Air conditioner user comfort evaluation method considering meteorological influence

Abstract

The invention discloses an air conditioner user comfort level evaluation method considering meteorological influence. The method comprises the following steps: S1, collecting comfort level data of an air conditioner user under different meteorological conditions; a data set is constructed and marked; s2, a scoring model is constructed, the model comprises a feature extraction module and a prediction module based on a collaborative firefly algorithm, data of the data set in the step S1 is imported into the feature extraction module, and preliminary features are obtained; s3, importing the preliminary features into a prediction module, and obtaining a predicted comfort evaluation result; and S4, the air conditioner control parameters are dynamically adjusted based on the comfort degree evaluation result fed back in real time. Based on the air conditioner setting parameters, the meteorological data and the physiological response data of the user, a complete comfort level evaluation system is constructed, causal inference modeling is adopted, the influence path of the weather on the comfort level is defined, and the reliability of the data is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of artificial intelligence model training methods, and particularly relates to a method for evaluating the comfort of air-conditioning users considering meteorological impacts. Background Art

[0002] With the rapid development of smart home technologies, the intelligent control of air-conditioning systems has gradually become a research hotspot. However, existing air-conditioning control strategies mainly rely on fixed rules or simple machine learning methods, making it difficult to accurately adapt to the comfort requirements of different users under different environmental conditions. Especially under complex meteorological conditions, traditional methods often neglect the comprehensive impacts of environmental factors on air-conditioning settings and human physiological states, resulting in poor accuracy and adaptability of comfort prediction. In addition, although most current comfort evaluation methods based on deep learning can extract high-dimensional features from data, they often lack causal relationship modeling, which may lead to pseudo-correlation problems and insufficient model generalization ability. Meanwhile, deep learning models are prone to falling into local optima during the training process, leading to unstable training and affecting the reliability of comfort prediction.

[0003] Chinese Patent with Publication No. CN119509013A proposes a comfort evaluation method, device, equipment, and storage medium for air conditioners. The present invention obtains the current operation data and user operation behavior information of the air conditioner, and uses a trained comfort evaluation model to evaluate the comfort of the current operation data and user operation behavior information to obtain comfort evaluation data. Finally, based on a pre-set comfort metric interval and the comfort evaluation data, the current comfort evaluation result of the air conditioner is determined, avoiding the technical problem of low accuracy in evaluating the comfort of users when using air conditioners in the prior art and improving the comprehensiveness during comfort evaluation.

[0004] Chinese Patent with Publication No. CN119477038A proposes a method for evaluating the energy conservation of a wind-thermal energy storage system based on game theory. The method includes the following steps: establishing an evaluation index system model, including three types of indicators: energy efficiency indicators, energy structure indicators, and pollutant emissions per unit of electricity; establishing a GTCE-TOPSIS evaluation method model, using the subjective weight method AHP and the objective weight method EWM to determine the weight values of the evaluation index values, and aggregating the weight results; constructing a weighted matrix, calculating the positive ideal solution and negative ideal solution of each object to be evaluated in the weighted matrix, as well as the distances from the positive ideal solution and the negative ideal solution, then calculating the closeness value of the object to be evaluated, and finally classifying the closeness value to obtain the final grade. On the premise of providing a comfortable indoor environment for users, energy is saved, the consumption of clean energy is promoted, and thus the clean development goal of the energy system is further achieved.

[0005] The Chinese invention patent with the publication number CN119469648A proposes a cruise ship theater temperature and air flow organization evaluation system, including: an adjustment module for adjusting the air supply speed, air supply volume, and air supply temperature in the cruise ship theater; a data collection module for collecting the air supply speed, air supply volume, and air supply temperature data of the measuring points in the passenger sightseeing area and the performance area; a data recording module for recording in real time the collected air supply speed, air supply volume, and air supply temperature data to obtain the change process of temperature and air flow over time; a data evaluation module for obtaining the indoor temperature field and air flow organization distribution in the stable state according to the change process of temperature and air flow over time. A cruise ship theater temperature and air flow organization evaluation method is also disclosed. The present invention improves the measurement accuracy of the temperature and air flow field in the cruise ship theater, provides data support for the subsequent adjustment of the air conditioning air outlet position, and improves the environmental comfort and passenger satisfaction of the theater.

[0006] The prior art has the following deficiencies:

[0007] Only using neural networks for comfort prediction, without considering the causal relationship of data, there is a problem of spurious correlation; neural networks are prone to falling into local optima, the training is unstable, and there are problems of gradient disappearance or gradient explosion; existing methods mostly rely on black-box deep learning models and lack interpretability; there is a lack of an optimized feedback mechanism for air conditioning intelligent control; feature extraction is single, without considering the fusion of multi-source data; the neural network structure is fixed and it is difficult to adapt to different environments and data changes. Summary of the Invention

[0008] In view of the technical problems in the prior art, a method for evaluating the comfort of air conditioning users considering meteorological effects is provided, including: Step S1: Collect the comfort data of air conditioning users under different meteorological conditions; and construct a data set and label it based on this. Step S2: Construct a scoring model. The model includes a feature extraction module and a prediction module based on the collaborative firefly algorithm. Import the data of the data set in Step S1 into the feature extraction module to obtain preliminary features; in this process, a causal inference fusion gate is adopted, and a differentiable causal graph network is used to model the causal chain of comfort corresponding to different source collection information to suppress the interference of spurious correlation features. Step S3: Import the preliminary features into the prediction module to obtain the predicted comfort evaluation result. Step S4: Dynamically adjust the air conditioning control parameters based on the real-time feedback comfort evaluation result.

[0009] Further, Step S2 is specifically: Step S21: Initialize the parameters of the neural network of the feature extraction module, and the initialization method is expressed as: ; ; In the formula, is the initial weight matrix of the neural network of the feature extraction module, is the random initialization function, is the initial bias vector of the neural network of the feature extraction module, and are the dimensions of the input layer and the output layer of the neural network of the feature extraction module respectively; Step S22: Optimize the parameters of the neural network of the feature extraction module through the cooperative firefly algorithm. Each firefly represents a set of network parameters, and the network parameters include the weight matrix and the bias vector of the neural network of the feature extraction module; Move and adjust the loss function as the brightness of the fitness in the search space. Specifically, the way of optimizing the neural network of the feature extraction module by the cooperative firefly algorithm is expressed as: ; Among them, is the brightness of the th firefly, is the loss function of the neural network of the feature extraction module, and are the input comfort evaluation data and labels of the neural network of the feature extraction module corresponding to the th group of network parameters respectively, is the weight matrix of the neural network of the feature extraction module, is the bias vector of the neural network of the feature extraction module; Update the position of the firefly based on the brightness of the firefly, and the update method is expressed as: ; In the formula, is the position of the updated firefly, corresponding to the updated weight of the neural network of the feature extraction module; is the attraction coefficient, is the light absorption coefficient; is brighter than the position of other fireflies, characterized as brighter than the corresponding brightness function calculation value is higher; is the th firefly position, is the th firefly position, ; Step S23: The input comfort evaluation data of the neural network corresponding to the th group of network parameters Collect information from different sources, where the source-collected information includes meteorological data, air-conditioning settings, and physiological response data; use a causal inference fusion gate, and use a differentiable causal graph network to model the causal chain of meteorological data → air-conditioning settings → physiological response data → comfort, suppressing the interference of pseudo-correlated features; thereby obtaining preliminary features. Step S24: Through neural network architecture search, automatically test the network architectures of different neural networks serving as feature extraction modules. The network architectures include the number of layers, the number of neurons, and the type of activation function to determine the optimal structure. The neural network architecture search determines the optimal structure through cross-validation, expressed as: ; In the formula, is the network structure search function, is the comfort evaluation data set after segmentation, is the number of folds of cross-validation, represents the performance evaluation value of the k-th fold cross-validation under the given weight matrix Wp and the k-th fold comfort evaluation data set , that is, the loss of the -th fold cross-validation, represents the minimum evaluation function; Step S25: Evaluate the optimized neural network serving as the feature extraction module. Specifically, by comparing the correlation degree between the extracted preliminary features and the labels, the evaluation method is expressed as: ; In the formula, is the feature quality index, is the neural network feature extraction function, is the similarity calculation function, represents the preliminary features, represents the preset label; Step S26: According to the feature quality index, adjust the parameters of the firefly algorithm to more finely adjust the parameters of the neural network. The adjustment method is expressed as: ; In the formula, is the adjusted attraction coefficient; is the attraction learning rate, represents the exponential function; Step S27: Repeat steps S21 to S26 iteratively until the preset stop iteration condition is met.

[0010] Furthermore, the loss function in step S22 is expressed as: ; In the formula, is the number of neural networks in the input feature extraction module for the current batch. represents the category with the highest probability obtained by calculating the neural network prediction output of the feature extraction module through a preset Softmax classification function; represents the true label of the comfort data; is the counterfactual regularization term; is the regularization parameter of the neural network in the feature extraction module; is the weight matrix is the Frobenius norm of.

[0011] Furthermore, in step S24, determining the optimal structure through cross-validation is specifically achieved through the -fold cross-validation, denoted as: ; In the formula, represents the performance evaluation value of the k-fold cross-validation under the given weight matrix Wp and the k-fold comfort evaluation dataset ; represents the number of comfort evaluation data points in the k-fold comfort evaluation dataset; represents the loss function for neural network architecture search; is the output of the neural network for the input under the given weight matrix after being calculated by a preset Softmax classification function to obtain the category with the highest probability, represents the output of the neural network.

[0012] Furthermore, step S23 is specifically as follows: construct a differentiable causal graph network, define the structured causal equation between multi-source data for collecting information at the source, and explicitly model the causal path through a gating mechanism; specifically: Obtain and preset the meteorological data as and the air conditioner setting parameters as and the physiological response data as ; Among them, is the dimension of the meteorological data; is the air conditioner setting dimension, represents the dimension of the physiological response data; represents the set of real numbers; The causal propagation process of the differentiable causal graph network is modeled as: ; ; ; ; In the formula, is the weight matrix of the causal path from meteorological data to air conditioner settings, is the bias term of the causal path from meteorological data to air conditioner settings, is the weight matrix of the causal path from air conditioner settings to physiological response data, is the bias term of the causal path from air conditioner settings to physiological response data, is the gating weight matrix, is the gating dimension, is the gating bias term, is the ReLU activation function, is the Sigmoid function, is the vector concatenation operation, is the element-wise multiplication, is the dynamic gating vector, which is used to control the indirect influence weight of meteorological data on physiological response data; represents the causal inference fusion gate; The pseudo-correlation suppression mechanism constrains the feature space through causal intervention loss to eliminate the influence of confounding factors. The counterfactual regularization term is defined as: ; In the formula, is the comfort prediction value of the neural network, is the meteorological data The Jacobian matrix of the preliminary features, is the air conditioner setting The Jacobian matrix of the preliminary features, is the Frobenius norm, which is the square root of the sum of the squares of the matrix elements, is the regularization penalty coefficient, represents the set of positive real numbers; By constraining the direct gradient contributions of meteorological data and air conditioner settings to the output preliminary features, the differentiable causal graph network makes predictions only through the causal path of physiological response data to block non-causal association paths; Perform multi-source feature fusion. Specifically, the output of the differentiable causal graph network is connected with the original features in a residual manner, retaining the complementary information of explicit causal features and implicit association features. Through the dynamic fusion coefficient, the contributions of explicit causal reasoning and data-driven implicit features are balanced, avoiding over-simplifying the real scenario in causal modeling, which is expressed as: ; In the formula, is the original input feature, is the multi-layer perceptron, which is used to extract implicit features; is the learnable dynamic fusion coefficient, and the update method is , is the learning rate, is the feature quality index, is the final output of the differentiable causal graph network, that is, the preliminary feature.

[0013] Furthermore, step S3 is specifically as follows: Input the preliminary feature into the prediction module. The prediction module is a fully connected layer, and the prediction probabilities of each comfort level are output through the Softmax classification function. The category with the highest probability is used as the final comfort evaluation result to reflect the user's comfort state in the current air-conditioning usage environment.

[0014] The positive and progressive effects of the present invention are as follows:

[0015] 1. Based on the air-conditioning set parameters, meteorological data, and the user's physiological response data, the present invention constructs a complete comfort evaluation system, uses causal inference modeling to clarify the influence path of meteorology on comfort, and improves the reliability of data.

[0016] 2. The present invention optimizes the neural network based on the collaborative firefly algorithm. The global search ability of the firefly algorithm is used to optimize the weights and biases of the neural network, overcoming the problems of gradient disappearance, gradient explosion, and local optimum. Moreover, neural network architecture search is adopted to automatically optimize the neural network structure and improve the model generalization ability.

[0017] 3. The present invention adopts a causal inference fusion mechanism to model the causal chain of "meteorological data → air-conditioning setting → physiological response data → comfort", thereby eliminating the influence of spurious correlation and improving the interpretability and reliability of the model. At the same time, through the causal intervention loss constraint (counterfactual regularization term), the overfitting problem caused by data-driven methods is avoided.

[0018] 4. The present invention combines causal features and data-driven implicit features, uses residual connection for feature fusion to improve the prediction accuracy, and through an intelligent feedback mechanism, dynamically adjusts the air-conditioning parameters according to the comfort prediction result to enhance the user experience. Description of the Drawings

[0019] Figure 1 is the step flow chart of an air-conditioning user comfort evaluation method considering meteorological influence according to the present invention.

[0020] Figure 2 is the accuracy comparison chart of different algorithms of an air-conditioning user comfort evaluation method considering meteorological influence according to the present invention and the prior art under different sizes of training samples.

[0021] Figure 3A method for evaluating the comfort of air - conditioner users considering meteorological impacts is different from the prior art in terms of the comparison chart of the characteristic quality of the average trend line under different noise levels for different algorithms. Specific implementation manners

[0022] The following uses specific examples to illustrate the implementation manners of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific implementation manners, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention.

[0023] A method for evaluating the comfort of air - conditioner users considering meteorological impacts includes: Step S1: Collect the comfort data of air - conditioner users under different meteorological conditions; and construct a data set and label it accordingly. Step S2: Construct a scoring model. The model includes a feature extraction module and a prediction module based on the collaborative firefly algorithm. Import the data of the data set in Step S1 into the feature extraction module to obtain preliminary features. In this process, a causal inference fusion gate is adopted, and a differentiable causal graph network is used to model the causal chain of the comfort corresponding to different source - collected information, suppressing the interference of pseudo - relevant features. Step S3: Import the preliminary features into the prediction module to obtain the predicted comfort evaluation result. Step S4: Dynamically adjust the air - conditioner control parameters based on the real - time feedback of the comfort evaluation result.

[0024] Further, Step S2 is specifically as follows: Step S21: Initialize the parameters of the neural network of the feature extraction module. The initialization method is expressed as: ; ; In the formula, is the initial weight matrix of the neural network of the feature extraction module, is the random initialization function, is the initial bias vector of the neural network of the feature extraction module, and are the dimensions of the input layer and the output layer of the neural network of the feature extraction module respectively. Step S22: Optimize the parameters of the neural network of the feature extraction module through a collaborative firefly algorithm. Each firefly represents a set of network parameters, and the network parameters include the weight matrix and bias vector of the neural network of the feature extraction module; Move and adjust the loss function as the brightness of fitness in the search space. Specifically, the way of optimizing the neural network of the feature extraction module by the collaborative firefly algorithm is expressed as: ; where is the brightness of the -th firefly, is the loss function of the neural network of the feature extraction module, and are the input comfort evaluation data and label of the neural network of the feature extraction module corresponding to the -th set of network parameters respectively, is the weight matrix of the neural network of the feature extraction module, is the bias vector of the neural network of the feature extraction module; Update the position of the firefly based on the brightness of the firefly, and the update method is expressed as: ; In the formula, is the position of the updated firefly, corresponding to the weight of the updated neural network of the feature extraction module; is the attraction coefficient, is the light absorption coefficient; is the position of other fireflies brighter than , characterized by having a higher calculated value of the corresponding brightness function than ; is the position of the -th firefly, is the position of the -th firefly, ; Step S23: The input comfort evaluation data of the neural network corresponding to the -th set of network parameters collect information for different sources. The source collection information includes meteorological data, air-conditioning settings, and physiological response data; Use a causal inference fusion gate to model the causal chain of meteorological data → air-conditioning settings → physiological response data → comfort using a differentiable causal graph network, and suppress the interference of pseudo-correlated features; Thus, preliminary features are obtained. Step S24: Through neural network architecture search, automatically test different network architectures of the neural network serving as the feature extraction module. The network architecture includes the number of layers, the number of neurons, and the type of activation function to determine the optimal structure. The neural network architecture search determines the optimal structure through cross-validation, expressed as: ; In the formula, is the network structure search function, is the comfort evaluation data set after segmentation, is the number of folds of cross-validation, represents the performance evaluation value of the k-th fold cross-validation under the given weight matrix Wp and the k-th fold comfort evaluation data set , that is, the loss of the -th fold cross-validation, represents the minimum evaluation function; Step S25: Evaluate the neural network optimized as the feature extraction module, specifically by comparing the correlation between the extracted preliminary features and the labels. The evaluation method is expressed as: ; In the formula, is the feature quality index, is the neural network feature extraction function, is the similarity calculation function, represents the preliminary features, represents the preset label; Step S26: According to the feature quality index, adjust the parameters of the firefly algorithm to more finely adjust the parameters of the neural network. The adjustment method is expressed as: ; In the formula, is the adjusted attraction coefficient; is the attraction learning rate, represents the exponential function; Step S27: Repeat the iterative steps S21~S26 until the preset stop iteration condition is met.

[0025] Furthermore, the loss function in step S22 is expressed as: ; In the formula, is the number of neural networks in the current batch input to the feature extraction module, represents the category with the highest probability calculated by the neural network prediction output of the feature extraction module through the preset Softmax classification function; represents the true label of the comfort data; is the counterfactual regularization term; is the regularization parameter of the neural network of the feature extraction module; is the weight matrix Frobenius norm.

[0026] Further, the specific process of determining the optimal structure through cross-validation in step S24 is specifically through the -fold cross-validation, which is expressed as: ; In the formula, represents the performance evaluation value of the k-fold cross-validation under the given weight matrix Wp and the k-fold comfort evaluation data set . represents the number of comfort evaluation data points in the k-fold comfort evaluation data set; represents the loss function of the neural network architecture search; is the category with the highest probability calculated by the preset Softmax classification function for the output of the neural network for the input under the given weight matrix , represents the output of the neural network.

[0027] Further, step S23 is specifically: constructing a differentiable causal graph network, defining a structured causal equation between multi-source data for collecting information at the source, and explicitly modeling the causal path through a gating mechanism; specifically: Obtain and preset the meteorological data as , the air conditioner setting parameter as , and the physiological response data as ; Among them, is the dimension of the meteorological data; is the air conditioner setting dimension, represents the dimension of the physiological response data; represents the set of real numbers; The causal propagation process of the differentiable causal graph network is modeled as: ; ; ; ; In the formula, is the weight matrix of the causal path from the meteorological data to the air conditioner setting, is the bias term of the causal path from the meteorological data to the air conditioner setting, is the weight matrix of the causal path from the air conditioner setting to the physiological response data, is the bias term of the causal path from the air conditioner setting to the physiological response data, is the gating weight matrix, is the gating dimension, is the gating bias term, is the ReLU activation function, is the Sigmoid function, is the vector concatenation operation, is the element-wise multiplication, is the dynamic gating vector, which is used to control the weight of the indirect influence of meteorological data on physiological response data; represents the causal inference fusion gate; The pseudo-correlation suppression mechanism constrains the feature space through the causal intervention loss to eliminate the influence of confounding factors. The counterfactual regularization term is defined as: ; In the formula, is the predicted comfort value of the neural network, is the meteorological data is the Jacobian matrix of the preliminary features, is the air conditioner setting is the Jacobian matrix of the preliminary features, is the Frobenius norm, which is the square root of the sum of the squares of the matrix elements, is the regularization penalty coefficient, represents the set of positive real numbers; By constraining the direct gradient contributions of meteorological data and air conditioner settings to the output preliminary features, the differentiable causal graph network makes predictions only through the causal path of physiological response data to block non-causal association paths; Perform multi-source feature fusion. Specifically, the output of the differentiable causal graph network is connected with the original features in a residual manner, retaining the complementary information of explicit causal features and implicit association features. Through the dynamic fusion coefficient, the contributions of explicit causal reasoning and data-driven implicit features are balanced, avoiding over-simplifying the real scenario in causal modeling, which is expressed as: ; In the formula, is the original input feature, is the multi-layer perceptron, which is used to extract implicit features; is the learnable dynamic fusion coefficient, and the update method is , is the learning rate, is the feature quality index, is the final output of the differentiable causal graph network, that is, the preliminary features.

[0028] Furthermore, step S3 is specifically as follows: Input the preliminary features into the prediction module. The prediction module is a fully connected layer, and the predicted probabilities of each comfort level are output through the Softmax classification function. The category with the highest probability is used as the final comfort evaluation result to reflect the comfort state of the user in the current air conditioner usage environment.

[0029] To verify the effectiveness of the technology of the present invention, the following experiments were conducted:

[0030] Referring to Figure 2 , to verify the adaptability and prediction accuracy of the proposed method under different data scales, by comparing with conventional techniques such as traditional neural networks, random forests, and support vector machines, the advantages of the algorithm in complex multi-source data fusion were revealed. The experiment took the training data volume as a variable and compared the changing trends of the accuracy rates of each method for user comfort evaluation. The results showed that the present method demonstrated higher learning efficiency and stable performance under different data scales. Especially under the condition of limited data, it could still maintain a high prediction accuracy, indicating that the parameter search mechanism optimized based on the collaborative firefly algorithm and the design of the causal inference fusion gate, compared with the simple dependence on explicit features in traditional methods, could effectively capture the deep associations among meteorological data, air-conditioning settings, and physiological response data by dynamically balancing causal path modeling and implicit feature extraction. Thus, more significant performance improvements were shown when the data scale expanded, verifying the superiority of the algorithm in feature expression and pattern mining.

[0031] Referring to Figure 3 , to evaluate the feature extraction quality of different methods under noise interference, the robustness performances of key features such as meteorological parameters (i.e., meteorological data), equipment settings (i.e., air-conditioning settings), and physiological signals (i.e., physiological response data) were mainly compared. The experiment simulated sensor acquisition errors and user feedback noise and compared the ability of each method to maintain the causal relationships among features. The results showed that the present technology could still stably extract highly discriminative features under noise interference. The explicit modeling of the "meteorological data - air-conditioning setting - physiological response data" causal chain by the differentiable causal graph network and the suppression mechanism of pseudo-related features by counterfactual regularization, compared with the dependence of traditional neural networks on surface statistical associations, the present method strengthened the propagation weights of causal paths through a dynamic gating mechanism, making the quality indicators of core features such as physiological signals significantly better than the comparison methods. Even under extreme noise conditions, the model could still retain the effective information in the original data through residual connections, reflecting the practical value of the algorithm in complex real-world scenarios.

[0032] The present invention has been described in detail with reference to the embodiments accompanied by drawings. Those of ordinary skill in the art can make various variations to the present invention according to the above description. Therefore, certain details in the embodiments should not constitute a limitation to the present invention, and the present invention will take the scope defined by the appended claims as the protection scope.

Claims

1. A method for evaluating air-conditioning user comfort considering meteorological influences, characterized in that: include: Step S1: collecting comfort data of air-conditioning users under different meteorological conditions; And use this to build and annotate the data set; Step S2: Construct a scoring model, which includes a feature extraction module and a prediction module based on the collaborative firefly algorithm. Import the data of the data set in step S1 into the feature extraction module to obtain preliminary features. In this process, a causal inference fusion gate is used to use a differentiable causal graph network to model the causal chain of comfort corresponding to different source collection information, thereby suppressing the interference of pseudo-correlation features. Step S3: importing preliminary features into the prediction module to obtain predicted comfort evaluation results; Step S4: Dynamically adjust the air conditioning control parameters based on the real-time feedback comfort evaluation results.

2. The air conditioning user comfort evaluation method considering meteorological influence as claimed in claim 1, characterized in that: Step S2 is specifically as follows: Step S21: Initialize the parameters of the neural network of the feature extraction module. The initialization method is expressed as: ； ； In the formula, is the initial weight matrix of the neural network of the feature extraction module, is the random initialization function, is the initial bias vector of the neural network of the feature extraction module, and are the dimensions of the input and output layers of the neural network of the feature extraction module, respectively; Step S22: Optimize the parameters of the neural network of the feature extraction module by the collaborative firefly algorithm, where each firefly represents a set of network parameters, including the weight matrix and bias vector of the neural network of the feature extraction module; move and adjust the loss function as the brightness of the fitness in the search space. Specifically, the collaborative firefly algorithm optimizes the neural network of the feature extraction module as follows: ； in, For the The brightness of a firefly, is the loss function of the neural network of the feature extraction module, and are the input comfort evaluation data and labels of the neural network of the feature extraction module corresponding to the i-th group of network parameters, is the weight matrix of the neural network of the feature extraction module, is the bias vector of the neural network of the feature extraction module; The position of the firefly is updated based on the brightness of the firefly. The update method is expressed as: ； In the formula, is the updated position of the firefly, corresponding to the updated weight of the neural network of the feature extraction module; is the attraction coefficient, is the light absorption coefficient; For comparison The positions of other brighter fireflies are represented by Compare The corresponding brightness function calculation value is higher; For the The location of the fireflies. For the The location of the fireflies. ; Step S23: The input comfort evaluation data of the neural network corresponding to the group network parameters Collect information from different sources, including meteorological data, air conditioning settings, and physiological response data; adopt causal inference fusion gates and use differentiable causal graph networks to model the causal chain of meteorological data → air conditioning settings → physiological response data → comfort, suppressing the interference of pseudo-correlation features; thereby obtaining preliminary features; Step S24: Through the neural network architecture search, the network architecture of different neural networks as feature extraction modules is automatically tested, and the network architecture includes the number of layers, the number of neurons and the type of activation function to determine the optimal structure. The neural network architecture search determines the optimal structure through cross-validation, which is expressed as: ； In the formula, Search function for network structure, is the segmented comfort evaluation dataset, is the number of cross validation folds, Represents the k-fold cross validation under a given weight matrix Wp and the k-fold comfort evaluation dataset The performance evaluation value under The loss of the fold cross validation, represents the minimum evaluation function; Step S25: Evaluate the optimized neural network as a feature extraction module, specifically by comparing the correlation between the extracted preliminary features and the labels. The evaluation method is expressed as: ； In the formula, is the characteristic quality indicator, is the neural network feature extraction function, is the similarity calculation function, Represents preliminary features, Indicates a preset label; Step S26: According to the feature quality index, the parameters of the firefly algorithm are adjusted to more finely adjust the parameters of the neural network. The adjustment method is expressed as: ； In the formula, is the adjusted attractiveness coefficient; is the attractive learning rate, represents the exponential function; Step S27: Repeat the iteration of steps S21 to S26 until a preset stop iteration condition is met.

3. The air conditioning user comfort evaluation method considering meteorological influence as claimed in claim 2, characterized in that: Loss function in step S22 It is expressed as: ； In the formula, The number of neural networks that feed the feature extraction module for the current batch, It indicates the category with the highest probability calculated by the preset Softmax classification function after the neural network prediction output of the feature extraction module; Indicates the true label of comfort data; is the counterfactual regularization term; is the regularization parameter of the neural network of the feature extraction module; is the weight matrix The Frobenius norm of .

4. The method for evaluating air-conditioning user comfort considering meteorological influence as claimed in claim 2, characterized in that: In step S24, the optimal structure is determined by cross-validation. Fold cross validation, expressed as: ； In the formula, Represents the k-fold cross validation under a given weight matrix Wp and the k-fold comfort evaluation dataset The performance evaluation value under Indicates The number of comfort evaluation data points in the comfort evaluation dataset; Represents the loss function for neural network architecture search; is the neural network under a given weight matrix Next pair input The output of is calculated by the preset Softmax classification function to obtain the category with the highest probability. Represents the output of the neural network.

5. The method for evaluating air-conditioning user comfort considering meteorological influence as claimed in claim 2, characterized in that: Step S23 is specifically: constructing a differentiable causal graph network, defining a structured causal equation between multiple source data of source collected information, and explicitly modeling the causal path through a gating mechanism; specifically: Get and preset weather data for , the air conditioning setting parameters are , the physiological response data is ; in, is the meteorological data dimension; Set the dimensions for the air conditioner, Dimensions representing physiological response data; represents the set of real numbers; The causal propagation process of the differentiable causal graph network is modeled as: ； ； ； ； In the formula, The weight matrix of the causal path from meteorological data to air conditioning, The bias term for the causal path from meteorological data to air conditioning, The weight matrix for the causal paths from conditioning to the physiological response data, A bias term for the causal path from conditioning to the physiological response data, is the gating weight matrix, is the gating dimension, is the gate bias term, is the ReLU activation function, is the Sigmoid function, is the vector concatenation operation, is element-wise multiplication, is a dynamic gating vector used to control the weight of the indirect impact of meteorological data on physiological response data; represents the causal inference fusion gate; The pseudo-correlation suppression mechanism constrains the feature space through causal intervention loss, eliminates the influence of confounding factors, and defines the counterfactual regularization term as: ； In the formula, is the comfort prediction value of the neural network, For weather data The Jacobian matrix of the preliminary features, Set for air conditioner The Jacobian matrix of the preliminary features, is the Frobenius norm, the square root of the sum of the squares of the matrix elements, is the regularization penalty coefficient, represents the set of positive real numbers; By constraining the direct gradient contribution of meteorological data and air conditioning settings to the initial features of the output, the differentiable causal graph network is made to be only based on physiological response data. The causal pathway is predicted to block the non-causal association pathway; Multi-source feature fusion is performed by residually connecting the output of the differentiable causal graph network with the original features, retaining the complementary information of explicit causal features and implicit association features, and balancing the contribution of explicit causal reasoning and data-driven implicit features through dynamic fusion coefficients to avoid over-simplification of real scenarios by causal modeling. It is expressed as: ； In the formula, is the original input feature, It is a multi-layer perceptron, used to extract implicit features; is the learnable dynamic fusion coefficient, and the update method is , is the learning rate, is the characteristic quality indicator, It is the final output of the differentiable causal graph network, that is, the preliminary feature.

6. The method for evaluating air-conditioning user comfort considering meteorological influence as claimed in claim 2, characterized in that: Step S3 is specifically as follows: The preliminary features are input into the prediction module, which is a fully connected layer. The prediction probability of each comfort level is output through the Softmax classification function. The category with the highest probability is used as the final comfort evaluation result to reflect the user's comfort state in the current air-conditioning environment.

Images