Building indoor environment heat and humidity load decoupling prediction method and device

Abstract

The invention discloses a building indoor environment heat and humidity load decoupling prediction method and device. The method comprises the steps of collecting data of a target building indoor environment in real time; the collected data is preprocessed; constructing an FEEMD (fast ensemble empirical mode decomposition) model, wherein the FEEMD is fast ensemble empirical mode decomposition; the FEEMD method is adopted to decompose the time sequence of the data set, and the complex time sequence is converted into stable data; carrying out OrdinalEncoder feature coding, and constructing a training data set required by training of a neural network model CNN-GRU; a CNN-GRU model is built; evaluating model performance indexes; according to the air conditioner heat and humidity load decoupling prediction method based on the CNN-GRU neural network model, noise errors of high-frequency parts of humidity and temperature parameters are effectively reduced through an FEEMD decomposition method, so that the precision and stability of a prediction result are improved, facial expression features and facial temperature features of indoor people are introduced, and the prediction accuracy is improved. And the robustness of heat and humidity load model prediction is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence prediction technology and indoor environmental quality control, and in particular to a method and device for decoupling prediction of heat and moisture loads in a building indoor environment. Background Art

[0002] At present, the proportion of energy consumption in the construction industry has gradually increased, accounting for 36% of global energy demand and 37% of energy-related carbon emissions. The main energy consumption of building operation is HVAC, which accounts for about 2 / 3 of the total operating energy consumption. And as climate conditions deteriorate, the global average temperature continues to warm. HVAC has gradually become an indispensable part of life and is one of the lifelines under extreme humid weather and heat waves. Accurate control of the air conditioning system can reduce cooling waste, improve air conditioning operation efficiency and save building energy consumption.

[0003] As people's production and living standards continue to improve, the requirements for indoor air temperature and humidity are also gradually increasing, which requires more accurate control methods for the operation of building air conditioning systems. Especially in the high temperature and high humidity areas of South China, there is a large demand for dehumidification inside buildings. Low-temperature chilled water and condensation dehumidification are usually used to dehumidify the air to meet the requirements for indoor temperature and humidity control accuracy. For this process of coupling heat load and moisture load (referred to as "heat and moisture load"), the air supply of the air conditioning system generally needs to be reheated, resulting in the cancellation of cold and heat, and increasing the operating energy consumption of the air conditioner. In order to solve the technical problem of low air conditioning energy efficiency caused by the coupling of heat and moisture loads, people have tried to improve it. Chinese patent document 201410522631.X provides a "control method and device for an air conditioning system", including: obtaining the real-time sensible heat load and real-time latent heat load of the building, and obtaining the rated sensible heat and rated latent heat of the air conditioning system terminal of the building; obtaining the sensible heat load percentage of the real-time sensible heat load and the rated sensible heat according to the real-time sensible heat load and the rated sensible heat; obtaining the latent heat load percentage of the real-time latent heat load and the rated latent heat according to the real-time latent heat load and the rated latent heat; determining a real-time chilled water boundary temperature according to the real-time sensible heat load percentage and the real-time latent heat load percentage; adjusting the average chilled water temperature of the air conditioning system according to the real-time chilled water boundary temperature, so that the average chilled water temperature is less than or equal to the real-time chilled water boundary temperature. This cycle is repeated continuously, thereby solving the problems of affecting the cooling and dehumidification capacity of the air conditioning system and the problem of high energy consumption caused by the existing air conditioning system control strategy.

[0004] However, this technology still has shortcomings and still cannot get rid of its inherent control lag. The reason is that the energy efficiency level of the current air-conditioning system operation depends on the pre-established operation control strategy. Most of the control strategies are aimed at matching the system output load and the real-time demand load of the building, thereby controlling the operation of the air-conditioning system. However, the control method based on the real-time demand load target is a negative feedback regulation, which has an inherent control lag, limiting the improvement of the energy efficiency level of the air-conditioning system. Therefore, the use of advanced air-conditioning load forecasting technology can know the load in advance, and realize efficient and energy-saving operation of the air-conditioning system through advance control.

[0005] Commonly used air conditioning cooling load prediction methods are: physical model method, mathematical statistical analysis method and artificial neural network method. The physical model method is suitable for long-term load prediction in the design stage, but it has high requirements for modeling accuracy and is difficult to obtain data; based on mathematical statistical analysis methods, such as multivariate linear regression, although it is easy to use and develop and takes into account the temporal relationship, it cannot handle nonlinear problems well, and has low generalization and can only be used for a single building type; artificial neural network is an emerging prediction method, such as support vector machine, BP neural network, etc., which has strong generalization and can achieve high prediction accuracy when given sufficient samples. The existing artificial intelligence algorithm has the error problem of noise in the high-frequency part of humidity and temperature parameters, and cannot achieve the accuracy and stability of the prediction results. In addition, the existing technology does not introduce the temperature and humidity facial expression characteristics and temperature and humidity voice expression characteristics of indoor people as prediction indicators, and the prediction results lack robustness. Summary of the invention

[0006] The purpose of the present invention is to provide a method and device for decoupling prediction of heat and moisture loads in a building indoor environment, so as to solve the above-mentioned problems existing in the prior art.

[0007] The specific application is as follows:

[0008] A method for decoupling prediction of heat and humidity loads in indoor environment of a building, comprising the following steps:

[0009] S1, collecting first indoor data of the indoor environment of a target building, first physiological data of indoor personnel of the target building, and second outdoor data of the outdoor environment where the target building is located, wherein the first indoor data, the first physiological data, and the second outdoor data all include historical data and real-time data;

[0010] S2, cleaning, preprocessing, clustering analysis, and encoding the collected first indoor data, first physiological data, and second outdoor data in turn, and generating first indoor data features, second outdoor data features, and first physiological data features, respectively;

[0011] S3, using the FEEMD method to decompose the time series of the first indoor data set and the second outdoor data set to obtain IMF components;

[0012] S4, performing OrdinalEncoder feature encoding on the discrete IMF components in the first indoor data set and the second outdoor data set, so that the discrete IMF component data is converted into a number between 0 and n-1, where n is the number of types; the remaining continuous IMF component data is Z-Score standardized and encoded, and the encoded data is subjected to time series window processing using a window sliding method to generate corresponding first indoor data features and second outdoor data features, wherein the first indoor data features and the second outdoor data features are used to construct a training data set required for training a neural network model CNN-GRU, and the training data set is divided into a training set and a test set, wherein the training data set also includes a first physiological data feature;

[0013] S5. Analyze the training set through the CNN-GRU model to obtain the final model after the CNN-GRU model training, wherein the CNN-GRU model uses a convolutional neural network CNN combined with a gated recurrent neural network GRU;

[0014] S6. Input the test set into the final model after CNN-GRU model training to obtain the predicted value of the final model; evaluate the performance of the model in predicting the indoor heat and humidity load of the target building based on the predicted value and the true value combined with the mean square error (MSE);

[0015] S7. Preset the comfort threshold of the indoor facial temperature data of the target building. If the prediction result does not meet the comfort threshold, use the genetic algorithm to dynamically adjust the CNN-GRU model parameters to obtain the final prediction result of the indoor heat and humidity load of the target building.

[0016] Furthermore, the specific implementation process of preprocessing the first physiological data in S2 is:

[0017] S21, real-time recording of the user's facial expression file and facial temperature data file, and aligning the timestamps of the two, wherein the first indoor data includes indoor air dry-bulb temperature, indoor air relative humidity, chilled water supply temperature of the air-conditioning system, chilled water return temperature, chilled water flow, fresh air outlet wind speed, fresh air outlet air temperature, fresh air outlet air humidity, exhaust air temperature, exhaust air humidity, return air outlet air temperature, return air outlet air humidity and air conditioning load data; the second outdoor data includes outdoor air dry-bulb temperature, outdoor air relative humidity and outdoor air wind speed; the first physiological data includes the user's facial expression features and facial temperature features;

[0018] S22, performing Gaussian denoising on the real-time recorded user facial expressions, and extracting speech feature sequences using a spectrogram algorithm to generate output speech expression feature sequences;

[0019] S23, based on the video file of facial temperature data, using an image recognition algorithm to perform face detection, and extracting a facial temperature feature sequence therefrom;

[0020] S24. Classify the speech expression feature sequence and the facial temperature feature sequence through a three-dimensional convolutional neural network algorithm, generate expression features and facial temperature features respectively, and classify the obtained expression features and facial temperature features as the first physiological data features.

[0021] Furthermore, the specific implementation process of performing cluster analysis on the first indoor data and the second outdoor data in S2 is as follows:

[0022] The first indoor data and the second outdoor data are classified and processed respectively through the K-prototype clustering algorithm, and the data are divided into two categories: main category data and other category data; the main category data is retained and the other category data is removed; after the data is classified by the clustering algorithm, the category with the most dense distribution and the highest score after classification is taken as the main category data, and all other categories are taken as other category data.

[0023] Furthermore, the specific implementation process of S3 is as follows:

[0024] S31, the white noise sequence g i (t) is combined with the original heat and moisture load data signal x(t) to obtain a new heat and moisture load data sequence x i (t): x i (t) = x(t) + g i (t), i=1, 2, ... n, where n is the number of times the white noise sequence is generated, and the original heat and moisture load data signal includes a first indoor data signal and a second outdoor data signal;

[0025] S32. Decompose the new heat and moisture load data time series x according to the FEEMD method i (t), the decomposed components include IMF component a i,j (t) and the trend term u i (t), that is Among them, a i,j (t) represents the jth IMF component obtained after the introduction of white noise i times, and K is the number of IMF components;

[0026] S33, for n groups of IMF components a i,j (t) and the trend term u i (t) The average calculation is performed to obtain the final component Ij (t) and trend term P(t):

[0027]

[0028] Furthermore, the OrdinalEncoder feature encoding in S4 includes: OrdinalEncoder feature encoding is a feature encoding method for processing categorical data; OrdinalEncoder feature encoding is used to convert categorical labels into integer values ​​so that the model can process non-numerical data; the remaining continuous IMF component data is subjected to Z-Score standardized encoding, including: after the continuous IMF component data features are subjected to Z-Score standardized encoding, the continuous IMF component data is mapped to between 0 and 1.

[0029] Furthermore, the specific implementation of S5 is as follows:

[0030] The CNN-GRU model architecture and processing process are as follows: the CNN model and the GRU model are connected in series, the first physiological data feature is input into the CNN model, the first CNN processed data set is output through CNN model feature processing, and the first CNN processed data set, the first indoor data feature, and the second outdoor data feature are input into the GRU model;

[0031] The processing process of the CNN model is:

[0032] Applying kernel K-means clustering analysis with adaptive weight allocation to the first physiological data feature Y, feature enhancement and dimension reduction are performed to obtain the enhanced feature vector Y', which is expressed as:

[0033]

[0034] in, is the kernel function, W' is the dimension reduction weight matrix;

[0035] The enhanced feature vector Y' is input into a multi-layer fusion and convolution network, and combined with the nonlinear transformation function φ, the fused feature vector Z is expressed as:

[0036] where α i is the weight of the i-th mode, Q and S represent the temperature and humidity voice expression features and temperature and humidity facial expression features respectively;

[0037] Input the fused feature Z into the two-dimensional convolution layer and calculate the output value of the two-dimensional convolution layer. The expression is:

[0038] C=tanh(W*Z+b), where tanh is the rectified linear activation function, W is the weight matrix of the two-dimensional convolutional layer, and b is the bias term of the two-dimensional convolutional layer.

[0039] Furthermore, the GRU model processing process is:

[0040] S51, the GRU model is divided into three layers, marked as the first layer, the second layer and the third layer, the first CNN processed data set is used to input the first layer, the first indoor data feature is used to input the second layer, and the second outdoor data feature is used to input the third layer;

[0041] S52, the calculation formula of the GRU model is:

[0042] z t =σ(W z .[h t-1 , x t ]+b z ),

[0043] r t =σ(W r .[h t-1 , x t ]+b r ),

[0044]

[0045] Among them, x t represents the input feature vectors of the first, second and third layers, z t 、r t , h t Respectively represent the output data of the update gate, reset gate, candidate hidden state, and current hidden state in each layer of the GRU model, W z , W r , W represents the weight matrix of the update gate, reset gate, and candidate hidden state corresponding to the GRU model, b z , b r , b represents the update gate, reset gate, and bias term corresponding to the candidate hidden state of the GRU model, σ() represents the activation function, and h t and h t-1 Respectively represent the hidden states of the current moment and the previous moment;

[0046] S53, the output heat and moisture load prediction value calculation formula of the GRU model is:

[0047] Q=σ(W1.y GRU1 +W2.y GRU2 +W3.y GRU3 +b e ),

[0048] Among them, W1, W2, and W3 represent the weight matrices of the fully connected layers corresponding to the first, second, and third layers respectively; b e represents the bias term of the fully connected layer; y GRU1 ,y GRU2 ,y GRU3 They represent the output features corresponding to the first, second and third layers respectively; Q represents the predicted value of heat and moisture load output by the GRU model.

[0049] Furthermore, the evaluation performance index in S6 includes: inputting the test set into the final model after CNN-GRU training to obtain the model's predicted value for the training set result; evaluating the performance of the model in heat and moisture load prediction based on the model's predicted value for heat and moisture load data and the corresponding true value of heat and moisture load data in the training set combined with the mean square error MSE; the mean square error MSE calculation formula is:

[0050] Where n is the number of heat and moisture load data test sets, Q j is the true value of the heat and moisture load data corresponding to the training set, is the model's predicted value for the heat and moisture load data.

[0051] A device for decoupling prediction of heat and humidity loads in indoor environment of a building, the device comprising a memory and a processor;

[0052] The storage device is used to store a program of a method for decoupling prediction of heat and moisture loads of indoor environment of a building;

[0053] The processor is used to execute the program of the building indoor environment heat and moisture load decoupling prediction method to implement the building indoor environment heat and moisture load decoupling prediction method described in the present invention.

[0054] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects:

[0055] The embodiment of the present invention provides real-time data collection of the indoor environment of a target building; preprocessing the collected data; constructing a FEEMD model, wherein the FEEMD is a fast ensemble empirical mode decomposition; decomposing the time series of a data set using the FEEMD method to convert the complex time series into stable data; OrdinalEncoder feature encoding to construct a training data set required for training a neural network model CNN-GRU; building a CNN-GRU model; and evaluating model performance indicators. The present invention provides an air conditioning heat and humidity load decoupling prediction method based on the CNN-GRU neural network model, which effectively reduces the error of noise in the high-frequency part of humidity and temperature parameters by using the FEEMD decomposition method, thereby improving the accuracy and stability of the prediction result, and introducing indoor facial expression features and facial temperature data of people to improve the robustness of model prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a flow chart of a method for decoupling prediction of heat and humidity loads in indoor environment of a building provided by an embodiment of the present invention;

[0057] Figure 2 is a schematic diagram of a device for decoupling prediction of heat and humidity loads in a building indoor environment provided by an embodiment of the present invention;

[0058] Figure 3 It is a schematic diagram of a CNN-GRU model of a building indoor environment heat and humidity load decoupling prediction device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The present invention will be described in detail below in conjunction with the accompanying drawings.

[0060] Example 1

[0061] First, the technical terms involved in the embodiments of the present application are explained.

[0062] (1) Gated Recurrent Neural Network (GRU)

[0063] The Gated Recurrent Neural Network (GRU) is a variant of the Recurrent Neural Network (RNN) for processing sequence data. It solves the gradient vanishing and gradient exploding problems in traditional RNNs by introducing a gating mechanism. The GRU has a simpler structure, fewer parameters, and is usually faster to train.

[0064] The core of GRU lies in two gates: Reset Gate and Update Gate. The Reset Gate determines how the new input information is combined with the previous memory, while the Update Gate controls the amount of memory from the previous moment saved to the current moment. Both gates are calculated by the sigmoid function, and their output values ​​are between 0 and 1, which are used to control the flow of information.

[0065] The advantages of GRU include powerful time series data processing capabilities, high efficiency and good interpretability. GRU is widely used in many fields such as natural language processing, speech recognition, and recommendation systems.

[0066] The embodiment of the present invention provides a method for decoupling prediction of heat and humidity loads in indoor environment of a building. Figure 1 , including the following steps:

[0067] S1, collecting first indoor data of the indoor environment of a target building, first physiological data of indoor personnel of the target building, and second outdoor data of the outdoor environment where the target building is located, wherein the first indoor data, the first physiological data, and the second outdoor data all include historical data and real-time data;

[0068] S2, cleaning, preprocessing, clustering analysis, and encoding the collected first indoor data, first physiological data, and second outdoor data in turn, and generating first indoor data features, second outdoor data features, and first physiological data features, respectively;

[0069] S3, using the FEEMD method to decompose the time series of the first indoor data set and the second outdoor data set to obtain IMF components;

[0070] Specifically, the IMF component is a modal function with time characteristics, which converts complex time series into stable data;

[0071] S4, performing OrdinalEncoder feature encoding on the discrete IMF components in the first indoor data set and the second outdoor data set, so that the discrete IMF component data is converted into a number between 0 and n-1, where n is the number of types; the remaining continuous IMF component data is Z-Score standardized and encoded, and the encoded data is subjected to time series window processing using a window sliding method to generate corresponding first indoor data features and second outdoor data features, wherein the first indoor data features and the second outdoor data features are used to construct a training data set required for training a neural network model CNN-GRU, and the training data set is divided into a training set and a test set, wherein the training data set also includes a first physiological data feature;

[0072] S5. Analyze the training set through the CNN-GRU model to obtain the final model after the CNN-GRU model training, wherein the CNN-GRU model uses a convolutional neural network CNN combined with a gated recurrent neural network GRU;

[0073] S6. Input the test set into the final model after CNN-GRU model training to obtain the predicted value of the final model; evaluate the performance of the model in predicting the indoor heat and humidity load of the target building based on the predicted value and the true value combined with the mean square error (MSE);

[0074] S7. Preset the comfort threshold of the indoor facial temperature data of the target building. If the prediction result does not meet the comfort threshold, use the genetic algorithm to dynamically adjust the CNN-GRU model parameters to obtain the final prediction result of the indoor heat and humidity load of the target building.

[0075] Specifically, the first indoor data, the first physiological data and the second outdoor data of the indoor environment of the target building are collected; the collected first indoor data, the first physiological data and the second outdoor data are cleaned, preprocessed, clustered and encoded in turn to generate the first indoor data feature, the second outdoor data feature and the first physiological data feature respectively; the FEEMD method is used to decompose the time series of the first indoor data and the second outdoor data, and the complex time series is converted into stable data; OrdinalEncoder feature encoding is performed to construct the training data set required for training the neural network model CNN-GRU; a CNN-GRU model is built, and the heat and humidity load of the indoor environment of the target building is predicted by the CNN-GRU model based on the first indoor data feature, the second outdoor data feature and the first physiological data feature; the model performance indicators are evaluated and re-optimized; the air conditioning heat and humidity load decoupling prediction method based on the CNN-GRU neural network model effectively reduces the noise error in the high-frequency part of the humidity and temperature parameters through the FEEMD decomposition method, thereby improving the accuracy and stability of the prediction results, and the facial expression features and facial temperature features of indoor people are introduced to improve the robustness of the model prediction.

[0076] In the above embodiment, specifically, the specific implementation process of preprocessing the first physiological data in S2 is:

[0077] S21, real-time recording of the user's facial expression file and facial temperature data file, and aligning the timestamps of the two, wherein the first indoor data includes indoor air dry-bulb temperature, indoor air relative humidity, chilled water supply temperature of the air-conditioning system, chilled water return temperature, chilled water flow, fresh air outlet wind speed, fresh air outlet air temperature, fresh air outlet air humidity, exhaust air temperature, exhaust air humidity, return air outlet air temperature, return air outlet air humidity and air conditioning load data; the second outdoor data includes outdoor air dry-bulb temperature, outdoor air relative humidity and outdoor air wind speed; the first physiological data includes the user's facial expression features and facial temperature features;

[0078] S22, performing Gaussian denoising on the real-time recorded user facial expressions, and extracting speech feature sequences using a spectrogram algorithm to generate output speech expression feature sequences;

[0079] S23, based on the video file of facial temperature data, using an image recognition algorithm to perform face detection, and extracting a facial temperature feature sequence therefrom;

[0080] S24. Classify the speech expression feature sequence and the facial temperature feature sequence through a three-dimensional convolutional neural network algorithm, generate expression features and facial temperature features respectively, and classify the obtained expression features and facial temperature features as the first physiological data features.

[0081] Specifically, the user's facial expression features include happiness, like, disgust, hate, surprise, comfort, and disgust; the user's facial temperature features include happiness, like, disgust, hate, surprise, comfort, temperature is too high, temperature is too low, temperature is comfortable, and humidity is comfortable.

[0082] In the above embodiment, specifically, the specific implementation process of performing cluster analysis on the first indoor data and the second outdoor data in S2 is:

[0083] The first indoor data and the second outdoor data are classified and processed by the K-prototype clustering algorithm, and the data are divided into two categories: main category data and other category data; the main category data is retained and the other category data is removed; after the data is classified by the clustering algorithm, the category with the most dense distribution and the highest score after classification is taken as the main category data, and all other categories are taken as other category data;

[0084] Specifically, the K-prototype clustering algorithm performs clustering on the sample set D = {y1, y2, ...y n The classification method is: by continuously updating the central object, the cluster division C = {C1, C2, ...C k}, C1, C2, C k Each represents a single cluster, where each cluster includes multiple single samples in the same cluster; the clustering rule is to minimize the mean square error, and the mean square error formula is: Among them, y represents being classified into cluster C i The sample data in , δ represents the Hamming distance, w i It is cluster C i The mean vector, P d The smaller the value of is, the higher the similarity of samples y within the cluster.

[0085] In the above embodiment, specifically, the specific implementation process of S3 is:

[0086] S31, the white noise sequence g i (t) is combined with the original heat and moisture load data signal x(t) to obtain a new heat and moisture load data sequence x i (t): x i (t) = x(t) + g i (t), i=1, 2, ... n, where n is the number of times the white noise sequence is generated, and the original heat and moisture load data signal includes a first indoor data signal and a second outdoor data signal;

[0087] S32. Decompose the new heat and moisture load data time series x according to the FEEMD method i (t), the decomposed components include IMF component a i,j (t) and the trend term ui (t), that is Among them, a i,j (t) represents the jth IMF component obtained after the introduction of white noise i times, and K is the number of IMF components;

[0088] S33, for n groups of IMF components a i,j (t) and the trend term u i (t) The average calculation is performed to obtain the final component I j (t) and trend term P(t):

[0089] Furthermore, the OrdinalEncoder feature encoding in S4 includes: OrdinalEncoder feature encoding is a feature encoding method for processing categorical data; OrdinalEncoder feature encoding is used to convert categorical labels into integer values ​​so that the model can process non-numerical data; the remaining continuous IMF component data is subjected to Z-Score standardized encoding, including: after the continuous IMF component data features are subjected to Z-Score standardized encoding, the continuous IMF component data is mapped to between 0 and 1.

[0090] In the above embodiment, specifically, the specific implementation of S5 is:

[0091] The CNN-GRU model architecture and processing process are as follows: the CNN model and the GRU model are connected in series, the first physiological data feature is input into the CNN model, and the first CNN processed data set is output through CNN model feature processing, and the first CNN processed data set, the first indoor data feature, and the second outdoor data feature are input into the GRU model, such as Figure 3 As shown;

[0092] The processing process of the CNN model is:

[0093] Applying kernel K-means clustering analysis with adaptive weight allocation to the first physiological data feature Y, feature enhancement and dimension reduction are performed to obtain the enhanced feature vector Y', which is expressed as:

[0094]

[0095] in, is the kernel function, W' is the dimension reduction weight matrix;

[0096] The enhanced feature vector Y' is input into a multi-layer fusion and convolution network, and combined with the nonlinear transformation function φ, the fused feature vector Z is expressed as:

[0097] where α iis the weight of the i-th mode, Q and S represent the temperature and humidity voice expression features and temperature and humidity facial expression features respectively;

[0098] Input the fused feature Z into the two-dimensional convolution layer and calculate the output value of the two-dimensional convolution layer. The expression is:

[0099] C=tanh(W*Z+b), where tanh is the rectified linear activation function, W is the weight matrix of the two-dimensional convolutional layer, and b is the bias term of the two-dimensional convolutional layer.

[0100] In the above embodiment, specifically, the GRU model processing process is:

[0101] S51, the GRU model is divided into three layers, marked as the first layer, the second layer and the third layer, the first CNN processed data set is used to input the first layer, the first indoor data feature is used to input the second layer, and the second outdoor data feature is used to input the third layer;

[0102] S52, the calculation formula of the GRU model is:

[0103] z t =σ(W z .[h t-1 , x t ]+b z ),

[0104] r t =σ(W r .[h t-1 , x t ]+b r ),

[0105]

[0106] Among them, x t represents the input feature vectors of the first, second and third layers, z t 、r t , h t Respectively represent the output data of the update gate, reset gate, candidate hidden state, and current hidden state in each layer of the GRU model, W z , W r , W represents the weight matrix of the update gate, reset gate, and candidate hidden state corresponding to the GRU model, b z , b r , b represents the update gate, reset gate, and bias term corresponding to the candidate hidden state of the GRU model, σ() represents the activation function, and h t and h t-1 Respectively represent the hidden states of the current moment and the previous moment;

[0107] S53, the output heat and moisture load prediction value calculation formula of the GRU model is:

[0108] Q=σ(W1.y GRU1 +W2.y GRU2 +W3.y GRU3 +b e ),

[0109] Among them, W1, W2, and W3 represent the weight matrices of the fully connected layers corresponding to the first, second, and third layers respectively; b e represents the bias term of the fully connected layer; y GRU1 ,y GRU2 ,y GRU3 They represent the output features corresponding to the first, second and third layers respectively; Q represents the predicted value of heat and moisture load output by the GRU model.

[0110] In the above embodiment, specifically, the evaluation performance index in S6 includes: inputting the test set into the final model after CNN-GRU training, and obtaining the prediction value of the model for the training set result; evaluating the performance of the model in heat and moisture load prediction according to the prediction value of the model for the heat and moisture load data and the true value of the corresponding heat and moisture load data in the training set combined with the mean square error MSE; the mean square error MSE calculation formula is:

[0111] Where n is the number of heat and moisture load data test sets, Q j is the true value of the heat and moisture load data corresponding to the training set, is the model's predicted value for the heat and moisture load data.

[0112] Example 2

[0113] Step 1: Use an intelligent monitoring device to collect outdoor air dry-bulb temperature, outdoor air relative humidity, outdoor air wind speed, direct solar radiation, solar scattered radiation, indoor air dry-bulb temperature, indoor air relative humidity, number of personnel, equipment utilization rate, chilled water supply temperature of the air-conditioning system, chilled water return temperature, chilled water flow, fresh air inlet wind speed, fresh air inlet air temperature, fresh air inlet air humidity, exhaust outlet air temperature, exhaust outlet air humidity, return air outlet air temperature, return air outlet air humidity and air conditioning load data. The collection frequency is once every hour.

[0114] Preprocessing the collected data, dividing the collected data into sensible heat load data and latent heat load data, wherein the sensible heat load represents temperature change, and the latent heat load represents humidity change;

[0115] Identification of abnormal data, the air conditioning load is divided into 24 sequences according to time, from 0:00 to 23:00. Through the box plot, it can be intuitively seen that the values ​​that exceed the upper limit set by the box plot or are lower than the lower limit set are abnormal values.

[0116] To deal with abnormal data, missing values ​​and abnormal values ​​were replaced by the mean of the data at the same time point in a week.

[0117] The collected data are normalized to eliminate the impact of dimensions between different units.

[0118] The formula for calculating normalization is:

[0119]

[0120] Where: X' is the normalized data value, X is the original data set, X max and X min are the maximum and minimum values ​​in the original data set, respectively.

[0121] Step 2: This embodiment includes but is not limited to using cluster analysis for feature selection. This embodiment uses principal component analysis to cluster and select features for variables affecting air conditioning heat and humidity loads, and divides the data set into sensible heat load training set Text_Seen 1, test set Text_Seen 2 and latent heat load training set Text_Latent 1, test set Text_Latent 2, specifically including the following steps:

[0122] First, assume that there are n data samples, each of which contains p data x1, x2, x3…x n , then the representation matrix of the original data sample is:

[0123]

[0124] The above p original data are transformed into p new data using the linear transformation of the matrix, denoted as F, and expressed as follows:

[0125]

[0126] Then the variance contribution rate of each main component is calculated by the following formula:

[0127]

[0128] Select the principal components whose features are greater than the preset threshold and whose cumulative variance contribution rate reaches more than 80% as the input of the air conditioning load prediction model. This input includes the sensible heat load training set Text_Seen1, the test set Text_Seen2 and the latent heat load training set Text_Seen1, the test set Text_Seen2;

[0129] The test set data is input into the CNN-GRU model for training, and the training is stopped when the preset conditions are met.

[0130] Furthermore, the evaluation performance index includes but is not limited to the following one or a combination of two or more, including: mean square error MSE, mean absolute error Mae, mean absolute error percentage Mape, root mean square error Rmse. The calculation formula is as follows:

[0131]

[0132] Among them, m is the number of samples, CL p,j is the load forecast value, CL a，j is the measured load value, CL a，mean is the average value of the measured load.

[0133] Example 3

[0134] A device for decoupling prediction of heat and humidity loads in indoor environment of a building, the device comprising a memory and a processor, such as Figure 2 As shown;

[0135] The storage device 10 is used to store a program of a method for predicting the decoupling of heat and moisture loads of indoor environments of buildings;

[0136] The processor 20 is used to execute the program of the building indoor environment heat and moisture load decoupling prediction method to implement the building indoor environment heat and moisture load decoupling prediction method described in the present invention.

[0137] The processor 20 is further used to collect first indoor data of the indoor environment of the target building, first physiological data of the indoor personnel of the target building, and second outdoor data of the outdoor environment where the target building is located, wherein the first indoor data, the first physiological data, and the second outdoor data all include historical data and real-time data;

[0138] The processor 20 is further used to sequentially clean, pre-process, cluster analyze, and encode the collected first indoor data, the first physiological data, and the second outdoor data, and to generate first indoor data features, second outdoor data features, and first physiological data features, respectively;

[0139] The processor 20 is further configured to decompose the time series of the first indoor data set and the second outdoor data set using a FEEMD method to obtain an IMF component;

[0140] The processor 20 is further used to perform OrdinalEncoder feature encoding on the discrete IMF components in the first indoor data set and the second outdoor data set, so that the discrete IMF component data is converted into a number between 0 and n-1, where n is the number of types; the remaining continuous IMF component data is Z-Score standardized and encoded, and the encoded data is subjected to time series window processing using a window sliding method to generate corresponding first indoor data features and second outdoor data features, wherein the first indoor data features and the second outdoor data features are used to construct a training data set required for training a neural network model CNN-GRU, and the training data set is divided into a training set and a test set, wherein the training data set also includes a first physiological data feature;

[0141] The processor 20 is further used to analyze the training set through the CNN-GRU model to obtain a final model after the CNN-GRU model is trained. The CNN-GRU model uses a convolutional neural network CNN combined with a gated recurrent neural network GRU;

[0142] The processor 20 is also used to input the test set into the final model after the CNN-GRU model training to obtain the predicted value of the final model; and evaluate the performance of the model in predicting the indoor heat and humidity load of the target building according to the predicted value and the true value combined with the mean square error MSE;

[0143] The processor 20 is also used to preset a comfort threshold of the facial temperature data of indoor personnel in the target building. If the prediction result does not meet the comfort threshold, a genetic algorithm is used to dynamically adjust the CNN-GRU model parameters to obtain a final prediction result of the indoor heat and humidity load of the target building.

[0144] The algorithm and display provided herein are not inherently related to any particular computer, virtual system or other device. Various general purpose systems can also be used together with the teachings based on this. According to the above description, it is obvious that the structure required for constructing such systems. In addition, the present invention is not directed to any specific programming language either. It should be understood that various programming languages ​​can be utilized to realize the content of the present invention described herein, and the description of the above specific languages ​​is for disclosing the best mode of the present invention.

[0145] In the description provided herein, a large number of specific details are described. However, it is understood that embodiments of the present invention can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.

[0146] Similarly, it should be understood that in order to streamline the present disclosure and aid in understanding one or more of the various inventive aspects, in the above description of exemplary embodiments of the present invention, various features of the present invention are sometimes grouped together into a single embodiment, figure, or description thereof. However, this disclosed method should not be interpreted as reflecting the intention that the claimed invention requires more features than those explicitly recited in each claim. More specifically, as reflected in the claims below, inventive aspects lie in less than all the features of the individual embodiments disclosed above. Therefore, the claims that follow the specific embodiment are hereby expressly incorporated into the specific embodiment, with each claim itself serving as a separate embodiment of the present invention.

[0147] Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component, and in addition they may be divided into a plurality of submodules or subunits or subcomponents. Except that at least some of such features and / or processes or units are mutually exclusive, all features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed in this manner may be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature providing the same, equivalent or similar purpose.

[0148] In addition, those skilled in the art will appreciate that, although some embodiments herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the present invention and form different embodiments. For example, in the claims below, any one of the claimed embodiments may be used in any combination.

[0149] The various component embodiments of the present invention may be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It should be understood by those skilled in the art that a microprocessor or digital signal processor (DSP) may be used in practice to implement some or all of the functions of some or all of the components in the apparatus according to an embodiment of the present invention. The present invention may also be implemented as a device or device program (e.g., a computer program and a computer program product) for executing part or all of the methods described herein. Such a program implementing the present invention may be stored on a computer-readable medium, or may have the form of one or more signals. Such a signal may be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

Claims

1. A decoupling prediction method for heat and humidity load of building indoor environment, characterized in that: The following steps are involved: S1, collecting first indoor data of the indoor environment of a target building, first physiological data of indoor personnel of the target building, and second outdoor data of the outdoor environment where the target building is located, wherein the first indoor data, the first physiological data, and the second outdoor data all include historical data and real-time data; S2, cleaning, preprocessing, clustering analysis, and encoding the collected first indoor data, first physiological data, and second outdoor data in turn, and generating first indoor data features, second outdoor data features, and first physiological data features, respectively; S3, using the FEEMD method to decompose the time series of the first indoor data set and the second outdoor data set to obtain IMF components; S4, performing OrdinalEncoder feature encoding on the discrete IMF components in the first indoor data set and the second outdoor data set, so that the discrete IMF component data is converted into a number between 0 and n-1, where n is the number of types; the remaining continuous IMF component data is Z-Score standardized and encoded, and the encoded data is subjected to time series window processing using a window sliding method to generate corresponding first indoor data features and second outdoor data features, wherein the first indoor data features and the second outdoor data features are used to construct a training data set required for training a neural network model CNN-GRU, and the training data set is divided into a training set and a test set, wherein the training data set also includes a first physiological data feature; S5. Analyze the training set through the CNN-GRU model to obtain the final model after the CNN-GRU model training, wherein the CNN-GRU model uses a convolutional neural network CNN combined with a gated recurrent neural network GRU; S6. Input the test set into the final model after CNN-GRU model training to obtain the predicted value of the final model; evaluate the performance of the model in predicting the indoor heat and humidity load of the target building based on the predicted value and the true value combined with the mean square error (MSE); S7. Preset the comfort threshold of the indoor facial temperature data of the target building. If the prediction result does not meet the comfort threshold, use the genetic algorithm to dynamically adjust the CNN-GRU model parameters to obtain the final prediction result of the indoor heat and humidity load of the target building.

2. A method for decoupling prediction of heat and humidity loads in indoor building environment according to claim 1, characterized in that: The specific implementation process of preprocessing the first physiological data in S2 is: S21, real-time recording of the user's facial expression file and facial temperature data file, and aligning the timestamps of the two, the first indoor data includes indoor air dry-bulb temperature, indoor air relative humidity, chilled water supply temperature of the air-conditioning system, chilled water return temperature, chilled water flow, fresh air outlet wind speed, fresh air outlet air temperature, fresh air outlet air humidity, exhaust air temperature, exhaust air humidity, return air outlet air temperature, return air outlet air humidity and air conditioning load data; the second outdoor data includes outdoor air dry-bulb temperature, outdoor air relative humidity and outdoor air wind speed; The first physiological data includes facial expression characteristics and facial temperature characteristics of the user; S22, performing Gaussian denoising on the real-time recorded user facial expressions, and extracting speech feature sequences using a spectrogram algorithm to generate output speech expression feature sequences; S23, based on the video file of facial temperature data, using an image recognition algorithm to perform face detection, and extracting a facial temperature feature sequence therefrom; S24. Classify the speech expression feature sequence and the facial temperature feature sequence through a three-dimensional convolutional neural network algorithm, generate expression features and facial temperature features respectively, and classify the obtained expression features and facial temperature features as the first physiological data features.

3. A method for decoupling prediction of heat and humidity loads in indoor environment of a building according to claim 1, characterized in that: The specific implementation process of performing cluster analysis on the first indoor data and the second outdoor data in S2 is as follows: The first indoor data and the second outdoor data are classified and processed respectively through the K-prototype clustering algorithm, and the data are divided into two categories: main category data and other category data; the main category data is retained and the other category data is removed; after the data is classified by the clustering algorithm, the category with the most dense distribution and the highest score after classification is taken as the main category data, and all other categories are taken as other category data.

4. A method for decoupling prediction of heat and humidity loads in building indoor environment according to claim 1, characterized in that: The specific implementation process of S3 is as follows: S31, the white noise sequence g i (t) is combined with the original heat and moisture load data signal x(t) to obtain a new heat and moisture load data sequence x i (t): x i (t) = x(t) + g i (t), i=1, 2, ... n, where n is the number of times the white noise sequence is generated, and the original heat and moisture load data signal includes a first indoor data signal and a second outdoor data signal; S32. Decompose the new heat and moisture load data time series x according to the FEEMD method i (t), the decomposed components include IMF component a i,j (t) and the trend term u i (t), that is Among them, a i,j (t) represents the jth IMF component obtained after the introduction of white noise i times, and K is the number of IMF components; S33, for n groups of IMF components a i,j (t) and the trend term u i (t) The average calculation is performed to obtain the final component I j (t) and trend term P(t):

5. The method for decoupling prediction of heat and humidity loads in indoor environment of a building according to claim 1 is characterized in that: The OrdinalEncoder feature encoding in S4 includes: OrdinalEncoder feature encoding is a feature encoding method for processing categorical data; OrdinalEncoder feature encoding is used to convert categorical labels into integer values ​​so that the model can process non-numerical data; the remaining continuous IMF component data is subjected to Z-Score standardized encoding, including: after the continuous IMF component data features are subjected to Z-Score standardized encoding, the continuous IMF component data is mapped to between 0 and 1.

6. A method for decoupling prediction of heat and humidity loads in building indoor environment according to claim 2, characterized in that: The specific implementation of S5 is: The CNN-GRU model architecture and processing process are as follows: the CNN model and the GRU model are connected in series, the first physiological data feature is input into the CNN model, the first CNN processed data set is output through CNN model feature processing, and the first CNN processed data set, the first indoor data feature, and the second outdoor data feature are input into the GRU model; The processing process of the CNN model is: Applying kernel K-means clustering analysis with adaptive weight allocation to the first physiological data feature Y, feature enhancement and dimension reduction are performed to obtain the enhanced feature vector Y', which is expressed as: in, is the kernel function, W' is the dimension reduction weight matrix; The enhanced feature vector Y' is input into a multi-layer fusion and convolution network, and combined with the nonlinear transformation function φ, the fused feature vector Z is expressed as: where α i is the weight of the i-th mode, Q and S represent the temperature and humidity voice expression features and temperature and humidity facial expression features, respectively; Input the fused feature Z into the two-dimensional convolution layer and calculate the output value of the two-dimensional convolution layer. The expression is: C=tanh(W*Z+b), where tanh is the rectified linear activation function, W is the weight matrix of the two-dimensional convolutional layer, and b is the bias term of the two-dimensional convolutional layer.

7. A method for decoupling prediction of heat and humidity loads in indoor environment of a building according to claim 6, characterized in that: The GRU model processing process is: S51, the GRU model is divided into three layers, marked as the first layer, the second layer and the third layer, the first CNN processed data set is used to input the first layer, the first indoor data feature is used to input the second layer, and the second outdoor data feature is used to input the third layer; S52, the calculation formula of the GRU model is: z t =σ(W z .[h t-1 ,x t ]+b z ), r t =σ(W r .[h t-1 ,x t ]+b r ), Among them, x t represents the input feature vectors of the first, second and third layers, z t 、r t , h t Respectively represent the output data of the update gate, reset gate, candidate hidden state, and current hidden state in each layer of the GRU model, W z , W r , W represents the weight matrix of the update gate, reset gate, and candidate hidden state corresponding to the GRU model, b z 、b r , b represents the update gate, reset gate, and bias term corresponding to the candidate hidden state of the GRU model, σ() represents the activation function, and h t and h t-1 Respectively represent the hidden states of the current moment and the previous moment; S53, the output heat and moisture load prediction value calculation formula of the GRU model is: Q=σ(W1.y GRU1 +W2.y GRU2 +W3.y GRU3 +b e ), Among them, W1, W2, and W3 represent the weight matrices of the fully connected layers corresponding to the first, second, and third layers respectively; b e represents the bias term of the fully connected layer; y GRU1 ,y GRU2 ,y GRU3 They represent the output features corresponding to the first, second and third layers respectively; Q represents the predicted value of heat and moisture load output by the GRU model.

8. A method for decoupling prediction of heat and humidity loads in indoor environment of a building according to claim 7, characterized in that: The evaluation performance index in S6 includes: inputting the test set into the final model after CNN-GRU training to obtain the model's predicted value for the training set result; evaluating the performance of the model in heat and moisture load prediction based on the model's predicted value for heat and moisture load data and the corresponding true value of heat and moisture load data in the training set combined with the mean square error MSE; the mean square error MSE calculation formula is: Where n is the number of heat and moisture load data test sets, Q j is the true value of the heat and moisture load data corresponding to the training set, is the model's predicted value for the heat and moisture load data.

9. A device for predicting the thermal and humid load decoupling of a building indoor environment, used to execute the method for predicting the thermal and humid load decoupling of a building indoor environment according to any one of claims 1 to 8, characterized in that: The device includes a memory and a processor; The storage device is used to store a program of a method for decoupling prediction of heat and moisture loads of indoor environment of a building; The processor is used to execute the program of the building indoor environment heat and moisture load decoupling prediction method to implement the building indoor environment heat and moisture load decoupling prediction method described in the present invention.

Images