters, a span of 45 meters, and an enclosure structure of Low-E double-glazed curtain wall with a thermal insulation coefficient of 0.35 W / (㎡・K); through on-site measurements, data on the sealing performance of the baggage sorting area walls was supplemented, determining the air permeability to be 0.5 m³ / (㎡・h).

[0021] Establish a dual-dimensional environmental data collection system encompassing both outdoor and indoor environments: outdoor meteorological parameters focus on core indicators such as temperature, humidity, solar radiation intensity, wind speed, and precipitation; indoor environmental parameters cover comfort-related data such as real-time room temperature, relative humidity, CO2 concentration, and air velocity in various areas. Example: Outdoor data is collected through the airport meteorological observation station, recording hourly data from 8:00 AM to 8:00 PM daily, including temperature (e.g., summer peak 35℃, winter minimum -2℃), humidity (summer average 65%), and solar radiation intensity (midday peak 800W / ㎡); temperature and humidity sensors are deployed in waiting areas and corridors, collecting data every 15 minutes, resulting in a real-time room temperature of 25℃ and a CO2 concentration of 420ppm in the No. 1 boarding gate area of ​​the waiting area.

[0022] Integrating dynamic passenger data from multiple channels, the core data includes passenger flow, dwell time in different areas, passenger density distribution, and peak-hour characteristics: The total number of passengers arriving and departing is obtained through the airport ticketing system and security checkpoint counters; the dwell time of passengers in each area is analyzed using video surveillance image recognition technology; and peak passenger flow periods are correlated with flight schedules. Example: The ticketing system extracts a total of 12,000 passengers arriving and departing on a certain day, with 3,500 arriving during the morning peak (7:00-9:00); the hourly throughput of security checkpoint A in Terminal 2 is obtained through security checkpoint counters (e.g., 850 passengers between 9:00-10:00); and video image analysis determines that the average dwell time of passengers in the central area of ​​the departure hall is 15 minutes, with a peak passenger density of 0.8 people / ㎡ during peak hours (14:00-16:00).

[0023] The system collects operational parameters across the entire air conditioning system equipment chain, covering core equipment such as chillers, fan coil units, cooling towers, and water pumps. This includes key operating parameters such as equipment start / stop status, operating load rate, output power, inlet / outlet water temperature / airflow, and speed. It also records auxiliary data such as equipment operating time and maintenance records. Example: Through the air conditioning system SCADA monitoring platform, the system obtains data showing that chiller unit 1 has an operating load rate of 75%, an output power of 1200kW, a chilled water supply temperature of 7℃, and a return water temperature of 12℃. It also records the fan coil unit area 3's speed at 900 r / min and airflow of 500 m³ / h. Finally, it supplements the cleaning and maintenance records of cooling tower 2 with equipment operation and maintenance logs, confirming its current heat exchange efficiency at 92%.

[0024] S102 processes multi-source raw data from the airport, integrating multi-dimensional historical data on outdoor weather, indoor environment, passenger trajectories, and equipment operation, and using recurrent neural network technology to mine patterns in time-series data to generate standardized cooling demand-related feature data.

[0025] In one implementation, a "spatiotemporal dimension alignment + data quality grading" integration mechanism is established to address the heterogeneous characteristics of outdoor meteorological (temperature, humidity, etc.), indoor environmental (room temperature, CO2 concentration, etc.), passenger trajectory (crowd density, dwell time, etc.), and equipment operation (load rate, speed, etc.) data in terms of structure, dimensions, and collection frequency. Spatiotemporal dimension alignment uses the physical partitions of the terminal building (e.g., areas A / B / C of Terminal 2) as the spatial benchmark and 15 minutes as the unified time granularity to synchronize data with different collection frequencies (e.g., one meteorological data per hour, one equipment operation data per minute) to a unified time axis, achieving precise spatiotemporal matching of multi-source data. Data quality grading is based on data completeness, accuracy, and timeliness, classifying data into four levels: high-quality (completeness ≥ 98%), qualified (completeness 90%-97%), requiring correction (completeness 80%-89%), and invalid (completeness < 80%). Synchronous execution of outlier removal and intelligent missing value completion: The 3σ principle is used to remove outlier data that exceeds a reasonable range (such as outdoor temperature of -40℃ or 60℃), and missing data is completed by interpolation algorithms based on airport operation patterns (such as combining passenger flow trends during peak flight periods).

[0026] The outdoor temperature (32℃, collected hourly), indoor temperature (24℃, collected every 15 minutes), and passenger density (0.6 people / ㎡, collected every 30 minutes) of Area A in Terminal 2 from 15:00 to 15:15 were aligned to the 15:00-15:15 time window. It was found that the rotation speed data of a certain fan coil unit was missing during this period. Combined with the frequency of take-off and landing of flights during the same period (2 flights landed in this area from 15:00 to 15:15) and historical data of similar scenarios, the rotation speed was completed to 850 r / min. The outdoor humidity of 98% (exceeding the reasonable humidity range of 50%-85% for the same period in the local area) falsely reported by a certain sensor during the period of 15:30-15:45 was removed. Finally, a complete multi-source airport data set covering all elements of "building structure - environmental status - personnel flow - equipment operation" was formed.

[0027] To address the significant impact of peak flight times (e.g., 7:00-9:00 and 14:00-16:00 daily) and seasonal changes (e.g., summer (June-August) and winter (December-February) on airport air conditioning load, an improved recurrent neural network model (based on LSTM architecture) is constructed. An "airport operation characteristic adaptation layer" is added to the model structure, using airport-specific time-series information such as peak passenger flow periods, flight arrival and departure frequencies, and route type (domestic / international) as key input gating parameters. By adjusting the weights of the gating units, the model's ability to capture patterns in airport air conditioning load data with large lags and strong fluctuations is enhanced. Based on integrated multi-dimensional historical data, the model is trained to achieve in-depth mining and accurate extraction of time-series patterns.

[0028] Using historical data from an airport over the past year as training samples, information such as peak passenger flow (average 3,000 passengers / hour) and flight takeoff and landing frequency (average 15 flights / hour) from 7:00 to 9:00 daily is input into the adaptation layer. The model focuses on the load change patterns during this period through a gating mechanism. It has uncovered temporal patterns such as "during the summer from 14:00 to 16:00, when domestic flights are landing intensively, the cooling demand in the departure hall increases by 40% compared to off-peak hours" and "during winter rain and snow, the cooling loss in the glass curtain wall area of ​​the terminal increases by 25%, and the load demand rises one hour earlier." This breaks through the limitations of traditional recurrent neural networks that rely solely on general temporal features.

[0029] A "load demand-data feature" mapping model is constructed to transform the unstructured time-series patterns discovered into quantifiable structured feature indicators. Considering the architectural characteristics of airports with their large, open spaces, "airport building mechanism correction factors" (such as regional spatial volume correction coefficients and building envelope thermal resistance correction coefficients) are introduced to perform scenario-based calibration of the structured feature indicators, eliminating the influence of building physics on feature accuracy. Finally, in conjunction with data standardization specifications, the calibrated feature indicators undergo format unification (e.g., uniformly adopting a "value + unit" format) and range normalization (mapping indicator values ​​to the [0,1] interval), generating standardized cooling demand-related feature data that combines data standardization with airport scenario adaptability.

[0030] The unstructured pattern of "surge in cooling demand in densely populated passenger areas from 14:00 to 16:00 in summer" is transformed into structured indicators such as "time-period load fluctuation weight (0.8 for 14:00-16:00)" and "regional passenger flow load coefficient (0.9 for densely populated areas)" by a mapping model. For the spatial differences between the departure hall (18 meters high, 20,000㎡ volume) and the corridor (3 meters high, 5,000㎡ volume), a volume correction coefficient is introduced (1.2 for the departure hall, 0.8 for the corridor) to calibrate the regional passenger flow load coefficient. The calibrated indicators, such as "time-period load fluctuation weight 0.8" and "calibrated regional passenger flow load coefficient 1.08", are formatted as "0.8 (unitless)" and "1.08 (unitless)" according to standardized format, and normalized to the [0,1] interval, ultimately generating standardized feature data.

[0031] S103 processes standardized cooling demand-related characteristic data to construct a multi-regional room temperature neural network large-lag prediction model and a dynamic energy consumption trend prediction model for central air conditioning systems. Through multi-modal large-scale model fusion analysis, accurate cooling demand prediction results are generated for each spatiotemporal dimension.

[0032] In one implementation, standardized cooling demand-related feature data is dimensionally split to extract building mechanism-related features, including regional spatial parameters and heat transfer characteristics, as well as dynamic demand features related to time-period load fluctuations and pedestrian density, forming a feature subset suitable for dual-model training. Specifically, the standardized cooling demand-related feature data is systematically dimensionally split, clearly defining two core feature categories: building mechanism-related features and dynamic demand features. Building mechanism-related features focus on the physical characteristics of the high-ceilinged spaces of airport terminals, extracting regional spatial parameters (such as floor height, building area, and space volume of each functional area) and heat transfer characteristics (such as thermal resistance of the building envelope, insulation coefficient, and air permeability). Dynamic demand features revolve around the temporal sequence of load changes and factors influencing people, extracting time-period load fluctuations (such as peak / valley load values, fluctuation amplitude, and duration at different times) and pedestrian density-related features (such as regional pedestrian density, average dwell time, and peak-period pedestrian proportion). The two types of features are screened and deduplicated, and redundant indicators (such as features with a correlation of less than 0.3 with cooling demand) are removed to form a feature subset suitable for training the two models (multi-region room temperature neural network large lag prediction model and central air conditioning system dynamic energy consumption trend prediction model).

[0033] The architectural features of Terminal 2's A-zone waiting hall (18 meters high, 20,000㎡ volume, thermal resistance of the building envelope 0.5㎡・K / W) and B-zone corridor (3 meters high, 5,000㎡ volume, thermal resistance of the building envelope 0.4㎡・K / W) were extracted from the standardized feature data. Dynamic demand features were extracted, including a peak load of 1200kW and a fluctuation range of 30% from 7:00 to 9:00 daily, a passenger flow density of 0.8 people / ㎡ from 14:00 to 16:00, and an average stay of 15 minutes. After screening, a feature subset containing 12 core indicators was formed, including 5 architectural features and 7 dynamic demand features, which are adapted to the input requirements of dual-model training data.

[0034] To address the significant lag in multi-regional room temperature control, a neural network-based large-lag prediction model is constructed. A dynamic lag time calibration algorithm is introduced, using the delay in cold air transfer in different regions as a hyperparameter, and iteratively optimizing the model weights based on historical room temperature response data. Specifically, to address the significant lag (cold air transfer delay of 5-30 minutes) in room temperature control across multiple airport areas (waiting area, departure hall, baggage sorting area, etc.), a neural network-based large-lag prediction model using LSTM (Long Short-Term Memory) is constructed. The model structure includes an input layer, hidden layers (3 layers of LSTM units, 128 neurons per layer), a dropout layer (dropout rate 0.2 to prevent overfitting), and an output layer (fully connected layer, outputting the predicted room temperature for each region). A lag time dynamic calibration algorithm is introduced, using the delay time of cold energy transfer in different areas as the model hyperparameter (initial values ​​are based on historical data statistics, such as 20 minutes in the waiting area and 8 minutes in the corridor). Iterative optimization is performed using historical room temperature response data (cold energy adjustment instructions and corresponding room temperature change curves for each area in the past 6 months). After each training, the mean square error (MSE) between the predicted room temperature and the actual room temperature is calculated. The hyperparameter (delay time) and model weights are dynamically adjusted based on the error feedback until the MSE is lower than the preset threshold (0.5℃²).

[0035] Using Terminal 2's A and B waiting areas as prediction targets, hyperparameters were initialized (20 minutes delay for A and 8 minutes for B). A subset of feature data from the past six months (building mechanism features + dynamic demand features) and corresponding actual room temperature data were input for model training. After the first training, the MSE for A was 1.2℃². By adjusting the delay time to 25 minutes and optimizing the LSTM unit weights, the MSE decreased to 0.45℃² after the second training, meeting the accuracy requirements. The final model accurately captures the hysteresis response pattern of "25 minutes after the cooling adjustment command is issued, the room temperature in A drops from 26℃ to 24℃."

[0036] This paper constructs a dynamic energy consumption trend prediction model based on the coupled operation characteristics of multiple devices in a central air conditioning system. The model uses system operation characteristics such as equipment load rate, energy efficiency parameters, and cooling capacity delivery path as core inputs, and captures the dynamic matching pattern between cooling capacity supply and demand through time-series correlation analysis. Specifically, considering the coupled operation characteristics of equipment such as chillers, fan coil units, cooling towers, and water pumps in the central air conditioning system (e.g., chiller load rate affects chilled water temperature, which in turn affects fan coil unit heat exchange efficiency), a dynamic energy consumption trend prediction model based on GRU (Gated Circulation Unit) is constructed. The model input layer selects core indicators of equipment operation characteristics: equipment load rate (current load / rated load of chiller, fan coil unit load ratio), energy efficiency parameters (chiller COP value, cooling tower heat exchange efficiency), and cooling capacity delivery path characteristics (resistance of chilled water supply and return pipe network, water flow velocity, and delivery distance). The hidden layer consists of two layers of GRU units (100 neurons per layer), introducing an attention mechanism to strengthen the focus on key equipment operation characteristics. The output layer outputs the predicted value of the matching degree between the system's cooling capacity supply and demand. By using time-series correlation analysis (calculating the Pearson correlation coefficient between equipment operating characteristics and cooling demand), we can capture the dynamic matching pattern of cooling supply and demand at different times (e.g., during peak hours, the chiller unit load rate needs to reach more than 80% to meet the demand).

[0037] Inputting operating characteristic data such as chiller unit load rate of 75%, COP value of 4.2, cooling tower heat exchange efficiency of 92%, chilled water supply and return water network resistance of 0.02MPa, and water flow velocity of 1.2m / s, the model focuses on the chiller unit load rate and COP value through an attention mechanism. Time-series correlation analysis reveals that "when the chiller unit load rate increases from 75% to 85%, the cooling capacity increases by 18%, matching the 20% increase in cooling demand during peak hours in Area A with a 90% match," accurately capturing the dynamic correlation between cooling supply and demand.

[0038] A multimodal large-scale model fusion mechanism is introduced, iteratively adjusting the weight ratios based on the prediction accuracy performance of the two models under different scenarios. Simultaneously, multimodal auxiliary information, including building structural characteristics and real-time environmental parameters, is integrated to correct prediction biases, generating accurate cooling demand prediction results covering all regions and all time periods across various spatiotemporal dimensions. Specifically, a multimodal large-scale model fusion mechanism (based on a weighted fusion algorithm) is introduced, using the prediction accuracy of the two models (a large-lag prediction model and a dynamic energy consumption trend prediction model) as the core basis for iteratively adjusting their weight ratios. First, typical airport operating scenarios were defined (morning peak 7:00-9:00, midday off-peak 11:00-13:00, evening peak 14:00-16:00, and nighttime off-peak 22:00-6:00 the next day). Historical data was used to verify the prediction accuracy of the dual models in each scenario (e.g., the accuracy of the morning peak large lag model was 88%, and the accuracy of the dynamic energy consumption model was 82%; the accuracy of the nighttime off-peak large lag model was 75%, and the accuracy of the dynamic energy consumption model was 90%). Weights were dynamically allocated according to the formula "weight = single model accuracy / sum of dual model accuracy" (0.52 for the morning peak large lag model and 0.48 for the dynamic energy consumption model; 0.46 for the nighttime off-peak large lag model and 0.54 for the dynamic energy consumption model). Simultaneously, it integrates multimodal auxiliary information, including building structural characteristics (such as regional spatial volume and envelope type) and real-time environmental parameters (real-time outdoor temperature, humidity, and solar radiation intensity). The residual correction algorithm corrects the deviation of the dual-model fusion results (e.g., when the outdoor temperature suddenly rises by 5°C, the predicted value is corrected by 10% based on the building's thermal insulation characteristics). Finally, it generates accurate cooling demand prediction results covering the entire area (each functional zone) and all time periods (15-minute granularity) in all spatiotemporal dimensions.

[0039] During the morning rush hour, the large lag prediction model outputs a cooling demand of 1200kW for Zone A, while the dynamic energy consumption trend prediction model outputs 1150kW. The preliminary fusion result, calculated with weights of 0.52 and 0.48, is 1176kW. Combining real-time environmental parameters (outdoor temperature 32℃, 3℃ higher than the same period in history) and the building characteristics of Zone A (60% glass curtain wall, significant impact from sunlight), the prediction result is corrected to 1250kW using a residual correction algorithm. The error between this prediction and the actual cooling demand of 1230kW is only 1.6%, achieving accurate prediction.

[0040] S104 processes the accurate cooling demand forecast results and constructs an air conditioning energy consumption standard model that includes peak and valley time period division, regional load level classification, and equipment energy efficiency threshold setting. Based on the strategy of first determining the evaluation standard and then making dynamic scheduling, it generates equipment group collaborative optimization control constraints.

[0041] In one implementation, the spatiotemporal distribution characteristics, load fluctuation amplitude, and peak demand period data in the accurate cooling demand forecast results are extracted and processed. Combined with airport operation peak-valley patterns, regional functional differences, and equipment operating limit parameters, the criteria for dividing peak-valley periods, regional load level judgment standards, and equipment energy efficiency threshold ranges are determined. The accurate cooling demand forecast results in each spatiotemporal dimension are analyzed in depth to extract three core data points: spatiotemporal distribution characteristics (cooling demand values ​​for each functional area at different times, and differences in demand distribution between areas), load fluctuation amplitude (demand difference between adjacent periods in a single area, and the ratio of maximum to minimum demand on a given day), and peak demand periods (the specific time when cooling demand in each area reaches its peak, and the duration of the peak). The standards were developed by combining three key reference bases: airport operation peak and valley patterns (peak / off-peak / low peak periods of passenger flow determined by flight arrival and departure schedules and passenger arrival and departure data), regional functional differences (comfort requirements and energy consumption characteristics of waiting areas, departure halls, baggage sorting areas, etc.), and equipment operating limit parameters (rated load, maximum energy efficiency, and safe operation thresholds of equipment such as chillers and fan coil units). The final standards clarified the basis for peak and valley period division, regional load level determination criteria, and equipment energy efficiency threshold ranges.

[0042] Extract forecast data from Terminal 2's A-zone waiting hall (cooling demand 1250kW, fluctuation range 40% from 14:00-16:00) and B-zone corridor (cooling demand 300kW, fluctuation range 15% from 10:00-11:00); combine this with airport operational patterns (peak passenger flow 7:00-9:00 and 14:00-16:00), area functions (waiting area needs to maintain 24-26℃, baggage sorting area only needs 22-28℃), and equipment parameters (chilled water unit rated load). The peak and valley time periods are determined based on the following criteria: peak hours are defined as periods with a population density of ≥0.6 people / m², off-peak hours are defined as periods with a population density of 0.3-0.6 people / m², and low-peak hours are defined as periods with a population density of <0.3 people / m². The regional load level is determined based on the following criteria: peak hour demand ≥1000kW is classified as Level 1 load, 500-1000kW is classified as Level 2 load, and <500kW is classified as Level 3 load. The equipment energy efficiency threshold range is defined as: chiller COP value ≥3.8 and fan coil unit energy efficiency ratio ≥2.5.

[0043] Based on the extracted core data and established judgment criteria, a "three-dimensional" air conditioning energy consumption standard model is constructed. The model architecture includes an input layer, a rule layer, and an output layer: the input layer receives accurate cooling demand forecast data, airport operation scenario data, and equipment parameter data; the rule layer embeds three core logics: peak and valley time period division rules, regional load level judgment rules, and equipment energy efficiency constraint rules; the output layer outputs energy consumption control standards for each scenario. Clearly define the quantitative boundaries of evaluation indicators for each dimension to ensure that the model can be implemented: peak and off-peak periods are quantified as "peak 7:00-9:00, 14:00-16:00, off-peak 9:00-14:00, 16:00-22:00, and low-peak 22:00-7:00 the next day"; regional load levels are quantified as "level 1 load has the highest priority for cooling capacity supply, level 2 is next, and level 3 is the lowest"; equipment energy efficiency thresholds are quantified as "the operating range of chiller load rate is 40%-85% (corresponding to COP value of 3.8-5.0), and the fan coil unit speed range is 600-1200 r / min (corresponding to energy efficiency ratio of 2.5-3.2)".

[0044] The model takes into account the predicted cooling demand of the baggage sorting area in Zone C of Terminal 2 (800kW during peak hours, 500kW during off-peak hours, and 300kW during low-peak hours). After calculation by the rule layer: the 800kW peak hour is determined to be a level 2 load, and the cooling supply must meet priority 2. The load rate of the chiller units for this area should be controlled at 60%-70% (COP value 4.2-4.5), and the fan coil unit speed should be 800-1000r / min; the 500kW off-peak hour is determined to be a level 3 load, and the load rate of the chiller units should be 40%-50%, with the fan coil unit speed at 600-800r / min; the 300kW low-peak hour is determined to be a level 3 load, and the load rate of the chiller units can be reduced to below 40%, with the fan coil unit speed maintained at 600r / min. The model outputs the energy consumption control standards for this area at different times, with clear and executable quantification boundaries.

[0045] Following the core strategy of "first defining evaluation standards and then implementing dynamic scheduling," the three core rules in the energy consumption standard model are integrated and correlated: peak-valley time period rules as time-dimensional constraints, regional load level rules as spatial-dimensional constraints, and equipment energy efficiency threshold rules as equipment operation constraints, thus constructing a constraint system adapted to the coordinated control of equipment groups. This system clarifies three types of core control conditions: the upper limit of cooling supply for each time period and region (determined based on load level and equipment rated load to avoid oversupply), equipment operating energy efficiency compliance requirements (based on energy efficiency threshold ranges to ensure equipment operates within the high-efficiency range), and load dynamic allocation thresholds (based on regional priority and equipment coupling characteristics to determine the load allocation ratio range for each device), ultimately generating coordinated optimization control constraints for equipment groups.

[0046] After integrating the model rules, the following constraints are generated for the Terminal 2 A waiting hall (level 1 load during peak hours): Cooling capacity supply limit of 1500kW (not exceeding the rated load of the chiller units), equipment operating energy efficiency requirements of chiller unit COP≥4.0 and fan coil unit energy efficiency ratio≥2.8, and load dynamic allocation threshold of "chiller unit 1 bears 60% of the load, chiller unit 2 bears 40%, and fan coil units 1-10 in A area operate at 1000-1200 r / min"; For the B corridor (level 3 load during off-peak hours), the following constraints are generated: Cooling capacity supply limit of 500kW, equipment operating energy efficiency requirements of chiller unit COP≥3.8 and fan coil unit energy efficiency ratio≥2.5, and load dynamic allocation threshold of "chiller unit 1 bears 30% of the load, and fan coil units 1-5 in B area operate at 600-800 r / min", ensuring that the equipment group meets demand while achieving optimal energy consumption and coordinated operation.

[0047] S105 processes the control constraints of the equipment group for collaborative optimization, and combines the cooling demand forecast data to perform collaborative optimization calculations on the air conditioning equipment group, which includes chillers, fan coil units, and cooling towers, to generate the target equipment operation plan.

[0048] In one implementation, an equipment characteristic analysis algorithm is used to deeply decompose the upper limit of cooling capacity supply, energy efficiency requirements, and load allocation thresholds in the collaborative optimization control constraints of the equipment group, generating baseline rules for equipment operation characteristics and parameter adaptation threshold ranges suitable for chillers, fan coil units, and cooling towers. For the constraints of the air conditioning equipment group in Terminal 2 (peak cooling capacity supply limit of 1500kW in Area A, chiller COP ≥ 3.8, load allocation threshold of 40%-80%), the algorithm decomposes them as follows: the baseline rule for chillers is "dynamically adjust the load rate according to cooling demand, prioritizing maintaining COP in the 3.8-5.0 range," and the parameter adaptation threshold range is "load rate 40%-85%, chilled water supply temperature 6-8℃, return water temperature 10-12℃"; fan coil units... The baseline rule for the cooling tower is "adjust the rotation speed according to the regional cooling demand, taking into account the balance between wind speed and energy efficiency", and the parameter adaptation threshold range is "rotation speed 600-1200r / min, air volume 300-800m³ / h, energy efficiency ratio ≥2.5"; the baseline rule for the cooling tower is "adjust the rotation speed according to the load change of the chiller unit to ensure heat exchange efficiency", and the parameter adaptation threshold range is "rotation speed 800-1500r / min, water supply flow 5-15m³ / h, heat exchange efficiency ≥85%".

[0049] By connecting the air conditioning equipment group's operational efficiency database (which stores historical operating parameters, energy efficiency data, fault records, etc.) with cooling demand forecast data across various spatiotemporal dimensions, and with "supply-demand balance + optimal energy efficiency" as the core objective, a benchmark for coordinated equipment control is constructed—clarifying the overall energy efficiency targets and collaborative logic for the equipment group's operation under different cooling demand scenarios. The optimization rules for each equipment's operating parameters are determined through supply-demand matching logic: based on the matching degree between the predicted cooling demand and the equipment's rated capacity, rules for parameter adjustment priorities and magnitudes are established. A multi-device coupled collaborative model (a hybrid model based on mechanistic modeling and data-driven approaches) is introduced. The model architecture includes an input layer (cooling demand, equipment status, and constraints), a coupling layer (simulating the mutual influence between devices, such as the impact of chiller load changes on cooling tower heat exchange demand), and an output layer (optimized operating parameter suggestions for each device). By combining intelligent optimization algorithms (such as an improved particle swarm optimization algorithm with a dynamic adjustment strategy for inertia weights to improve convergence speed) and airport air conditioning system operation scenario profiles (covering peak / off-peak / low-peak, extreme / normal weather, and other scenario characteristics), a dynamic computation mechanism of "demand-constraint-optimization" collaboration is established to achieve dynamic optimization of equipment operating parameters based on changes in demand and constraints.

[0050] Historical data of chiller unit 1 (COP=4.3 at 70% load and COP=3.7 at 90% load) was retrieved from the equipment operating efficiency database. Combined with the predicted peak cooling demand of 1250kW in Area A, the collaborative control benchmark was constructed as "comprehensive energy efficiency of equipment group ≥ 4.0, cooling supply error ≤ 5%". The optimization rule was determined through supply and demand matching logic: "Chillers prioritize adjusting load rate, fan coil units adjust speed synchronously, and cooling towers adjust according to chiller load changes". Simulation using a multi-equipment coupled collaborative model showed that "when the chiller load rate is 75%, the cooling tower speed needs to be 1200r / min and the fan coil unit speed 1000r / min to maintain system balance". After iterative optimization using an improved particle swarm optimization algorithm, the optimal parameter combination was determined, and a dynamic calculation mechanism was established—when cooling demand fluctuates by ±10%, the chiller load rate is adjusted by ±8%, the fan coil unit speed by ±100r / min, and the cooling tower speed by ±150r / min.

[0051] Using equipment type as the core coverage dimension, relevant data from chillers, fan coil units, and cooling towers are integrated: chillers integrate cooling capacity demand allocation data (75% peak cooling capacity allocation for area A, 25% for area B) and operating parameter ranges (load rate, water temperature, etc.); fan coil units integrate corresponding area cooling capacity demand data and parameter ranges such as speed and airflow; cooling towers integrate chiller load-related data and parameter ranges such as speed and flow rate. The integrated data is compared with equipment operating characteristic baseline rules and parameter adaptation threshold ranges. Valid data that meets the rules and threshold requirements is filtered out, while abnormal data exceeding the thresholds is removed, ensuring that the data input for subsequent calculations is compliant and valid.

[0052] The data from chiller units (cooling demand of 937.5kW in Area A, load rate threshold of 40%-85%), fan coil units (cooling demand of 937.5kW for fan coil units 1-10 in Area A, speed threshold of 600-1200r / min), and cooling towers (corresponding to a chiller unit load rate of 75% and a speed threshold of 800-1500r / min) were compared with the baseline rules and threshold ranges. Abnormal data of "speed 1300r / min" for fan coil unit 3 (exceeding the upper limit of the threshold) was removed, and valid data was retained for subsequent calculations.

[0053] By normalizing parameters (mapping parameters of different dimensions, such as load rate, temperature, and speed, to the [0,1] range) and preprocessing supply and demand data (establishing a mapping relationship between cooling demand and equipment operating parameters, such as an increase of approximately 6% in chiller load rate for every 100kW increase in cooling demand), the system integrates equipment group collaborative optimization control constraints with cooling demand forecast data to eliminate the impact of data heterogeneity and insufficient correlation. The preprocessed data is then input into a multi-equipment coupled collaborative model. The model simulates the dynamic influence between equipment through a coupling layer and performs multiple rounds of iterative calculations using intelligent optimization algorithms, continuously adjusting the operating parameters of each device until the dual objectives of cooling demand and optimal energy efficiency are met, thus enhancing the accuracy of the optimization calculations. By integrating the characteristics of a two-layer architecture of "dynamic adjustment of demand-side forecasting + collaborative response of supply-side equipment," the optimized equipment operating parameters, inter-equipment collaborative logic (such as the linkage adjustment sequence of cooling towers and fan coil units after chiller load adjustment), and energy-saving benefit estimates (such as the energy consumption reduction ratio and annual electricity savings compared to traditional operating schemes) are integrated to generate a target equipment operating scheme.

[0054] After preprocessing the peak scenario data in Area A, the data is input into a multi-device coupled collaborative model. The model simulates that when the chiller unit load rate is 75% (chilled water supply temperature 7℃, return water temperature 12℃), the cooling tower needs to maintain a speed of 1200 r / min and the fan coil units 1-10 need to maintain a speed of 1000 r / min in order to meet the cooling demand of 1250kW. After 100 iterations of the improved particle swarm optimization algorithm, the parameters converge to the optimal solution. Integrating the characteristics of a dual-layer architecture, the following target equipment operation scheme is generated: Chiller Unit 1 load rate 75%, chilled water supply 7℃, return water 12℃; Fan coil units A zone 1-10 speed 1000r / min, air supply volume 600m³ / h; Cooling tower 2 speed 1200r / min, makeup water flow 10m³ / h; The collaborative logic is "the chiller unit first adjusts its load rate, the cooling tower adjusts its speed after a 30-second delay, and the fan coil units adjust synchronously"; The estimated energy-saving benefit is "this scheme reduces energy consumption by 18% compared to the traditional fixed-frequency operation scheme, saving approximately 120,000 kWh of electricity per year".

[0055] S106 processes the target equipment operation plan and related data, and combines a two-layer architecture of precise demand-side forecasting and intelligent supply-side regulation to generate early warning information for airport air conditioning systems that takes into account both comfort and energy-saving goals.

[0056] In one implementation, based on preset dual-objective verification rules for comfort and energy saving, the system incorporates parameter execution deviation, energy consumption compliance rate, and room temperature fluctuation range from the target equipment operation plan. This allows for feature identification of the equipment operation data and the adaptability of the dual-layer architecture, generating a set of early warning anomaly features. The comfort objective focuses on maintaining room temperature in each area at 22-26℃ (waiting area) and 22-28℃ (baggage sorting area). The energy saving objective focuses on a comprehensive energy efficiency of ≥4.0 for the equipment group and a unit cooling energy consumption of ≤0.8kWh / kW. Three key evaluation indicators from the target equipment operation plan are introduced: parameter execution deviation (the difference and percentage between actual operating parameters and the parameters set in the plan), energy consumption compliance rate (the ratio of actual energy consumption to baseline energy consumption), and room temperature fluctuation range (the difference between the maximum and minimum room temperature values ​​per unit time in a single area). Based on these rules and indicators, the system identifies features of the adaptability of the equipment operation data to the dual-layer architecture of "precise demand-side forecasting and intelligent supply-side control," filtering out anomalies that do not meet the dual-objective requirements to form a set of early warning anomaly features.

[0057] The target operating plan for Terminal 2, Area A waiting hall, was set at a chiller unit load rate of 75% and an ambient temperature of 24℃. Actual operating data showed a load rate of 85% (parameter execution deviation of 13.3%), an energy consumption compliance rate of 115% (exceeding the energy-saving target), and an ambient temperature fluctuation range of 3℃ (exceeding the allowable ±1℃ fluctuation for comfort). Actual operating data for Area B corridor showed a fan coil unit speed of 1300 r / min (parameter execution deviation of 8.3%), an energy consumption compliance rate of 108%, and an ambient temperature fluctuation range of 2.5℃. Based on feature identification, a set of abnormal warning features was generated: "Chiller unit load rate execution deviation exceeding 10%", "Energy consumption compliance rate exceeding 105%", "Ambient temperature fluctuation range exceeding 2℃", and "Fan coil unit speed execution deviation exceeding 5%".

[0058] By comparing the relevant data of the target equipment operation plan with the preset optimal operating parameter library and energy consumption benchmark library, and verifying it in different airport operation scenarios, the deviation of the equipment operation status from the dual target requirements is determined, and an operation deviation assessment result is generated. The optimal operating parameter library stores the optimal parameters of different equipment in various scenarios, such as the optimal load rate of chillers during peak hours of 70%-75%, and the optimal speed of fan coil units of 800-1000 r / min. The energy consumption benchmark library stores the unit cooling energy consumption benchmark under different scenarios, such as ≤0.8kWh / kW during peak hours and ≤0.7kWh / kW during off-peak hours. The relevant data of the target equipment operation plan (equipment operating parameters, actual energy consumption, room temperature data) are compared with the data in the preset database, and verified in different airport operation scenarios (peak / off-peak / low-peak, extreme weather / normal weather): the peak scenario focuses on verifying the sufficiency of cooling supply and energy efficiency balance, the off-peak scenario focuses on verifying the accuracy of energy consumption control, and the low-load operation stability of the equipment focuses on verifying the low-load operation stability of the equipment. By calculating the deviation formula "(actual value - benchmark value) / benchmark value × 100%", the degree of deviation between the equipment operating status and the dual target requirements is determined, and the operating deviation assessment result is generated.

[0059] The actual data of the waiting hall in Area A during peak hours was compared with the database: the chiller unit load rate was 85% (13.3% deviation from the optimal range), the actual unit cooling energy consumption was 0.92 kWh / kW (15% deviation from the benchmark), and the room temperature was 24-27℃ (4.2% deviation from the upper limit of the comfort target). Combined with the peak scenario verification (comfort should be given priority, and a small deviation in energy consumption is allowed), the comprehensive operational deviation was calculated to be 12.1%, and the evaluation result was generated: "The operational deviation of the waiting hall in Area A is 12.1%, which is a moderate deviation. It is mainly due to the excessive load rate of the chiller unit, which leads to excessive energy consumption. The room temperature fluctuation is within the acceptable range."

[0060] The deviation assessment results are correlated with the set of early warning anomalies. Combining demand-side cooling capacity forecast accuracy feedback and supply-side equipment collaborative response efficiency, the deviation assessment results are weighted to generate an early warning optimization vector that integrates forecasting and control collaborative logic. The correspondence between anomalies and deviations is clearly defined (e.g., "chiller load rate deviation exceeding 10%" corresponds to an energy consumption deviation of 15%, and "room temperature fluctuation exceeding 2℃" corresponds to a comfort deviation of 4.2%). Two key weighting factors are introduced: demand-side cooling capacity forecast accuracy feedback (error rate between predicted and actual values, with lower error rates carrying higher weight) and supply-side equipment collaborative response efficiency (equipment parameter adjustment response time, with shorter response times carrying higher weight). A weighted summation algorithm is used to process the deviation assessment results, generating an early warning optimization vector that integrates forecasting and control collaborative logic—the vector dimension corresponds to the optimization priority and adjustment magnitude of each anomaly feature; a larger value indicates that the feature needs priority optimization.

[0061] The demand-side cooling capacity forecast accuracy for the waiting hall in Area A is 1.6% (weight 0.8), and the supply-side equipment coordination response efficiency is 10 seconds (weight 0.7). Correlation analysis shows that "chiller load rate deviation" has the highest correlation with the deviation degree (0.9), followed by "room temperature fluctuation" (0.6). Through weighted calculation: (15%×0.8×0.9)+(4.2%×0.7×0.6)=10.8%+1.764%=12.564%, an early warning optimization vector is generated: [0.6 (chiller load rate adjustment priority), 0.3 (room temperature fluctuation optimization priority), 0.1 (fan coil unit parameter adjustment priority)], clearly prioritizing the reduction of chiller load rate to optimize energy consumption deviation.

[0062] The operational deviation assessment results and early warning optimization vectors are normalized and comprehensively calculated. Combined with the real-time control and response requirements of the airport air conditioning system, early warning information for the airport air conditioning system is generated, including early warning levels, abnormal parameter identifiers, and energy-saving and comfort balance suggestions. Specifically, the operational deviation assessment results (e.g., 12.1%) and early warning optimization vectors (e.g., [0.6, 0.3, 0.1]) are normalized: the deviation is mapped to the [0, 1] interval (12.1% corresponds to 0.121), and the values ​​of each dimension of the optimization vector are normalized while maintaining relative priority. Using the comprehensive calculation formula of "deviation × vector weight," combined with the real-time control and response requirements of the airport air conditioning system (response time ≤ 5 minutes during peak hours, ≤ 10 minutes during off-peak / low-peak hours), the early warning levels (mild deviation: deviation < 8%; moderate deviation: 8%-15%; severe deviation: > 15%), abnormal parameter identifiers (identifying the core parameters causing the deviation), and energy-saving and comfort balance suggestions (providing specific adjustment directions based on the priority of the optimization vectors) are determined, ultimately generating early warning information for the airport air conditioning system. Example: After normalization and comprehensive calculation, the comprehensive score of the waiting hall in Area A is 0.121 × 0.6 = 0.0726. Combined with the real-time control requirements during peak hours, the warning level is determined to be a moderate warning. The abnormal parameters are identified as "chiller load rate (85%), unit cooling energy consumption (0.92 kWh / kW)". The energy-saving and comfort balance recommendations are: "1. Reduce the chiller load rate to 75%, with an expected energy consumption reduction of 13%; 2. Fine-tune the fan coil speed to 1000 r / min to maintain a stable room temperature of 24-25℃; 3. Continuously monitor changes in cooling demand and provide feedback on the adjustment effect every 5 minutes", thus generating complete warning information.

[0063] In one implementation, such as Figure 2 As shown, this application also provides an air conditioning load prediction and feedforward energy-saving control device that integrates airport building mechanisms and multimodal large models, including:

[0064] The acquisition module 201 is used to acquire multi-source raw airport data, including airport building characteristics, environmental parameters, personnel flow, and air conditioning system operating conditions.

[0065] Processing module 202 is used to process multi-source raw data from the airport. By integrating multi-dimensional historical data on outdoor weather, indoor environment, passenger trajectories, and equipment operation, it uses recurrent neural network technology to mine patterns in time-series data and generate standardized cooling demand-related feature data. It then processes this standardized cooling demand-related feature data to construct a multi-regional room temperature neural network large-lag prediction model and a dynamic energy consumption trend prediction model for the central air conditioning system. Through multi-modal large-scale model fusion analysis, it generates accurate cooling demand prediction results for each spatiotemporal dimension. Finally, it processes these accurate cooling demand prediction results to construct a model covering peak and valley time... The system employs a standard model for air conditioning energy consumption, which includes segment division, regional load level classification, and equipment energy efficiency threshold setting. Based on a strategy of first defining evaluation standards and then implementing dynamic scheduling, it generates collaborative optimization control constraints for equipment groups. These constraints are then processed, and combined with cooling demand forecast data, collaborative optimization calculations are performed on the air conditioning equipment group, including chillers, fan coil units, and cooling towers, to generate target equipment operation plans. Finally, the target equipment operation plans and related data are processed, and a two-tiered architecture combining precise demand-side forecasting and intelligent supply-side control is used to generate early warning information for airport air conditioning systems that balances comfort and energy-saving goals.

[0066] The computer-readable storage medium provided in the above embodiments of this application and the air conditioning load prediction and feedforward energy-saving control method integrating airport building mechanism and multimodal large model provided in the embodiments of this application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application stored therein.

[0067] The various embodiments in this application are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for evaluating the air conditioning load prediction and feedforward energy-saving control method, electronic device, electronic device, and readable storage medium based on the integration of airport building mechanisms and multimodal large-scale models are basically similar to the embodiments of the air conditioning load prediction and feedforward energy-saving control method based on the integration of airport building mechanisms and multimodal large-scale models described above. Therefore, the description is relatively simple, and relevant parts can be referred to in the description of the embodiments of the air conditioning load prediction and feedforward energy-saving control method based on the integration of airport building mechanisms and multimodal large-scale models described above.

Claims

1. A method for predicting air conditioning load and feedforward energy-saving control that integrates airport building mechanisms and multimodal large-scale models, characterized in that, include: Acquire multi-source raw airport data, including airport building characteristics, environmental parameters, passenger flow, and air conditioning system operating conditions; The airport's multi-source raw data is processed by integrating multi-dimensional historical data on outdoor weather, indoor environment, passenger trajectories, and equipment operation. Recurrent neural network technology is used to mine the patterns in time-series data and generate standardized cooling demand-related feature data. Standardized cooling demand-related characteristic data are processed to construct a multi-regional room temperature neural network large lag prediction model and a central air conditioning system dynamic energy consumption trend prediction model. Through multi-modal large model fusion analysis, accurate cooling demand prediction results in various spatiotemporal dimensions are generated. The accurate cooling demand forecast results are processed to construct an air conditioning energy consumption standard model that includes peak and valley time period division, regional load level classification, and equipment energy efficiency threshold setting. Based on the strategy of first determining the evaluation standard and then making dynamic scheduling, the collaborative optimization control constraints of the equipment group are generated. The constraints of the collaborative optimization control of the equipment group are processed, and combined with the cooling demand forecast data, the collaborative optimization calculation of the air conditioning equipment group including chillers, fan coil units and cooling towers is performed to generate the target equipment operation plan. The system processes the target equipment operation plan and related data, and combines a two-tier architecture of precise demand-side forecasting and intelligent supply-side regulation to generate early warning information for airport air conditioning systems that balances comfort and energy-saving goals.

2. The method as described in claim 1, characterized in that, The airport's multi-source raw data is processed by integrating multi-dimensional historical data on outdoor weather, indoor environment, passenger trajectories, and equipment operation. Recurrent neural network technology is used to mine patterns in time-series data, generating standardized cooling demand-related feature data, including: In response to the heterogeneous characteristics of outdoor meteorological, indoor environmental, passenger trajectory and equipment operation data, an integration mechanism of spatiotemporal dimension alignment and data quality classification is established to achieve spatiotemporal matching of multi-source data. Then, through outlier removal and intelligent missing value completion, a complete airport multi-source data set covering all elements of buildings, environment, personnel and equipment is formed. To address the significant impact of peak flight times and seasonal changes on airport air conditioning load, an improved recurrent neural network model was constructed. An airport operation characteristic adaptation layer was added to the model structure to deeply mine and accurately extract the temporal patterns of multi-dimensional historical data. A load demand-data feature mapping model is constructed to transform unstructured time-series patterns into structured feature indicators. At the same time, an airport building mechanism correction factor is introduced to perform scenario-based calibration of the feature indicators. Combined with data standardization specifications, the calibrated feature indicators are processed to unify the format and normalize the range, generating standardized cooling demand-related feature data that combines data standardization and airport scenario adaptability.

3. The method as described in claim 1, characterized in that, Standardized cooling demand-related characteristic data are processed to construct a multi-regional room temperature neural network large-lag prediction model and a dynamic energy consumption trend prediction model for central air conditioning systems. Through multimodal large-scale model fusion analysis, accurate cooling demand prediction results are generated for each spatiotemporal dimension, including: Standardized cooling demand-related feature data are dimensionally split to extract building mechanism-related features including regional spatial parameters and heat transfer characteristics, as well as dynamic demand features related to time-period load fluctuations and pedestrian density, forming a feature subset suitable for dual-model training; to address the large lag in multi-regional room temperature regulation, a neural network large lag prediction model is constructed, and a lag time dynamic calibration algorithm is introduced, using the cooling transfer delay time of different regions as the model hyperparameter, and iteratively optimizing the model weights by combining historical room temperature response data; Combining the multi-equipment coupled operation characteristics of central air conditioning systems, a dynamic energy consumption trend prediction model is constructed. The system operation characteristics of equipment load rate, energy efficiency parameters, and cooling load delivery path are used as core inputs. Through time series correlation analysis, the dynamic matching pattern of cooling load supply and demand is captured. A multimodal large model fusion mechanism is introduced, and the weight ratio is iteratively adjusted based on the prediction accuracy performance of the two models under different scenarios. At the same time, multimodal auxiliary information such as building structural characteristics and real-time environmental parameters are integrated to correct prediction deviations, and accurate cooling demand prediction results covering all regions and all time periods are generated in all spatiotemporal dimensions.

4. The method as described in claim 3, characterized in that, The accurate cooling demand forecast results are processed to construct a standard air conditioning energy consumption model that includes peak and valley time periods, regional load level classification, and equipment energy efficiency threshold settings. Based on a strategy of first defining evaluation standards and then implementing dynamic scheduling, collaborative optimization control constraints for the equipment group are generated, including: Extract and process the spatiotemporal distribution characteristics, load fluctuation amplitude, and peak demand period data from the accurate cooling demand forecast results. Combine the airport operation peak and valley patterns, regional functional differences, and equipment operating limit parameters to determine the basis for peak and valley period division, regional load level judgment criteria, and equipment energy efficiency threshold range. Based on the extracted core data and judgment criteria, an air conditioning energy consumption standard model is constructed that covers peak and valley time period division, regional load level classification, and equipment energy efficiency threshold setting, clarifying the quantitative boundaries of evaluation indicators in each dimension. Following the core strategy of first defining evaluation standards and then implementing dynamic scheduling, the peak-valley time period rules, load level corresponding energy consumption upper limits, and equipment energy efficiency constraints in the energy consumption standard model are linked and integrated to generate a constraint system that adapts to the coordinated control of equipment groups. This system includes the upper limit of cooling supply for each time period and region, the requirements for equipment operating energy efficiency compliance, and the equipment group coordinated optimization control constraints for load dynamic allocation thresholds.

5. The method as described in claim 1, characterized in that, The constraints of the collaborative optimization control of the equipment group are processed, and combined with the cooling demand forecast data, collaborative optimization calculations are performed on the air conditioning equipment group, including chillers, fan coil units, and cooling towers, to generate the target equipment operation plan, including: The equipment characteristic analysis algorithm is used to deeply decompose the upper limit of cooling supply, energy efficiency compliance requirements and load distribution threshold in the collaborative optimization control conditions of equipment group, and generate equipment operation characteristic baseline rules and parameter adaptation threshold ranges that are suitable for chillers, fan coil units and cooling towers. By connecting the air conditioning equipment group's operating efficiency database with cooling demand forecast data, a benchmark for equipment collaborative control is constructed. The operating parameter optimization rules for different equipment are determined through supply and demand matching logic. At the same time, a multi-equipment coupling and collaboration model is introduced. Combined with intelligent optimization algorithms and airport air conditioning system operation scenario profiles, a dynamic computer mechanism for demand-constraint-optimization collaboration is established. Based on equipment type as the coverage dimension, the system integrates the operating parameter range and cooling demand allocation data corresponding to chillers, fan coil units, and cooling towers, and combines equipment operating characteristic baseline rules and parameter adaptation thresholds. By integrating constraints and forecast data through parameter normalization and supply-demand data correlation preprocessing mechanisms, and combining multi-device coupled collaborative models to enhance the accuracy of optimization calculations, the system integrates the characteristics of a two-layer architecture of dynamic adjustment of demand-side forecasts and collaborative response of supply-side equipment to generate target equipment operation schemes with accompanying equipment operating parameters, collaborative logic descriptions, and energy-saving benefit estimates.

6. The method as described in claim 5, characterized in that, The system processes the target equipment's operation plan and related data, combining a two-tiered architecture of precise demand-side forecasting and intelligent supply-side control to generate early warning information for airport air conditioning systems that balances comfort and energy-saving goals. This includes: Based on the preset dual-objective verification rules of comfort and energy saving, the parameter execution deviation, energy consumption compliance rate and room temperature fluctuation range in the target equipment operation scheme are introduced to identify the features of the equipment operation data and the adaptability of the dual-layer architecture, and generate a set of early warning abnormal features. By comparing the relevant data of the target equipment operation plan with the preset optimal equipment operation parameter library and energy consumption benchmark value library, and combining the verification of different airport operation scenarios, the deviation of the equipment operation status from the dual target requirements is determined, and the operation deviation assessment result is generated. The deviation assessment results are correlated with the set of early warning anomalies. Combined with the feedback of the cooling capacity forecast accuracy on the demand side and the collaborative response efficiency of the equipment on the supply side, the deviation assessment results are weighted to generate an early warning optimization vector that integrates forecasting and control collaborative logic. The operational deviation assessment results and early warning optimization vectors are normalized and comprehensively calculated. Combined with the real-time control and response requirements of the airport air conditioning system, early warning information for the airport air conditioning system is generated, which includes early warning level, abnormal parameter identification, and energy-saving comfort balance suggestions.

7. A device for predicting air conditioning load and feeding forward energy-saving control that integrates airport building mechanisms and multimodal large-scale models, characterized in that, The device includes: The acquisition module is used to acquire multi-source raw airport data, including airport building characteristics, environmental parameters, personnel flow, and air conditioning system operating conditions. The processing module is used to process multi-source raw data from the airport. By integrating multi-dimensional historical data on outdoor weather, indoor environment, passenger trajectories, and equipment operation, it uses recurrent neural network technology to mine patterns in time-series data and generate standardized cooling demand-related feature data. This standardized cooling demand-related feature data is then processed to construct a multi-regional room temperature neural network large-lag prediction model and a dynamic energy consumption trend prediction model for the central air conditioning system. Through multi-modal large-scale model fusion analysis, accurate cooling demand prediction results are generated for each spatiotemporal dimension. Finally, these accurate cooling demand prediction results are processed to construct a system covering peak and off-peak periods. A standard model for air conditioning energy consumption, based on load level classification by region and equipment energy efficiency threshold setting, is used to generate control constraints for collaborative optimization of equipment groups. These constraints are then processed and combined with cooling demand forecast data to perform collaborative optimization calculations on air conditioning equipment groups including chillers, fan coil units, and cooling towers, generating target equipment operation plans. Finally, the target equipment operation plans and related data are processed, and a two-tiered architecture combining precise demand-side forecasting and intelligent supply-side control is used to generate early warning information for airport air conditioning systems that balances comfort and energy-saving goals.

8. An electronic device, characterized in that, include: First processor; and memory for storing executable instructions of the first processor; The first processor is configured to execute the air conditioning load prediction and feedforward energy-saving control method according to any one of claims 1 to 6 by executing the executable instructions.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the second processor, it implements the air conditioning load prediction and feedforward energy-saving control method that integrates airport building mechanisms and multimodal large models as described in any one of claims 1 to 6.

Images