Temperature and Humidity Data Evaluation of Tight Sportswear during Motion Based on Intelligent Modeling

In this paper, the Long Short-Term Memory(LSTM) neural network is used to establish a time series model of temperature and humidity. The LSTM neural network prediction model is a time recursive neural network which can solve the problem of long-term dependence. So far, LSTM has been widely used in transportation, finance and other fields, but not including clothing. We will adopt the LSTM model to predict the temperature and humidity data of different parts of the human body.

The LSTM cell structure consists of the Input Gate, Output Gate, Forget Gate and Cell State. At time t, the LSTM structure is updated as follows. ft=σ(Wfxt+Ufht−1+bf)it=σ(Wixt+Uiht−1+bi)c˜t=tanh(Wcxt+Ucht−1+bc)ot=σ(Woxt+Uoht−1+bo)ct=ftct−1+itc˜tht=ot⋅tanh(ct) Where, f_t, i_t, o_t are the outputs of the forgetting gate, input gate and output gate at time t; c_t and c_t are the contents stored in the t time memory unit; x_t and h_t are the input vector and the output of the hidden layer at time t, respectively; σ represents the Sigmoid function; W_f, U_f and b_f are the weight and deviation of the forgetting gate respectively. W_i, U_i and b_i are the weight and deviation of the input door, respectively; W_o, U_o and b_o are the weights and deviations of the output gates, respectively; and W_c, U_c and b_c are the weight and deviation of the contents stored in the memory unit, respectively.

The gating structure in the neural unit structure of the LSTM neural network uses the Sigmoid function and tanh function. Among them, the Sigmoid function: f(x)=11+e−x tanh function: f(x)=ex−e−xex+e−x The main difference between the Sigmoid function and tanh function is that the value range of the Sigmoid function is (0,1), while that of the tanh function is (-1,1). Both functions increase with an increase in variable x, and finally realize the function of the gating structure.

In order to study the thermal and wet comfort of sports tights for the human body, a DHT22 digital temperature and humidity sensor is mainly used in the temperature and humidity data acquisition module. DHT22 is a temperature and humidity composite sensor of low cost and long-term stable operation, with relative humidity and temperature measurement, of quick response, small size, low power consumption, high cost performance, strong anti-interference ability, and with a calibrated digital signal output [23-24].

The acquisition device is mainly embodied in the hardware design and software program design centered on Arduino Nano 33 BLE. It processes, compares, analyzes and comprehensively judges the real-time data collected by the temperature and humidity sensor embedded in the tights, then sends and receives the data through the Bluetooth function, and stores all the data during the exercise.

Seven tight sportswear long-sleeved tops were purchased for 3 participants who are male, Their age and body size are as follows, respectively: age: 25±1, height: 175.1±2.0cm, weight: 66.3±3.1kg, bust girth: 91.6±1.9cm, shoulder width: 41.1±0.2. Parameters of the tight sportswear are shown in Table 1.

In order to study the thermal and humid comfort of the upper body, the temperature and humidity of the chest, back, waist and abdomen of the upper body were measured, as shown in Figure 1. DHT22 was attached to these four parts, and the temperature and humidity data of these parts were collected during running.

Fig. 1

These participants were requested to run at a speed of 6km/h on a treadmill for 30minutes, as shown in Figure 2. The experiment was conducted in a room with a temperature of (20±2)°C and relative humidity of (60±5)%. All the subjects wore the same style tight pants.

Fig. 2

We used SPSS 23.0 to analyze the Spearman correlation of the collected data, shown in Table 2 and Table 3. From Table 2 and Table 3, it can be seen that there is a high correlation between the humidity of the waist, chest, back and abdomen, thus for humidity data collection, only one part of the data can be used to predict the humidity of other parts through the LSTM model. However, through the analysis of Spearman correlation results, it is found that the temperature correlation between the abdomen and chest is low, as well as that between the back and chest is low, while the correlation between the abdomen, back and waist is high. Therefore, for temperature collection, only the temperature of the chest and waist is needed. The purpose of this paper is only to verify the feasibility of the LSTM model, thus it only analyzes the temperature and humidity of the waist, and then predicts the humidity and temperature of the chest, back and abdomen.

In this paper, the LSTM model uses a hidden layer to predict the comfort, and takes the comfort and temperature on the waist surface as input variables, and the prediction result is the comfort on the abdominal surface, the temperature on the abdomen surface, the humidity on the back surface, the temperature on the back surface, the humidity on the chest surface, and the temperature on the chest surface. The number of neurons in the input layer and in the output layer is 2 and 6, respectively. The number of neurons in the hidden layer represents the number of nodes used for memory, which is selected as 30. Construct the collected temperature and humidity data into a training set matrix, as follows: TrainData={(yA−humd(1),yB-humd(1),yC-humd(1),xW−humd(1),)(yA−humd(2),yB-humd(2),yC-humd(2),xW−humd(2)…,(yA-humd(m),yB−humd(m),yC-humd(m),xW−humd(m))}TrainData={(yA-temp(1),yB-temp(1),yC-temp(1),xW-temp(1),)(yA−temp(2),yB-temp(2),yC-temp(2),xW−temp(2)),…,(yA-temp(m),yB-temp(m),yC-temp(m),xW-temp(m))} $$\matrix{ \matrix{ {\rm{T}}rainData\, = \{ (y_{{\rm{A}} - \,{\rm{humd}}\,}^{(1)},y_{{\rm{B - humd}}\,}^{(1)},y_{{\rm{C - humd}}\,}^{(1)},x_{W - \,humd\,}^{(1)},) \hfill \cr (y_{{\rm{A}} - \,{\rm{humd}}\,}^{(2)},y_{{\rm{B - humd}}\,}^{(2)},y_{{\rm{C - humd}}\,}^{(2)},x_{W - \,humd\,}^{(2)} \ldots ,(y_{{\rm{A - humd}}\,}^{({\rm{m}})}, \hfill \cr y_{{\rm{B}} - \,{\rm{humd}}\,}^{({\rm{m}})},y_{{\rm{C - humd}}\,}^{({\rm{m}})},x_{W - \,humd\,}^{(m)})\} \hfill \cr} \hfill \cr \matrix{ {\rm{T}}rainData\, = \{ (y_{{\rm{A - temp}}\,}^{(1)},y_{{\rm{B - temp}}\,}^{(1)},y_{{\rm{C - temp}}\,}^{(1)},x_{W{\rm{ - }}temp\,}^{(1)},) \hfill \cr (y_{{\rm{A}} - \,{\rm{temp}}\,}^{(2)},y_{{\rm{B - temp}}\,}^{(2)},y_{{\rm{C - temp}}\,}^{(2)},x_{W - temp}^{(2)}), \ldots ,(y_{{\rm{A - temp}}\,}^{({\rm{m}})}, \hfill \cr y_{{\rm{B - temp}}\,}^{({\rm{m}})},y_{{\rm{C - temp}}\,}^{({\rm{m}})},x_{W{\rm{ - temp}}\,}^{(m)})\} \hfill \cr} \hfill \cr } $$ where, A-humd means the humidity on the abdomen surface. A-temp the temperature on the abdomen surface, B-humd the humidity on the back surface, B-temp the temperature on the back surface, C-humd the humidity on the chest surface, C-temp the temperature on the chest surface, W-humd the humidity on the waist surface, and W-temp means the temperature on the waist surface.

The specific process is as shown in Figure3.

Fig. 3

The artificial neural network library Keras in Python is used to model the LSTM neural network, the key parts of which are introduced below.

The Min Max Scaler is used to normalize the training data and test data:

scaler = Min Max Scaler()

train X = scaler.fit_transform(train X)

train Y = scaler.fit_transform(train Y)

test X = scaler.transform(test X)

Use the Input in Keras to establish the input layer, LSTM the LSTM layer and Dense the output layer. An LSTM neural network model with a structure of 2-30-6 and a time step of 10 is created by using the Model. The activation function is the ReLU function, the learning algorithm

- the Adam algorithm, the loss function

Compile the model for training with model.compile:

model.compile(optimizer=‘adam’, loss=‘mse’)

Use model.fit to learn the data of the training set, with a number of iterations of 300: history = model.fit (trainx, trainy, epochs = 300).

Use the model.predict to predict the test set data:

lstmt = model.predict(test X)

Inverse normalization is performed on the prediction result to obtain the final result:

lstm_predict = scaler.inverse_transform(lstmt)

The mean absolute error (MAE), mean absolute percentage error (MAPE) and root mean square error (RMSE) are used to measure the performance of the model in predicting temperature and humidity. MAE can well reflect the error of the predicted value; MAPE is a percentage value, indicating the average deviation degree of the predicted value from the real value; RMSE represents the deviation between the predicted value and the real value, which is often used as an evaluation index for time series prediction problems. Their calculation formula is as follows. MAE=1N∑i=1N| xi−ti |MAPE=100%N∑i=1N|xi−ti|tiRMSE=1N∑i=1N(xi−ti)2 Where, x_i and t_i represent the predicted value and actual value of the i-th sample, respectively, and n represents the number of samples. The lower the values of these three indicators, the higher the prediction accuracy and the better the performance of the model.