An indoor thermal comfort data generation method, system, device and medium

Abstract

The present invention belongs to the field of building thermal comfort, and specifically relates to a method, system, device and medium for generating indoor thermal comfort data. The method includes the following steps: collecting a number of groups of real data within a period of time, each group of real data including indoor temperature, indoor relative humidity and indoor globe temperature; inputting the current indoor temperature, indoor relative humidity and indoor globe temperature into the trained generator, and the generator outputs the predicted indoor temperature, indoor relative humidity and indoor globe temperature; obtaining the clothing thermal resistance, human metabolic rate and indoor wind speed, as well as the predicted indoor temperature, indoor relative humidity and indoor globe temperature, and matching the corresponding indoor thermal comfort data. By accurately predicting PMV through measuring the three factors of temperature, relative humidity and mean radiant temperature, the time-consuming level in the training stage is reduced, the stability is improved, and at the same time, the calculation complexity of the PMV index is reduced.

Description

Technical Field

[0001] The present invention belongs to the field of building thermal comfort, and particularly relates to a method, a system, a device and a medium for generating indoor thermal comfort data. Background Art

[0002] The thermal comfort index PMV was proposed by Danish scientist Professor Fanger in the 1970s as an evaluation index for characterizing human thermal reactions. At the same time, the PMV index is also the most widely used and recognized thermal comfort evaluation index in the world. It has a complex non-linear relationship with a variety of environmental variables and human parameters and cannot be directly measured. In the actual working process, due to the complexity of the HVAC system itself, the certain coupling between the four variables, and the complex non-linear iterative relationship of the four variables in the PMV index calculation process, it is extremely difficult technically to take them all as control variables at the same time.

[0003] The BP (Back Propagation) neural network was proposed by a group of scientists led by Rumelhart and McClelland in 1986. It is a multi-layer feedforward network with the error backpropagation algorithm as the training algorithm and is also one of the most mature and widely used neural networks. It can learn and represent a large number of input-output mapping relationships without explicitly knowing the mathematical equations of this mapping relationship in advance.

[0004] The learning process of the BP neural network is a process of repeatedly training the network with training samples and reducing the error between the actual output and the expected output by continuously changing the weights and thresholds of the network. When using the BP neural network to generate indoor thermal comfort data, it is found that this method has a slow convergence speed and is prone to falling into local minima; at the same time, this method is relatively sensitive to network initial values, learning efficiency, etc., and the training results are unstable. Summary of the Invention

[0005] Aiming at the problem of slow convergence speed when using the BP neural network to generate indoor thermal comfort data in the prior art, the present invention provides a method, a system, a device and a medium for generating indoor thermal comfort data.

[0006] To achieve the above object, the technical solution of the present invention is as follows:

[0007] First aspect: A method for generating indoor thermal comfort data, comprising the following steps:

[0008] S1: Collect a number of groups of real data within a period of time, and each group of real data includes indoor temperature, indoor relative humidity, and indoor globe temperature;

[0009] S2: Divide the several groups of real data into training data and discriminant data according to the time sequence, and add redundancy to obtain the original random noise limit condition;

[0010] S3: Input the training data and the original random noise limit condition into the generator, and the generator outputs generated data;

[0011] S4: Input the original random noise limit condition, discriminant data and generated data into the discriminator, and the discriminator outputs a discrimination result;

[0012] S5: According to the discrimination result, update the discriminator parameters, and the discriminator backpropagates the error to the generator to update the generator parameters;

[0013] S6: When the discrimination result shows that the error reaches the preset accuracy or the number of learning times is greater than the preset training iteration times, terminate the algorithm and complete the training of the generator;

[0014] S7: Input the current indoor temperature, indoor relative humidity and indoor globe temperature into the trained generator, and the generator outputs the predicted indoor temperature, indoor relative humidity and indoor globe temperature;

[0015] S8: Obtain the clothing thermal resistance, human metabolic rate and indoor wind speed, as well as the predicted indoor temperature, indoor relative humidity and indoor globe temperature in S7, and match the corresponding indoor thermal comfort data.

[0016] Further, the original random noise limit condition is the data interval obtained by adding 20% redundancy to both the upper and lower limits of the real data.

[0017] Further, the number of training iterations is 10000.

[0018] Further, the preset accuracy is that the error is close to 10 -3 .

[0019] Further, the activation functions of the hidden layer and output layer of the generator and discriminator are: f(x) = max(0, x). When x is 1, the result is 1; when x is 0, the result is 0.

[0020] Further, in step S5, according to the loss functions of the discriminator D and the generator G, update the corresponding parameters.

[0021] Further, the loss functions are:

[0022]

[0023] where z is a random noise following a Gaussian distribution; x is the real data; G represents the generator; D represents the discriminator; Pdata(x) represents the probability distribution of the real data; Pz(x) represents the probability distribution of the random noise; x ∼ Pdata means randomly sampling x from the distribution of the real data; z ∼ Pz means sampling the noise z from the random noise of the Gaussian distribution; x|y means obtaining x under the constraint condition y, and z|y means obtaining z under the constraint condition y; D(x|y) represents the vector output by the discriminator D after receiving the input within the brackets; G(z|y) represents the vector output by the generator G after receiving the input within the brackets.

[0024] In a second aspect, an indoor thermal comfort data generation system includes:

[0025] A data acquisition unit: used to acquire several groups of real data within a period of time, and each group of real data includes indoor temperature, indoor relative humidity, and indoor globe temperature;

[0026] A data preprocessing unit: used to divide the several groups of real data into training data and discriminant data in chronological order, and add redundancy to obtain the original random noise constraint condition;

[0027] A generation unit: used to input the training data and the original random noise constraint condition into the generator, and the generator outputs the generated data;

[0028] A discriminant unit: used to input the original random noise constraint condition, discriminant data, and generated data into the discriminator, and the discriminator outputs the discriminant result;

[0029] A backpropagation unit: used to update the discriminator parameters according to the discriminant result, and the discriminator backpropagates the error to the generator to update the generator parameters;

[0030] A termination unit: used to terminate the algorithm and complete the training of the generator when the discriminant result shows that the error reaches the preset accuracy or the number of learning times is greater than the preset training iteration times;

[0031] A prediction unit: inputs the current indoor temperature, indoor relative humidity, and indoor globe temperature into the generator after training, and the generator outputs the predicted indoor temperature, indoor relative humidity, and indoor globe temperature;

[0032] A thermal comfort unit: used to obtain the clothing thermal resistance, human metabolic rate, and indoor wind speed, as well as the predicted indoor temperature, indoor relative humidity, and indoor globe temperature in S7, and match the corresponding indoor thermal comfort data.

[0033] In a third aspect, a computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor, when executing the computer program, implements the indoor thermal comfort data generation method according to any one of the above.

[0034] In a fourth aspect, a computer-readable storage medium stores a computer program. The computer program, when executed by a processor, implements the indoor thermal comfort data generation method according to any one of the above.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] First, by measuring three factors, namely temperature, relative humidity, and mean radiant temperature, to accurately predict PMV, the time-consuming level in the training stage is reduced, stability is improved, and at the same time, the computational complexity of the PMV index is reduced; according to a large number of on-site test results in rural areas of Guanzhong, two human subjective factors affecting the PMV index, metabolic rate and human clothing thermal resistance, are set as fixed values, with a faster convergence speed, stronger stability, and more convenient implementation; additional limiting conditions are added to the generation process of the random noise signal, and the fluctuation range of the actual PMV index with an additional 20% redundancy is used as an additional limiting condition, which speeds up the convergence speed of the network.

[0037] Second, the present invention reduces the sensitivity of the network to initial parameters and does not require such a large cost of hardware equipment as compared with the GAN network.

[0038] Third, the training process of the present invention has a higher similarity to the way people learn and grow themselves, and a new route for humans to guide AI to complete high-level tasks is expanded. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings forming a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention.

[0040] In the drawings:

[0041] Figure 1 is a schematic flow chart of a method for generating indoor thermal comfort data according to the present invention;

[0042] Figure 2 is a flow block diagram of a training model of a method for generating indoor thermal comfort data according to the present invention;

[0043] Figure 3 is a comparison of the number of iterations of different prediction models of a method for generating indoor thermal comfort data according to the present invention;

[0044] Figure 4It is the network structure diagram of the training model of a method for generating indoor thermal comfort data according to the present invention;

[0045] Figure 5 It is the generated effect diagram of PMV based on the BP neural network;

[0046] Figure 6 It is the generated effect diagram of PMV based on the GAN network;

[0047] Figure 7 It is the generated effect diagram of PMV of a method for generating indoor thermal comfort data according to the present invention. Detailed implementation manners

[0048] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments. It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.

[0049] The following detailed descriptions are all exemplary descriptions, aiming to provide further detailed descriptions of the present invention. Unless otherwise specified, all technical terms adopted by the present invention have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs. The terms used in the present invention are only for describing specific implementation manners, and are not intended to limit the exemplary embodiments according to the present invention.

[0050] The generative adversarial network GAN (Generative Adversarial Networks) is a new framework for estimating generative models through an adversarial process proposed by Ian J. Goodfellow et al. in October 2014. The well-known Nash equilibrium in game theory is the core idea source of the generative adversarial network. In GAN, the two participants in the game are respectively a generator and a discriminator. The game goal of the generator is to learn the distribution of real data as much as possible, while the goal of the discriminator is to be able to correctly distinguish whether the data input into it comes from real data or the generator. In order to finally win the game, these two game participants need to continuously optimize themselves to improve their own generative ability or discriminative ability, and this learning and optimization process is to find a Nash equilibrium between the two.

[0051] To achieve the simulation of the real data distribution, generative adversarial networks (GANs) require sampling. However, the excessive freedom in sampling may lead to certain uncontrollability in the convergence process of GANs when the data volume is extremely large. Moreover, the training method of GANs has extremely low dependence on prior knowledge (the initially sampled samples). One of the drawbacks of this method is that it exacerbates the sensitivity of the network to the initial parameters. Although the dependence is extremely low, there is still some dependence. Finally, although GANs have powerful performance, the performance of the GAN model has a strong positive correlation with the performance of the hardware devices used in the training process. A large model requires the use of top-of-the-line GPUs in the industry. Therefore, high costs are required to train GANs strongly.

[0052] Example 1

[0053] The present invention provides a method for generating indoor thermal comfort data of rural residences in Guanzhong based on GANs, which does not require too much prior knowledge, can be effectively applied to non-professional operators, and significantly improves the authenticity of the generated data.

[0054] A method for generating indoor thermal comfort data of rural residences in Guanzhong based on GANs includes the following steps:

[0055] Step 1: Collect real data.

[0056] The collected data includes multiple groups of indoor temperature, indoor relative humidity, and indoor globe temperature over a period of time. The collected data includes training data and discriminant data.

[0057] Among the six variables affecting PMV thermal comfort, both clothing thermal resistance and human metabolic rate are subjective factors. These two factors have different values when the human body is in different times, different locations, and different activity states, and are difficult to determine.

[0058] Usually, in indoor thermal comfort control, appropriate values are directly assigned to these two subjective factors based on prior knowledge and the specific environment without treating them as control variables. Since the wind speed can be defaulted to the static wind speed indoors, the collected data only includes these three categories. After obtaining the real PMV data, the multiple groups of collected data are divided into training data and discriminant data. And the additional information captured from the prior knowledge is used as a limiting condition by separately feeding the additional information to the generative model and the discriminant model as part of the input layer.

[0059] Improvement points: Compared with the BP network, it reduces the time-consuming level in the training stage and can improve its stability. Because the network structure of the BP neural network is related to the time-consuming level and its stability in the training stage (the network structure also becomes more complex, which not only increases the time-consuming level in the training stage but also affects its stability), there is a contradiction (that is, the lower the error accuracy, the more complex the network structure).

[0060] Step 2, set the limiting conditions.

[0061] Use the data interval obtained by adding 20% redundancy to the upper and lower ranges of the three types of collected data as the original random noise limiting conditions;

[0062] This original random noise limiting condition will be used as the input of the generator together with the training data.

[0063] The generated data corresponding to the output of the generator, together with the original random noise limiting conditions and the discriminant data, make a discriminant result under the action of the discriminator.

[0064] Improvement point: In the standard GAN network structure, the test data is the input of the original generator in the standard GAN network, while the discriminant data and the generated data are the inputs of the original discriminator. Although the training data may be scarce, the input signal is random, resulting in a wide and random generated data by the generator. The consequence of this is that it takes a long time to train the network to converge. When the limiting conditions are added, the range of the training data can be narrowed when input, thereby shortening the convergence time.

[0065] Step 3, set the GAN hyperparameters.

[0066] The hyperparameters of GAN include the number of original random noise, the original random noise limiting conditions, and the number of training iterations. Define the three factors collected in Step 1 as the original random noise, the data collected in Step 2 as the original noise limiting conditions. The original random noise is the indoor temperature, indoor relative humidity, and indoor black globe temperature, a total of three; the original random noise limiting conditions are the intervals obtained by adding 20% redundancy to the upper and lower of the actually measured indoor temperature t, indoor relative humidity h, and indoor black globe temperature tmr.

[0067] The number of GAN training iterations is related to the accuracy. The process of using the data once to update the model parameters of GAN is one iteration.

[0068] Improvement point: There is no original limiting condition in the GAN network. Since the input data of the generator is too free, the generated data is numerous and random, which makes the discriminator quite time-consuming. The purpose of the present invention is to output the PMV comfort level felt by the intelligent agent in the research room faster and more accurately.

[0069] Step 4, construct the generator and discriminator of GAN, and determine the activation functions of their hidden layers and output layers.

[0070] The purpose of the discriminator is to complete the task of distinguishing the input data, that is, to distinguish between real data and forged data. The generator is committed to optimizing the performance of the forged data on the discriminator and reducing the difference between the performance of the generated data on the discriminator and the performance of the real data on the discriminator.

[0071] Improvement: The generator and the discriminator will be attached with additional information (restriction conditions) captured from prior knowledge. By separately feeding the additional information to the generator and the discriminator as part of the input layer, the GAN that integrates prior knowledge is realized, improving the convergence controllability and convergence speed of the GAN network.

[0072] Step 5, calculate the loss functions of the generator and the discriminator and backpropagate the error to train the GAN network.

[0073] The discriminator and the generator will continuously conduct adversarial training, and their respective performances will be continuously upgraded during the adversarial process. Finally, even when the performance of the discriminator is quite high, it is difficult to correctly distinguish the source of the data samples. At this time, it should be determined that the generator has thoroughly mastered the distribution law of the real data samples. Judge whether the network error meets the conditions. When the error reaches the preset accuracy or the number of learning times is greater than the preset maximum number, the algorithm is terminated. Otherwise, select the next learning sample and the corresponding expected output for the next round of learning.

[0074] In the process of obtaining the minimum sum of squared errors, the GAN network often uses the method of error backpropagation to change its weights and thresholds. Its training samples consist of two parts: the input and the corresponding expected output. The GAN network uses the training samples to repeatedly train the network, and it reduces the error between its output and the expected output by continuously changing the network weights and thresholds.

[0075] Improvement: Due to the addition of restriction conditions, the original formula for calculating probability in the generative adversarial network (GAN) calculation formula becomes the form of conditional probability.

[0076] Step 6, compare the PMV data output by the generator in the model with the actual measured value to determine the quality of the performance.

[0077] According to the specific situation of the indoor thermal comfort prediction problem, the present invention starts from modifying the restriction conditions for generating random samples and proposes a method for generating indoor thermal comfort data of rural residences in Guanzhong based on GAN. This method integrates the characteristics of the physical parameter changes in the indoor environment of buildings in rural areas of Guanzhong captured by previous on-site tests, adds restriction conditions, and controls the relationship between the output and the input to obtain generated data under strict restriction conditions.

[0078] Embodiment 2

[0079] In this embodiment, 160 sets of measured data on indoor thermal comfort of a rural residence in Guanzhong during winter are selected. The input features include:

[0080] Indoor air temperature, indoor relative humidity, indoor globe temperature;

[0081] The output feature is: indoor PMV.

[0082] The first 100 sets of data are used as training data, and the last 60 sets of data are used as discriminant data. This example mainly tests the generation effect of this method and compares it with the effect of the data generation method based on the BP neural network.

[0083] Step 1: Collect real data.

[0084] The collected data includes 160 sets of indoor temperature t, indoor relative humidity h, and indoor globe temperature tmr over a period of time.

[0085] Step 2: Use the data interval obtained by adding 20% redundancy to both the upper and lower ranges of the collected data as the original random noise limit condition.

[0086] Step 3: Set the hyperparameters of the GAN. The number of original random noise points is 3, the original random noise limit condition, and the number of training iterations is 10,000. According to Step 1, the number of original random noise points is 3, the original random noise limit condition is set according to Step 2, and when the number of GAN training iterations is 10,000 times, the accuracy of the validation set will rise to 60%.

[0087] Step 4: Construct the generator and discriminator of the GAN, and the activation functions of their hidden layers and output layers are ReLU.

[0088] The ReLU (Rectified Linear Unit) function has been very popular in recent years. Its calculation formula is very simple: f(x) = max(0, x). During the process of backpropagating the error, when x is 1, the result is 1; when x is 0, the result is 0.

[0089] Step 5: Calculate the loss functions of the generator and discriminator and backpropagate the error to train the GAN-L network. When the error reaches the preset accuracy, the error is close to 10 -3 , or the number of learning times is greater than the preset maximum number of 10,000, then terminate the algorithm.

[0090] The process of backpropagation is as follows: After generating fake samples from the original test data array, set the labels of these fake samples to 1, that is, consider these fake samples as real samples during the training of the generator. Since the discriminator is used to generate the error at this time, and the purpose of error backpropagation is to make the fake samples generated by the generator gradually approach real samples. When the fake samples are not real but the labels are 1, the error given by the discriminator will be very large, which forces the generator to make a large adjustment; conversely, when the fake samples are real enough and the labels are 1, the error given by the discriminator will decrease, which completes the process of the fake samples gradually approaching real samples and serves the purpose of confusing the discriminator.

[0091] According to the loss functions of the discriminative model and the generative model, the backpropagation algorithm can be used to update the parameters of the model. First, update the parameters of the discriminative model, and then update the parameters of the generator using the noise data obtained by resampling.

[0092] Calculate the loss functions of the generator and the discriminator according to the following formula and backpropagate the error:

[0093]

[0094] In the formula, z is a random noise obeying the Gaussian distribution; x is the discriminative data; G represents the generator; D represents the discriminator; P data (x) represents the probability distribution of the discriminative data; P z (x) represents the probability distribution of the random noise; x ∼ P data means randomly extracting x from the distribution of the discriminative data; z ∼ P z means extracting the noise z from the random noise of the Gaussian distribution; x|y means obtaining x under the constraint condition y, and z|y means obtaining z under the constraint condition y, so the probability expressions have changed from the states in the original GAN network to the current conditional probability formulas. Both D(x|y) and G(z|y) represent the vectors output by the discriminator and the generator after receiving the input in the brackets. For the generator G, using the random noise z as the input, the generator G hopes that the samples it generates can deceive the discriminator D as much as possible, so it is necessary to maximize the discrimination probability D(G(z|y)). Therefore, for the generator G, its objective function is to minimize log(1 - D(G(z|y))). For the discriminator D, in order to distinguish the discriminative data and the fake generated data as much as possible, it hopes to minimize the discrimination probability D(G(z|y)) while maximizing the discrimination probability D(x|y).

[0095] In the generative adversarial network that fuses prior knowledge, the probability calculation expression will evolve into a conditional probability. The input of the generator changes from random noise to its own additional constraint conditions that fuse prior knowledge, and the input of the discriminator changes from the generated data and the discriminative data to both of them plus the constraint conditions.

[0096] The training set of the prior knowledge fusion type GAN network becomes: {real data, matching constraint conditions}, {real data, non-matching constraint conditions}, {generated data, matching constraint conditions}; for only the first data pair, the discriminator D should determine it as correct.

[0097] Step 6: Input the data to be predicted into the trained generator, and the generator outputs the generated thermal comfort data, which is compared with the actual measured value to determine the quality of the performance.

[0098] The experimental results show that the thermal comfort prediction model based on the prior knowledge fusion type GAN reaches the convergence state in only about one-third of the time of the thermal comfort prediction model based on the original GAN, and has the smallest prediction error among the three. In summary, it can be seen that the prior knowledge fusion type GAN network is closer to the actual thermal comfort value PMV in network training than the original GAN network, and consumes less time to train to convergence than the original GAN network and the BP network.

[0099] The present invention mainly solves the problem of collecting multiple factors affecting PMV to more quickly and accurately predict the true PMV. For the present invention, it can be applied to accurately predict the PMV felt by the human body inside the research room in practice, so that some real-time measures can be taken to make the intelligent body reach a comfortable level.

[0100] Embodiment 3

[0101] An indoor thermal comfort data generation system, comprising:

[0102] A data acquisition unit: used to acquire several groups of real data within a period of time, and each group of real data includes indoor temperature, indoor relative humidity, and indoor globe temperature;

[0103] A data preprocessing unit: used to divide the several groups of real data into training data and discriminant data in chronological order, and add redundancy to obtain the original random noise constraint conditions;

[0104] A generation unit: used to input the training data and the original random noise constraint conditions into the generator, and the generator outputs the generated data;

[0105] A discriminant unit: used to input the original random noise constraint conditions, discriminant data, and generated data into the discriminator, and the discriminator outputs the discriminant result;

[0106] A backpropagation unit: used to update the discriminator parameters according to the discriminant result, and the discriminator backpropagates the error to the generator to update the generator parameters;

[0107] A termination unit: used to terminate the algorithm and complete the training of the generator when the discriminant result shows that the error reaches the preset accuracy or the number of learning times is greater than the preset training iteration times;

[0108] Prediction unit: Input the current indoor temperature, indoor relative humidity, and indoor globe temperature into the generator after training is completed, and the generator outputs the predicted indoor temperature, indoor relative humidity, and indoor globe temperature;

[0109] Thermal comfort unit: Used to obtain the clothing thermal resistance, human metabolic rate, and indoor wind speed, as well as the indoor temperature, indoor relative humidity, and indoor globe temperature predicted in S7, and match the corresponding indoor thermal comfort data.

[0110] Example 4

[0111] A computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, where when the processor executes the computer program, it implements the indoor thermal comfort data generation method described in Embodiments 1 and 2.

[0112] Example 5

[0113] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the indoor thermal comfort data generation method described in Embodiments 1 and 2.

[0114] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0115] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, and the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0116] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction device that implements the functions specified in the process(es) Figure 1 step(s) or multiple step(s) and / or block(s) Figure 1 block(s) or multiple block(s).

[0117] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in the process(es) Figure 1 step(s) or multiple step(s) and / or block(s) Figure 1 block(s) or multiple block(s).

[0118] As is known to those skilled in the art, the present invention can be implemented by other embodiments without departing from its spirit or essential characteristics. Therefore, the above-described disclosed embodiments are illustrative in all respects and not exclusive. All changes within the scope of the present invention or equivalent to the scope of the present invention are encompassed by the present invention.

[0119] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications or equivalent substitutions can still be made to the specific embodiments of the present invention, and any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

Claims

1. An indoor thermal comfort data generation method, characterized in that, it includes the following steps: S1: Collect several groups of real data within a period of time, and each group of real data includes indoor temperature, indoor relative humidity, and indoor globe temperature; S2: Divide the several groups of real data into training data and discriminant data in chronological order, and add redundancy to obtain the original random noise limit condition. The original random noise limit condition is the data interval obtained by adding 20% redundancy above and below the real data; S3: Input the training data and the original random noise limit condition into the generator, and the generator outputs generated data; S4: Input the original random noise limit condition, discriminant data, and generated data into the discriminator, and the discriminator outputs a discriminant result; S5: According to the discriminant result and the loss functions of discriminator D and generator G, update the discriminator parameters, and the discriminator backpropagates the error to the generator to update the generator parameters, where the loss function is where z is a random noise following a Gaussian distribution; x is real data; G represents the generator; D represents the discriminator; P data P(x) represents the probability distribution of real data; P z P(x) represents the probability distribution of random noise; x ~ P data denotes randomly sampling x from the distribution of real data; z ~ P z denotes sampling noise z from the random noise of the Gaussian distribution; x|y means obtaining x under the constraint condition y, while z|y means obtaining z under the constraint condition y; D(x|y) represents the vector output by the discriminator D after receiving the input within the parentheses; G(z|y) represents the vector output by the generator G after receiving the input within the parentheses; S6: When the discriminant result shows that the error reaches the preset accuracy or the number of learning times is greater than the preset training iteration times, terminate the algorithm and complete the training of the generator; S7: Input the current indoor temperature, indoor relative humidity, and indoor globe temperature into the trained generator, and the generator outputs the predicted indoor temperature, indoor relative humidity, and indoor globe temperature; S8: Obtain the clothing thermal resistance, human metabolic rate, and indoor wind speed, as well as the predicted indoor temperature, indoor relative humidity, and indoor globe temperature in S7, and match the corresponding indoor thermal comfort data.

2. The indoor thermal comfort data generation method according to claim 1, characterized in that, the number of training iterations is 10,000.

3. The indoor thermal comfort data generation method according to claim 1, characterized in that, The preset accuracy is error approximation .

4. The indoor thermal comfort data generation method according to claim 1, characterized in that, the activation functions of the hidden layer and output layer of the generator and discriminator are: f(x) = max(0,x), when x is 1, the obtained result is 1; when x is 0, the obtained result is 0.

5. An indoor thermal comfort data generation system, characterized in that, it includes: Data acquisition unit: used to collect several groups of real data within a period of time, and each group of real data includes indoor temperature, indoor relative humidity, and indoor globe temperature; Data preprocessing unit: used to divide the several groups of real data into training data and discriminant data in chronological order, and add redundancy to obtain the original random noise limit condition; Generation unit: used to input the training data and the original random noise limit condition into the generator, and the generator outputs generated data; Discrimination unit: used to input the original random noise limit condition, discriminant data, and generated data into the discriminator, and the discriminator outputs a discriminant result; Backpropagation unit: used to update the discriminator parameters according to the discriminant result, and the discriminator backpropagates the error to the generator to update the generator parameters; Termination unit: used to terminate the algorithm and complete the training of the generator when the discriminant result shows that the error reaches the preset accuracy or the number of learning times is greater than the preset training iteration times. Prediction unit: Input the current indoor temperature, indoor relative humidity, and indoor globe temperature into the generator after training is completed, and the generator outputs the predicted indoor temperature, indoor relative humidity, and indoor globe temperature; Thermal comfort unit: Used to obtain the clothing thermal resistance, human metabolic rate, and indoor wind speed, as well as the indoor temperature, indoor relative humidity, and indoor globe temperature predicted in S7, and match the corresponding indoor thermal comfort data.

6. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein, when the processor executes the computer program, it implements the indoor thermal comfort data generation method according to any one of claims 1 to 5.

7. A computer-readable storage medium storing a computer program, wherein, when the computer program is executed by a processor, it implements the indoor thermal comfort data generation method according to any one of claims 1 to 5.

Images