Intelligent wearable device and method for rapidly identifying personal thermal comfort

Abstract

The invention discloses an intelligent wearable device and method for rapidly identifying personal thermal comfort. The intelligent wearable device is composed of an environment temperature and humidity detection module, a human body skin temperature detection module, a human body heart rate detection module, a thermal comfort voting module, a wireless communication module, a power supply module, a display module and the like. (2) the intelligent wearable device collects small batches of environment / human body physiological parameters, including environment temperature and humidity, human body skin temperature, human body heart rate and thermal comfort voting values, to construct a target domain data set; and (3) the control computer selects similar data from the public pseudo source domain data pool to construct a source domain data set according to target domain data uploaded by the intelligent wearable device. And (4) the control computer constructs a thermal comfort classifier. The system has the advantages of high sensor integration level, small data acquisition amount, high prediction precision and the like, and is beneficial to improving the indoor thermal comfort level and reducing the energy consumption of indoor buildings.

Description

Technical Field

[0001] The present invention belongs to the technical field of indoor thermal environment control, and specifically belongs to an intelligent wearable device and method for rapid identification of thermal comfort of a crowd. Background Art

[0002] Creating good indoor space thermal comfort can not only save energy consumption in buildings, but also improve people's happiness, productivity and work efficiency. Traditional indoor environment prediction methods include but are not limited to PMV (Predicted Mean Vote) method, adaptive comfort model prediction method, etc. The PMV model uses mathematical expressions of environmental factors (air temperature, radiant temperature, wind speed and humidity) and personal factors (metabolic rate and clothing thermal resistance) to represent the occupant's thermal sensation index. Among them, personal factors are mainly roughly estimated, which cannot reflect the differences in thermal comfort of people, regions, etc.; the adaptive model uses linear regression between thermal sensation and working temperature to define an acceptable thermal environment, and there is also a problem of poor generalization performance. Patent CN114322230A discloses a thermal environment adjustment system and method based on a smart bracelet. The smart bracelet in this patent collects two parameters, human body temperature and ambient temperature, and does not involve the collection of human heart rate and ambient humidity described in this patent. Studies have shown that human body temperature and heart rate as well as ambient temperature and humidity are closely related to personal thermal comfort. Secondly, the patent mentions the use of traditional data-driven models such as neural networks or support vector machines for thermal comfort prediction. In contrast, the prediction model described in this patent only needs to collect a small batch of field data and identify thermal comfort through a transfer learning strategy, with a small amount of data collected and good speed. Summary of the invention

[0003] The purpose of the present invention is to provide an intelligent wearable device and method for rapid identification of thermal comfort of a crowd, which can predict the overall thermal comfort level of indoor crowds, improve the automation level of indoor thermal environment regulation, and thus reduce building energy consumption.

[0004] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: an intelligent wearable device for rapid identification of personal thermal comfort, wherein the main components of the intelligent wearable device are arranged on both sides of a double-layer PCB board of the wrist width, and the main components include: an ambient temperature and humidity detection module for detecting ambient temperature and humidity; a skin temperature detection module for obtaining skin temperature; a human heart rate detection module for detecting human heart rate signals; a thermal comfort voting module for users to vote on subjective thermal comfort, a wireless communication module for reading data from each module and transmitting it wirelessly to a control computer; a power supply module for DC power supply of the device; an OLED display module for displaying the collected data, and the ambient temperature and humidity detection module, the skin temperature detection module, the human heart rate detection module, the thermal comfort voting module, the display module, the wireless communication module and the power supply module are all connected to the intelligent wearable device.

[0005] The present invention provides an intelligent wearable method for rapid identification of personal thermal comfort, comprising the following steps: (1) The intelligent wearable device is composed of an environmental temperature and humidity detection module, a human skin temperature detection module, a human heart rate detection module, a thermal comfort voting module, a wireless communication module, a power supply module, an OLED display module, etc. (2) The intelligent wearable device collects a small batch of environmental / human physiological parameters, including environmental temperature and humidity, human skin temperature, human heart rate, and thermal comfort voting values ​​to construct a target domain data set. (3) The control computer selects similar data from a public pseudo-source domain data pool to construct a source domain data set based on the target domain data uploaded by the intelligent wearable device. (4) The control computer constructs a thermal comfort classifier. The classifier parameters are first pre-trained using the source domain data, and then the classifier parameters are fine-tuned using the target domain data. The resulting classifier can be used for rapid identification of personal thermal comfort.

[0006] Furthermore, the method used by the sensor to collect heart rate in step (1) is photoplethysmography (PPG) to measure heart rate. The basic principle of PPG is to use the absorption and reflection characteristics of light, and use photoelectric sensors to detect the light intensity passing through the skin, tissues and blood vessels to calculate the heart rate. These changes are reflected in the PPG signal, and the fluctuation of the signal is synchronized with the heart cycle. Assuming I(t) is the light intensity measured at time t, the PPG signal can be expressed as:

[0007] I(t)=I0+ΔI(t)

[0008] Where I0 is the base light intensity (the constant part reflected by the skin and tissue), ΔI(t) is the part of the light intensity change caused by the change in blood volume. During the sensor detection process, peak values ​​can be clearly detected. By detecting these peak values, the heart rate can be calculated. The heart rate (HR) can be expressed as:

[0009]

[0010] Where N is the number of peaks, T is the time interval (seconds), and 60 is used to convert seconds to minutes. The final HR is the heart rate per minute.

[0011] Furthermore, the smart wearable device is worn on the user's wrist, the heart rate sensor and the skin temperature sensor are in close contact with the user's skin, and the control computer maintains communication with the wireless communication module of the smart wearable device. During the data collection process, the smart wearable device display module is used to display the information collected by the sensor in real time, including: human skin temperature, human heart rate, environmental temperature and humidity, and thermal perception voting value; the thermal comfort voting module is used to vote for the user's thermal perception under the current environmental state.

[0012] Furthermore, in step (2), the subject is required to vote for the thermal value in a stable indoor thermal environment, and use the smart wearable device to vote for the current thermal value. After the voting is completed, press the button to indicate confirmation. At this time, the OLED display screen on the smart wearable device will display the message sending status. After the sending is completed, the display screen will display "OK", indicating that one set of information collection is completed. The smart wearable device is used to collect the subject's physiological state information, surrounding environment information, and seven-level thermal voting values. 50 to 60 groups of data are collected in small batches to construct the target domain data set to prepare for the transfer learning of the model.

[0013] Furthermore, in step (3), due to the influence of factors such as region, climate, and subjects, it is found that the data distribution difference between the target dataset and the public dataset is too large, so it is necessary to use a nearest neighbor search algorithm to find data in the public dataset that is similar to the target domain for transfer learning. The main process of this method is as follows: Assuming that all data are of unknown category, based on the Euclidean distance, find the data with the smallest distance to the given query item in a search space.

[0014] The Euclidean distance is defined as:

[0015]

[0016] where x i =(x1,x2,...x n ) and y i =(y1,y2,...y n ) are two given n-dimensional data points, which represent the data points in the target domain and the source domain data set in the present invention. Indoor temperature, humidity, skin temperature, and human heart rate are selected as features, and the source domain (data in the public data set) is used as the search space for neighbor search. The calculation formula is as follows:

[0017]

[0018] Y=unique(y1,y2,…,y m )

[0019] where dist(q t ,x) is the calculation of q t The Euclidean distance from x, q t is the data query point in the target domain dataset, t represents the number of features, x is the data point in the source domain dataset, and the argmin operation is used to find the point with the smallest Euclidean distance as y t Then, the unique function is used to exclude duplicate objects, and the searched data is used as the source domain data set.

[0020] Furthermore, the classifier model in step (4) is constructed based on a one-dimensional convolutional neural network 1D-CNN model; wherein the 1D-CNN model includes two convolutional layers and one fully connected layer, and a ReLU activation function is applied after each convolutional layer to enhance the nonlinear expression ability of the model. After the convolutional layer, the model uses a fully connected layer containing 50 neurons, which is used to combine and transform the features extracted by the convolutional layer to further extract high-level features; in the process of using the target domain to train the personal thermal comfort classifier, the parameters in the fully connected layer 2 of the source domain training neural network model are migrated to the fully connected layer 2 of the target domain model. Finally, the output from the fully connected layer is passed to the output layer, and the softmax activation function is used to improve the classification efficiency and accuracy of the 1D-CNN model, thereby improving the accuracy of the model.

[0021] Furthermore, the model is deployed on a control computer. At this time, the subject wears a smart wearable device, which will continuously collect data such as human physiological parameters and environmental information parameters without affecting normal office work, and send the data to the control computer through the wireless communication module. The control computer runs a personal thermal comfort classifier for prediction, and then obtains the thermal comfort status of the subject in the current state.

[0022] Beneficial effects of the present invention:

[0023] In view of the fact that only small batches of data can be collected or the data is insufficient in an online environment, the present invention introduces the model transfer learning method. By using smart wearable devices in an online environment, the thermal comfort of the crowd in a small sample situation can be quickly predicted. Finally, the measured thermal comfort can be applied to the HVAC system to improve the level of automatic temperature regulation in the system while reducing building energy consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is the overall flow chart of the method

[0025] Figure 2 The composition diagram of the smart wearable device

[0026] Figure 3 This is the circuit PCB structure diagram of the smart wearable device in the present invention

[0027] Figure 4 is the thermal comfort voting value range;

[0028] In the figure, 1. smart wearable device; 2. thermal comfort voting module; 3. human heart rate detection module; 4. ambient temperature and humidity detection module; 5. human skin temperature detection module; 6. display module; 7. wireless communication module; 8. power supply module. DETAILED DESCRIPTION

[0029] The present invention proposes a smart wearable device and method for quickly identifying the thermal comfort of a crowd. The purpose of the present invention is to use the smart wearable device to quickly predict the thermal comfort of the crowd in the current state, which is helpful to improve the automatic adjustment performance of the HVAC system. The overall flow chart is as follows Figure 1 As shown, the following steps are included:

[0030] Step 1: Encapsulate the thermal comfort voting module 2, the human heart rate detection module 3, the ambient temperature and humidity detection module 4, the skin temperature detection module 5, the display module 6, the wireless communication module 7, and the power supply module 8 into a smart wearable device such as Figure 2 .

[0031] The present invention adopts a double-layer PCB board structure, including an upper layer and a lower layer, and its overall dimensions are 65*33*1.6MM. Figure 3 As shown, various sensors and communication components are arranged separately.

[0032] The thermal comfort voting module device uses a GPIO port to connect to a communication device. The device used is two 4.5*4.5*5 vertical patch 4-pin micro-button switches. The right switch is used to confirm to the user that the data will be sent to the control computer. When the button is pressed, the data will be packaged and sent to the control through the communication module.

[0033] The human heart rate detection module adopts the MAX30102 human heart rate acquisition module and the communication module adopts IIC to exchange data, and the heart rate detection module adopts the photoplethysmography method to detect the human heart rate;

[0034] The environmental temperature and humidity detection module adopts the DHT11 module to use temperature-sensitive and humidity-sensitive resistors to detect environmental parameters in real time, and obtains corresponding temperature and humidity data through the amplifier circuit and filter circuit in the circuit. The module can detect the temperature and humidity of the environment in real time, and transmit the data to the main controller, and exchange data with the communication module through the GPIO port.

[0035] The human skin temperature and humidity detection module adopts AHT10, uses SMD packaging, and uses an improved MEMS semiconductor capacitive humidity sensor and a standard on-chip temperature sensor component to measure the skin surface temperature. The module exchanges data with the communication module through the GPIO port;

[0036] The OLED display module uses a 0.96-inch OLED display LCD serial port screen module, which uses IIC for communication;

[0037] The wireless communication module adopts ESP8266 NODEMCU, which is composed of ESP8266WiFi module, USB to TTL chip, 5V to 3.3V LDO, and automatic download circuit;

[0038] The power supply module is powered by a 3.7V-300mAh polymer lithium battery. The lithium battery charging and discharging circuit uses a tp4056 lithium battery management chip. The chip uses SOP-8 packaging technology, with a size of 5.2*4.9*1.2mm, an operating temperature of -40°C to 130°C, and an operating voltage of 1 to 8V.

[0039] Step 2: The smart wearable device collects a small batch of environmental / human physiological parameters, including environmental temperature and humidity, human skin temperature, human heart rate, and thermal comfort voting values ​​to construct the target domain dataset.

[0040] The basic process of collecting small batch thermal comfort data on site is as follows: first, keep the target indoor thermal environment stable, the subject wears the smart wearable device to enter the room, and the control computer maintains a stable communication state with the smart wearable device in real time. After the subject's thermal sensation state is stable, enter his or her 7-level thermal comfort state in the smart wearable device. The thermal comfort value range is as follows: Figure 4 As shown in the figure, when the user presses the confirmation button, the screen displays "OK", indicating that the smart wearable device has packaged the data and sent it to the control computer, and a set of data has been collected. After that, the subjects adjust the air-conditioning temperature according to their own thermal preferences. After the current thermal environment stabilizes, the subjects vote for thermal sensation again on the smart wearable device. There is no limit to the number of collection times and collection time, and 10 to 50 groups of experimental samples can be collected on site. The control computer preprocesses these data sets, including outlier detection and normalization processing, and the processed data sets are used as the target domain data sets.

[0041] Step 3: The control computer selects similar data from the public pseudo source domain data pool to construct a source domain data set based on the target domain data uploaded by the smart wearable device.

[0042] The RP-884 database of the American Society of Heating, Refrigerating and Air-Conditioning Engineers is one of the public databases widely used in the study of human thermal comfort. The RP-884 dataset involves more than 25,000 observations collected from 52 studies and 26 cities in different climate zones around the world. The target domain dataset generated by step 2 uses the nearest neighbor search algorithm (NNS) to filter similar data in the RP-884 dataset. The data features include: ambient temperature, ambient humidity, human heart rate, skin temperature, and thermal perception voting. The filtered data constitute the source domain dataset in this method.

[0043] Step 4: Control the computer to build a thermal comfort classifier.

[0044] First, the classifier parameters are pre-trained using source domain data, and then the classifier parameters are fine-tuned using target domain data. The resulting classifier can be used to quickly identify the thermal comfort of people. The parameters for constructing the classifier are shown in Table 1. The Dropout parameter can handle the overfitting problem by randomly reducing the number of neurons in the hidden layer during training. The batch size is related to the generalization of the model performance. By training the entire data set and dividing them into small parts, the memory cost can be reduced, the speed of the classifier's prediction can be accelerated, and the real-time performance can be improved.

[0045] Table 1

[0046]

[0047] Step 5: Use smart wearable devices to collect physiological and environmental information online, and quickly identify and predict personal thermal comfort based on the established thermal comfort model.

[0048] The sensors on the wearable device collect human physiological signals and surrounding environmental parameters online in real time, and upload the data to the control computer through the communication component. The control computer inputs the collected data into the constructed thermal comfort classifier, and the predicted result is the personal thermal comfort level in the current thermal environment.

Claims

1. A smart wearable device for rapid identification of personal thermal comfort, characterized in that: The main components of the smart wearable device are arranged on both sides of a double-layer PCB board with a wrist width. The main components include: an environmental temperature and humidity detection module, which is used to detect environmental temperature and humidity; a skin temperature detection module, which is used to obtain skin temperature; a human heart rate detection module, which is used to detect human heart rate signals; a thermal comfort voting module, which is used for users to vote on subjective thermal comfort, and a wireless communication module, which is used to read data from each module and transmit it wirelessly to a control computer; a power supply module, which is used for DC power supply of the device; an OLED display module, which is used to display the collected data. The environmental temperature and humidity detection module, the skin temperature detection module, the human heart rate detection module, the thermal comfort voting module, the display module, the wireless communication module and the power supply module are all connected to the smart wearable device.

2. According to claim 1, a smart wearable method for rapid identification of personal thermal comfort is characterized in that: The method comprises the following steps: (1) the intelligent wearable device comprises an environmental temperature and humidity detection module, a human skin temperature detection module, a human heart rate detection module, a thermal comfort voting module, a wireless communication module, a power supply module, and a display module; (2) the intelligent wearable device collects a small batch of environmental / human physiological parameters, including environmental temperature and humidity, human skin temperature, human heart rate, and thermal comfort voting values ​​to construct a target domain data set; (3) The control computer selects similar data from the public pseudo-source domain data pool to construct a source domain data set based on the target domain data uploaded by the smart wearable device; (4) The control computer constructs a thermal comfort classifier, first pre-trains the classifier parameters using the source domain data, and then fine-tunes the classifier parameters using the target domain data. The resulting classifier can be used to quickly identify personal thermal comfort.

3. The intelligent wearable method for rapid identification of personal thermal comfort according to claim 2, characterized in that: In the step (1), the method used by the human heart rate detection module to collect heart rate is photoplethysmography (PPG) to measure heart rate. The basic principle of PPG is to use the absorption and reflection characteristics of light and use photoelectric sensors to detect the light intensity passing through the skin, tissues and blood vessels to calculate the heart rate. These changes are reflected in the PPG signal, and the fluctuation of the signal is synchronized with the heart cycle. Assuming I(t) is the light intensity measured at time t, the PPG signal can be expressed as: I(t)=I0+ΔI(t) Among them, I0 is the basic light intensity, the constant part reflected by the skin and tissue, ΔI(t) is the part of the light intensity change caused by the change in blood volume. During the sensor detection process, peak values ​​can be clearly detected. By detecting these peak values, the heart rate can be calculated. The heart rate HR can be expressed as: Where N is the number of peaks, T is the time interval, and 60 is used to convert seconds to minutes. The final HR is the heart rate per minute.

4. The intelligent wearable method for rapid identification of personal thermal comfort according to claim 2 is characterized in that: The smart wearable device is worn on the user's wrist, the human heart rate detection module and the skin temperature detection module are close to the user's skin, the control computer maintains communication with the wireless communication module of the smart wearable device, and during the data collection process, the display module is used to display the information collected by the sensor in real time, including: human skin temperature, human heart rate, ambient temperature and humidity, and thermal perception voting value; the thermal comfort voting module is used to vote for the user's thermal perception under the current environmental conditions.

5. The intelligent wearable method for rapid identification of personal thermal comfort according to claim 2 is characterized in that: In step (2), the subject is required to vote on the thermal sensation value in a stable indoor thermal environment, and use the smart wearable device to vote on the current thermal sensation value. After the voting is completed, press the button to indicate confirmation. At this time, the display module on the smart wearable device will display the message sending status. After the sending is completed, the display screen of the display module will display "OK", indicating that a set of information collection is completed. The smart wearable device is used to collect the subject's physiological state information, surrounding environment information, and seven-level thermal sensation voting values. 50 to 60 groups of data are collected in small batches to construct the target domain data set to prepare for the transfer learning of the model.

6. The intelligent wearable method for rapid identification of personal thermal comfort according to claim 2, characterized in that: In step (3), due to the influence of geographical, climatic, and subject-related factors, it is found that the data distribution difference between the target dataset and the public dataset is too large, and it is necessary to use a nearest neighbor search algorithm to find data in the public dataset that is similar to the target domain for transfer learning. The main process of this method is as follows: Assuming that all data are of unknown categories, based on the Euclidean distance, find the data with the smallest distance to a given query item in a search space; The Euclidean distance is defined as: where x i =(x1,x2,...x n ) and y i =(y1,y2,...y n ) are two given n-dimensional data points, which respectively represent the data points in the target domain and the source domain data set in the present invention. Indoor temperature, humidity, skin temperature, and human heart rate are selected as features, and the source domain is used as the search space for nearest neighbor search. The calculation formula is as follows: And unique(y1,y2,…,y m ) where dist(q t ,x) is the calculation of q t The Euclidean distance from x, q t is the data query point in the target domain dataset, t represents the number of features, x is the data point in the source domain dataset, and the argmin operation is used to find the point with the smallest Euclidean distance as y t , then use the unique function to exclude duplicate objects and use the searched data as the source domain data set.

7. The intelligent wearable method for rapid identification of personal thermal comfort according to claim 2, characterized in that: The classifier model in the step (4) is constructed based on a one-dimensional convolutional neural network 1D-CNN model; wherein the 1D-CNN model includes two convolutional layers and one fully connected layer, and a ReLU activation function is applied after each convolutional layer to enhance the nonlinear expression ability of the model. After the convolutional layer, the model uses a fully connected layer containing 50 neurons, which is used to combine and transform the features extracted by the convolutional layer, thereby further extracting high-level features; in the process of using the target domain to train the personal thermal comfort classifier, the parameters in the fully connected layer 2 of the source domain training neural network model are migrated to the fully connected layer 2 of the target domain model. Finally, the output from the fully connected layer is passed to the output layer, and the softmax activation function is used to improve the classification efficiency and accuracy of the 1D-CNN model, thereby improving the accuracy of the model.

8. According to the intelligent wearable method for rapid identification of personal thermal comfort as described in claim 1, it is characterized in that: In order to solve the problem of unstable accuracy of the human heart rate detection module due to external interference and noise, a mean filtering algorithm is used to obtain the arithmetic mean of the sampled heart rate in a fixed time window as the heart rate value. t , the calculation formula is: Among them, 2k+1 is the window width, and x is the original sampling value of the heart rate sensor.

Images