A method for measuring wall thermal resistance based on artificial neural network

Abstract

The present invention relates to the technical field of wall thermal resistance measurement, and discloses a wall thermal resistance measurement method based on an artificial neural network. The wall thermal resistance measurement method based on an artificial neural network predicts the internal thermal conduction heat flow of the measured wall through the artificial neural network, and can accurately measure the thermal conduction heat flow inside the enclosure structure without destroying the wall, and greatly reduce the time cost of field experiments, thereby reducing the impact of environmental restrictive factors on thermal resistance measurement. The wall thermal resistance value can be obtained according to the temperature difference between the inner and outer surfaces of the wall and the thermal conduction heat flow inside the wall. The method is also applicable to active, passive and phase change material-containing buildings.

Description

Technical Field

[0001] The invention relates to the technical field of wall thermal resistance measurement, and in particular to a wall thermal resistance measurement method based on an artificial neural network. Background Art

[0002] In building energy-saving applications, in addition to energy-saving design for new buildings, existing buildings also need to be renovated for energy saving. For some older buildings, much of their building envelope data is missing, or some thermal physical parameters have deviated from the original design initial values ​​due to years of wind and rain, so it is very important to determine the current actual value of their thermal resistance.

[0003] The patent publication number CN112229869A discloses a device and method for testing the thermal resistance of a building wall on site. The testing device includes a heating plate, a temperature controller, and two temperature sensors, two heat flux sensors, and two insulation plates. The first temperature sensor, the first heat flux sensor, the heating plate, and the first insulation plate are all attached to the hot side of the wall to be tested, and the second temperature sensor, the second heat flux sensor, and the second insulation plate are all attached to the cold side of the wall to be tested. A through hole is provided in the middle of the second insulation plate as a one-dimensional heat flux channel, and the second temperature sensor and the second heat flux sensor are both located in the through hole. The testing method is to quantitatively determine the size of the one-dimensional heat transfer area on both the hot and cold sides of the wall, and to calculate the equivalent thermal resistance of the wall by measuring and recording the surface temperature of the one-dimensional heat transfer area on both the hot and cold sides of the wall and the heat flux density passing through the one-dimensional heat transfer area.

[0004] The prior art has the following defects:

[0005] 1. Rainy and snowy weather must be avoided. The uncertainty of climate fluctuations will bring certain difficulties to on-site measurements, thus extending the experimental period;

[0006] 2. Highly dependent on temperature and heat flow environment, the temperature difference between indoor and outdoor must be greater than 10°C to stabilize heat transfer;

[0007] 3. Additional heating plates are required for heating to stabilize the heat transfer, which increases manpower and financial costs;

[0008] 4. It takes at least 2 weeks, requires air conditioning operating conditions, a long time period and high time cost;

[0009] 5. The applicability is not comprehensive enough, and the situation where the enclosure structure contains phase change materials cannot be accurately analyzed;

[0010] 6. The operation is complicated and requires the combined use of insulation board, temperature sensor, heat flow sensor and heating plate. Summary of the invention

[0011] In view of the above problems in the prior art, a wall thermal resistance measurement method based on artificial neural network is proposed.

[0012] The present application provides a wall thermal resistance measurement method based on an artificial neural network, the purpose of which is to accurately measure the heat flow inside the enclosure structure without destroying the wall, and greatly reduce the time cost of field experiments, thereby reducing the impact of environmental restrictive factors on thermal resistance measurement. The thermal resistance value of the wall can be obtained based on the temperature difference between the inner and outer surfaces of the wall and the heat flow inside the wall.

[0013] The technical solution of the present invention is: a wall thermal resistance measurement method based on an artificial neural network, comprising the following steps:

[0014] Step 1: Use the building simulation software to establish a preset room according to the thermal physical parameters of the building envelope structure; if the preset room envelope structure only contains sensible heat materials, then establish the preset room Room1; if the preset room envelope structure contains phase change materials, then establish the preset room Room2. Input the thermal physical parameters of the room envelope structure and meteorological parameters into the building simulation software, and output the data of each thermal performance index of the room in real time;

[0015] Step 2: Process the input layer and output layer of the preset rooms Room1 and Room2;

[0016] Step 3: Train the multi-layer feedforward neural network in the artificial neural network, preset room Room1 to obtain the artificial neural network ANNfit1, and preset room Room2 to obtain the artificial neural network ANNfit2;

[0017] Step 4: Use the fitted network structures ANNfit1 and ANNfit2 to continue inputting the thermal performance index data of the preset room models Room1 and Room2 in the training set at different times to obtain the internal heat conduction heat flow of the tested wall A;

[0018] Step 5: According to the thermal resistance formula, obtain the thermal resistance of wall A of the preset room models Room1 and Room2;

[0019] Step 6: Input the thermal performance index parameters of Room 3 whose enclosure structure only contains sensible heat materials and Room 4 whose enclosure structure contains phase change materials into the fitted network structures ANNfit1 and ANNfit2 respectively, obtain the internal heat conduction heat flow of the measured wall B through training, and then obtain the thermal resistance of wall B according to the thermal resistance formula.

[0020] Furthermore, in step 1, the thermal performance index data of room Room1 are output in real time, including the inner and outer surface temperatures of each wall and roof, the floor temperature, the indoor air temperature, the internal thermal conduction heat flow of the measured wall A, the air humidity and the air conditioning switch status; the thermal performance index data of room Room2 are output in real time, including the inner and outer surface temperatures of each wall and roof, the floor temperature, the indoor air temperature, the internal thermal conduction heat flow of the measured wall A, the cross-sectional temperature of the phase change material and the sensible heat material, the air humidity and the air conditioning switch status.

[0021] Furthermore, in step 2, the thermal performance index data of at least two days output in real time by the preset room Room1, the inner and outer surface temperatures of each wall and roof, the floor temperature and the indoor air temperature are used as the input layer data, and the thermal conduction heat flux of the measured wall A is used as the output layer data; the thermal performance index data of at least two days output in real time by the preset room Room2, the inner and outer surface temperatures of each wall and roof, the floor temperature, the cross-sectional temperatures of the phase change material and the sensible heat material and the indoor air temperature are used as the input layer data, and the thermal conduction heat flux of the measured wall A is used as the output layer data.

[0022] Furthermore, in step 5, the fitted thermal resistance value is compared with the original wall thermal resistance data, and the root mean square error coefficient of deviation CV (RMSE) should be less than 30%; the thermal resistance formula of the enclosure structure containing only sensible heat materials is:

[0023] ;

[0024] Among them, R r is the wall thermal resistance; q d is the measured wall heat flux density; T e is the ambient temperature; T i is the indoor air temperature, and the thermal resistance formula of the building envelope containing phase change materials is:

[0025] ;

[0026] Wherein, Td is the cross-sectional temperature at the junction of the phase change material and the sensible heat material.

[0027] Furthermore, in steps 1, 2 and 3, a multi-layer feedforward neural network is established, the number of hidden layers of which is set to 1, and the number of neural nodes is determined by trial and error, and the number of neural nodes when the regression coefficient R value is closest to 1 is taken as the final result.

[0028] Furthermore, the input layer of the training set should increase the air humidity and the air conditioning switch status. If the outer surface of the measured wall of the preset room contains phase change materials or a structure with phase change material characteristics, the input layer of the training set should increase the cross-sectional temperature of the phase change material and the sensible heat material.

[0029] Furthermore, for the network verification in step 5, the calculated thermal resistance value is compared with the original wall thermal resistance data, and the root mean square error coefficient of deviation CV (RMSE) should be less than 30%, otherwise the amount of data input during the preset should be increased by the number of days, or the number of hidden layers and nodes of the artificial neural network should be adjusted so that the CV (RMSE) is less than 30%, indicating that the result is accurate.

[0030] Furthermore, if the room to be measured is a room whose enclosing structure contains only sensible heat materials, the preset room must also be a room whose enclosing structure contains only sensible heat materials; if the room to be measured is a room whose enclosing structure contains only phase change materials, the preset room must also be a room whose enclosing structure contains phase change materials.

[0031] Beneficial effects of the present invention:

[0032] 1. The thermal resistance of the wall can be quickly obtained without destroying the wall structure and without the need for additional heating equipment or sunshades, thus reducing financial and labor costs;

[0033] 2. Reduced computing costs and engineering time costs. The artificial neural network only takes 1-2 seconds to execute, while the traditional method requires more than 2 weeks of measurement, and must avoid rainy and snowy weather;

[0034] 3. The method is widely applicable, not only to the case where the enclosure structure contains only sensible heat materials, but also to the case where the enclosure structure contains phase change materials;

[0035] 4. The traditional calculation method must have air-conditioning operation conditions, while this method is applicable not only to buildings with air-conditioning operation conditions, but also to passive buildings without air-conditioning operation conditions;

[0036] 5. The operation is simple and convenient. There is a pre-programmed fitting toolbox in Matlab software, namely the multi-layer feedforward neural network, and an instruction manual is attached for the use of the toolbox, which can be used directly. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a diagram of an experimental device for measuring the thermal resistance of a wall of an enclosure structure containing only sensible heat materials in the wall thermal resistance measurement method based on an artificial neural network of the present invention;

[0038] Figure 2 This is a diagram of an experimental device for measuring the thermal resistance of a wall containing a phase change material in an enclosure structure in the wall thermal resistance measurement method based on an artificial neural network of the present invention;

[0039] Figure 3 It is a structural diagram of a multi-layer feedforward neural network in the wall thermal resistance measurement method based on an artificial neural network of the present invention;

[0040] Figure 4This is a measurement principle diagram of a wall thermal resistance measurement method based on an artificial neural network according to the present invention;

[0041] Figure 5 It is a diagram of artificial neural network fitting results for measuring the thermal resistance of a wall of an enclosure structure containing only sensible heat materials in the wall thermal resistance measurement method based on artificial neural network of the present invention;

[0042] Figure 6 This is a diagram of artificial neural network fitting results for measuring the thermal resistance of a wall containing a phase change material in an enclosure structure in the wall thermal resistance measurement method based on an artificial neural network of the present invention; DETAILED DESCRIPTION

[0043] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0044] Reference Figure 1-Figure 4 , a wall thermal resistance measurement method based on an artificial neural network is provided. The wall thermal resistance measurement method based on an artificial neural network of the present invention comprises the following steps:

[0045] Preset process:

[0046] If you want to measure the thermal resistance of the wall B of the building Room3 whose envelope structure contains only sensible heat materials and the building Room4 whose envelope structure contains phase change materials, you must first have a preset process to collect the training set data of the artificial neural network. Through commercial building simulation software, such as Energyplus, Trnsys, etc., select the meteorological parameters of the typical representative month or typical representative day in the area where the measured buildings Room3 and Room4 are located to input. According to the thermal physical parameters of the building envelope structure given in the building design standard GB50176-2016 or the thermal physical parameters of the envelope structure materials of any square experimental room, establish a preset room. If the preset room envelope structure only contains sensible heat materials, establish a preset room Room1, input the thermal physical parameters of the room envelope structure and meteorological parameters and other data to make the building simulation software run, and output the data of each room's thermal performance index in real time, including the internal and external surface temperature of each wall, roof, floor temperature, indoor air temperature, internal heat conduction heat flow of the measured wall A, air humidity, and air conditioning switch status. For the inner surface of the tested wall A of Room 1, the heat transfer is mainly affected by the convection heat transfer of the indoor air temperature, the radiation heat transfer of other inner surfaces of the wall, the inner surface of the roof and the floor, and the heat conduction heat transfer of the interior of the wall A. In terms of the building thermal environment, it is also affected by the indoor air humidity. In addition, whether the air conditioner is running or not also has a certain impact on the heat flow of the inner surface of the tested wall A. This is the case where the building envelope structure only contains sensible heat materials. The main mapping relationships are as follows: Figure 1If the default room enclosure structure contains phase change materials, then create a default room Room2, such as Figure 2 As shown, the source of heat flow on the inner surface of wall A increases the influence of phase change material on the heat conduction heat flow of the surface. Inputting the thermal physical parameters of the room enclosure structure and meteorological parameters and other data enables the building simulation software to run and output the data of various thermal performance indicators of the room in real time, including the temperature of the inner and outer surfaces of each wall and roof, floor temperature, indoor air temperature, internal heat conduction heat flow of the measured wall A, cross-sectional temperature of phase change material and sensible heat material, air humidity and air conditioning switch status (Room1, Room2, Room3, Room4 only represent the model code of the modeled room, which is only used to distinguish the preset building from the measured building. Walls A and B also only represent the code of the wall, which can represent any wall in the building room).

[0047] For nonlinear relationships in architecture, artificial neural networks can handle this relationship very well. Multilayer feedforward neural network is a type of artificial neural network, which is universal and widely used in the field of architecture. The artificial neural network fitting toolbox in Matlab software is a multilayer feedforward neural network, which is trained using the Levenberg-Marquardt back-propagation algorithm, sigmoid (tansig) hidden neurons and linear (purelin) output neurons. The samples are randomly divided into 70% training, 15% validation, and 15% testing. And the performance of the network is judged by using the regression coefficient (R) and MSE as evaluation criteria. When the MSE reaches stability or the maximum number of iterations is 1000, the iteration is terminated. The closer the R value is to 1, the stronger the correlation between the predicted value and the measured value, and the better the network. People can directly use this pre-written code, which is more convenient than other networks. Multilayer feedforward neural network consists of three parts, input layer, output layer and one or more hidden layers between the input layer and the output layer. The size of the input and output layers matches the number of inputs and outputs of the problem, the neurons in each layer only take the output signal of the previous layer as input, and the nodes in each layer are connected to other nodes in the adjacent previous layer. In order to obtain a better fit, the input and output data and their structure and parameters must be carefully set.

[0048] Figure 3It is the structure of a multi-layer feedforward neural network. According to the mapping relationship, the input layer parameters x1, x2…xn of the training set need to input the indoor air temperature, the inner surface temperature of other walls except wall A, the inner surface temperature of the roof, the floor temperature, the inner surface heat flow of wall A, and the air humidity. These parameters can be obtained by real-time output of the commercial building simulation software in step 1. In addition, for active buildings (with air conditioning or heating running) and passive buildings, they also need to be input in the training set. Active buildings input 1, and passive buildings input 0. The output layer is the heat conduction heat flow of wall A. At least 2 days of data are input for fitting in the input and output layers. The number of hidden layers is set to 1, because one hidden layer is enough to find the correlation between the input and output variables. Setting too many hidden layers may even lead to overfitting of the model. The number of neuron nodes can also be determined by trial and error. The number of neuron nodes when the R value is closest to 1 is taken as the final result. The fitted artificial neural network is named ANNfit1. The outer surface of the tested wall contains materials and structures of fixed phase change materials. The input layer parameters of the training set need to add a phase change material sensible heat material cross-sectional temperature, and other parameters remain unchanged. The output layer is the heat flow of wall A. The artificial neural network of the case where the enclosure structure contains phase change materials after fitting is named ANNfit2 (ANNfit1 and ANNfit2 only represent the codes of the network structures after fitting in the two cases where the enclosure structure contains only sensible heat materials and contains phase change materials. Here, it is only to distinguish the two cases where the enclosure structure contains only sensible heat materials and contains phase change materials).

[0049] Calibration process:

[0050] For the case where the enclosure structure only contains sensible heat materials, the input layer parameters of the Room1 model training set are input into the preset and fitted ANNfit1 again, which are the same as the input layer parameters in step 2, but the data at different times are selected. This is to perform network fitting calculations to obtain output layer data at another time, that is, the internal heat conduction heat flow of wall A at another time. The thermal resistance value of wall A is then calculated by the internal heat conduction heat flow of wall A obtained after fitting and the internal and external surface temperatures of wall A:

[0051] (1)

[0052] R r is the wall thermal resistance; q d is the measured wall heat flux density; T e is the ambient temperature; T i is the indoor air temperature.

[0053] In the case where the enclosure structure contains phase change materials, the input layer parameters of the Room2 model training set are input into the preset and fitted ANNfit2 again, which are the same as the input layer parameters in step 2, but the data at different times are selected. This is to perform network fitting calculations to obtain output layer data at another time, that is, the heat conduction heat flow of the sensible heat material inside wall A at another time. The thermal resistance value of the sensible heat material inside wall A is then calculated by the heat conduction heat flow of the sensible heat material inside wall A, the inner surface temperature of wall A, and the "phase change material sensible heat material cross-sectional temperature".

[0054] (2)

[0055] T d It is the cross-sectional temperature of the phase change material, in °C.

[0056] Compared with the original wall thermal resistance data, the root mean square error coefficient of deviation CV (RMSE) should be less than 30%. Otherwise, the amount of data input during the preset should be increased, which can be increased by days, or the number of hidden layers and nodes of the artificial neural network should be adjusted.

[0057] Measurement process:

[0058] If you need to measure the thermal resistance of wall B in Room 3 of the building under test, you can use the calibrated ANNfit1 to achieve this. Measure the inner surface temperature of each wall in Room 2, the wall surface heat flow, air temperature and air humidity as the training set. The input layer parameters of the training set are the same as the input layer parameters in step 1. Through ANNfit1 fitting, you can get the output layer parameters, that is, the internal heat flow of wall B. Then, according to the thermal resistance formula (1), you can get the thermal resistance value of wall B.

[0059] If you need to measure the thermal resistance of wall B of Room 4 of the building under test, you can use the calibrated ANNfit2 to achieve this. Measure the inner surface temperature of each wall in Room 4, the wall surface heat flow, the cross-sectional temperature of the phase change material sensible heat material, the air temperature and the air humidity as the training set. The input layer parameters of the training set are the same as the input layer parameters in step 1. Through ANNfit2 fitting, you can get the output layer parameters, that is, the heat flow of wall B. Then, according to the thermal resistance formula (2), you can get the thermal resistance value of wall B.

[0060] Example 1

[0061] First, a room model with a 0.5×0.5×0.5 m enclosure structure containing only sensible heat materials was established in the commercial building simulation software EnergyPlus. The enclosure structure parameters were set to the thermal physical parameters of the building enclosure structure given in the building design standard GB50176-2016. The meteorological data was selected from the October meteorological data in Hefei in autumn. The EnergyPlus official website has a meteorological data file package. The air conditioner was set to off. The building simulation software runs and outputs the data of various thermal performance indicators of the room in real time, including the surface temperature of each wall, roof, floor temperature, indoor air temperature, internal heat conduction heat flux of the measured wall A, and air humidity.

[0062] The network is then built using the artificial neural network fitting toolbox (multi-layer feedforward neural network) in Matlab. The input layer parameters x1, x2…xn of the training set input the indoor air temperature, the inner surface temperature of other walls except wall A, the inner surface temperature of the roof, the floor temperature, the inner surface heat flow of wall A, the air humidity, and the input 0 represents that the air conditioner is not running. The output layer is the heat conduction heat flow of wall A. The input layer and the output layer input 2 days of data for fitting. The number of hidden layers is set to 1, and the number of neuron nodes is determined to be 4 by trial and error. Run Matlab to obtain the network ANNfit1, and the regression coefficient R value can reach 0.97, as shown Figure 5 , proving the network ANNfit1. Then the calibration process is carried out. The model established in the commercial building simulation software EnergyPlus is set to change only the meteorological data to the meteorological data of November in the autumn of Hefei, and then the real-time output of the thermal performance index data of each room. The other steps are the same as above. The network ANNfit1 is obtained. The heat flow obtained by the output layer is compared with the thermal resistance of wall A obtained by the thermal resistance formula (1). The CV (RMSE) is 8%, which is much less than 30%, which proves the excellence of the method.

[0063] Example 2

[0064] First, a room model with a 0.5×0.5×0.5 m enclosure structure containing sensible heat materials was established in the commercial building simulation software EnergyPlus. The enclosure structure parameter settings and meteorological parameters were the same as in Example 1. The air conditioner was set to off. The building simulation software ran and output the data of various thermal performance indicators of the room in real time, including the temperature of the inner and outer surfaces of each wall and roof, the floor temperature, the indoor air temperature, the cross-sectional temperature of the phase change material sensible heat material, the internal heat conduction heat flow of the measured wall A, and the air humidity.

[0065] The network is then built using the artificial neural network fitting toolbox (multi-layer feedforward neural network) in Matlab. The input layer parameters x1, x2…xn of the training set input the indoor air temperature, the inner surface temperature of other walls except wall A, the inner surface temperature of the roof, the floor temperature, the inner surface heat flow of wall A, the air humidity, the cross-sectional temperature of the phase change material, and the input 0 represents that the air conditioner is not running. The output layer is the heat conduction heat flow of wall A. The input and output layers input 2 days of data for fitting. The number of hidden layers is set to 1, and the number of neuron nodes is determined to be 4 by trial and error. Run Matlab to obtain the network ANNfit2, and the regression coefficient R value can reach 0.99, such as Figure 6 , proving the excellence of the network ANNfit2. Then the calibration process was carried out. The model established in the commercial building simulation software EnergyPlus was set to change only the meteorological data to the meteorological data of November in Hefei in autumn, and then output the data of each room's thermal performance index in real time. The other steps were the same as above. The network ANNfit2 obtained, the heat flow obtained in the output layer, and the thermal resistance of wall A obtained according to the thermal resistance formula (1) were compared with the data set in the original EnergyPlus. The CV (RMSE) was 5%, which was much less than 30%, proving the excellence of the method.

[0066] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A wall thermal resistance measurement method based on artificial neural network, characterized in that: The following steps are involved: Step 1: Use the building simulation software to establish a preset room according to the thermal physical parameters of the building envelope structure; if the preset room envelope structure only contains sensible heat materials, then establish the preset room Room1; if the preset room envelope structure contains phase change materials, then establish the preset room Room2. Input the thermal physical parameters of the room envelope structure and meteorological parameters into the building simulation software, and output the data of each thermal performance index of the room in real time; Step 2: Process the input layer and output layer of the preset rooms Room1 and Room2. The preset room Room1 outputs at least two days of thermal performance index data in real time, the inner and outer surface temperatures of each wall and roof, the floor temperature and the indoor air temperature as the input layer data, and the heat conduction heat flux of the measured wall A as the output layer data; the preset room Room2 outputs at least two days of thermal performance index data in real time, the inner and outer surface temperatures of each wall and roof, the floor temperature, the cross-sectional temperature of the phase change material and the sensible heat material and the indoor air temperature as the input layer data, and the heat conduction heat flux of the measured wall A as the output layer data; Step 3: Train the multi-layer feedforward neural network in the artificial neural network, preset room Room1 to obtain the artificial neural network ANNfit1, and preset room Room2 to obtain the artificial neural network ANNfit2; Step 4: Continue to input the thermal performance index data of the preset room models Room1 and Room2 of the training set at different times through the fitted network structures ANNfit1 and ANNfit2 to obtain the thermal conduction heat flow inside the measured wall A; Step 5: According to the thermal resistance formula, obtain the thermal resistance of wall A of the preset room models Room1 and Room2; Step 6: Input the thermal performance index parameters of Room 3 whose enclosure structure only contains sensible heat materials and Room 4 whose enclosure structure contains phase change materials into the fitted network structures ANNfit1 and ANNfit2 respectively, obtain the internal heat conduction heat flow of the measured wall B through training, and then obtain the thermal resistance of wall B according to the thermal resistance formula.

2. The wall thermal resistance measurement method based on artificial neural network according to claim 1 is characterized in that: In step 1, the thermal performance index data of room Room1 are output in real time, including the inner and outer surface temperatures of each wall and roof, the floor temperature, the indoor air temperature, the internal thermal conduction heat flow of the measured wall A, the air humidity and the air conditioning switch status; the thermal performance index data of room Room2 are output in real time, including the inner and outer surface temperatures of each wall and roof, the floor temperature, the indoor air temperature, the internal thermal conduction heat flow of the measured wall A, the cross-sectional temperature of the phase change material and the sensible heat material, the air humidity and the air conditioning switch status.

3. The wall thermal resistance measurement method based on artificial neural network according to claim 1 is characterized in that: In step 5, the fitted thermal resistance value is compared with the original wall thermal resistance data, and the root mean square error deviation coefficient should be less than 30%; the thermal resistance formula of the enclosure structure containing only sensible heat materials is: ； in, R r is the wall thermal resistance; q d is the measured wall heat flux density; T e is the ambient temperature; T i is the indoor air temperature, and the thermal resistance formula of the building envelope containing phase change materials is: ； Wherein, Td is the cross-sectional temperature at the junction of the phase change material and the sensible heat material.

4. The wall thermal resistance measurement method based on artificial neural network according to claim 1 is characterized in that: In steps 1, 2, and 3, a multilayer feedforward neural network is established, the number of hidden layers is set to 1, the number of neuron nodes is determined by trial and error, and the number of neuron nodes when the regression coefficient R value is closest to 1 is taken as the final result.

5. The wall thermal resistance measurement method based on artificial neural network according to claim 1 is characterized in that: The input layer of the training set should add air humidity and air conditioning switch status. If the outer surface of the measured wall of the preset room contains phase change materials or structures with phase change material characteristics, the input layer of the training set should add the cross-sectional temperature of the phase change materials and sensible heat materials.

6. The wall thermal resistance measurement method based on artificial neural network according to claim 1 is characterized in that: In the network verification in step 5, the calculated thermal resistance value is compared with the original wall thermal resistance data. The root mean square error deviation coefficient should be less than 30%. Otherwise, the amount of data input during the preset should be increased by the number of days, or the number of hidden layers and nodes of the artificial neural network should be adjusted to be less than 30%, indicating that the result is accurate.

7. The wall thermal resistance measurement method based on artificial neural network according to claim 1 is characterized in that: If the room to be measured has an enclosure structure containing only sensible heat materials, the preset room must also be an enclosure structure containing only sensible heat materials; if the room to be measured has an enclosure structure containing only phase change materials, the preset room must also be an enclosure structure containing phase change materials.

Images