Outdoor thermal comfort intelligent evaluation method and system taking psychological state as medium

Abstract

The application discloses an outdoor thermal comfort intelligent evaluation method and system taking a psychological state as a medium. The method specifically comprises the following steps: acquiring psychological index data and multi-scale parameters, wherein the multi-scale parameters comprise environmental physical parameters and physiological response parameters; inputting the environmental physical parameters and the physiological response parameters into a dynamic feature extraction module, performing environmental physical parameter coding and physiological time sequence feature extraction, and performing cross-modal feature fusion to generate a fusion feature vector; constructing a psychological intermediary prediction module according to the psychological index data, inputting the fusion feature vector into the psychological intermediary prediction module, and generating a dynamic adjustment coefficient; and inputting the fusion feature vector and the dynamic adjustment coefficient into a thermal comfort degree evaluation module, and outputting a multi-scale prediction result. The application solves the spatiotemporal alignment problem of environmental parameter transience and physiological response hysteresis, and significantly improves the individual adaptability and prediction reliability of outdoor thermal comfort degree evaluation.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of thermal environment evaluation and artificial intelligence, and particularly relates to an outdoor thermal comfort intelligent evaluation method and system taking psychological state as a medium. BACKGROUND

[0002] Under the dual pressures of global climate change and urbanization, it is urgent to create a high-quality, comfortable and healthy outdoor thermal environment. How to scientifically, accurately and comprehensively evaluate the thermal comfort of people in the outdoor environment has become a challenging and cutting-edge research topic in the field of thermal environment perception. At present, although the evaluation technology of human thermal comfort is constantly developing and improving, there are still two key bottlenecks, which seriously restrict the accuracy and reliability of the evaluation results.

[0003] Firstly, in the construction process of traditional thermal comfort evaluation models, single physical index or physiological signal analysis methods are mainly used. Among them, the widely used Predicted Mean Vote (PMV) model mainly predicts the thermal sensation of the human body under steady-state conditions based on physical environmental parameters such as air temperature, humidity, wind speed and average radiant temperature. After considering physical and physiological indicators, a batch of classic steady-state heat transfer mechanism models such as PET and SET have emerged. However, the perception of the human body to the outdoor thermal environment is a complex and dynamic process, and ignoring the mediating role of psychological state in it will cause a large deviation in the evaluation or prediction of actual human thermal comfort.

[0004] Secondly, although the application of existing multi-modal fusion methods in thermal comfort evaluation aims to integrate multiple data sources to improve the comprehensiveness and accuracy of evaluation, it still faces many problems in actual operation. At present, many multi-modal fusion methods often rely on static weight distribution strategies, i.e., fixed weights are set for different data sources (such as physical environmental parameters, physiological signals, psychological questionnaire data, etc.) in advance, and then a simple weighted sum is performed. However, this static weight distribution method ignores the importance changes of different data sources under different environmental and individual states. In fact, there is a complex dynamic coupling relationship between environmental stimuli, physiological responses and psychological feedback. Static weight distribution cannot capture this dynamic coupling relationship, resulting in that the evaluation results after fusion cannot accurately reflect the real thermal comfort of the human body in complex outdoor environments. In addition, some multi-modal fusion methods only use shallow feature splicing methods to simply splice the features of different data sources together, without deeply mining the internal relationship and potential law between them, which also limits the improvement of evaluation performance. SUMMARY

[0005] The present application aims to provide an outdoor thermal comfort intelligent evaluation method and system mediated by psychological state, solve the spatio-temporal alignment problem of environmental parameter transience and physiological response hysteresis, significantly improve the individual adaptability and prediction reliability of human thermal comfort evaluation in complex outdoor thermal environment, and solve at least one of the above technical problems.

[0006] In a first aspect, the present application provides an outdoor thermal comfort intelligent evaluation method mediated by psychological state, which specifically comprises:

[0007] Obtaining psychological index data and multi-scale parameters, the psychological index data including multiple emotional factor data convertible into overall emotional disorder degree data, and the multi-scale parameters including environmental physical parameters and physiological response parameters;

[0008] Inputting the environmental physical parameters and physiological response parameters into a dynamic feature extraction module, performing environmental physical parameter coding and physiological time sequence feature extraction, and performing cross-modal feature fusion to generate a fusion feature vector;

[0009] Constructing a psychological intermediary prediction module according to the psychological index data, inputting the fusion feature vector into the psychological intermediary prediction module, performing fusion emotional state prediction to generate a dynamic adjustment coefficient;

[0010] Inputting the fusion feature vector and the dynamic adjustment coefficient into a thermal comfort evaluation module, integrating spatio-temporal feature interaction modes and loss function optimization to output multi-scale prediction results.

[0011] In a second aspect, the present application provides an outdoor thermal comfort intelligent evaluation system mediated by psychological state, which specifically comprises:

[0012] A data acquisition module for obtaining psychological index data and multi-scale parameters, the psychological index data including multiple emotional factor data convertible into overall emotional disorder degree data, and the multi-scale parameters including environmental physical parameters and physiological response parameters;

[0013] A feature fusion module for inputting the environmental physical parameters and physiological response parameters into a dynamic feature extraction module, performing environmental physical parameter coding and physiological time sequence feature extraction, and performing cross-modal feature fusion to generate a fusion feature vector;

[0014] A dynamic adjustment module for constructing a psychological intermediary prediction module according to the psychological index data, inputting the fusion feature vector into the psychological intermediary prediction module, performing emotional disorder degree prediction to generate a dynamic adjustment coefficient;

[0015] A prediction output module for inputting the fusion feature vector and the dynamic adjustment coefficient into a thermal comfort evaluation module, integrating spatio-temporal feature interaction modes and loss function optimization to output multi-scale prediction results.

[0016] In a third aspect, the present application provides a computer device, comprising a memory and a processor, and a computer program stored in the memory, which, when executed on the processor, implements the method for intelligent outdoor thermal comfort assessment mediated by psychological state as described above.

[0017] In a fourth aspect, the present application provides a computer readable storage medium having stored thereon a computer program, which, when executed on a processor, implements the method for intelligent outdoor thermal comfort assessment mediated by psychological state as described above.

[0018] Compared with the prior art, the present application has at least one of the following technical effects:

[0019] 1. The present application innovatively proposes a new evaluation paradigm driven by psychological mediation, and establishes a nonlinear mapping mechanism between multi-modal parameters by designing a three-stage deep model of dynamic extraction of environmental-physiological characteristics, joint reasoning of psychological state and probabilistic prediction of thermal comfort. Compared with traditional methods, the present application innovatively introduces psychological state as a dynamic adjustment variable, adopts cross-modal attention mechanism and gating fusion technology, solves the spatio-temporal alignment problem of environmental parameter transience and physiological response hysteresis, significantly improves the individual adaptability and prediction reliability of thermal comfort assessment in complex outdoor scenes, and provides a scientific decision basis for the field of thermal environment monitoring.

[0020] 2. The present application innovatively introduces psychological state as a key intermediate variable to solve the technical bottleneck of thermal comfort assessment in outdoor environment, and constructs a multi-level deep learning model integrating physical environmental parameters, physiological indicators and psychological state. The model not only realizes intelligent and accurate prediction of individual thermal comfort perception, but also systematically reveals the mechanism of psychological state in thermal perception regulation, breaks through the evaluation limitation of traditional methods relying only on physical or physiological indicators, and has important scientific value and wide engineering application prospect.

[0021] 3. The present application is based on the real-time weight distribution mechanism of the double-channel neural network, and constructs a dynamic feature fusion framework through the collaborative analysis of environmental physical parameters and human physiological indicators. The technology breaks through the traditional static weighting mode and adopts the joint degree of environmental perception confidence and physiological feedback sensitivity.

[0022] 4. The present application designs three-level feature processing. Firstly, the nonlinear thermal stress features are extracted through the environmental parameter encoder, the bidirectional LSTM captures the physiological time sequence dependence, and the gated attention network realizes the cross-modal dynamic interaction. By decoupling the spatial distribution characteristics of physical environment and the time sequence hysteresis effect of physiological response in stages, the scale conflict of multi-source data is eliminated, and the feature fusion efficiency is improved.

[0023] 5, The application innovatively designs a psychological-thermal comfort double-path loss function. For the thermal comfort path, the psychological calibration true value constraint is introduced to ensure the physical interpretability of the dynamic weight. For the psychological path, variable sparsification is achieved through KL divergence and L1 regularization, and the system maintains prediction accuracy while outputting three interpretable indicators: thermal comfort score, psychological influence factor contribution, and prediction confidence interval.

[0024] 6, The application realizes a technical breakthrough in the field of intelligent environment perception by constructing a psychological intermediary driven thermal comfort evaluation system. The psychological state is introduced as a dynamic adjustment variable to real-time correct the contribution weight of environment and physiological characteristics through the psychological intermediary module, solving the evaluation deviation problem caused by ignoring psychological fluctuations in traditional models. A bidirectional gated fusion network is designed to model the transient characteristics of environmental parameters and the lag effect of physiological responses simultaneously, solving the problem of time and space asynchrony of multi-source data. A bidirectional feedback mechanism of thermal comfort regression and psychological modeling is constructed to realize multi-dimensional traceability of evaluation results through variable sparsification constraint and dynamic adjustment coefficient true value supervision. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 is a flowchart of an outdoor thermal comfort intelligent evaluation method provided by an embodiment of the application, which takes psychological state as a medium;

[0026] Figure 2 is a structural diagram of a multi-level deep learning model of an outdoor thermal comfort intelligent evaluation method provided by an embodiment of the application, which takes psychological state as a medium;

[0027] Figure 3 is a structural diagram of an outdoor thermal comfort intelligent evaluation system provided by an embodiment of the application, which takes psychological state as a medium;

[0028] Figure 4 is a structural diagram of a computer device provided by an embodiment of the application. DETAILED DESCRIPTION

[0029] In the following description, specific details are set forth such as particular system configurations, techniques, etc., in order to provide a thorough understanding of the embodiments of the application. However, it will be apparent to one skilled in the art that the application can be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the application with unnecessary detail.

[0030] It should be understood that the term "include" as used in the specification and throughout the claims means the presence of the described features, integers, steps, operations, elements, and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0031] It should also be understood that the term "and / or" as used in the specification and throughout the claims, means any one or more of the associated listed items, as well as all possible combinations of the items.

[0032] As used in the specification and throughout the claims, the term "if" can be interpreted as meaning "when" or "once" or "in response to a determination" or "in response to a detection" depending on the context. Similarly, the phrase "if it is determined" or "if [the described condition or event] is detected" can be interpreted as meaning "once it is determined" or "in response to the determination" or "once [the described condition or event] is detected" or "in response to the detection of [the described condition or event]" depending on the context.

[0033] In addition, in the description of the application and the appended claims, the terms "first", "second", "third", etc. are used only to distinguish descriptions, and cannot be understood as indicating or implying relative importance.

[0034] In the present application, the reference "one embodiment" or "some embodiments" and the like means that the specific features, structures or characteristics described in connection with the embodiment are included in one or more embodiments of the application. Therefore, the statements "in one embodiment", "in some embodiments", "in other some embodiments", "in further some embodiments" and the like appearing in different places in the specification are not necessarily all referring to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized. The terms "include", "contain", "have" and their variants mean "include but not limited to", unless otherwise specifically emphasized.

[0035] In the embodiments of the present application, the execution subject of the flow includes a terminal device. The terminal device includes but is not limited to a server, a computer, a smart phone, a tablet computer and other devices capable of executing the method disclosed in the present application. Figure 1 The flowchart of the outdoor thermal comfort intelligent evaluation method mediated by the psychological state disclosed in an embodiment of the present application is shown, and the details are as follows:

[0036] S101, acquiring psychological index data and multi-scale parameters, the psychological index data including a plurality of emotional factor data which can be converted into overall emotional disorder degree data, and the multi-scale parameters including environmental physical parameters and physiological response parameters.

[0037] In the present embodiment, in order to ensure the scientific quantification of the psychological indicators and the high correlation with thermal comfort, the BPOMS scale is used for psychological state representation, and the indicator system ensures the fine expression of the psychological state, thereby providing a solid foundation for the scientific construction of the model intermediate layer. At the same time of collecting the psychological indicators, multi-dimensional parameter information is synchronously collected, thereby providing rich input information for the model and improving the dynamic response capability of the prediction.

[0038] Specifically, the collected data includes:

[0039] (1) Psychological indicator data: The six emotional factors (BPOMS scale is used to measure the individual subjective psychological state) of tension (T), anger (A), fatigue (F), confusion (C), depression (D) and vigor (V) are evaluated by the Likert 5-level scale, and the adjustment effect of emotional tendency on thermal comfort evaluation is revealed.

[0040] (2) Environmental physical parameters: Four key variables of air temperature, relative humidity, radiation temperature and wind speed comprehensively reflect the physical characteristics of the thermal environment.

[0041] (3) Physiological response parameters: Skin temperature, heart rate variability (HRV) and local sweat rate are synchronously collected to accurately quantify the physiological adaptability and thermal stress response of individuals.

[0042] The collected data is preprocessed, feature extracted and normalized to form a data set, and the content of the data set includes:

[0043] (1) Environmental physical parameter matrix: A 4-dimensional environmental physical feature vector composed of air temperature (℃), relative humidity (%), radiation temperature (℃) and wind speed (m / s). The matrix constitutes the basic data of the dynamic change of the environment.

[0044] (2) Physiological indicator multi-channel time series matrix: Including skin temperature (℃), heart rate variability (low / high frequency ratio calculated based on RR interval) and local sweat rate, the matrix constitutes the physiological response data of the human body to thermal stimulation.

[0045] (3) Psychological state quantification matrix: According to the six emotional factors, the structured score data of the degree of total mood disturbance (TMD) can be calculated, and the result is stored in the form of a numerical vector.

[0046] (4) Thermal comfort score vector: Based on the thermal comfort score of the subjects in the test environment, a 1-10 point system is used to reflect the subjective thermal comfort perception intensity of individuals. The data is collected through the subjective evaluation of the subjects, and supports the supervised learning training. ​

[0047] S102, input the environmental physical parameters and physiological response parameters into the dynamic feature extraction module, encode the environmental physical parameters and extract physiological time sequence features, and perform cross-modal feature fusion to generate a fusion feature vector.

[0048] In this embodiment, the dynamic feature extraction module aims to extract effective features from the environmental physical parameters and physiological response parameters respectively, and fuse these features to generate a feature vector that can comprehensively reflect the state of the human body in an outdoor thermal environment.

[0049] The collected environmental physical parameters (such as air temperature, humidity, wind speed, and mean radiant temperature, etc.) are subjected to data cleaning to remove abnormal values and noise data. For example, for the data collected by the air temperature sensor, if the temperature value at a certain time exceeds the reasonable range (such as below -50℃ or above 80℃), it is considered as an abnormal value, which can be removed or replaced by the average value of adjacent time data. Then, the cleaned data is subjected to normalization processing to map different dimensional environmental physical parameters into a unified numerical range, facilitating subsequent feature extraction and fusion. The normalization method can use the maximum-minimum normalization, i.e. subtracting the minimum value of each parameter from the value of the parameter, and then dividing by the difference between the maximum and minimum values of the parameter. Next, the environmental physical parameters are encoded using hierarchical encoding. The environmental physical parameters are grouped according to their physical meaning and correlation, for example, air temperature and mean radiant temperature are grouped into one group, and humidity and wind speed are grouped into another group. For each group of parameters, their statistical features (such as mean, variance, maximum, minimum, etc.) and time sequence features (such as rate of change, trend, etc.) are extracted. Statistical features can reflect the overall situation of the parameters within a certain time, while time sequence features can reflect the change law of the parameters over time. The extracted statistical features and time sequence features are combined to form a feature vector for each group of environmental physical parameters. These feature vectors are spliced to obtain a complete environmental physical parameter encoding vector. This encoding vector can comprehensively represent the physical state of the outdoor thermal environment.

[0050] The collected physiological response parameters (such as heart rate, skin conductivity, and other physiological signals) are filtered to remove high-frequency noise and baseline drift in the signals. A digital filter (such as a Butterworth filter) can be used to filter the physiological signals, and the appropriate cutoff frequency can be set according to the frequency characteristics of the physiological signals. Then, the filtered physiological signals are segmented, and the continuous physiological signals are divided into several time windows. The length of each time window can be set according to actual needs, for example, 1 minute or 5 minutes. The purpose of this is to facilitate the extraction of the time sequence characteristics of the physiological signals and to reflect the physiological changes of the human body in different time periods. Next, for the physiological signals in each time window, the time domain features (such as mean, standard deviation, peak value, etc.) and frequency domain features (such as power spectral density, dominant frequency, etc.) are extracted. The time domain features can reflect the fluctuations of the physiological signals in time, while the frequency domain features can reveal the frequency components and energy distribution of the physiological signals. The sliding window method is used to extract features from the physiological signals, that is, after a set of features is extracted for each time window, the window slides forward by a certain step size, and the features of the next time window are extracted. In this way, the feature sequence of the physiological signals in the time sequence can be obtained. The extracted feature sequence is further processed, such as using principal component analysis (PCA) to reduce the dimensionality of the features, remove redundant features, and retain the most representative features, thereby obtaining the physiological time sequence feature vector. This feature vector can reflect the dynamic changes in the physiological response of the human body in the outdoor thermal environment.

[0051] Since the collection frequency and sampling time of environmental physical parameters and physiological response parameters may not be consistent, feature alignment operations are needed. Interpolation or resampling methods can be used to adjust the two types of feature sequences to the same time scale, ensuring that each time point has corresponding environmental physical parameter features and physiological time sequence features. First, the importance weights of each feature in the environmental physical parameter feature vector and the physiological time sequence feature vector are calculated respectively. By calculating the correlation between features or using a self-attention mechanism to learn the internal relationship between features, the importance score of each feature is obtained. Then, the environmental physical parameter feature vector and the physiological time sequence feature vector are weighted and summed according to the importance weights to obtain the fused feature vector. In the weighted sum process, features with higher importance weights occupy a larger proportion in the fused feature vector, thereby better reflecting the correlation and complementary relationship between different modal features. To further improve the quality of the fused features, the fused feature vector is normalized to make its numerical range meet the requirements of subsequent modules. The final generated fused feature vector can comprehensively and comprehensively reflect the outdoor thermal environment and the physiological and psychological state of the human body in it, providing effective feature representation for subsequent thermal comfort assessment.

[0052] Through the above implementation steps of environmental physical parameter coding, physiological timing feature extraction, and cross-modal feature fusion, the dynamic feature extraction module can generate high-quality fusion feature vectors, laying a solid foundation for subsequent thermal comfort assessment.

[0053] In S103, a psychological intermediary prediction module is constructed according to the psychological index data, the fusion feature vector is input into the psychological intermediary prediction module, and a dynamic adjustment coefficient is generated through emotion disturbance prediction.

[0054] In this embodiment, the psychological intermediary prediction module innovatively establishes a thermal comfort assessment mechanism based on emotion disturbance dynamic compensation, and constructs a cross-modal causal reasoning chain of “environment-physiological features→psychological intermediary→thermal comfort” by introducing the quantitative index of the Brief Profile of Mood States (BPOMS), i.e., Total Mood Disturbance (TMD).

[0055] First, an emotion disturbance degree prediction mechanism is constructed. A hierarchical analysis method is used to first perform feature clustering analysis on different levels of emotion disturbance degree according to the division of emotion disturbance degree, and find out the typical feature mode of emotion disturbance degree at each level. Then, the emotion disturbance degree prediction model is established using these typical feature modes. This branch uses a rule-based reasoning method to match the typical feature mode according to the relevant information in the input fusion feature vector (such as environmental stimuli and physiological state reflected by environmental physical parameters and physiological response parameters), and predicts the emotion disturbance degree of the human body under the current situation. For example, when the environmental temperature is high and the heart rate of the human body is accelerated, it may indicate that the emotional state has a tendency to fluctuate greatly. This branch will reason according to these information to give a prediction value of the emotion disturbance degree.

[0056] Then, according to the potential relationship between the psychological state evaluation value and the thermal comfort perception, the generation rule of the dynamic adjustment coefficient is established. For example, when the psychological state evaluation value indicates that the human body is in a stable psychological state, the tolerance to the thermal environment may be improved to some extent, and at this time the dynamic adjustment coefficient can be appropriately reduced to adjust the output result of the subsequent thermal comfort assessment module; on the contrary, when the psychological state evaluation value shows that the human body is in a fluctuating psychological state, the dynamic adjustment coefficient can be appropriately increased, so that the thermal comfort assessment result is more inclined to reflect the discomfort perception of the human body. In this way, the adjustment coefficient that can dynamically reflect the influence of the psychological state of the human body on the thermal comfort perception is generated.

[0057] Through the above architecture design of the psychological intermediary prediction module and the implementation steps of emotion disturbance degree prediction, an effective psychological intermediary prediction module can be constructed, and an accurate dynamic adjustment coefficient can be generated, providing key support for subsequent thermal comfort assessment.

[0058] S104, input the fusion feature vector and the dynamic adjustment coefficient into the thermal comfort evaluation module, output multi-scale prediction results by integrating the spatio-temporal feature interaction mode and loss function optimization.

[0059] In this embodiment, the thermal comfort evaluation module aims to comprehensively fuse the fusion feature vector and the dynamic adjustment coefficient, and output accurate and multi-scale thermal comfort prediction results by integrating the spatio-temporal feature interaction mode and loss function optimization, thereby providing strong support for outdoor thermal environment research and application.

[0060] In some embodiments, with reference to Figure 2 , the dynamic feature extraction module includes an environmental physical parameter encoder, a physiological time sequence feature extractor, and a cross-modal feature fusion unit; in the above step S102, the environmental physical parameters and the physiological response parameters are input into the dynamic feature extraction module, the environmental physical parameters are encoded and the physiological time sequence features are extracted, and the cross-modal feature fusion is performed to generate a fusion feature vector, which specifically includes:

[0061] The environmental physical parameters are input into the environmental physical parameter encoder, and the air temperature parameters, the relative humidity parameters, the radiation temperature parameters, and the wind speed parameters are processed by a four-layer fully connected neural network to generate an environmental feature vector;

[0062] The physiological response parameters are input into the physiological time sequence feature extractor, and the physiological time sequence signals in multiple time windows including skin temperature, heart rate variability HRV, and local sweat rate are processed by a bidirectional long short-term memory network to obtain a physiological feature vector;

[0063] The environmental feature vector and the physiological feature vector are input into the cross-modal feature fusion unit, the environmental feature vector and the physiological feature vector are independently projected by a double path, and the importance of each feature dimension is determined by similarity calculation to generate an attention weight;

[0064] Based on the attention weight, the environmental feature vector and the physiological feature vector are element-wise multiplied and added to obtain a fusion feature vector.

[0065] In this embodiment, the environmental physical parameter encoder processes the input 4-dimensional environmental parameter vector , including air temperature ( ), relative humidity (RH), radiation temperature ( ), and wind speed ( ) by a four-layer fully connected neural network. The Gaussian error linear unit (GeLU) is used as the activation function between network layers:

[0066]

[0067] wherein is an error function, GeLU (Gaussian exponential linear unit) is used to represent the input value of a single neuron in the neural network before activation. GeLU has smoother gradient characteristics than ReLU, which is beneficial for modeling complex interactions between environmental parameters.

[0068] The network structure is designed as a bottleneck architecture with layer-by-layer expansion and compression. The first layer maps the 4-dimensional input to a 64-dimensional space (weight matrix ), the second layer expands to 128 dimensions ( ) to capture high-order nonlinear relationships, the third layer compresses back to 64 dimensions ( ) to eliminate redundancy, and finally outputs a 32-dimensional environmental feature vector ( ) that satisfies the following expression:

[0069]

[0070] where , , , are the bias vectors of the four fully connected layers, respectively, , , , are the weight matrices of the four fully connected layers, respectively.

[0071] This network structure design breaks through the limitation of traditional linear combination and captures the nonlinear coupling relationship between temperature, humidity, radiation, and wind speed, such as the synergistic effect in high temperature and high humidity environments. The output feature vector can represent the equivalent thermal stress level.

[0072] The physiological time series feature extractor uses a bidirectional long short-term memory network (BiLSTM) to process the physiological time series signals within a 30-second time window , with an input dimension of 3 (skin temperature, heart rate variability HRV, local sweat rate) and a time step T=30. First, forward propagation is performed to calculate the hidden state at time step in chronological order, capturing the lag effect of physiological response. Then, reverse propagation is performed to calculate in reverse order to model the forward-looking features of physiological regulation, with the following calculation method:

[0073]

[0074]

[0075] where , are the weight matrices of forward and backward propagation, respectively. The former learns the dynamic association pattern between physiological signals and historical states, and the latter allows the reverse LSTM to learn different feature patterns from forward propagation, , are the bias terms for forward and backward propagation, respectively; is the physiological signal vector at time step ; [] denotes vector concatenation operation, which encodes the current input and historical state or future state jointly.

[0076] On this basis, feature fusion is performed, and the forward state at the final time is concatenated with the initial reverse state , and a 128-dimensional feature vector is output through layer normalization (LayerNorm). The formula is as follows:

[0077]

[0078] where is the hidden state of the forward LSTM at time step , which encodes the physiological signal time sequence dependence feature from to . is the hidden state of the reverse LSTM at time step , which encodes the reverse time sequence feature from to . , are trainable weight matrices and bias terms, respectively, the former is used to map the concatenated 256-dimensional vector to a 128-dimensional feature space, and the latter is used to adjust the mean of the feature distribution.

[0079] This content captures the dynamic changes of skin temperature, HRV spectrum migration (reflecting autonomic nervous regulation), and the time dependence of sweat rate accumulation effect. The bidirectional structure can model the lag and forward-looking features of physiological responses, such as sweat delay and heart rate recovery process. Finally, through layer normalization, the feature scale difference is eliminated, and high-quality physiological representation suitable for downstream tasks is output.

[0080] The cross-modal feature fusion unit designs an interpretable attention mechanism to dynamically fuse environmental and physiological features. First, feature interaction modeling is performed, and 32-dimensional environmental features and 128-dimensional physiological features are projected to a 64-dimensional common space. The double paths are independently projected, allowing the environment and physiological features to retain their modality specificity and avoiding information loss caused by early fusion:

[0081]

[0082] where denotes linear projection transformation; is an input feature vector or ; is a trainable matrix, dim corresponds to the input feature dimension (when , , , ); is a bias vector, used to shift the feature distribution.

[0083] On this basis, the attention weight is generated, the importance of each feature dimension is determined through similarity calculation, and the calculation method is as follows:

[0084]

[0085] wherein is a normalized exponential function, which converts the original score into a probability distribution and ensures weight normalization; is a learnable attention vector, is its transpose; is a hyperbolic tangent function, which is used to compress the feature value to interval, and enhance the nonlinear expression ability; and are the projection results of the environmental features and physiological features after the above linear projection transformation, respectively.

[0086] After the attention weight is generated, the feature fusion is performed, and the environmental and physiological features are finely fused according to the attention weight, and element-by-element multiplication and addition are performed:

[0087]

[0088] wherein denotes Hadamard product, i.e. multiplication of corresponding elements; finally, the fused feature vector is transmitted to the downstream module.

[0089] By dynamically adjusting the contribution weight of multi-modal features, this mechanism enables the model to adapt to the dominant influencing factors under different environmental-physiological states, overcoming the limitations of fixed weight coefficients in traditional methods.

[0090] In some embodiments, with reference to Figure 2 , the psychological intermediary prediction module includes a feature decoupling compression network and a dynamic adjustment coefficient generation unit; in the above step S103, the fused feature vector is input into the psychological intermediary prediction module, and a dynamic adjustment coefficient is generated through overall emotional disorder degree prediction, specifically including:

[0091] The fusion feature vector is input into a feature decoupling compression network, and through feature compression and emotion sensitive factor mapping, overall emotion disorder degree data is obtained;

[0092] The overall emotion disorder degree data is input into a dynamic adjustment coefficient generation unit, and through hyperbolic tangent function constraint range, a dynamic adjustment coefficient is generated.

[0093] Further, the fusion feature vector is input into the feature decoupling compression network, and through feature compression and emotion sensitive factor mapping, a TMD value is obtained, specifically including:

[0094] The fusion feature vector is processed by the bottleneck type full connection layer of the feature decoupling compression network to perform feature dimension reduction and noise suppression, and a primary compression feature is generated;

[0095] Based on the layer normalization-full connection joint distillation layer of the feature decoupling compression network, an emotion sensitive factor highly related to the overall emotion disorder degree is extracted from the primary compression feature;

[0096] Based on the final layer of the feature decoupling compression network, the emotion sensitive factor is mapped to a normalized overall emotion disorder degree scalar through affine transformation and Tanh activation, and an original overall emotion disorder degree prediction value is obtained;

[0097] Through normalization processing on the original overall emotion disorder degree prediction value, a target overall emotion disorder degree prediction value is obtained.

[0098] In this embodiment, in order to eliminate the collinearity interference between the environmental physical parameters and the physiological signals, a two-stage feature decoupling compression algorithm is proposed. The input layer receives a 64-dimensional mixed feature vector from the cross-modal fusion module , first realizes feature space dimension reduction and noise suppression through a bottleneck type full connection layer:

[0099]

[0100] wherein is a trainable weight matrix responsible for projecting the 64-dimensional input feature to a 32-dimensional hidden space, is a corresponding bias term; is a Gaussian error linear activation function, which has a smoother gradient characteristic than ReLU and is more suitable for modeling the continuous distribution of emotion-related features. The core role of this layer is to generate a feature decoupling mask and cross-modal noise filtering. Through the sparse training of the weight matrix , the inhibition mask of the steady-state component in the environmental parameter is automatically learned, and the activation strength of the physiological transient feature is enhanced; the gating effect of the GELU function can effectively filter the coupling interference of environmental sensor noise and physiological signal drift.

[0101] In the primary compression feature On this basis, a layer normalization-full connection joint distillation layer is introduced to extract emotion-sensitive factors highly related to TMD from the mixed features

[0102]

[0103] where is the feature vector after layer normalization and full connection layer processing, and represents the abstract features highly related to emotional disorder extracted from the mixed features, is a trainable weight matrix, is the corresponding bias term; is the layer normalization operation. In the specific operation, normalization is performed along the feature dimension:

[0104]

[0105] where, is the th component of , representing the normalized value of the th dimension in the feature vector. and are the mean and variance respectively, and the normalization parameter can learn to adjust the feature scale. This layer automatically strengthens the weight coefficients of the emotion disorder-related features through backpropagation.

[0106] The final layer adopts an affine transformation-Tanh activation joint structure to map the emotion-sensitive factors to a normalized TMD scalar:

[0107]

[0108] where is a trainable weight matrix, is the corresponding bias term; the output is constrained to the interval, where the specific definition of , and represent the exponential functions of and respectively, corresponding to the normalization processing of the actual measurement value to obtain the normalized value :

[0109]

[0110] where, is the actual measurement value of the total emotional disorder score of BPOMS, and are the total number of samples in the dataset min and max of the sigmoid function, for linear mapping of actual measurement values to the interval [0, 1].

[0111] Compared with the sigmoid function, the zero-centering property of the tanh activation allows the backpropagation gradient to remain valid for a wider range of values, biasing the item to dynamically compensate for individual psychological trait differences, setting the prior distribution through Bayesian optimization , making the model universally applicable to subject groups with different emotional baselines.

[0112] Further, a dynamic adjustment coefficient is generated:

[0113]

[0114] The output is constrained by the hyperbolic tangent function, which controls the contribution weight of environmental features and physiological features. When , the environmental feature contribution is strengthened (suitable for stable physiological state scenarios), and when , the physiological feature weight is increased (suitable for individual abnormal state detection), further implementing psychological state compensation and analyzing and correcting thermal comfort assessment bias.

[0115] The psychological intermediary prediction module quantifies the discretization level of overall emotional disorder based on a two-stage feature decoupling compression algorithm through a feature decoupling compression network, and finally generates a dynamic adjustment coefficient with physiological physical significance, dynamically balancing the contribution weights of environmental parameters and physiological signals.

[0116] In some embodiments, with reference Figure 2 , the thermal comfort assessment module includes a spatiotemporal feature integration unit, a multi-scale prediction unit, and a cross-module joint loss function; in the above step S104, the fusion feature vector and the dynamic adjustment coefficient are input into the thermal comfort assessment module, and through the integration of spatiotemporal feature interaction patterns and loss function optimization, a multi-scale prediction result is output, specifically including:

[0117] Based on the spatiotemporal feature integration unit, the contribution weights of environmental features and physiological features in the fusion feature vector are adjusted in real time through the dynamic adjustment coefficient, the spatiotemporal feature interaction patterns are integrated, and the final hidden state is obtained;

[0118] Based on the multi-scale prediction unit, the thermal comfort score and its confidence are obtained by performing multi-scale prediction on the final hidden state;

[0119] Based on the cross-module joint loss function, loss optimization is performed according to the prediction values output by the multi-scale prediction unit and the psychological intermediary module.

[0120] In this embodiment, the thermal comfort assessment module is composed of three parts: spatio-temporal feature integration unit, multi-scale prediction unit, and cross-module joint loss function. Its core function is to receive the fusion feature vector and dynamic adjustment coefficient, integrate the spatio-temporal feature interaction mode, and optimize the loss function, finally output the multi-scale thermal comfort prediction results, providing a comprehensive and accurate basis for outdoor thermal environment assessment.

[0121] The influence degree of environmental features and physiological features in the fusion feature vector on thermal comfort perception will change with the changes of individual state and environmental conditions. Dynamic adjustment coefficient plays a key role here, which can adjust the contribution weight of the two features in the evaluation process in real time. Specifically, an initial weight is set for environmental features and physiological features respectively, and the weight is adjusted according to the size and direction of the dynamic adjustment coefficient. After the weight adjustment, the spatio-temporal feature interaction mode is integrated for the fusion feature vector. In the time dimension, the trend of feature change over time is considered, for example, for the fusion feature vector collected at multiple time points, the change rate of each feature in the time series is calculated, and whether it is rising, falling or stable is analyzed. Through the analysis of these trends, the dynamic adaptation process of the human body in the thermal environment can be understood. In the spatial dimension, the data of monitoring points in different geographical locations are combined. If multiple monitoring points are set in different areas of the city, the fusion feature vector of each monitoring point is classified according to the geographical location. The correlation and difference between features in different areas are analyzed, for example, the environmental features of some areas are similar, but the physiological features are different, which may be related to the characteristics of the population in this area (such as age distribution, activity type, etc.). The feature vectors of different areas at each time point are spliced to form a comprehensive feature vector containing spatio-temporal information. Then, the key features in the comprehensive feature vector are weighted through attention mechanism, highlighting the features that have greater influence on thermal comfort perception, so as to obtain the final hidden state. Attention mechanism can automatically assign different weights according to the correlation and importance between features, so that the final hidden state can more accurately reflect the thermal comfort state of the human body in complex outdoor environment.

[0122] Based on the final hidden state output by the spatio-temporal feature integration unit, the overall scale thermal comfort prediction is carried out. The overall scale prediction aims to evaluate the average thermal comfort level of the whole study area (such as a city area). The thermal comfort score can adopt a grading system of 1 - 10 points, and the higher the score, the higher the thermal comfort. At the same time, according to the prediction confidence evaluation method of the model (such as based on the number of samples and feature consistency), the confidence of the overall prediction result is given. The confidence can be expressed as a percentage, for example, 80% confidence means that the prediction result has an 80% chance of being accurate.

[0123] Local scale prediction focuses on the thermal comfort of specific locations or small areas within the region of interest, such as the perimeter of a single building, a park, etc. For each local area, its corresponding subset of final hidden states is extracted. A more refined analysis method, such as case-based reasoning, is employed. A large number of thermal comfort cases of different local areas under different environmental conditions are collected, including environmental parameters, physiological responses, psychological indicators, and actual thermal comfort evaluations in the cases. The subset of final hidden states of the current local area is matched with the cases in the case library to find the most similar case. According to the thermal comfort evaluation results of the similar case, the thermal comfort score and confidence of the current local area are predicted. Since the characteristics of local areas may be more complex and diverse, the case-based reasoning method can better capture these subtle differences and improve the accuracy of local prediction. Through the prediction of overall scale and local scale, multi-scale thermal comfort scores and their confidence are obtained, providing a basis for decision-making at different levels.

[0124] A cross-module joint loss function is constructed, which comprehensively considers the difference between the predicted values of the multi-scale prediction unit and the psychological mediation module and the actual thermal comfort evaluation. The loss function can be designed as a weighted sum of multiple parts.

[0125] For the multi-scale prediction unit, the loss part includes the overall prediction error and the local prediction error. The overall prediction error measures the difference between the predicted overall thermal comfort score and the actual overall thermal comfort evaluation, and the local prediction error measures the difference between the predicted local thermal comfort score and the actual local thermal comfort evaluation. For the psychological mediation module, the loss part measures the difference between the predicted influence of the psychological state on thermal comfort and the actual role of the psychological state in thermal comfort. Appropriate weights are assigned to each loss part, and the weights can be adjusted according to the importance of different parts to the thermal comfort evaluation. An iterative optimization method is used to adjust the parameters in the thermal comfort evaluation module according to the value of the cross-module joint loss function. In each iteration, the current loss function value is calculated. According to the gradient information of the loss function or other optimization strategies (such as heuristic search algorithms), the weight distribution parameters in the spatio-temporal feature integration unit, the model parameters in the multi-scale prediction unit, etc. are updated. Through multiple iterations, the parameters are constantly optimized, and the value of the cross-module joint loss function gradually decreases, thereby improving the overall prediction performance of the thermal comfort evaluation module and ensuring that the multi-scale prediction results are more accurate and reliable.

[0126] Through the specific implementation steps of the above spatio-temporal feature integration unit, multi-scale prediction unit and cross-module joint loss function, the thermal comfort evaluation module can fully utilize the fusion feature vector and dynamic adjustment coefficient, integrate spatio-temporal feature interaction patterns, and output accurate and multi-scale thermal comfort prediction results through loss function optimization, providing strong support for outdoor thermal environment research and application.

[0127] Further, the contribution weight of the environmental features and physiological features in the fusion feature vector is adjusted in real time by dynamically adjusting the coefficient, the spatio-temporal feature interaction mode is integrated, and a final hidden state is obtained, specifically including:

[0128] A gated recurrent network is constructed to dynamically adjust the coefficient as a conditional constraint, the contribution weight of the environmental features and physiological features is adjusted in real time by the gated recurrent network, and integrated features are obtained;

[0129] The historical hidden state is obtained, the overall emotional disorder prediction result is projected according to the integrated features, the candidate state weight matrix and the emotional state projection matrix, and a candidate state vector is obtained;

[0130] The emotional enhancement gate is constructed, the integrated features and the candidate state vector are input into the emotional enhancement gate, the overall emotional disorder prediction result is adjusted by the gating through the reference weight matrix and the modulation matrix, and an updated gating vector is obtained, which is used to control the fusion ratio of the historical hidden state and the current candidate state;

[0131] Based on the updated gating vector, the final hidden state is obtained by fusing the candidate state vector and the historical hidden state.

[0132] In this embodiment, the gated recurrent network is designed to fuse the time sequence features and the psychological state feedback, and the dynamic adjustment coefficient generated by the psychological intermediary module is introduced as a conditional constraint, first, the environmental-physiological feature dynamic weighting is performed, and the contribution weight of the environmental features and physiological features is adjusted in real time through to output integrated features :

[0133]

[0134] Among them is the environmental feature, is the physiological feature (projected into a 32-dimensional space). Then, the emotional enhancement gate is designed,

[0135] is the updated gating vector, which is used to control the fusion ratio of the historical hidden state and the current candidate state , and the designed gating mechanism can retain important historical states:

[0136]

[0137] Among them is the reference weight matrix, which learns the general mode of feature interaction; is the overall emotional disorder prediction result ​modulation matrix, which is mapped to the weight correction amount, dynamically adjusts the gating decision logic; Sigmoid compresses the output to interval, realizing a soft selection mechanism.

[0138] On the basis of the above, the emotional state projection, is a candidate state vector, representing the new feature representation of the current time step, which is calculated as follows:

[0139]

[0140] wherein is a candidate state weight matrix, encoding the spatiotemporal feature interaction pattern. is an emotional state projection matrix, which maps to a 256-dimensional feature space, realizing explanatory correction of psychological state on physiological response. represents a function for constraining the output to (-1, 1), alleviating the problem of gradient explosion. This part corrects the candidate state by suppressing the feature deviation caused by emotional fluctuations.

[0141] Further, the thermal comfort score and its confidence are obtained by multi-scale prediction of the final hidden state, specifically including:

[0142] A probability prediction network based on psychological-physiological joint constraint is constructed, which includes a psychological modulation prediction head, an uncertainty quantization unit, and an interpretable sampling unit.

[0143] Based on the probability prediction network, the weight distribution of the final hidden state is dynamically adjusted by dynamically adjusting the coefficient, and the mean prediction result of psychological modulation is output.

[0144] Based on the uncertainty quantization unit, the final hidden state is processed according to the first amplification coefficient to obtain an emotional enhancement uncertainty quantization result, wherein the first amplification coefficient represents the amplification coefficient of the overall emotional disturbance degree on uncertainty.

[0145] Based on the mean prediction result of psychological modulation and the emotional enhancement uncertainty quantization result, the Gaussian distribution processing is performed through the interpretable sampling unit to obtain the thermal comfort score and its confidence.

[0146] In this embodiment, a probability prediction network based on psychological-physiological joint constraint is constructed to realize the output of thermal comfort score. First, the psychological modulation prediction head is designed to dynamically adjust the weight distribution of the final hidden state after spatiotemporal feature integration, output the mean prediction of psychological modulation: ​

[0147]

[0148] and are the weight matrix and bias term of linear transformation, respectively, to output the mean of thermal comfort score. is the weight matrix of dynamic adjustment coefficient, which extends the scalar to 256-dimensional space. represents function to constrain to to prevent the adjustment amplitude from being too large. is the Hadamard product, which realizes personalized modulation of hidden state .

[0149] Secondly, uncertainty quantification design is performed, which is the uncertainty quantification with emotional enhancement, and its calculation method is as follows:

[0150]

[0151] where , are the linear transformation parameters (weight matrix and bias term) to generate the standard deviation base value. is to ensure that the standard deviation is within a certain range, while maintaining the smoothness and differentiability of the function. is the overall emotional disorder amplification coefficient of uncertainty. The overall emotional disorder amplifies the predicted standard deviation, reflecting the impact of psychological fluctuations on the stability of evaluation. The output uncertainty quantification result is used to assist users in understanding the prediction reliability.

[0152] Finally, interpretable sampling is performed to output the thermal comfort score and its confidence:

[0153]

[0154] where is the standard Gaussian noise to realize the reparameterization trick. is the variational posterior distribution conditioned on the feature . is the contribution coefficient of the variable , which balances the influence of psychological traits and random noise. By , the differentiability of the random sampling process is realized, supporting the gradient backpropagation to the psychological intermediary module. This part establishes the mapping relationship between psychological traits and physiological responses by introducing the variable .

[0155] Further, the cross-module joint loss function includes a thermal comfort regression loss function and a psychological modeling loss function; the cross-module joint loss function is used to perform loss optimization on the predicted values output by the multi-scale prediction unit and the psychological intermediary module, and specifically includes:

[0156] Based on the thermal comfort regression loss function, the thermal comfort predicted value output by the multi-scale prediction unit and the thermal comfort true value are used to perform loss optimization processing on the output value of the multi-scale prediction unit.

[0157] Based on the psychological modeling loss function, the cross-entropy loss KL divergence loss of the emotional disorder degree prediction task is used to perform loss optimization processing on the output value of the psychological intermediary module.

[0158] In this embodiment, a cross-module joint loss function is designed to realize end-to-end multi-task optimization. First, the thermal comfort regression loss is defined, and the mean square error (MSE) loss is used to ensure the thermal comfort prediction accuracy:

[0159]

[0160] wherein is the number of batch training samples, and are the thermal comfort predicted value and the true value of the i-th sample, respectively; and are the dynamic adjustment coefficient predicted value and the true value of the i-th sample, respectively, is calibrated by experimental psychology, and is used to improve the physical meaning of the adjustment coefficient, is a weight parameter of the adjustment coefficient constraint term. The first term MSE forces the model output to approximate the true thermal comfort score

[0161] , ensuring the reliability of the core function. The second term uses weighted constraint to approach the true value , preventing the psychological intermediary module from outputting meaningless adjustment coefficients. This loss jointly optimizes thermal comfort prediction and psychological adjustment coefficient generation, avoiding feature bias caused by module fragmentation. Secondly, the psychological modeling loss is introduced, and the CE+KL loss maintains the rationality of the psychological state modeling:

[0162]

[0163]

[0164] ​​

[0165]

[0166] in Cross-entropy loss for a task predicting the degree of emotional disturbance. For the first True labels (one-hot encoding) for the degree of overall mood disorder. For the output of the first The probability of overall emotional disturbance at a certain degree. The KL (Kullback-Leibler) divergence loss is used. The output of the psychological mediation module Mean of the variable For the first Standard deviation of the variable. Then the variable generated by the psychological mediation module is the first one. dimension; , , These are the weighting coefficients for each loss term. This represents the L1 regularization function.

[0167] In this loss definition, Ensure that the overall emotional disturbance prediction is consistent with the thermal comfort score logic; Constraint variables Follows a standard normal distribution To prevent overfitting; (L1 regularization) forces non-critical dimensions to approach zero, retains core psychological factors, and improves the interpretability of psychological traits.

[0168] Finally, an annealing training strategy was adopted to match the human cognitive pattern (perceiving the environment first and then understanding psychology) to improve the model's convergence stability.

[0169]

[0170] in For the first The total weight of the mental modeling loss during step training. For the initial total weight ( ), This is the number of warm-up training steps (approximately 5 epochs).

[0171] Early training phase ( Gradually increase The weighting of psychological prediction tasks should be adjusted to avoid premature interference with the thermal comfort regression of the main task. hour, Fully utilize psychological modeling loss. Balance feature extraction through progressive weight adjustment. ) and mental modeling ( The gradient update direction is determined to prevent optimization oscillations.

[0172] Reference Figure 3 An embodiment of the present invention provides an intelligent outdoor thermal comfort assessment system 3 mediated by psychological state, the system 3 specifically comprising:

[0173] The data acquisition module 301 acquires psychological index data and multi-scale parameters. The psychological index data includes multiple emotional factor data that can be converted into overall emotional disorder data. The multi-scale parameters include environmental physical parameters and physiological response parameters.

[0174] The feature fusion module 302 is used to input environmental physical parameters and physiological response parameters into the dynamic feature extraction module, and generate a fused feature vector by encoding environmental physical parameters and extracting physiological time-series features, and performing cross-modal feature fusion.

[0175] The dynamic adjustment module 303 is used to construct a psychological mediation prediction module based on psychological index data. The fused feature vector is input into the psychological mediation prediction module, and a dynamic adjustment coefficient is generated by predicting the degree of emotional disorder.

[0176] The prediction output module 304 is used to input the fused feature vector and dynamic adjustment coefficient into the thermal comfort assessment module, and output multi-scale prediction results by integrating the spatiotemporal feature interaction mode and loss function optimization.

[0177] It is understandable that, such as Figure 1 The content of the illustrated embodiment of the outdoor thermal comfort intelligent assessment method mediated by psychological state is applicable to the embodiment of this outdoor thermal comfort intelligent assessment system mediated by psychological state. The specific functions implemented by the embodiment of this outdoor thermal comfort intelligent assessment system mediated by psychological state are as follows: Figure 1 The embodiment of the intelligent assessment method for outdoor thermal comfort mediated by psychological state shown is the same, and the beneficial effects achieved are the same as those described above. Figure 1 The beneficial effects achieved by the illustrated embodiment of the intelligent assessment method for outdoor thermal comfort mediated by psychological state are also the same.

[0178] It should be noted that the information interaction and execution process between the above systems are based on the same concept as the method embodiments of the present invention. For details on their specific functions and technical effects, please refer to the method embodiments section, which will not be repeated here.

[0179] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above functional units and modules is exemplified, and in actual application, the above functions can be completed by different functional units and modules according to needs, that is, the internal structure of the system is divided into different functional units or modules to complete all or part of the above described functions. The functional units and modules in the embodiments can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit, and the integrated unit can be realized in the form of hardware or software. In addition, the specific names of the functional units and modules are only for easy distinction, and do not limit the protection scope of the application. The specific working process of the units and modules in the system can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0180] With reference to Figure 4 The embodiment of the present application also provides a computer device 4, which comprises a memory 402, a processor 401 and a computer program 403 stored in the memory 402, and when the computer program 403 is executed on the processor 401, the computer program 403 implements the outdoor thermal comfort intelligent evaluation method based on psychological state as described in any one of the above methods.

[0181] The computer device 4 can be a desktop computer, a notebook computer, a palm computer, a cloud server and the like. The computer device 4 can include, but is not limited to, a processor 401 and a memory 402. Those skilled in the art can understand that, Figure 4 The computer device 4 is only an example and does not constitute a limitation on the computer device 4, and can include more or fewer components than those shown, or combine certain components, or different components, for example, can also include an input / output device, a network access device and the like.

[0182] The processor 401 can be a central processing unit (CPU), and the processor 401 can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0183] The memory 402 may, in some embodiments, be an internal storage unit of the computer device 4, such as a hard disk or a memory of the computer device 4. The memory 402 may, in other embodiments, also be an external storage device of the computer device 4, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, and the like, equipped on the computer device 4. Further, the memory 402 may, in addition, include both an internal storage unit and an external storage device of the computer device 4. The memory 402 is used to store an operating system, application programs, a Boot Loader, data, and other programs, such as program codes of the computer program, and the like. The memory 402 may, in addition, be used to temporarily store data that has been output or is to be output.

[0184] The embodiment of the present application further provides a computer readable storage medium, which stores a computer program. When the computer program is run by a processor, the computer program implements the outdoor thermal comfort intelligent evaluation method based on a psychological state as a medium.

[0185] In the embodiment, the integrated unit, if implemented in the form of a software functional unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the present application implements all or part of the processes in the above-mentioned embodiment methods, which can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer readable storage medium, and the computer program, when executed by a processor, can implement the steps of the above-mentioned method embodiments. The computer program includes computer program codes, which can be in the form of source code, object code, executable files, or some intermediate forms, etc. The computer readable medium at least includes any entity or device capable of carrying the computer program code to a photographing device / terminal equipment, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. For example, a U disk, a mobile hard disk, a magnetic disk or an optical disk, etc. In some jurisdictions, according to legislation and patent practice, the computer readable medium cannot be an electrical carrier signal and a telecommunication signal.

[0186] In the above-mentioned embodiments, the description of each embodiment has its own focus, and the parts not described or recorded in detail in a certain embodiment can be referred to the relevant description of other embodiments.

[0187] Those skilled in the art can understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0188] In the embodiments disclosed in the present application, it should be understood that the disclosed apparatus / terminal device and method can be implemented in other ways. For example, the apparatus / terminal device embodiments described above are only schematic. The division of the modules or units is only a logical function division, and there can be another division in actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual coupling or direct coupling or communication connection between the units can be indirect coupling or communication connection through some interface, device or unit, and can be electrical, mechanical or in other forms.

[0189] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, i.e. they can be located in one place or distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiments.

Claims

1. A method for outdoor thermal comfort intelligent assessment mediated by psychological state, characterized in that, The method specifically comprises: Obtaining psychological index data and multi-scale parameters, the psychological index data comprising multiple emotional factor data convertible into overall emotional disorder degree data, and the multi-scale parameters comprising environmental physical parameters and physiological response parameters; Inputting the environmental physical parameters and the physiological response parameters into a dynamic feature extraction module, performing environmental physical parameter coding and physiological time sequence feature extraction, and performing cross-modal feature fusion to generate a fusion feature vector; Constructing a psychological intermediary prediction module according to the psychological index data, inputting the fusion feature vector into the psychological intermediary prediction module, performing emotional disorder degree prediction to generate a dynamic adjustment coefficient; Inputting the fusion feature vector and the dynamic adjustment coefficient into a thermal comfort evaluation module, performing integration of space-time feature interaction modes and loss function optimization to output multi-scale prediction results; The psychological intermediary prediction module comprises a feature decoupling compression network and a dynamic adjustment coefficient generation unit; the inputting of the fusion feature vector into the psychological intermediary prediction module, the overall emotional disorder degree prediction, and the generation of the dynamic adjustment coefficient specifically comprise: Inputting the fusion feature vector into the feature decoupling compression network, performing feature compression and emotional sensitive factor mapping to obtain overall emotional disorder degree data; Inputting the overall emotional disorder degree data into the dynamic adjustment coefficient generation unit, performing hyperbolic tangent function range constraint to generate the dynamic adjustment coefficient; The inputting of the fusion feature vector into the feature decoupling compression network, the feature compression, and the emotional sensitive factor mapping to obtain the overall emotional disorder degree data specifically comprise: Performing feature dimension reduction and noise suppression on the fusion feature vector through a bottleneck type full connection layer of the feature decoupling compression network to generate primary compression features; Extracting emotional sensitive factors highly related to the overall emotional disorder degree from the primary compression features based on a layer normalization-full connection joint distillation layer of the feature decoupling compression network; Mapping the emotional sensitive factors to a normalized overall emotional disorder degree scalar through affine transformation and Tanh activation based on a final layer of the feature decoupling compression network to obtain an original overall emotional disorder degree prediction value; Obtaining a target overall emotional disorder degree prediction value by performing normalization processing on the original overall emotional disorder degree prediction value.

2. The method of claim 1, wherein, The dynamic feature extraction module comprises an environmental physical parameter encoder, a physiological time sequence feature extractor, and a cross-modal feature fusion unit; the inputting of the environmental physical parameters and the physiological response parameters into the dynamic feature extraction module, the environmental physical parameter coding, the physiological time sequence feature extraction, and the cross-modal feature fusion to generate the fusion feature vector specifically comprise: Inputting the environmental physical parameters into the environmental physical parameter encoder, processing air temperature parameters, relative humidity parameters, radiation temperature parameters, and wind speed parameters through a four-layer full connection neural network to generate an environmental feature vector; Inputting the physiological response parameters into the physiological time sequence feature extractor, processing physiological time sequence signals in multiple time windows through a bidirectional long short-term memory network to obtain a physiological feature vector, the physiological time sequence signals comprising skin temperature, heart rate variability HRV, and local sweat rate; The environmental feature vector and the physiological feature vector are input into a cross-modal feature fusion unit, the environmental feature vector and the physiological feature vector are independently projected in a double-path mode, the importance of each feature dimension is determined through similarity calculation, and an attention weight is generated; Based on the attention weight, the environmental feature vector and the physiological feature vector are element-wise multiplied and added to obtain a fusion feature vector.

3. The method of claim 1, wherein, The thermal comfort evaluation module includes a spatio-temporal feature integration unit, a multi-scale prediction unit, and a cross-module joint loss function; the fusion feature vector and the dynamic adjustment coefficient are input into the thermal comfort evaluation module, the spatio-temporal feature interaction mode is integrated and the loss function is optimized, and a multi-scale prediction result is output, specifically including: Based on the spatio-temporal feature integration unit, the contribution weights of the environmental features and the physiological features in the fusion feature vector are adjusted in real time through the dynamic adjustment coefficient, the spatio-temporal feature interaction mode is integrated, and a final hidden state is obtained; Based on the multi-scale prediction unit, the thermal comfort score and the confidence of the thermal comfort score are obtained by performing multi-scale prediction on the final hidden state; Based on the cross-module joint loss function, loss optimization is performed according to the prediction values output by the multi-scale prediction unit and the psychological intermediary module.

4. The method of claim 3, wherein, The contribution weights of the environmental features and the physiological features in the fusion feature vector are adjusted in real time through the dynamic adjustment coefficient, the spatio-temporal feature interaction mode is integrated, and a final hidden state is obtained, specifically including: A gated recurrent network is constructed, the dynamic adjustment coefficient is used as a conditional constraint, the contribution weights of the environmental features and the physiological features are adjusted in real time through the gated recurrent network, and an integrated feature is obtained; The historical hidden state is obtained, the overall emotional disorder prediction result is projected according to the integrated feature, the candidate state weight matrix, and the emotional state projection matrix, and a candidate state vector is obtained; An emotional enhancement gate is constructed, the integrated feature and the candidate state vector are input into the emotional enhancement gate, the overall emotional disorder prediction result is adjusted through the benchmark weight matrix and the modulation matrix, an updated gate vector is obtained, and the updated gate vector is used to control the fusion ratio of the historical hidden state and the current candidate state; Based on the updated gate vector, the final hidden state is obtained by fusing the candidate state vector and the historical hidden state.

5. The method of claim 3, wherein, The thermal comfort score and its confidence are obtained by performing multi-scale prediction on the final hidden state, specifically including: A psychological-physiological joint constraint-based probability prediction network is constructed, the probability prediction network includes a psychological modulation prediction head, an uncertainty quantization unit, and an interpretable sampling unit; Based on the probability prediction network, the weight distribution of the final hidden state is dynamically adjusted through the dynamic adjustment coefficient, and a psychological modulation mean prediction result is output; Based on the uncertainty quantization unit, the final hidden state is processed according to the first amplification coefficient to obtain an emotional enhancement uncertainty quantization result, and the first amplification coefficient represents the amplification coefficient of the overall emotional disorder degree to uncertainty; Based on the psychological modulation mean prediction result and the emotional enhancement uncertainty quantization result, the thermal comfort score and its confidence are obtained by Gaussian distribution processing through the interpretable sampling unit.

6. The method of claim 3, wherein, The cross-module joint loss function comprises a thermal comfort regression loss function and a psychological modeling loss function; the loss optimization is performed according to the predicted values output by the multi-scale prediction unit and the psychological intermediary module based on the cross-module joint loss function, and specifically comprises: Based on the thermal comfort regression loss function, the output value of the multi-scale prediction unit is loss optimized according to the thermal comfort predicted value and the thermal comfort real value output by the multi-scale prediction unit; Based on the psychological modeling loss function, the output value of the psychological intermediary module is loss optimized according to the cross entropy loss KL divergence loss of the emotional disorder degree prediction task.

7. A mental state-mediated outdoor thermal comfort intelligent assessment system for implementing the method of any one of claims 1-6, characterized in that, The system specifically comprises: The data acquisition module is used for acquiring psychological index data and multi-scale parameters, the psychological index data comprises a plurality of emotional factor data which can be converted into overall emotional disorder degree data, and the multi-scale parameters comprise environmental physical parameters and physiological response parameters; The feature fusion module is used for inputting the environmental physical parameters and the physiological response parameters into the dynamic feature extraction module, performing environmental physical parameter coding and physiological time sequence feature extraction, and performing cross-modal feature fusion to generate a fusion feature vector; The dynamic adjustment module is used for constructing a psychological intermediary prediction module according to the psychological index data, inputting the fusion feature vector into the psychological intermediary prediction module, and generating a dynamic adjustment coefficient through emotional disorder degree prediction; The prediction output module is used for inputting the fusion feature vector and the dynamic adjustment coefficient into the thermal comfort evaluation module, integrating the spatio-temporal feature interaction mode and the loss function optimization, and outputting a multi-scale prediction result.

8. A computer device, comprising: It comprises: A memory and a processor and a computer program stored in the memory, when the computer program is executed on the processor, the outdoor thermal comfort intelligent evaluation method based on psychological state as a medium is realized as claimed in any one of claims 1 to 6.

Images