Thermal comfort control method and device based on visual perception and global sensitivity analysis

Abstract

The present invention relates to the technical field of intelligent building environment control and energy management, and specifically to a thermal comfort control method and device based on visual perception and global sensitivity analysis. An attribute recognition model based on the collaborative optimization of a lightweight visual perception network and an attention mechanism is constructed, and high-precision dynamic capture of human attributes is achieved through the model; and based on the recognized human attributes, human metabolic rate information is calculated, and then a thermal comfort model based on global sensitivity analysis and a hybrid intelligent algorithm is constructed. The human thermal comfort is calculated through the model, and finally the operating parameters of the air conditioner or heat pump are adjusted based on the human thermal comfort, which effectively solves the contradiction between real-time response of environmental parameters and energy consumption optimization, and provides an innovative solution for building energy saving and personalized thermal comfort management.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent building environment control and energy management, and in particular to a thermal comfort control method and device based on visual perception and global sensitivity analysis. Background Art

[0002] With the rapid development of the global economy, people are focusing on energy conservation, emission reduction, and green and low-carbon living, while also placing higher demands on the performance of air conditioning systems and the comfort of their living environments. Today's air conditioning systems must conserve energy, protect the environment, and create a pleasant indoor living environment. Therefore, energy-saving design and operational control of air conditioning systems are particularly important.

[0003] With the deepening of the concepts of smart buildings and healthy living, precise control of indoor environments faces multiple challenges: 1) Most data on environmental variables and human body variables of subjects and venues are collected through various monitoring devices. However, the method of wearing monitoring devices is not easy to promote, and traditional convolutional networks cannot balance fine-grained feature extraction and computational efficiency, resulting in low recognition accuracy of key attributes such as clothing thermal resistance; 2) Thermal comfort assessment relies on empirical formulas, ignoring the coupling effect of multiple factors, making it difficult to quantify the nonlinear impact of variables such as season and metabolic rate on thermal perception, and there is a lag in response to individual differences and environmental mutations; 3) Energy consumption control strategies are single and lack dynamic adjustment capabilities based on real-time thermal comfort feedback, making it difficult to balance comfort and energy efficiency. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to propose a thermal comfort control method and device based on visual perception and global sensitivity analysis to solve the above technical problems.

[0005] Based on the above objectives, the present invention provides a thermal comfort control method based on visual perception and global sensitivity analysis, comprising:

[0006] Obtain environmental information and indoor character image information;

[0007] Inputting the indoor person image information into a pre-trained attribute recognition model to identify human body attribute information;

[0008] Calculating metabolic rate information of a person in the indoor person image information based on the human body attribute information;

[0009] Inputting the environmental information, human attribute information, and metabolic rate information into a pre-trained thermal comfort model to calculate human thermal comfort;

[0010] Based on the calculation result of the human body thermal comfort, the operating parameters of the air conditioner or heat pump are adjusted.

[0011] As an optional implementation manner, the environmental information includes indoor temperature information, outdoor temperature information, indoor humidity information, and indoor wind speed information;

[0012] The human attribute information includes gender information, age information, movement posture information, and clothing thermal resistance information;

[0013] The metabolic rate information of the person in the indoor person image information is identified based on the gender information, age information and motion posture information.

[0014] As an optional implementation, the training process of the attribute recognition model is as follows:

[0015] Acquire multi-dimensional human image data based on standardized indoor office scenes;

[0016] Preprocessing the multi-dimensional human body image data;

[0017] Construct an initial attribute recognition model with PP-LCNet as the backbone network;

[0018] The pre-processed multi-dimensional human image data is input into the initial attribute recognition model for convolution calculation to obtain a feature map;

[0019] The feature map is input into a channel attention module and a spatial attention module for calculation, wherein the channel attention module is used to enhance the clothing texture features in the feature map, and the spatial attention module is used to enhance the key areas of the character's body in the recognition feature map;

[0020] The feature maps after strengthening the clothing texture features and strengthening the recognition of the key areas of the human body are input into the average pooling layer for average pooling, and then the feature maps after average pooling are input into the fully connected layer for expansion to obtain the expanded feature maps;

[0021] Input the expanded feature map into the classifier to obtain image data with human attribute labels;

[0022] Calculate the loss function and repeat the above training process until the calculation result of the loss function reaches the threshold, that is, the model converges, and a trained attribute recognition model is obtained.

[0023] As an optional implementation, the preprocessing includes normalization, horizontal flipping, and random cropping.

[0024] As an optional implementation, the training process of the thermal comfort model is:

[0025] Acquire a historical thermal comfort dataset, wherein the historical thermal comfort dataset includes historical gender information, historical age information, historical clothing thermal resistance information, historical metabolic rate information, historical indoor temperature information, historical outdoor temperature information, historical indoor humidity information, and historical indoor wind speed information;

[0026] Preprocessing the data in the historical thermal comfort dataset;

[0027] An initial thermal comfort model based on an FNN network is constructed, wherein the initial thermal comfort model uses gender, age, clothing thermal resistance, metabolic rate, indoor air temperature, outdoor temperature, relative humidity, and wind speed as input feature parameters and uses a thermal comfort target as an output feature parameter;

[0028] The preprocessed thermal comfort data set is input into the initial thermal comfort model for training, and the model parameters are adjusted until the model converges to obtain a trained thermal comfort model.

[0029] As an optional implementation manner, the preprocessing of the data in the historical thermal comfort dataset includes:

[0030] Remove outliers, fill in missing data, and process non-numeric labeled columns with one-hot encoding.

[0031] As an optional implementation, the method for selecting the input characteristic parameters of the thermal comfort model is:

[0032] Acquire a historical data set, where the historical data set includes historical season information season, historical age information age, historical gender information gender, historical personnel height information ht, historical weight information wt, historical indoor temperature information ta, historical working temperature information top, historical radiation temperature information tr, historical outdoor temperature information tg, historical indoor humidity information rh, historical indoor wind speed information vel, historical human metabolic rate information met, historical clothing thermal resistance information clo, historical outdoor air temperature information t_out, historical outdoor humidity information rh_out, historical fan status information fan, historical window status information window, and historical door status information door;

[0033] Preprocessing the data information in the historical data set;

[0034] An initial Cubist model is constructed, wherein the initial Cubist model uses season information, age information, gender information, person height information, weight information, indoor temperature information, working temperature information, radiant temperature information, outdoor temperature information, indoor humidity information, indoor wind speed information, human metabolic rate information, clothing thermal resistance information, outdoor air temperature information, outdoor humidity information, fan status information, window status information, and door status information as input feature parameters, and uses thermal comfort thermal_sensation as an output feature parameter;

[0035] Input the preprocessed historical data set into the initial Cubist model to train the model, adjust the model parameters until the model converges, and obtain the trained Cubist model;

[0036] A Sobol sequence sampling method is used to extract multiple sets of feature sample combinations from the historical thermal comfort dataset;

[0037] Inputting multiple groups of feature sample combinations into the trained Cubist model to obtain thermal comfort prediction values ​​output by the model;

[0038] According to the variance decomposition theory, based on each thermal comfort prediction value output by the model, the main effect index S1 and the total effect index ST of the corresponding characteristic sample combination are calculated, and the influence of various factors on thermal comfort is evaluated based on the calculation results;

[0039] Based on the evaluation results, gender, age, clothing thermal resistance, metabolic rate, indoor air temperature, outdoor temperature, relative humidity, and wind speed were selected as input characteristic parameters of the thermal comfort model.

[0040] As an optional implementation manner, the preprocessing of the data information in the historical data set includes:

[0041] For the missing historical data sets, the historical indoor temperature information ta, historical working temperature information top, historical radiation temperature information tr, historical outdoor temperature information tg, historical indoor humidity information rh, historical outdoor air temperature information t_out, historical outdoor humidity information rh_out, historical clothing thermal resistance information clo, historical fan status information fan, historical window status information window, historical door status information door, and historical season information season are filled in by mode according to the season;

[0042] The historical gender information in the historical dataset is processed using one-hot encoding;

[0043] The missing historical age information age, historical personnel height information ht, historical weight information wt, historical human metabolic rate information met, and historical indoor wind speed information vel in the historical data set are filled with average values.

[0044] As an optional implementation manner, adjusting the operating parameters of the air conditioner or heat pump based on the calculation result of the human body thermal comfort includes:

[0045] subtracting an actual thermal comfort value obtained by calculation based on the human body thermal comfort from a target thermal comfort value;

[0046] If the difference is outside the preset range, adjust the operating parameters of the air conditioner or heat pump to make the actual thermal comfort value closer to the target thermal comfort value.

[0047] Corresponding to the thermal comfort control method, the present invention further provides a thermal comfort control device based on visual perception and global sensitivity analysis, comprising:

[0048] Acquisition module, used to obtain environmental information and indoor character image information;

[0049] A recognition module, configured to input the indoor person image information into a pre-trained attribute recognition model to identify human attribute information;

[0050] a first calculation module, configured to calculate metabolic rate information of a person in the indoor person image information based on the human body attribute information;

[0051] A second calculation module is used to input the environmental information, human attribute information and metabolic rate information into a pre-trained thermal comfort model to calculate human thermal comfort;

[0052] The adjustment module is used to adjust the operating parameters of the air conditioner or heat pump based on the calculation result of the human body thermal comfort.

[0053] The beneficial effects of the present invention are as follows: the present invention constructs an attribute recognition model based on the collaborative optimization of a lightweight visual perception network and an attention mechanism, and realizes high-precision dynamic capture of human attributes through the model; and based on the identified human attributes, calculates the human metabolic rate information, and then constructs a thermal comfort model based on global sensitivity analysis and a hybrid intelligent algorithm, and calculates the human thermal comfort through the model, and finally adjusts the working parameters of the air conditioner or heat pump based on the human thermal comfort, effectively solving the contradiction between real-time response of environmental parameters and energy consumption optimization, and providing innovative solutions for building energy conservation and personalized thermal comfort management. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0055] Figure 1 This is a flow chart of a thermal comfort control method according to an embodiment of the present invention;

[0056] Figure 2 This is the result calculated through sensitivity analysis in the embodiment of the present invention;

[0057] Figure 3 This is a flow chart of a thermal comfort control device according to an embodiment of the present invention;

[0058] Figure 4 A schematic diagram of a device according to a specific embodiment of the present invention;

[0059] Figure 5 A building plan diagram of a specific embodiment of the present invention;

[0060] Figure 6 A diagram of a building physical model according to a specific embodiment of the present invention;

[0061] Figure 7 2 is a comparison chart of energy consumption between the embodiment of the present invention and the prior art. DETAILED DESCRIPTION

[0062] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to specific embodiments.

[0063] It should be noted that, unless otherwise defined, the technical or scientific terms used in the present invention should have the usual meanings understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0064] As an embodiment of the present invention, a thermal comfort control method based on visual perception and global sensitivity analysis is provided, comprising:

[0065] Obtain environmental information and indoor character image information;

[0066] Inputting the indoor person image information into a pre-trained attribute recognition model to identify human body attribute information;

[0067] Calculating metabolic rate information of a person in the indoor person image information based on the human body attribute information;

[0068] Inputting the environmental information, human attribute information, and metabolic rate information into a pre-trained thermal comfort model to calculate human thermal comfort;

[0069] Based on the calculation result of the human body thermal comfort, the operating parameters of the air conditioner or heat pump are adjusted.

[0070] The present invention constructs an attribute recognition model based on the collaborative optimization of a lightweight visual perception network and an attention mechanism, through which high-precision dynamic capture of human attributes is achieved; and based on the identified human attributes, the human metabolic rate information is calculated, and then a thermal comfort model based on global sensitivity analysis and a hybrid intelligent algorithm is constructed. The human thermal comfort is calculated through this model, and finally the operating parameters of the air conditioner or heat pump are adjusted based on the human thermal comfort, which effectively solves the contradiction between real-time response of environmental parameters and energy consumption optimization, and provides innovative solutions for building energy conservation and personalized thermal comfort management.

[0071] As a specific embodiment of the present invention, Figure 1 As shown in FIG, a thermal comfort control method based on visual perception and global sensitivity analysis is provided, including:

[0072] S100, obtaining environmental information and indoor person image information;

[0073] The environmental information includes indoor temperature information, outdoor temperature information, indoor humidity information, and indoor wind speed information.

[0074] S200, inputting the indoor person image information into a pre-trained attribute recognition model to identify human attribute information;

[0075] The human body attribute information includes gender information, age information, movement posture information, and clothing thermal resistance information.

[0076] S300, calculating metabolic rate information of the person in the indoor person image information based on the human body attribute information;

[0077] Wherein, the metabolic rate information of the person in the indoor person image information is identified based on the gender information, age information and movement posture information.

[0078] S400, inputting the environmental information, human attribute information, and metabolic rate information into a pre-trained thermal comfort model to calculate human thermal comfort;

[0079] S500: Adjust operating parameters of an air conditioner or a heat pump based on the calculation result of the human body thermal comfort.

[0080] As an optional implementation, the training process of the attribute recognition model is as follows:

[0081] Acquire multi-dimensional human image data based on standardized indoor office scenes;

[0082] Preprocessing the multi-dimensional human image data, optionally including normalization, horizontal flipping, and random cropping, to ensure the model's robust recognition capability for complex human attributes;

[0083] Construct an initial attribute recognition model with PP-LCNet as the backbone network;

[0084] The pre-processed multi-dimensional human image data is input into the initial attribute recognition model for convolution calculation to obtain a feature map;

[0085] The feature map is input into a channel attention module and a spatial attention module for calculation, wherein the channel attention module is used to enhance the clothing texture features in the feature map, and the spatial attention module is used to enhance the key areas of the character's body in the recognition feature map;

[0086] The feature maps after strengthening the clothing texture features and strengthening the recognition of the key areas of the human body are input into the average pooling layer for average pooling, and then the feature maps after average pooling are input into the fully connected layer for expansion to obtain the expanded feature maps;

[0087] Input the expanded feature map into the classifier to obtain image data with human attribute labels;

[0088] Calculate the loss function and repeat the above training process until the calculation result of the loss function reaches the threshold, that is, the model converges, and a trained attribute recognition model is obtained.

[0089] Optionally, the dataset constructed in this paper is based on a standardized indoor office scene, using cameras to capture multi-dimensional human images. The experimental sample covers more than 30 subjects. Through a multi-batch combined experimental design, we systematically collect human image data of different genders, multiple postures (turning, standing, and sitting), and variable clothing thermal resistance states (including the act of putting on and taking off down jackets). To enhance data representation capabilities, standardized motion paradigms (including axial rotations and body transformations) are implemented during the construction process, and data augmentation strategies such as random cropping are used to ensure the model's robust recognition of complex human attributes.

[0090] The annotation format used is the same as that of the PA100k (pedestrian attribute dataset), with the image name on the left and attribute features on the right, represented by a sequence of 0 / 1 numbers. The dataset is divided into training, validation, and test sets in a 6:2:2 ratio.

[0091] The embodiment of the present invention uses the Baidu Feipang platform to train a deep learning model, adopts PP-LCNet as the backbone network to construct an attribute recognition model, sets the learning rate to 0.001, and the number of iterations to 50. This not only retains the high-precision characteristics of the currently popular backbone feature extraction network, but also reduces the amount of computation. At the same time, a CBAM attention network that combines convolution and attention mechanisms is used to analyze images from both spatial and channel perspectives. Spatial attention focuses on key areas of the human body, and channel attention strengthens clothing texture features. By using this attention network to actively learn important information from low-level features and integrate it with high-level features, the model can simultaneously obtain low-level fine-grained information and high-level semantic information, thereby improving accuracy.

[0092] For ease of understanding, the following provides a specific example to illustrate the specific training process of the attribute recognition model:

[0093] A Hikvision network camera (model DS-2CD3386FWDA4-LS) was used to capture RGB images of people indoors. The images were normalized to a tensor format (244*244, 3 channels). Data augmentation was then performed using horizontal flipping and random cropping to improve model robustness.

[0094] The input image passes through the first layer of 3*3 standard convolution (Stride=2), and the output feature tensor is 112*112*16, which extracts low-order edge and texture features;

[0095] The feature map passes through three depth-wise separable convolution modules (each module has two convolution layers) in sequence. Each depth-wise separable convolution kernel is 3*3, and the number of channels is doubled in sequence. At this time, the feature image size is 14*14*256;

[0096] Entering the last depth-wise separable convolution module, the convolution kernel size is 5*5, the step size is 2, and the feature map size is 7*7*512;

[0097] The CBAM attention module is added, and the feature maps are fed into the channel attention module and the spatial attention module respectively. In the channel attention module, the weights along the channel dimension are learned. First, the spatial dimension is compressed to generate a channel description vector with a feature map size of 1*1*512. The maximum activation value of each channel is then extracted. The two pooling results are input into the same multi-layer perceptron (MLP) and a sigmoid activation function is used to obtain channel weights, thereby enhancing clothing texture features. The spatial attention module compresses the feature information into a 7*7*1 matrix. After max pooling, the two pooling results are concatenated to obtain spatial attention weights, enhancing the ability to identify key body regions.

[0098] The image information after attention enhancement is input into the average pooling layer and then expanded into the fully connected layer. At this time, the size is 1*1*1280;

[0099] Finally, the classifier (Softmax classification layer) outputs the human attribute label;

[0100] Calculate the loss function, use the weighted cross entropy loss function, and then perform backpropagation to calculate the gradient of each layer and update the weight information until the model converges.

[0101] Through the above method, the neural network converged quickly within 50 iterations, the training loss curve decreased smoothly, and the accuracy of the validation set steadily improved; the final model achieved a model accuracy of f1: 0.96547, acc: 0.7818, prec: 0.96846, and recall: 0.96249, meeting the high-precision requirements of the building energy-saving system.

[0102] As an optional implementation, the training process of the thermal comfort model is:

[0103] Acquire a historical thermal comfort dataset, wherein the historical thermal comfort dataset includes historical gender information, historical age information, historical clothing thermal resistance information, historical metabolic rate information, historical indoor temperature information, historical outdoor temperature information, historical indoor humidity information, and historical indoor wind speed information;

[0104] Preprocessing the data in the historical thermal comfort dataset includes: removing outliers, filling missing data, and processing non-numeric labeled columns with one-hot encoding;

[0105] An initial thermal comfort model based on an FNN network is constructed, wherein the initial thermal comfort model uses gender, age, clothing thermal resistance, metabolic rate, indoor air temperature, outdoor temperature, relative humidity, and wind speed as input feature parameters and uses a thermal comfort target as an output feature parameter;

[0106] The preprocessed thermal comfort data set is input into the initial thermal comfort model for training, and the model parameters are adjusted until the model converges to obtain a trained thermal comfort model.

[0107] This paper constructs a thermal comfort prediction model based on a feedforward neural network (FNN) and uses the ASHRAE GTDB-II standard dataset for training. A hybrid intelligent optimization strategy is designed for the algorithm optimization phase: The particle swarm optimization (PSO) algorithm is used in the early stages of training to achieve rapid convergence, given its advantages of fast convergence, strong local optimization capabilities, and simple parameter adjustment. The gravitational search algorithm (GSA) is introduced in the later stages of training to perform a refined search, given its strong global exploration capabilities and high solution space traversal. The synergistic effect of these two optimization algorithms significantly improves the FNN model's prediction accuracy for indoor thermal comfort.

[0108] For ease of understanding, the following provides a specific example to illustrate the specific training process of the thermal comfort model:

[0109] Acquire a historical thermal comfort dataset, wherein the historical thermal comfort dataset includes historical gender information, historical age information, historical clothing thermal resistance information, historical metabolic rate information, historical indoor temperature information, historical outdoor temperature information, historical indoor humidity information, and historical indoor wind speed information;

[0110] The data in the dataset was preprocessed, including removing outliers, filling missing data, and using one-hot encoding for columns with non-numeric labels. A total of eight parameters were selected as input features: gender, age, clothing thermal resistance, metabolic rate, indoor air temperature, outdoor temperature, relative humidity, and wind speed. The thermal comfort target was used as the output, with seven categories: -3, -2, -1, 0, 1, 2, and 3.

[0111] The input features of the dataset are normalized using Min-Max, and the dataset is divided into training set, validation set, and test set according to the ratio of 6:2:2.

[0112] Construct an FNN network with 8 input nodes and 7 output nodes, and use the Softmax function to output probability distribution;

[0113] In the middle hidden layer design, 3 hidden layers are selected, with 20 neurons in each layer, the ReLU activation function is used, and the cross entropy loss function is used as the loss function.

[0114] All weights and biases of FNN are expanded into vectors. Each particle represents a set of network parameters. The loss function of the validation set is used as the fitness value function of the optimization algorithm to guide particle updates.

[0115] Initialize the FNN network weight parameters and input data into the FNN network; calculate the network output and loss function; update the particle speed and position according to the optimization algorithm; repeat this process until the maximum number of iterations, and finally obtain a model prediction accuracy of 0.88, which meets the actual application requirements.

[0116] As an optional implementation, the method for selecting the input characteristic parameters of the thermal comfort model is:

[0117] Acquire a historical data set, where the historical data set includes historical season information season, historical age information age, historical gender information gender, historical personnel height information ht, historical weight information wt, historical indoor temperature information ta, historical working temperature information top, historical radiation temperature information tr, historical outdoor temperature information tg, historical indoor humidity information rh, historical indoor wind speed information vel, historical human metabolic rate information met, historical clothing thermal resistance information clo, historical outdoor air temperature information t_out, historical outdoor humidity information rh_out, historical fan status information fan, historical window status information window, and historical door status information door;

[0118] Preprocessing the data information in the historical data set includes: performing mode filling according to season for the missing historical indoor temperature information ta, historical working temperature information top, historical radiation temperature information tr, historical outdoor temperature information tg, historical indoor humidity information rh, historical outdoor air temperature information t_out, historical outdoor humidity information rh_out, historical clothing thermal resistance information clo, historical fan status information fan, historical window status information window, historical door status information door, and historical season information season in the historical data set;

[0119] The historical gender information in the historical dataset is processed using one-hot encoding;

[0120] The missing historical age information age, historical personnel height information ht, historical weight information wt, historical human metabolic rate information met, and historical indoor wind speed information vel in the historical data set are filled with average values.

[0121] An initial Cubist model is constructed, wherein the initial Cubist model uses season information, age information, gender information, person height information, weight information, indoor temperature information, working temperature information, radiant temperature information, outdoor temperature information, indoor humidity information, indoor wind speed information, human metabolic rate information, clothing thermal resistance information, outdoor air temperature information, outdoor humidity information, fan status information, window status information, and door status information as input feature parameters, and uses thermal comfort thermal_sensation as an output feature parameter;

[0122] Input the preprocessed historical data set into the initial Cubist model to train the model, adjust the model parameters until the model converges, and obtain the trained Cubist model;

[0123] A Sobol sequence sampling method is used to extract multiple sets of feature sample combinations from the historical thermal comfort dataset;

[0124] Inputting multiple groups of feature sample combinations into the trained Cubist model to obtain thermal comfort prediction values ​​output by the model;

[0125] According to the variance decomposition theory, based on each thermal comfort prediction value output by the model, the main effect index S1 and the total effect index ST of the corresponding characteristic sample combination are calculated, and the influence of various factors on thermal comfort is evaluated based on the calculation results;

[0126] Based on the evaluation results, gender, age, clothing thermal resistance, metabolic rate, indoor air temperature, outdoor temperature, relative humidity, and wind speed were selected as input characteristic parameters of the thermal comfort model.

[0127] The present invention discloses a thermal comfort assessment method based on the ASHRAE GTDB-II dataset. Through data preprocessing (mode / mean filling, one-hot encoding), Cubist model construction and parameter optimization (MSE=1.03), combined with Sobol global sensitivity analysis (1000 samples), key influencing factors such as season, temperature and humidity are determined, thus achieving accurate prediction and control of the building thermal environment.

[0128] For ease of understanding, a specific embodiment is provided below to illustrate a method for selecting input characteristic parameters of a thermal comfort model:

[0129] The historical dataset obtained is the ASHRAE GTDB-II (ASHRAE Global Thermal Comfort Database II). The historical dataset includes historical season information season, historical age information age, historical gender information gender, historical personnel height information ht, historical weight information wt, historical indoor temperature information ta, historical working temperature information top, historical radiation temperature information tr, historical outdoor temperature information tg, historical indoor humidity information rh, historical indoor wind speed information vel, historical human metabolic rate information met, historical clothing thermal resistance information clo, historical outdoor air temperature information t_out, historical outdoor humidity information rh_out, historical fan status information fan, historical window status information window, and historical door status information door. tg is the local temperature obtained by the dataset creator from the meteorological bureau, and t_out is the temperature obtained by the dataset creator using a sensor installed outdoors.

[0130] Preprocess the data and fill in missing data. For data such as temperature, humidity, and clothing thermal resistance, fill in the mode according to season. For non-numeric gender columns, use one-hot encoding. For data unrelated to climate and season, such as age, height, weight, and metabolic rate, fill in the mean.

[0131] An initial Cubist model is constructed, wherein the initial Cubist model uses season information, age information, gender information, person height information, weight information, indoor temperature information, working temperature information, radiant temperature information, outdoor temperature information, indoor humidity information, indoor wind speed information, human metabolic rate information, clothing thermal resistance information, outdoor air temperature information, outdoor humidity information, fan status information, window status information, and door status information as input feature parameters, and uses thermal comfort thermal_sensation as an output feature parameter;

[0132] The preprocessed data is divided into training set and test set with an 8:2 ratio.

[0133] The training set was used to train the initial Cubist model, and model parameters were adjusted to optimize model performance. Cross-validation was used to determine the optimal parameter combination, resulting in a good model fit. The trained Cubist model was evaluated using the test set data, and the mean squared error (MSE) was calculated as the evaluation metric. The final result was 1.03288, demonstrating high model accuracy.

[0134] The constructed Cubist model was used to perform the Sobol global sensitivity analysis, where committees was set to 1, neighbors was set to 0, rules was set to 99, maxdepth was set to 1, and mincases was set to 10. Sobol sequence sampling was used to generate 1000 input feature sample combinations for subsequent sensitivity analysis calculations. For each input feature sample combination obtained by sampling, the corresponding output prediction value was calculated using the Cubist model. According to the variance decomposition theory, the main effect index (S1) and total effect index (ST) of each input feature were calculated.

[0135] The calculation results are as follows Figure 2 As shown in the figure, the total effect and main effect of the eight factors of gender, age, clothing thermal resistance, metabolic rate, indoor air temperature, outdoor temperature, indoor humidity, and indoor wind speed are greater than those of other factors. Therefore, gender, age, clothing thermal resistance, metabolic rate, indoor air temperature, outdoor temperature, indoor humidity, and indoor wind speed are selected as the calculation parameters of the thermal comfort model.

[0136] As an optional implementation manner, adjusting the operating parameters of the air conditioner or heat pump based on the calculation result of the human body thermal comfort includes:

[0137] subtracting an actual thermal comfort value obtained by calculation based on the human body thermal comfort from a target thermal comfort value;

[0138] If the difference is outside the preset range, adjust the operating parameters of the air conditioner or heat pump to make the actual thermal comfort value closer to the target thermal comfort value.

[0139] Corresponding to the thermal comfort control method, the present invention also provides a thermal comfort control device based on visual perception and global sensitivity analysis, such as Figure 3 Shown, including:

[0140] Acquisition module 10, used to obtain environmental information and indoor person image information;

[0141] The recognition module 20 is used to input the indoor person image information into a pre-trained attribute recognition model to identify human attribute information;

[0142] A first calculation module 30 is configured to calculate metabolic rate information of a person in the indoor person image information based on the human body attribute information;

[0143] A second calculation module 40 is configured to input the environmental information, human attribute information, and metabolic rate information into a pre-trained thermal comfort model to calculate human thermal comfort;

[0144] The adjustment module 50 is used to adjust the operating parameters of the air conditioner or heat pump based on the calculation result of the human body thermal comfort.

[0145] Those skilled in the art should understand that the technical effect of the thermal comfort control device is the same as that of the thermal comfort control method, which will not be described in detail here.

[0146] like Figure 4 As shown, an embodiment of the present invention also provides a schematic diagram of a device for applying this solution, in which a camera and a temperature and humidity sensor are used to collect data, and then the algorithm of the embodiment of the present invention is used to process the data.

[0147] In order to facilitate those skilled in the art to understand the technical solution of the present invention, an embodiment with a specific application scenario is provided below. Example

[0148] The research object is the first floor lobby of a university building in Tianjin. Its floor plan is as follows: Figure 5 As shown in Table 1, the dimensions of the building rooms are obtained based on the floor plan.

[0149] Table 1 Dimensions of building rooms

[0150]

[0151] The hall is connected to the kitchen, equipment room, bathroom, and an office. There is a staircase leading to the second floor in the northwest corner of the room. There are four independent fan coil units (heat pump output) in the hall. The fan coil units are 0.90 m × 0.50 m × 0.20 m in size, and their bottoms are 0.3 m from the ground. The four fan coil units are located on the south side of the two windows on the east wall, on the west side of the windows on the south wall, and on the west wall of the hall near the entrance. When the fan coil unit is turned on, the air supply outlet size is 0.90 m × 0.10 m. According to the actual situation of the fan coil unit blades, the air supply angle is set to 45° upward. The fan coil unit return air outlet is located at the bottom and measures 0.90 m × 0.10 m. Opposite the kitchen is a conference table with dimensions of 4.50 m × 3.00 m × 1.40 m. To facilitate the experiment, the staircase is simplified as a wall. The space connecting the equipment room, bathroom, office and hall is not built in the model. The physical model of the office is as follows: Figure 6 shown.

[0152] This experiment was conducted in the building. Rooftop surveillance cameras were used to collect real-time occupant attribute information, while temperature sensors distributed indoors and outdoors were used to obtain environmental parameters. Three students were invited to participate in the experiment, which was conducted over two days (outdoor weather conditions remained essentially the same during the experiment). The daily monitoring period was from 10:00 AM to 6:00 PM. On the first day, the natural state operation mode was adopted (meaning that the heat pump and air conditioner were not controlled by the present invention, and the indoor occupants remotely controlled the air conditioner and heat pump temperature according to their daily living habits). The system did not apply any control strategy. On the second day, the intelligent control system of the present invention was activated. Experimental data showed that ( Figure 7 In both operating modes, the system's hourly energy consumption was significantly lower than its natural state. Quantitative analysis showed that the intelligent control system reduced daily energy consumption by 5.6%, demonstrating the effectiveness of this invention in improving building energy conservation.

[0153] It should be understood by those skilled in the art that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present invention (including the claims) is limited to these examples. Within the scope of the present invention, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of the different aspects of the present invention as described above, which are not provided in detail for the sake of simplicity.

[0154] The present invention is intended to cover all such substitutions, modifications and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A thermal comfort control method based on visual perception and global sensitivity analysis, characterized in that: include: Obtain environmental information and indoor character image information; Inputting the indoor person image information into a pre-trained attribute recognition model to identify human body attribute information; Calculating metabolic rate information of a person in the indoor person image information based on the human body attribute information; Inputting the environmental information, human attribute information, and metabolic rate information into a pre-trained thermal comfort model to calculate human thermal comfort; Adjusting operating parameters of an air conditioner or heat pump based on the calculation result of the human body thermal comfort; The training process of the attribute recognition model is as follows: Acquire multi-dimensional human image data based on standardized indoor office scenes; Preprocessing the multi-dimensional human body image data; Construct an initial attribute recognition model with PP-LCNet as the backbone network; The pre-processed multi-dimensional human image data is input into the initial attribute recognition model for convolution calculation to obtain a feature map; The feature map is input into a channel attention module and a spatial attention module for calculation, wherein the channel attention module is used to enhance the clothing texture features in the feature map, and the spatial attention module is used to enhance the key areas of the character's body in the recognition feature map; The feature maps after strengthening the clothing texture features and strengthening the recognition of the key areas of the human body are input into the average pooling layer for average pooling, and then the feature maps after average pooling are input into the fully connected layer for expansion to obtain the expanded feature maps; Input the expanded feature map into the classifier to obtain image data with human attribute labels; Calculate the loss function and repeat the above training process until the calculation result of the loss function reaches the threshold, that is, the model converges, and a trained attribute recognition model is obtained; The environmental information includes indoor temperature information, outdoor temperature information, indoor humidity information, and indoor wind speed information; The human attribute information includes gender information, age information, movement posture information, and clothing thermal resistance information; Identifying metabolic rate information of a person in the indoor person image information based on the gender information, age information, and motion posture information; The training process of the thermal comfort model is as follows: Acquire a historical thermal comfort dataset, wherein the historical thermal comfort dataset includes historical gender information, historical age information, historical clothing thermal resistance information, historical metabolic rate information, historical indoor temperature information, historical outdoor temperature information, historical indoor humidity information, and historical indoor wind speed information; Preprocessing the data in the historical thermal comfort dataset; An initial thermal comfort model based on an FNN network is constructed, wherein the initial thermal comfort model uses gender, age, clothing thermal resistance, metabolic rate, indoor air temperature, outdoor temperature, relative humidity, and wind speed as input feature parameters and uses a thermal comfort target as an output feature parameter; The preprocessed thermal comfort data set is input into the initial thermal comfort model for training, and the model parameters are adjusted until the model converges to obtain a trained thermal comfort model; The method for selecting the input characteristic parameters of the thermal comfort model is as follows: Acquire a historical data set, where the historical data set includes historical season information season, historical age information age, historical gender information gender, historical personnel height information ht, historical weight information wt, historical indoor temperature information ta, historical working temperature information top, historical radiation temperature information tr, historical outdoor temperature information tg, historical indoor humidity information rh, historical indoor wind speed information vel, historical human metabolic rate information met, historical clothing thermal resistance information clo, historical outdoor air temperature information t_out, historical outdoor humidity information rh_out, historical fan status information fan, historical window status information window, and historical door status information door; Preprocessing the data information in the historical data set; An initial Cubist model is constructed, wherein the initial Cubist model uses season information, age information, gender information, person height information, weight information, indoor temperature information, working temperature information, radiant temperature information, outdoor temperature information, indoor humidity information, indoor wind speed information, human metabolic rate information, clothing thermal resistance information, outdoor air temperature information, outdoor humidity information, fan status information, window status information, and door status information as input feature parameters, and uses thermal comfort thermal_sensation as an output feature parameter; Input the preprocessed historical data set into the initial Cubist model to train the model, adjust the model parameters until the model converges, and obtain the trained Cubist model; A Sobol sequence sampling method is used to extract multiple sets of feature sample combinations from the historical thermal comfort dataset; Inputting multiple groups of feature sample combinations into the trained Cubist model to obtain thermal comfort prediction values ​​output by the model; According to the variance decomposition theory, based on each thermal comfort prediction value output by the model, the main effect index S1 and the total effect index ST of the corresponding characteristic sample combination are calculated, and the influence of various factors on thermal comfort is evaluated based on the calculation results; Based on the evaluation results, gender, age, clothing thermal resistance, metabolic rate, indoor air temperature, outdoor temperature, relative humidity, and wind speed were selected as input characteristic parameters of the thermal comfort model.

2. The thermal comfort control method based on visual perception and global sensitivity analysis according to claim 1 is characterized in that: The preprocessing includes normalization, horizontal flipping, and random cropping.

3. The thermal comfort control method based on visual perception and global sensitivity analysis according to claim 1 is characterized in that: The preprocessing of the data in the historical thermal comfort dataset includes: Remove outliers, fill in missing data, and process non-numeric labeled columns with one-hot encoding.

4. The thermal comfort control method based on visual perception and global sensitivity analysis according to claim 1 is characterized in that: The preprocessing of the data information in the historical data set includes: For the missing historical data sets, the historical indoor temperature information ta, historical working temperature information top, historical radiation temperature information tr, historical outdoor temperature information tg, historical indoor humidity information rh, historical outdoor air temperature information t_out, historical outdoor humidity information rh_out, historical clothing thermal resistance information clo, historical fan status information fan, historical window status information window, historical door status information door, and historical season information season are filled in by mode according to the season; The historical gender information in the historical dataset is processed using one-hot encoding; The missing historical age information age, historical personnel height information ht, historical weight information wt, historical human metabolic rate information met, and historical indoor wind speed information vel in the historical data set are filled with average values.

5. The thermal comfort control method based on visual perception and global sensitivity analysis according to claim 1 is characterized in that: The adjusting the operating parameters of the air conditioner or heat pump based on the calculation result of the human body thermal comfort includes: subtracting an actual thermal comfort value obtained by calculation based on the human body thermal comfort from a target thermal comfort value; If the difference is outside the preset range, adjust the operating parameters of the air conditioner or heat pump to make the actual thermal comfort value closer to the target thermal comfort value.

6. A thermal comfort control device based on visual perception and global sensitivity analysis, applied to the thermal comfort control method according to any one of claims 1 to 5, characterized in that: include: Acquisition module, used to obtain environmental information and indoor character image information; A recognition module, configured to input the indoor person image information into a pre-trained attribute recognition model to identify human attribute information; a first calculation module, configured to calculate metabolic rate information of a person in the indoor person image information based on the human body attribute information; A second calculation module is used to input the environmental information, human attribute information and metabolic rate information into a pre-trained thermal comfort model to calculate human thermal comfort; The adjustment module is used to adjust the operating parameters of the air conditioner or heat pump based on the calculation result of the human body thermal comfort.

Images