CN-122015234-A - Time-sharing partition AI temperature compensation and air conditioner energy-saving control method and equipment considering airport passenger comfort preference self-learning

CN122015234ACN 122015234 ACN122015234 ACN 122015234ACN-122015234-A

Abstract

The invention discloses a time-sharing partition AI temperature compensation and air conditioner energy-saving control method and equipment considering airport passenger comfort preference self-learning, which are applied to the technical field of data processing. And layering according to space partition and time granularity, constructing a target directed multi-view dynamic graph containing multi-class nodes, and generating node embedded information through a self-adaptive multi-view dynamic graph neural network. And finally, injecting the AI temperature compensation parameters into a reinforcement learning double network, combining parameter high-efficiency fine adjustment and double-target antagonism optimization, and generating AI temperature compensation parameters and air conditioner energy-saving control instructions of time-sharing partition through scene migration reasoning, thereby giving consideration to comfort and energy saving.

Inventors

WANG KUI
ZHANG ZONGYI
XU HUA
ZHOU NING
ZHAO SIYAO
Yu Fengjian
Hong xiaoxi
YAO YE
LIN SHAOJIA
YANG JIANZHONG
LIANG LULU
ZHAO XUEYING
LIAO HAITAO
PAN HONG
LIU QIMEI

Assignees

广西桂物节能集团有限公司
广西桂物金岸制冷空调技术有限责任公司
上海交通大学

Dates

Publication Date: 20260512
Application Date: 20260120

Claims (8)

1. A time-sharing partition AI temperature compensation and air conditioner energy-saving control method considering airport passenger comfort preference self-learning is characterized by comprising the following steps: Acquiring a target data set and a training sample set, wherein the training sample set comprises an airport passenger behavior track data set, a real-time environment parameter data set and a passenger comfort feedback data set; Based on the passenger behavior track data set, carrying out feature extraction on a passenger stay region, activity intensity and stay time by using a deep learning model, combining a real-time environment parameter data set, converting passenger behavior semantic features into comfort demand vectors by using a cross-mode feature fusion module, quantifying preference levels of passenger comfort feedback data into weighted features, and generating a multi-dimensional passenger comfort preference data set facing temperature compensation and energy-saving control; processing a multi-dimensional passenger comfort preference data set, layering according to space partition and time granularity, wherein each layer comprises area nodes, environment parameter nodes, passenger preference nodes and air conditioner operation nodes, and aggregating different-level real-time characteristics through dynamic perception windows to generate a target directed multi-view dynamic diagram consisting of environment states, passenger preferences and air conditioner operation states; modeling the target directed multi-view dynamic graph based on the self-adaptive multi-view dynamic graph neural network, introducing a passenger preference self-learning unit, dynamically updating comfort preference weights in different scenes, and generating node embedded information; Injecting node embedded information into each layer of a value network and a strategy network of the reinforcement learning model, optimizing model parameters by adopting a parameter efficient fine adjustment method, introducing an energy consumption-comfort dual-objective antagonism optimization module, and generating a decision interpretation chain through feature importance visualization to form semantic association information of fused environmental features, passenger preference features and air conditioner control strategies; And processing the target data set based on the reinforcement learning model and combining the fusion semantic association information, and simulating temperature compensation effects and energy consumption changes under different passenger flows and environment change scenes by using a scene migration reasoning module to generate AI temperature compensation parameters and air conditioner energy-saving control instructions of time-sharing partitions.
2. The method for time-sharing partition AI temperature compensation and air-conditioning energy-saving control considering airport passenger comfort preference self-learning according to claim 1, wherein based on a passenger behavior track data set, a deep learning model is utilized to perform feature extraction on a passenger stay region, activity intensity and stay time, and in combination with a real-time environment parameter data set, passenger behavior semantic features are converted into comfort demand vectors through a cross-modal feature fusion module, preference levels of passenger comfort feedback data are quantized into weighted features, and a multidimensional passenger comfort preference data set facing temperature compensation and energy-saving control is generated, and comprises: Analyzing key behavior elements of the passenger behavior track data set, and extracting behavior characteristics including space coordinates of a stay region, activity intensity levels and stay time intervals; Performing environmental factor disassembly on the real-time environmental parameter data set to obtain environmental characteristics including temperature, humidity, wind speed and illumination intensity; performing preference semantic conversion on the passenger comfort feedback data set, mapping subjective evaluation into comfort preference grade and sensitivity coefficient, and generating preference characteristics; Performing association alignment on behavior features, environment features and preference features through a cross-modal feature fusion module to construct a multidimensional feature association matrix; Introducing a dynamic weight allocation mechanism to give priority to the fusion features to generate weighted features; And generating a standardized comfort demand vector based on the weighted feature matrix, integrating the quantized weighted features, and constructing a multidimensional passenger comfort preference data set oriented to temperature compensation and energy saving control.
3. The method for time-division and zoned AI temperature compensation and air-conditioning energy-saving control taking into account airport passenger comfort preference self-learning according to claim 2, wherein the multi-dimensional passenger comfort preference data set is processed and layered according to space zoning and time granularity, each layer comprises zone nodes, environment parameter nodes, passenger preference nodes and air-conditioning operation nodes, different levels of real-time features are aggregated through dynamic perception windows, and a target directed multi-view dynamic diagram composed of environment states, passenger preferences and air-conditioning operation states is generated, which comprises: Based on characteristic distribution characteristics of the multidimensional passenger comfort preference data set and airport space-time association rules, negotiating and determining a hierarchical aggregation strategy by a data processing module and a graph structure generation engine, and determining node levels and association dimensions by space partition attribute analysis and time granularity division; Splitting the multi-dimensional passenger comfort preference data set by adopting a partition-time sharing two-dimensional partitioning mechanism, extracting corresponding characteristic data according to the terminal building functional partition and time interval, associating the regional, environmental, preference and air conditioner operation core elements, and generating a preliminary layered data block; Performing validity check on the preliminary layered data block, starting region calibration service aiming at space boundary fuzzy data, triggering a time sequence synchronization mechanism on data with inconsistent time stamps, executing a feature completion flow on node feature missing samples, and generating an optimized layered data scheme comprising a region calibration scheme, a time sequence synchronization strategy and a feature completion method; and integrating and executing the optimized layered data scheme, capturing characteristic changes of all layers in real time by combining a dynamic sensing window, synchronously generating node association weights and characteristic aggregation progress feedback, constructing a layer network comprising regional nodes, environment parameter nodes, passenger preference nodes and air conditioner operation nodes, and forming a target directed multi-view dynamic diagram consisting of environment states, passenger preference and air conditioner operation states.
4. The method for time-sharing partition AI temperature compensation and air-conditioning energy-saving control considering airport passenger comfort preference self-learning according to claim 1, wherein modeling processing is performed on a target directed multi-view dynamic graph based on a self-adaptive multi-view dynamic graph neural network, a passenger preference self-learning unit is introduced, comfort preference weights under different scenes are dynamically updated, and node embedded information is generated, comprising: Carrying out structural analysis on the target directed multi-view dynamic graph, layering according to the space partition type and the time granularity level, and extracting regional node environment sensitivity, preference node demand intensity and air conditioner node regulation response speed dynamic characteristics from each layer to generate node characteristic subsets of each level; The regional nodes are used as core hinges, the environment parameter nodes are used as influence factor nodes, the passenger preference nodes are used as demand guide nodes, the air conditioner operation nodes are used as execution feedback nodes, the dynamic attribute information of the nodes is generated by combining space topology mapping and preference type identification, and the side weight association information is generated according to the parameter influence degree and the preference-regulation response time sequence relation; Constructing a self-adaptive directed graph comprising full area nodes, multi-dimensional environment nodes, differentiated preference nodes and intelligent regulation nodes, dividing the self-adaptive directed graph into 3 views according to comfort requirement constraint, energy consumption optimization constraint and real-time response constraint, optimizing cross-view edge association strength through preference-regulation and control cooperative attention mechanism, and dynamically updating scene preference weights by a passenger preference self-learning unit; and carrying out hierarchical feature fusion and dynamic evolution modeling on the optimized multi-constraint directed associated graph based on the self-adaptive multi-view dynamic graph neural network, capturing a dynamic interaction rule among nodes through the space-time attention module, and outputting node embedded information containing environment-preference-regulation associated semantics.
5. The method for time-sharing partition AI temperature compensation and air-conditioning energy-saving control considering airport passenger comfort preference self-learning according to claim 1, wherein node embedded information is injected into each layer of a value network and a strategy network of a reinforcement learning model, model parameters are optimized by adopting a parameter efficient fine tuning method, an energy consumption-comfort dual-objective antagonism optimization module is introduced, a decision interpretation chain is generated through feature importance visualization at the same time, and semantic association information of fused environmental features, passenger preference features and air-conditioning control strategies is formed, comprising: Based on the double-objective optimization and semantic fusion requirements, constructing a fusion framework comprising a node embedded cross-layer injection module, a reinforced learning value and a strategy double network, and carrying out deep injection and associated semantic modeling on environmental features, passenger preference features and air-conditioning control features; Based on energy consumption-comfort collaborative optimization strategy design feature injection logic, determining injection dimension and feature fusion duty ratio of nodes embedded in a value network value evaluation layer and a strategy network decision output layer, and generating a network-feature adaptation rule; Setting a dynamic weight balance mechanism by combining the adaptive requirements of airport passenger flow density, environment fluctuation range and energy-saving target threshold, improving comfortable characteristic weight of a high-density passenger flow scene, strengthening energy consumption optimization characteristic duty ratio of a low-load operation scene, and balancing double-target weight distribution of an extreme environment scene; Introducing an energy consumption-comfort dual-objective antagonism optimization module, iteratively optimizing network parameters by a LoRA parameter efficient fine-tuning method, and synchronously starting a feature importance visualization tool to generate a full-link interpretation chain comprising feature-decision-making-effect; And integrating and executing the fusion framework, the injection rule, the dynamic weight mechanism and the resistance optimization module, and outputting time-sharing partition temperature compensation and energy-saving control decision basic data containing environment-preference-regulation dynamic correlation semantics.
6. A time-division AI temperature compensation and air-conditioning energy-saving control apparatus that considers airport passenger comfort preference self-learning, characterized in that the apparatus is configured to execute any one of the time-division AI temperature compensation and air-conditioning energy-saving control methods that consider airport passenger comfort preference self-learning.
7. An electronic device, comprising: and a memory for storing executable instructions of the first processor; wherein the first processor is configured to execute the time-division AI temperature compensation and air-conditioning energy-saving control method of any one of claims 1-5, taking into account airport passenger comfort preference self-learning, via execution of the executable instructions.
8. A computing device comprising a memory for storing computer program instructions and a second processor for executing the computer program instructions, wherein the computer program instructions, when executed by the second processor, trigger the device to perform the time-division-AI temperature compensation and air-conditioning energy-saving control method of any of claims 1-5, taking into account airport passenger comfort preference self-learning.

Description

Time-sharing partition AI temperature compensation and air conditioner energy-saving control method and equipment considering airport passenger comfort preference self-learning Technical Field The invention relates to the technical field of data processing, in particular to a time-sharing partition AI temperature compensation and air conditioner energy-saving control method and equipment considering self-learning of airport passenger comfort preference. Background Airport terminal building is as the public building of big space, high people's stream, and its air conditioning system needs to satisfy passenger comfort demand and energy-conserving operation target simultaneously, is the regulation and control difficult problem that is recognized in the industry, and current prior art mainly has following problem: The existing airport air conditioner control is mostly dependent on fixed parameters or manual experience adjustment, and is difficult to adapt to space-time dynamic change characteristics of the terminal building. The passenger density in the areas such as security check channels, boarding gates, waiting areas and the like is large in difference, local cold energy demand fluctuation is remarkable, the fixed regulation and control mode is easy to cause high temperature (or low) in the high-density areas and waste of energy consumption in the low-density areas, the passenger flow and environment (such as sunlight intensity) in the flight peak/flat peak and daytime/night change severely in the time dimension, an obvious temperature adjustment hysteresis (such as long time after full starting can reach target temperature) exists in an air conditioning system, real-time demand cannot be accurately matched, and energy consumption of an air conditioner per person is high. Taking industry practice data as an example, under the traditional manual regulation mode, the energy consumption of the air conditioner of a part of airport terminal buildings per person can reach 1.004 kilowatt-hours/half year, and the passenger comfort satisfaction fluctuation is large. The current air conditioning control system mostly takes a fixed temperature threshold (such as uniformly setting 26 ℃) as a regulation target, and does not consider individual comfort preference differences of passengers. On one hand, the prior art does not effectively integrate passenger behavior track (such as stay area, activity intensity and stay time) and subjective feedback (such as comfort level evaluation) data, passenger comfort demand weights under different scenes (such as peak passenger flow and extreme weather) cannot be quantified, and on the other hand, a dynamic self-learning mechanism is lacking, iterative optimization control strategies are difficult to optimize according to passenger feedback and environmental changes, so that the adaptability of a regulation scheme is poor, for example, the temperature preference difference between an old passenger gathering area and a young passenger gathering area cannot be identified, and the overall comfort experience is further reduced. In the prior art, local control is performed on air conditioning unit (such as a water chilling unit and a fan coil), and an environment-passenger-air conditioner multi-dimensional collaborative optimization system is not constructed. Firstly, the data of environmental parameters (temperature, humidity, wind speed), passenger behavior characteristics, air conditioner running states and the like are not deeply fused, and a global optimal control strategy is difficult to generate, secondly, the control algorithm lacks scene migration capability, cannot simulate the temperature compensation effect and energy consumption change under unconventional scenes such as passenger flow mutation, extreme weather and the like, has limited regulation precision (if the temperature prediction error is often more than +/-1 ℃), thirdly, the existing scheme is mostly a 'black box' decision, visual interpretation of feature importance and decision logic is not provided, and operation and maintenance personnel are difficult to trace back to regulate and control basis, so that fault investigation and strategy optimization are not facilitated. The existing airport air conditioner energy-saving control technology is mostly single-point innovation, and a closed loop system of 'demand prediction, supply regulation and control and effect evaluation' is not formed. A part of schemes are introduced into a simple prediction model, but sub-algorithms such as multi-region room temperature prediction, dynamic energy consumption trend prediction and the like are not integrated, prediction accuracy is not enough, and an energy consumption standard model (such as peak-valley period division and equipment energy efficiency threshold setting) is not available, so that energy saving effect and reasonable energy consumption boundary cannot be quantified, scientific decision basis is not available when t