Building thermal load calculation method based on GRU-Res network and giant regular ensemble

Abstract

The invention discloses a building thermal load calculation method based on a GRU-Res network and a giant regular ensemble. The method is realized based on a building thermal load calculation model based on a GRU-Res network and a giant regular ensemble. The construction method of the model comprises the following steps: S1, obtaining parameters of a building envelope, a meteorological data matrix and a building thermal load historical operation data matrix; s2, solving the one-dimensional steady-state heat transfer equation to obtain heat consumption of the building envelope, and constructing a heat consumption matrix of the building envelope; s3, constructing a penetration load calculation model based on a giant regular ensemble theory, solving the model to obtain penetration loads at all moments, and constructing a penetration load matrix; s4, constructing a feature matrix based on the heat consumption matrix, the penetration load matrix and the meteorological data matrix; a GRU-Res network model is built; and S5, training a GRU-Res network by taking the feature matrix as input and taking the building thermal load historical operation data matrix as a label, and finally obtaining a calculation model. According to the method, the precision and efficiency of building thermal load calculation can be remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the field of building load prediction calculation, and particularly relates to a building heat load calculation method integrating a gated recurrent unit residual network (GRU-Res) and the grand canonical ensemble theory. Background Art

[0002] Building energy consumption is an important part of global energy consumption and has become one of the main energy consumption fields. Accurately calculating and predicting building heat load is of crucial significance for optimizing the indoor thermal environment of buildings and reducing building energy consumption.

[0003] Traditional building heat load calculation methods are mainly divided into two categories: physical model-based calculation methods and data-driven calculation methods. Physical model-based calculation methods are based on basic principles such as heat transfer and fluid mechanics to establish a detailed mathematical model to describe the heat transfer process of buildings. They have clear physical meanings and can better reflect the actual heat transfer mechanism of buildings. However, they require a large number of building structure parameters, material parameters, meteorological parameters, etc. The modeling process is complex, the calculation amount is large, and the time consumption is long. Data-driven calculation methods use the historical operation data and meteorological data of buildings to establish a heat load prediction model through statistical or machine learning methods. They do not require a detailed physical model, the modeling process is relatively simple, the calculation speed is fast, and they are suitable for dealing with non-linear and time-varying heat load prediction problems. However, the prediction accuracy of the model depends on the quality and quantity of historical data. When the data volume is insufficient or the quality is not high, the generalization ability of the model is poor.

[0004] How to effectively integrate the prior knowledge of physical models into data-driven models, improve the prediction accuracy and generalization ability of the models, and at the same time endow the models with certain physical interpretability is an urgent problem to be solved at present. Summary of the Invention

[0005] In view of the limitations of traditional methods in dealing with building dynamic heat processes and infiltration heat transfer, the present invention innovatively integrates a gated recurrent unit residual network (GRU-Res) and the grand canonical ensemble theory, and proposes a building heat load calculation method based on the GRU-Res network and the grand canonical ensemble. This method can quickly calculate the building heat load by combining building characteristic data and weather data, and can effectively overcome the problems of insufficient accuracy and low efficiency in existing building heat load calculation methods.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A building heat load calculation method based on the GRU-Res network and the grand canonical ensemble is realized depending on a building heat load calculation model based on the GRU-Res network and the grand canonical ensemble. The construction method of this model specifically includes the following steps:

[0008] Step S1, parameter and necessary data acquisition: Obtain various parameters of the building envelope structure, local meteorological data, and the building's historical heat load to obtain a meteorological data matrix and a building heat load historical operation data matrix;

[0009] Step S2, calculation of the heat consumption of the building envelope structure: Construct a one-dimensional steady-state heat transfer equation, and solve the one-dimensional steady-state heat transfer equation based on various parameters of the building envelope structure to obtain the heat consumption of the building envelope structure at each moment, and then obtain the heat consumption matrix of the building envelope structure;

[0010] Step S3, infiltration load calculation: Based on the grand canonical ensemble theory, construct an infiltration load calculation model, solve the infiltration load calculation model to obtain the infiltration load at each moment, and then obtain the infiltration load matrix;

[0011] Step S4, construction of the feature matrix and GRU-Res network: Construct a feature matrix based on the heat consumption matrix of the building envelope structure, the infiltration load matrix, and the meteorological data matrix; Build a GRU-Res network, and the GRU-Res network includes a GRU layer and a residual connection Res;

[0012] Step S5, training and evaluation of the GRU-Res network: Use the feature matrix as the input and the building heat load historical operation data matrix as the label to train the GRU-Res network, and evaluate the performance of the model. Finally, obtain the building heat load calculation model based on the GRU-Res network and the grand canonical ensemble.

[0013] In the above technical solution, further, in step S1:

[0014] Step S11, determination of building envelope structure parameters: Obtain the thermal parameters of each part of the building envelope structure, specifically including the heat transfer coefficients of the exterior wall, exterior window, roof, floor, ground, and door.

[0015] Step S12, meteorological data acquisition: Obtain the hourly meteorological data of the area where the building is located through the local meteorological station, including the outdoor dry-bulb temperature t o (°C), outdoor wet-bulb temperature t s (°C), solar radiation I (W / m 2 ), atmospheric pressure P (Pa), and construct a meteorological data matrix x o .

[0016] Step S13, acquisition of building historical heat load: Obtain the hourly historical operation data of the building's heat load, and construct a building heat load historical operation data matrix Q history (W).

[0017] Further, step S2 includes the following steps:

[0018] Step S21: To consider the influence of different orientations and building components on the heat transfer coefficient and the resulting differences in the heat consumption of the building envelope, the building envelope is divided into multiple sub-regions according to the orientation, including the east external window, east external wall, west external window, west external wall, north external window, north external wall, south external window, south external wall, south external door, and roof;

[0019] Step S22: Establish a one-dimensional steady-state heat transfer equation for each part of the building envelope, and solve the one-dimensional steady-state heat transfer equation based on the parameters of the building envelope to obtain the heat consumption Q of the building envelope at each moment e , which can be calculated by the following formula:

[0020]

[0021] In the formula, m is the total number of sub-regions divided by the building envelope; n i is the orientation correction coefficient of sub-region i; K i is the heat transfer coefficient of sub-region i, W / (m 2 ·K); F i is the heat transfer area of sub-region i, m 2 ; t in and t out are the indoor temperature and outdoor temperature of sub-region i, °C; the heat transfer coefficient K i and the orientation correction coefficient n i are determined according to the heating design standard in combination with the building orientation and the sub-region structure.

[0022] Step S23: Combine the hourly heat consumption Q e of the building envelope into a heat consumption matrix Q e of the building envelope.

[0023] Furthermore, in step S3, for the thermodynamic characteristics of the building air infiltration process, an infiltration load calculation model is constructed based on the grand canonical ensemble theory, and the indoor space of the building is regarded as an open system with a fixed volume, which exchanges heat and mass with the outdoor environment. Solve the infiltration load calculation model to obtain the infiltration load at each moment, and then obtain the infiltration load matrix. Specifically, it includes the following steps:

[0024] Step S31: Based on the grand canonical ensemble theory, construct an infiltration load calculation model, and solve the infiltration load calculation model to obtain the infiltration load Q i , Q i The calculation formula of is as follows:

[0025]

[0026] Q i = η·ACH·ΔQ

[0027] In the formula, ΔQ represents the energy change during a single air particle heat and mass exchange process derived based on the grand canonical ensemble theory, describing the magnitude of the energy transfer accompanying the heat and mass exchange of indoor and outdoor air particles driven by the indoor-outdoor temperature difference; P is the atmospheric pressure, in Pa; m is the average mass of air molecules, in kg; k is the Boltzmann constant, 1.38×10-23 J / K; h is the Planck constant, 6.626×10 -34 J·s; V is the internal volume of the building, in m 3 ; T in and T out are the thermodynamic temperatures of the indoor temperature and the outdoor temperature respectively, in K; η is the building type correction coefficient; ACH is the building air change rate, indicating the number of times the indoor air is replaced by outdoor air per hour, in s -1 .

[0028] Step S32, combine the hourly infiltration loads Q i to form an infiltration load matrix Q i .

[0029] Furthermore, the said step S4 includes the following steps:

[0030] Step S41, input feature matrix construction, combine the heat loss matrix Q e of the building envelope obtained in step S2 and the infiltration load matrix Q i obtained in step S3 to form a building characteristic data matrix x b ; at the same time, splice the meteorological data matrix x o in step S1 with the building characteristic matrix x b to construct a feature matrix x t :

[0031] x b = [Q e , Q i

[0032] x o = [t o , t s , i]

[0033] x t = [x b , x o

[0034] Step S42, GRU-Res network construction, construct a neural network model including a GRU layer, a residual connection layer (Res layer) and a linear layer. The calculation process of the GRU layer is as follows:

[0035] z t = σ g ​​(W z ·[x t ,h t-1 +b z )

[0036] r t =σ g (W r ·[x t ,h t-1 +b r )

[0037]

[0038] y t =σ r (W t1 h t +b t )

[0039] The implementation method of the residual connection (Res) is as follows: the hidden state h of the GRU layer t is used to obtain the output y through a linear layer with the RELU as the activation function t , and added to the input feature matrix x t to obtain y res , thus implementing the residual connection to alleviate the problem of gradient disappearance and accelerate network convergence. The specific calculation process of the Res layer is as follows:

[0040] y res =y t +x t

[0041] The calculation process of the linear layer is as follows: y res is used to obtain y t2 through the first linear layer. Then y t2 is used to obtain the final prediction result y T , which is specifically expressed as:

[0042] y t2 =W t2 (y res )+b t2

[0043] y T =W T (y t2 )+b t3

[0044] In the formula, W z , W r , W h , W t1 , W t2 , W TFor updating the weight matrices of the update gate, reset gate, current moment memory information, current moment GRU layer output, first-layer linear layer output, and second-layer linear layer output; b z and b r and b h and b t1 and b t2 and b t3 are the bias terms of the update gate, reset gate, current moment memory information, current moment GRU layer output, first-layer linear layer output, and second-layer linear layer output, where b t3 is a scalar, and the others are all vectors; z t is the update gate, controlling information transmission; r t is the reset gate, controlling information forgetting; h t-1 and h t are the hidden states of the previous moment and the current moment respectively; * is the Hadamard product operation; is the current moment memory information; y t is the current moment GRU layer output; y t2 is the output of the first-layer linear layer, which is a vector, y T is the output of the second-layer linear layer, which is a scalar; σ g is the sigmoid activation function; σ r is the RELU activation function; y t +x t is the Res operation.

[0045] Furthermore, the step S5 includes the following steps:

[0046] Step S51, dataset division, dividing the input feature matrix x t and the building heat load historical operation data matrix Q history into a training set, a validation set, and a test set according to a ratio (such as 7:1:2).

[0047] Step S52, using the training set data to train the constructed GRU-Res network, and continuously adjusting the weight matrices and bias terms in the network through the backpropagation algorithm to minimize the error between the predicted heat load and the actual heat load. During the training process, the mean square error (MSE) is used as the loss function, and the Adam optimizer is used to update the network parameters:

[0048]

[0049] In the formula, y i is the actual heat load, is the heat load predicted by the model, and n is the number of samples.

[0050] Step S53, model validation: Use the validation set data to evaluate the performance of the model during the training process to prevent overfitting and optimize the model performance.

[0051] Step S54, model testing: Use the independent test set data to evaluate the performance of the finally trained model. The evaluation metrics are root mean square error (RMSE) and mean absolute error (MAE):

[0052]

[0053] Finally, a building heat load calculation model based on the GRU-Res network and the grand canonical ensemble is obtained.

[0054] Furthermore, by inputting the building envelope structure parameters and meteorological data into the building heat load calculation model based on the GRU-Res network and the grand canonical ensemble, the building heat load can be quickly calculated.

[0055] Compared with the prior art, the beneficial effects of the present invention are:

[0056] A building heat load calculation method based on the GRU-Res network and the grand canonical ensemble proposed by the present invention effectively integrates the advantages of data-driven and physical model methods and has significant beneficial effects. First, the GRU-Res network can efficiently process time series data such as building operation and meteorology, overcoming the deficiencies of traditional data-driven methods in capturing long-term dependencies, and significantly improving the accuracy and speed of heat load prediction. Second, the introduction of the grand canonical ensemble theory combines the indoor and outdoor air heat and mass transfer process of the building with the principles of statistical mechanics, endowing the model with certain physical meanings, enhancing the robustness and generalization ability of the model, and reducing the over-reliance on a large amount of historical data.

[0057] This method effectively integrates the ability of the ordinary GRU model to process time series data and the advantages of statistical mechanics theory, significantly improving the accuracy and efficiency of building heat load calculation, especially more accurate in the calculation of infiltration load. At the same time, the introduction of the GRU-Res network enhances the adaptability of the model to complex dynamic working conditions, helps to provide a more reliable basis for building energy conservation optimization and intelligent control, and contributes to the realization of the energy conservation and emission reduction goals in the building field.

[0058] Compared with the traditional physical model method, the present invention does not need to establish a complex building physical model, greatly reducing the calculation complexity and improving the calculation efficiency. In summary, the method of the present invention can realize the rapid and accurate calculation of building heat load, providing more reliable data support for building energy conservation optimization, HVAC system control and energy consumption assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The present invention will be further described below in conjunction with the drawings and embodiments.

[0060] Figure 1 It is the construction flow chart of the building heat load calculation model based on the GRU-Res network and the grand canonical ensemble of the present invention.

[0061] Figure 2 It is the GRU layer structure diagram.

[0062] Figure 3 It is the GRU-Res network structure diagram. Specific implementation manner

[0063] The present invention discloses a building heat load calculation method based on the GRU-Res network and the grand canonical ensemble. This method relies on a building heat load calculation model based on the GRU-Res network and the grand canonical ensemble to implement, as Figure 1 For the construction process of this model, it specifically includes the following steps:

[0064] Step S1:

[0065] 1) Determine the envelope structure parameters, collect parameters such as the materials, areas, window-wall ratios, and shape factors of the envelope structures of each part of the building, and thus determine the thermal parameters of each part of the envelope structure, specifically including the heat transfer coefficients of the exterior wall, exterior window, roof, floor, ground, and door.

[0066] 2) Obtain meteorological data, obtain the hourly outdoor dry-bulb temperature, outdoor wet-bulb temperature, solar radiation, and atmospheric pressure data of the area where the building is located through the local meteorological station, and construct a meteorological data matrix x o , for subsequent building heat load calculations.

[0067] 3) Obtain the building historical heat load, obtain the hourly historical operation data of the building heat load, and construct a building historical heat load operation data matrix Q history (W).

[0068] Step S2:

[0069] Step S21, divide the building envelope structure into multiple sub-regions according to the orientation, including the east exterior window, east exterior wall, west exterior window, west exterior wall, north exterior window, north exterior wall, south exterior window, south exterior wall, south exterior door, and roof;

[0070] Step S22, establish a one-dimensional steady-state heat transfer equation for each part of the building envelope structure, and solve the one-dimensional steady-state heat transfer equation based on the parameters of each part of the building envelope structure to obtain the heat consumption Q e of the building envelope structure at each moment, which can be calculated by the following formula:

[0071]

[0072] Wherein, m is the total number of sub-regions into which the building envelope is divided; n i is the orientation correction coefficient of sub-region i; K i is the heat transfer coefficient of sub-region i, W / (m 2 ·K); F i is the heat transfer area of sub-region i, m 2 ; t in and t out are the indoor temperature and outdoor temperature of sub-region i, °C; the heat transfer coefficient K i and the orientation correction coefficient n i are determined according to the heating design standard in combination with the building orientation and the structure of the sub-region.

[0073] Step S23: Combine the hourly heat consumption Q e of the building envelope into a heat consumption matrix Q e .

[0074] Step S3:

[0075] Step S31: Based on the grand canonical ensemble theory, construct a penetration load calculation model, and solve the penetration load calculation model to obtain the penetration load Q i , Q i at each moment. The calculation formula is as follows:

[0076]

[0077] Q i = η·ACH·ΔQ

[0078] Wherein, ΔQ represents the energy change amount in a single air particle heat and mass exchange process deduced based on the grand canonical ensemble theory, and describes the magnitude of the energy transfer accompanied when indoor and outdoor air particles undergo heat and mass exchange under the drive of the indoor and outdoor temperature difference; P is the atmospheric pressure, Pa; m is the average mass of air molecules, kg; k is the Boltzmann constant, 1.38×10-23 J / K; h is the Planck constant, 6.626×10 -34 J·s; V is the internal volume of the building, m 3 ; T in and T out are the thermodynamic temperatures of the indoor temperature and outdoor temperature respectively, K; η is the building type correction coefficient; ACH is the building air change rate, indicating the number of times the indoor air is replaced by outdoor air per hour, s -1 .

[0079] Step S32: Combine the hourly penetration load Q i into a penetration load matrix Q i .

[0080] Step S4:

[0081] In step S41, input feature matrix construction. Combine the heat consumption matrix Q of the enclosure structure obtained in step S2 e and the infiltration load matrix Q obtained in step S3 i to form the building characteristic data matrix x b ; at the same time, splice the meteorological data matrix x in step S1 o with the building characteristic matrix x b to construct the feature matrix x t :

[0082] x b = [Q e , Q i

[0083] x o = [t o , t s , I]

[0084] x t = [x b , x o

[0085] In step S42, construct the GRU-Res network (such as Figure 3 ), and construct a neural network model including a GRU layer (such as Figure 2 ), a residual connection (Res) layer, and a linear layer. The calculation process of the GRU layer is as follows:

[0086] z t = σ g (W z ·[x t , h t-1 + b z )

[0087] r t = σ g (W r ·[x t , h t-1 + b r )

[0088]

[0089] y t = σ r (W t1 h t + b t )

[0090] The implementation method of the Res layer is: the hidden state h of the GRU layer t ​​The output y is obtained through a linear layer with the RELU activation function t , and added to the input feature matrix x t to obtain y res , which is specifically expressed as:

[0091] y res = y t + x t

[0092] The calculation process of the linear layer is as follows: First, y res is passed through the first linear layer to obtain y t2 , and then y t2 is passed through the second linear layer to obtain the final prediction result y T , which is specifically expressed as:

[0093] y t2 = W t2 (y res ) + b t2

[0094] y T = W T (y t2 ) + b t3

[0095] In the formula, W z , W r , W h , W t1 , W t2 , W T are the weight matrices of the update gate, reset gate, memory information at the current moment, output of the GRU layer at the current moment, output of the first linear layer, and output of the second linear layer; b z , b r , b h , b t1 , b t2 , b t3 are the bias terms of the update gate, reset gate, memory information at the current moment, output of the GRU layer at the current moment, output of the first linear layer, and output of the second linear layer, where b t3 is a scalar, and the others are all vectors; z t is the update gate, controlling information transmission; r t is the reset gate, controlling information forgetting; h t-1 and h t are the hidden states at the previous moment and the current moment respectively; * is the Hadamard product operation; is the memory information at the current moment; y t is the output of the GRU layer at the current moment; y t2 is the output of the first linear layer, which is a vector, y Tis the output of the second - layer linear layer and is a scalar; σ g is the sigmoid activation function; σ r is the RELU activation function; y t +x t is the Res operation.

[0096] Step S5:

[0097] Step S51, dataset division. Divide the input feature matrix x t and the historical operation data matrix Q of building heat load history into a training set, a validation set and a test set according to the ratio of 7:1:2.

[0098] Step S52, use the training set data to train the constructed GRU - Res network. Continuously adjust the weight matrix and bias terms in the network through the back - propagation algorithm to minimize the error between the predicted heat load and the actual heat load. During the training process, the mean squared error (MSE) is used as the loss function, and the Adam optimizer is used to update the network parameters:

[0099]

[0100] In the formula, y i is the actual heat load, is the heat load predicted by the model, and n is the number of samples.

[0101] Step S53, model validation. Use the validation set data to evaluate the performance of the model during the training process to prevent overfitting and optimize the model performance.

[0102] Step S54, model testing. Use the independent test set data to evaluate the performance of the finally trained model. The evaluation metrics are the root mean squared error (RMSE) and the mean absolute error (MAE):

[0103]

[0104] Finally, a building heat load calculation model based on the GRU - Res network and the grand canonical ensemble is obtained.

[0105] Use the above - trained building heat load calculation model based on the GRU - Res network and the grand canonical ensemble to calculate the building heat load. By directly inputting the building envelope structure parameters and meteorological data into the model, the building heat load can be quickly calculated.

Claims

1. A building heat load calculation method based on GRU-Res network and grand canonical ensemble, characterized in that: It relies on a building heat load calculation model based on GRU-Res network and grand canonical ensemble. The construction method of the model specifically includes the following steps: Step S1, parameter and necessary data acquisition, obtain the parameters of the building envelope structure, local meteorological data and historical building heat load, obtain the meteorological data matrix and the building heat load historical operation data matrix; Step S2, calculating the heat consumption of the building envelope structure, constructing a one-dimensional steady-state heat transfer equation, solving the one-dimensional steady-state heat transfer equation based on various parameters of the building envelope structure to obtain the heat consumption of the building envelope structure at each moment, and then obtaining a heat consumption matrix of the building envelope structure; Step S3, seepage load calculation, based on the grand canonical ensemble theory, construct a seepage load calculation model, solve the seepage load calculation model to obtain the seepage load at each moment, and then obtain a seepage load matrix; Step S4, constructing a feature matrix and a GRU-Res network, and constructing a feature matrix based on the heat consumption matrix of the building envelope, the infiltration load matrix, and the meteorological data matrix; Building a GRU-Res network, wherein the GRU-Res network includes a GRU layer, a residual connection Res layer, and a linear layer; Step S5, training and evaluation of the GRU-Res network, taking the feature matrix as input and the building heat load historical operation data matrix as labels, training the GRU-Res network and evaluating the performance of the model, and finally obtaining the building heat load calculation model based on the GRU-Res network and the grand canonical ensemble.

2. A building heat load calculation method based on GRU-Res network and grand canonical ensemble according to claim 1, characterized in that: In the step S1: The method for obtaining the building envelope parameters is as follows: according to the actual building information, the thermal parameters of the building envelope structures of each part are obtained, specifically including the heat transfer coefficients of the exterior walls, exterior windows, roofs, floors, ground and doors; The method for obtaining the meteorological data is to obtain hourly meteorological data of the area where the building is located through the local meteorological station, including the outdoor dry-bulb temperature t o , unit is ℃, outdoor wet bulb temperature t s , unit is ℃, solar radiation I, unit is W / m 2 , atmospheric pressure P, unit is Pa.

3. The method for calculating building heat load based on GRU-Res network and grand canonical ensemble according to claim 2, characterized in that: The step S2 comprises the following steps: Step S21, dividing the building envelope into multiple sub-areas according to orientation, including east exterior windows, east exterior walls, west exterior windows, west exterior walls, north exterior windows, north exterior walls, south exterior windows, south exterior walls, south exterior doors, and roofs; Step S22, construct a one-dimensional steady-state heat transfer equation, and solve the one-dimensional steady-state heat transfer equation based on various parameters of the building envelope structure to obtain the heat consumption Q of the building envelope structure at each moment e , Q e Specifically calculated by the following formula: Where m is the total number of sub-areas divided by the building envelope; n i is the orientation correction coefficient of sub-area i; K i is the heat transfer coefficient of sub-region i, W / (m 2 ·K); f i is the heat transfer area of ​​sub-region i, m 2 ;t in and t out is the indoor temperature and outdoor temperature of sub-area i, ℃; the heat transfer coefficient K of sub-area i i and orientation correction factor n i The value of is determined according to the heating design standard and combined with the building orientation and sub-area structure; Step S23: Calculate the hourly heat consumption Q of the building envelope structure e Combined into the building envelope heat consumption matrix Q e .

4. The method for calculating building heat load based on GRU-Res network and grand canonical ensemble according to claim 3 is characterized in that: The step S3 comprises the following steps: Step S31: Based on the grand canonical ensemble theory, a seepage load calculation model is constructed, and the seepage load calculation model is solved to obtain the seepage load Q at each moment. i , Q i The calculation formula is as follows: Q i =η·ACH·ΔQ In the formula, ΔQ represents the energy change in a single air particle heat and mass exchange process derived from the grand canonical ensemble theory, which describes the energy transfer when the indoor and outdoor air particles exchange heat and mass driven by the indoor and outdoor temperature difference; P is the atmospheric pressure, Pa; m is the average mass of air molecules, kg; k is the Boltzmann constant, 1.38×10 -23 J / K; h is Planck's constant, 6.626×10 -34 J·s; V is the internal volume of the building, m 3 ; T in and T out are the thermodynamic temperatures of indoor and outdoor temperatures, K; η is the building type correction factor; ACH is the building ventilation rate, which means the number of times the indoor air is replaced by outdoor air per hour, s -1 ; Step S32: Calculate the hourly infiltration load Q i Combined into the penetration load matrix Q i .

5. The method for calculating building heat load based on GRU-Res network and grand canonical ensemble according to claim 4 is characterized in that: The step S4 comprises the following steps: Step S41, input the characteristic matrix to construct the heat consumption matrix Q of the enclosure structure obtained in step S2 e and the penetration load matrix Q obtained in step S3 i Combine to form the building characteristic data matrix x b ; The building characteristic matrix x b and the meteorological data matrix x o Concatenate to form the feature matrix x t : x b =[Q e ,Q i ] x o =[t o ,t s ,I] x t =[x b ,x o ] Step S42, GRU-Res network construction, constructing a neural network model including a GRU layer, a residual connection Res layer and a linear layer; the calculation process of the GRU layer is as follows: z t =σ g (W z ·[x t ,h t-1 ]+b z ) r t =σ g (W r ·[x t ,h t-1 ]+b r ) y t =s r (W t1 h t +b t ) The residual connection Res layer is implemented by: t The output y is obtained through a linear layer with RELU as the activation function t , and with the input feature matrix x t Add together to get y res , specifically expressed as: and res =and t +x t The calculation process of the linear layer is as follows: first, y res Through the first linear layer, we get y t2 , then y t2 The final prediction result y is obtained through the second linear layer T , specifically expressed as: y t2 =W t2 (y res )+b t2 y T =W T (y t2 )+b t3 Where W z , W r , W h , W t1 , W t2 , W T b is the weight matrix of the update gate, reset gate, current memory information, current GRU layer output, first linear layer output, and second linear layer output; b z , b r , b h , b t , b t2 , b t3 is the bias term of the update gate, reset gate, current memory information, current GRU layer output, first linear layer output, and second linear layer output, where b t3 is a scalar, the others are vectors; z t is the update gate, controlling information transmission; r t is the reset gate to control information forgetting; h t-1 and h t are the hidden states of the previous moment and the current moment respectively; * is the Hadamard product operation; Memory information for the current moment; y t is the GRU layer output at the current moment; y t2 is the output of the first linear layer, a vector, y t is the output of the second linear layer, which is a scalar; σ g is the sigmoid activation function; r is the RELU activation function; y t +x t It is the Res operation.

6. The method for calculating building heat load based on GRU-Res network and grand canonical ensemble according to claim 5 is characterized in that: In the step S5: Step S51, data set division, input feature matrix x t and the building heat load historical operation data matrix Q history Divide into training set, validation set and test set; Step S52, use the training set data to train the constructed GRU-Res network, and continuously adjust the weight matrix and bias term in the network through the back propagation algorithm to minimize the error between the predicted heat load and the actual heat load. During the training process, the mean square error MSE is used as the loss function, and the Adam optimizer is used to update the network parameters: In the formula, y i is the actual heat load, is the heat load predicted by the model, n is the number of samples; Step S53, model verification, using the verification set data to evaluate the model performance during the training process to prevent overfitting and optimize model performance; Step S54, model testing, uses independent test set data to evaluate the performance of the final trained model, and the evaluation indicators are root mean square error RMSE and mean absolute error MAE: Finally, a building heat load calculation model based on GRU-Res network and grand canonical ensemble was obtained.

7. The method for calculating building heat load based on GRU-Res network and grand canonical ensemble according to claim 1, characterized in that: By inputting building envelope parameters and meteorological data into the building heat load calculation model based on the GRU-Res network and the grand canonical ensemble, the building heat load can be calculated.

Images