Ventilation opening or skylight design method in folk house courtyard top capping reconstruction scene

Abstract

A ventilation opening or skylight design method in a folk house courtyard top capping reconstruction scene comprises the following specific steps that S1, a year-round natural ventilation quantity optimization target or a light environment and building energy consumption optimization target is set according to seasons, and the optimization target comprises a positive target and a negative target; s2, respectively executing an orthogonal experiment on the optimization targets in the step S1, calculating the optimization targets through a ventilation plug-in of a Grasshopper platform, and constructing a data set; s3, weighting the optimization target based on an entropy method to obtain an optimization target entropy weight; s4, establishing an optimization target prediction model based on a deep feedforward neural network, and performing training by using the data set obtained in the step S2; and S5, based on the optimization target prediction model trained in the step S4 and the optimization target entropy weight obtained in the step S3, generating a ventilation opening design scheme or a skylight design scheme by using an NSGA-II algorithm and a Pareto algorithm. According to the method, the entropy weight is endowed to the optimization target and is applied to multi-target optimization calculation.

Description

A method for designing ventilation openings or skylights in the scenario of adding or renovating the roof of a residential courtyard. Technical Field

[0001] This invention belongs to the field of green and low-carbon renovation technology of residential buildings, and relates to a design method for ventilation openings or skylights in the scenario of adding a roof to the courtyard of a residential building. Background Technology

[0002] The courtyard in traditional courtyard houses serves multiple functions, including providing shade, ventilation, and spatial organization. Currently, there is a real need for roof additions to the courtyards during modern renovations of these houses. These roof additions can provide rain shelter and enrich and expand the usable space of the courtyard. However, roof additions significantly impact ventilation in courtyard houses. Therefore, any roof addition renovation must consider its impact on natural ventilation. The design of ventilation openings in courtyard roof renovations affects the natural ventilation of courtyard houses differently depending on the season. Taking hot-summer, cold-winter regions as an example, in spring and autumn, the natural ventilation of courtyard-style residences needs to be maximized to remove excess indoor heat and improve indoor air quality; in winter, the natural ventilation needs to be minimized to reduce heat loss; in summer, there are two scenarios: sheltered from the wind and evening ventilation. The sheltered-from-the-wind scenario refers to daytime conditions in summer, where outdoor air temperature is higher than indoor air temperature, requiring a reduction in natural ventilation to decrease cooling energy consumption; the evening ventilation scenario is during the cooler, more suitable time to open windows, maximizing natural ventilation to reduce excess indoor heat and improve indoor air quality. It is evident that the objectives of natural ventilation in courtyard-style residences vary with the seasons, and there are complex coupling relationships between these seasonal objectives. Designing year-round natural ventilation openings for courtyard roof renovations based solely on a single seasonal natural ventilation objective is inevitably biased and may overlook other aspects. Therefore, the design of year-round natural ventilation openings for courtyard roof renovations is a multi-objective decision-making problem.

[0003] The addition of a roof to a courtyard space has a significant impact on the year-round natural lighting and energy consumption for cooling and heating within the courtyard. The impact of skylight design on year-round natural lighting and energy consumption for cooling and heating in residential courtyard renovations varies depending on the opening area. Reducing the window area helps reduce solar radiation entering the courtyard and reduces heat transfer from outside air through the roof envelope, thus lowering cooling energy consumption. However, reducing the window area also reduces the amount of natural light entering the courtyard, potentially leading to insufficient natural light. Increasing the window area increases the amount of natural light entering the courtyard. However, increasing the window area also increases solar radiation entering the courtyard and heat transfer from outside air through the roof envelope, thus increasing cooling energy consumption. The impact of skylight opening area on heating energy consumption is more complex. Increasing the window area increases solar radiation gain within the courtyard, reducing heating energy consumption, but it also increases heat transfer from inside the courtyard to the outside through the roof envelope. Therefore, it is necessary to calculate and analyze heating energy consumption based on the local climate characteristics and the specific building conditions. The impact of skylight size on glare in atriums is complex. Both excessively large and small skylights can lead to either excessive or insufficient lighting, increasing the risk of indoor glare. Therefore, it is necessary to calculate and analyze indoor glare based on the local light climate characteristics and the specific building conditions. The qualitative analysis above shows that skylight size simultaneously affects lighting, cooling energy consumption, heating energy consumption, and glare, and these four factors exhibit a complex, inversely related relationship.

[0004] There is considerable research and practice in multi-objective optimization design of building physical environments. Using various plugins on the Grasshopper platform for building environment simulation and multi-objective optimization calculations through plugins such as Wallacei and Octopus is a common technical solution for solving multi-objective optimization of building physical environments. This approach has advantages such as simple programming and visualized results. However, it suffers from two main drawbacks. First, the computation time is very long; a slightly complex multi-objective optimization calculation can take up to three weeks, severely impacting its feasibility. Second, while the importance of optimization objectives varies, and objective weights significantly influence the multi-objective optimization process and results, this approach often distributes optimization objective weights evenly, lacking comprehensive consideration and flexible weighting, thus affecting the reliability of the optimization process and results. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a design method for ventilation openings or skylights in the scenario of adding or renovating the roof of a residential courtyard. This method improves time efficiency, assigns entropy weights to the optimization objective and applies them in multi-objective optimization calculations, and provides a flexible optimization objective weighting scheme to adapt to actual engineering needs, including subjective weighting of optimization objectives.

[0006] The technical solution adopted in this invention is:

[0007] A method for designing ventilation openings or skylights in the scenario of adding or renovating the roof of a residential courtyard, the specific steps of which are as follows:

[0008] S1, set annual natural ventilation optimization targets according to the seasons or set light environment and building energy consumption optimization targets. The optimization targets include positive targets and negative targets. At the same time, set optimization variables and the value range of the optimization variables.

[0009] S2, Perform orthogonal experiments on the optimization objectives in step S1 respectively, calculate the optimization objectives using the Grasshopper platform plugin, and construct the dataset;

[0010] S3, The optimization objective is weighted based on the entropy method to obtain the optimization objective entropy weight;

[0011] S4. Establish an optimized target prediction model based on a deep feedforward neural network (DFNN) and train it using the dataset obtained in step S2.

[0012] S5. Based on the optimized target prediction model trained in step S4 and the optimized target entropy weights obtained in step S3, the NSGA-II algorithm and Pareto algorithm are used to generate the ventilation opening design scheme or skylight design scheme.

[0013] Furthermore, in step S1, the total number of objectives to be optimized is f, where the number of positive objectives is a and the number of negative objectives is b, and f = a + b. , , Indicates a positive goal, using , , This indicates a negative objective; a positive objective means the larger the target value, the better, and a negative objective means the smaller the target value, the better; set optimization variables. , , The number of optimization variables is d. The optimization variables refer to the design parameters of ventilation openings that affect natural ventilation or the design parameters of skylights that affect the light environment and building energy consumption.

[0014] Furthermore, the specific steps for constructing the dataset in step S2 are as follows:

[0015] S21, Set the number of factors for the orthogonal experiment to be... The number of levels for each factor is set to h, according to and Select an orthogonal experimental table and determine the number of experiments based on the orthogonal experimental table. ;

[0016] S22, determine the value of each factor for each experimental sample based on the uniform distribution of the level quantity and the value range of the factor;

[0017] S23, implement parametric geometric and physical modeling of buildings through programming on the Grasshopper platform;

[0018] S24. Using the corresponding plugins on the Grasshopper platform, calculate the optimization target values ​​according to the orthogonal experiment.

[0019] S25, The completed dataset, which includes: ( ), referring to the first The first optimization variable is the... The values ​​for this experiment are determined by the orthogonal experimental setup. ( (Refers to the first) The first positive goal The calculated values ​​for this experiment; ( (Refers to the first) The first negative target The calculated values ​​for this experiment.

[0020] Furthermore, the specific steps of assigning weights to all optimization indices based on the entropy method in step S3 are as follows:

[0021] S31, Use the dataset from step S2 as the sample dataset for the entropy method;

[0022] S32, using formula (1) to target the positive direction , , Data standardization is performed, and equation (2) is used for negative targets. , , , Data standardization is performed to obtain a standardized dataset that eliminates differences in measurement units;

[0023] (1)

[0024] (2)

[0025] In the formula, ( (Refers to the first) The first positive goal The standard value of this experiment;

[0026] ( (Refers to the first) The first negative target The standard value of this experiment;

[0027] S33, In order to avoid some data having low values ​​after standardization, equations (3) and (4) are used to perform translation processing on the positive and negative targets respectively;

[0028] (3)

[0029] (4)

[0030] In the formula,

[0031] H represents the magnitude of the index shift, with a value of 0.01.

[0032] ( (Refers to the first) The first positive goal The translation value of this experiment;

[0033] ( (Refers to the first) The first negative target The translation value of this experiment;

[0034] S34, use equations (5) and (6) to normalize the positive and negative targets respectively;

[0035] (5)

[0036] (6)

[0037] In the formula:

[0038] ( (Refers to the first) The first positive goal Normalized values ​​for this experiment;

[0039] ( (Refers to the first) The first negative target Normalized values ​​for this experiment;

[0040] S35, use equations (7) and (8) to calculate the information entropy of the optimized target for the positive and negative targets respectively;

[0041] (7)

[0042] (8)

[0043] In the formula:

[0044] (Refers to the first) Information entropy of a positive target

[0045] (Refers to the first) The information entropy of a negative target;

[0046] S36, use equations (9) and (10) to calculate the difference coefficients of the optimized target for the positive and negative targets respectively;

[0047] (9)

[0048] (10)

[0049] (Refers to the first) The difference coefficient of a positive objective

[0050] (Refers to the first) The difference coefficient of each negative target;

[0051] S37, use equations (11) and (12) to calculate the normalized entropy weights of the optimization target for the positive and negative targets respectively;

[0052] (11)

[0053] (12)

[0054] (Refers to the first) The normalized entropy weights of a positive objective.

[0055] (Refers to the first) Normalized entropy weights for each negative objective.

[0056] Furthermore, in step S4, the establishment and training of the target prediction model are performed using the Python platform and the Tensorflow toolkit.

[0057] Furthermore, the specific steps for establishing and training the optimized target prediction model in step S4 are as follows:

[0058] S41, Configure the environment: Import the relevant libraries on the Python platform;

[0059] S42, Read the dataset imported from step S2, and... , , , The column is the input data, and , , , , , The columns represent the output data;

[0060] S43, randomly split the dataset, with 80% used as the training set and 20% as the test set;

[0061] S44, standardize the optimization parameter values ​​in the dataset to a distribution with a mean of 0 and a standard deviation of 1, and standardize the optimization objective values ​​in the dataset to a distribution with a mean of 0 and a standard deviation of 1.

[0062] S45. Create an optimized target prediction model based on a deep feedforward neural network. Set the input layer dimension to d and the output layer dimension to f. Create neurons and use fully connected layers with ReLU activation function, for a total of five fully connected layers. The first two layers increase the number of neurons to expand the network's balancing ability, and the subsequent layers reduce the number of neurons to compress the feature dimension. Set the learning rate to 0.001. Set the mean squared error as the loss function. Set the mean absolute error as the overfitting evaluation metric. Set the training epochs of the model. Set the number of samples used for each parameter update.

[0063] S46. Input the dataset processed in step S44 into the optimized target prediction model for training. Divide 20% of the training set as the validation set and set an early stopping mechanism.

[0064] Furthermore, the libraries in step S41 include: pandas for data processing, numpy for numerical computation, sklearn for machine learning tools, tensorflow for building neural networks, and matplotlib for plotting.

[0065] Furthermore, step S45 also includes setting up the random discarding of 30% of neurons during the training process.

[0066] Furthermore, the specific steps for generating the ventilation opening or skylight design scheme in step S5 are as follows:

[0067] S51, Set the range of optimization variables and standardize the range of optimization variables to make it consistent with the format used when training the model;

[0068] S52, configured for multi-objective optimization problems, , , , Defined as a positive goal, , , , Defined as a negative target, , , , Define the optimization variables; call the optimization objective prediction model in step S4; input the entropy weights of the optimization objective. and ), set the optimization objective weights for the multi-objective optimization problem;

[0069] S53, set the population size, number of iterations, crossover rate, mutation rate, and set the number of solutions in the Pareto non-dominated solution set to q;

[0070] S54, calculate the overall priority weight of the schemes in the Pareto non-dominated solution set according to equation (13).

[0071] (13)

[0072] In the formula, ( ) is the comprehensive priority weight value of the j-th solution in the Pareto non-dominated solution set;

[0073] ( The value is the standardized value of the i-th positive objective value of the j-th solution in the Pareto non-dominated solution set, obtained by applying scaler_y.

[0074] ( The value is the standardized value of the i-th negative objective value of the j-th solution in the Pareto non-dominated solution set using scaler_y.

[0075] S55, the scheme corresponding to the maximum comprehensive priority weight value in the Pareto non-dominated solution set is used as the multi-objective optimization design scheme for year-round natural ventilation in the scenario of adding a roof to the courtyard of a residential building, or the multi-objective optimization design scheme for lighting and energy consumption in the scenario of adding a roof to the courtyard of a residential building.

[0076] The beneficial effects of this invention are: improving time efficiency, assigning entropy weights to optimization objectives and applying them in multi-objective optimization calculations, and providing a flexible optimization objective weighting scheme to adapt to practical engineering needs, including subjective weighting of optimization objectives. Attached Figure Description

[0077] Figure 1 is a schematic diagram of the technical solution of the present invention.

[0078] Figures 2-1 to 2-7 are architectural schematic diagrams of embodiments of the present invention.

[0079] Figure 3 is a schematic diagram of the programming for calculating the optimization target of the vent based on the Grasshopper platform.

[0080] Figure 4 is a schematic diagram of the calculation results of the ventilation opening optimization target in the orthogonal experiment.

[0081] Figures 5-1 and 5-2 are schematic diagrams of the ventilation opening design using the present invention and optimized target entropy weight.

[0082] Figure 6 is a schematic diagram of multi-objective optimization programming for vents on the Grasshopper platform.

[0083] Figure 7 shows the performance of multi-objective optimization programming for vents on the Grasshopper platform.

[0084] Figures 8-1 and 8-2 are schematic diagrams of the ventilation opening design using the present invention and optimized target average weighting.

[0085] Figure 9 is a schematic diagram of the Pareto solution set and optimal solution using the present invention and the optimization of the target entropy weight of the ventilation opening.

[0086] Figure 10 is a schematic diagram of the Pareto solution set and the optimal solution using the present invention and the average weighting of the ventilation opening optimization objective.

[0087] Figure 11 is a schematic diagram of the programming for calculating the target of skylight optimization based on the Grasshopper platform.

[0088] Figure 12 is a schematic diagram of the calculation results of the skylight optimization objective in the orthogonal experiment.

[0089] Figure 13 is a schematic diagram of a skylight design using the present invention and optimized target entropy weight.

[0090] Figure 14 is a schematic diagram of multi-objective optimization programming for skylights on the Grasshopper platform.

[0091] Figure 15 shows the performance of Skylight multi-objective optimization programming on the Grasshopper platform.

[0092] Figure 16 is a schematic diagram of a sunroof design using the present invention and optimized target average weighting.

[0093] Figure 17 is a schematic diagram of the Pareto solution set and optimal solution using the present invention and the skylight optimization of the target entropy weight.

[0094] Figure 18 is a schematic diagram of the Pareto solution set and the optimal solution using the present invention and the average weighting of the target of the skylight optimization. Detailed Implementation

[0095] The present invention will be further described below with reference to specific embodiments, but the invention is not limited to these specific embodiments. Those skilled in the art should recognize that the present invention covers all alternatives, improvements, and equivalents that may be included within the scope of the claims.

[0096] Using a courtyard-style residence in Hangzhou as a case study, this technical solution will be demonstrated. The aerial view of this case study building is shown in Figure 2-1. It has a depth of 18.6 meters, a width of 10.2 meters, a two-story main structure, a land area of ​​199.6 square meters, and a building area of ​​313.8 square meters.

[0097] The proposed solution is to add a roof to the atrium space of this building to provide rain shelter and enrich and expand the functionality of the atrium space. The preliminary design scheme for the roof addition is shown in Figures 2-2 to 2-6. A schematic diagram of the building renovation scheme is shown in Figure 2-2. The first-floor plan is shown in Figure 2-3, the second-floor plan in Figure 2-4, the ventilation opening design on the east facade is shown in Figure 2-5, the ventilation opening design on the west facade is shown in Figure 2-6, and the skylight design is shown in Figure 2-7.

[0098] Example 1

[0099] In addition to covering the atrium, this renovation plan proposes to install ventilation openings on the east and west sides of the building. The location, width, and height of these ventilation openings are crucial factors affecting the building's natural ventilation. This invention provides a year-round optimized design method for these ventilation openings to achieve natural ventilation under typical seasonal conditions and to find the optimal ventilation opening design. The preliminary design of the ventilation openings in this case is shown in Figures 2-5 and 2-6. The center of the ventilation opening is the center of the openable area. The minimum opening width and height are 1600mm x 250mm, and the maximum opening width and height are 6300mm x 950mm. This invention seeks the optimal opening scheme within the range between the minimum and maximum opening widths. The opening scheme is represented by the following optimization variables: East side opening width... Height of the opening on the east side , width of the opening on the west side Height of the opening on the west side The range between the minimum and maximum opening is the optimal range for the ventilation opening.

[0100] This embodiment provides a ventilation opening design method for a residential courtyard roof renovation project. Referring to Figure 1, which illustrates the technical solution of the present invention, the specific steps of this embodiment are as follows:

[0101] S1, set annual natural ventilation volume optimization targets by season, with a total number of targets f, of which a are positive optimization targets and b are negative optimization targets, f = a + b. , , Indicates a positive goal, using , , This indicates a negative objective; a positive objective means the target value should be as high as possible, and a negative objective means the target value should be as low as possible. (Set optimization variables) , , The number of optimization variables is d. The optimization variables refer to the design parameters of the ventilation openings that affect natural ventilation, and the range of values ​​for the optimization variables is set.

[0102] Specifically, this embodiment sets a target for optimizing ventilation in spring. Summer ventilation optimization goals Autumn ventilation optimization goals Summer wind shelter optimization targets Winter wind shelter optimization goals ;

[0103] The optimization indicators set in this embodiment , , , , All values ​​are average wind speeds measured at points evenly spaced at 0.5m intervals on a horizontal plane 1.5m above the ground inside the courtyard, in m / s.

[0104] Spring ventilation optimization goals With the positive objective of maximizing the value, the calculation method was carried out on the Grasshopper platform, using the Ladybug plugin to obtain meteorological parameters of Hangzhou. The statistics showed that the local spring maximum frequency wind direction was due north, and the average wind speed under the maximum frequency wind direction was 2.73 m / s. Under the calculation conditions, all the exterior windows of the case building were open.

[0105] Summer ventilation optimization goals With the positive objective of maximizing the value, the calculation method was carried out on the Grasshopper platform, using the Ladybug plugin to obtain meteorological parameters of Hangzhou. The statistics showed that the maximum frequency wind direction in the local summer from 6 pm to 9 pm was 22.5° east of south, and the average wind speed under the maximum frequency wind direction was 2.87 m / s. Under the calculation conditions, all the windows of the case building were open.

[0106] Autumn ventilation optimization goals With the positive objective of maximizing the value, the calculation method was carried out on the Grasshopper platform, using the Ladybug plugin to obtain meteorological parameters of Hangzhou. The statistics showed that the local autumn maximum frequency wind direction was due north, and the average wind speed under the maximum frequency wind direction was 2.59 m / s. Under the calculation conditions, all the exterior windows of the case building were open.

[0107] Summer wind shelter optimization targets The negative objective is that the smaller the value, the better. The calculation method is to use the Grasshopper platform and the Ladybug plugin to obtain meteorological parameters of Hangzhou. The statistics show that the local summer maximum frequency wind direction is 45° south of west, and the average wind speed under the maximum frequency wind direction is 2.55m / s. Under the calculation conditions, all the windows of the case building are closed.

[0108] Winter wind protection optimization goals The negative objective is that the smaller the value, the better. The calculation method is to use the Grasshopper platform and the Ladybug plugin to obtain meteorological parameters of Hangzhou. The statistics show that the local winter maximum frequency wind direction is due north, and the average wind speed under the maximum frequency wind direction is 2.50m / s. Under the calculation conditions, all the windows of the case building are closed.

[0109] Set the following four optimization variables: width of the east ventilation opening Height of the east ventilation opening Width of the west ventilation opening Height of the west ventilation opening All units are meters (m). and The value range is set to [1.6m, 6.3m]. and The value range is set to [0.25m, 0.95m].

[0110] S2, Perform orthogonal experiments on the optimization objectives in step S1 respectively, and use the ventilation plugin of the Grasshopper platform to calculate and analyze the optimization objectives and construct a dataset;

[0111] The specific steps for constructing the dataset are as follows:

[0112] S21, Set the number of factors for the orthogonal experiment to be... The number of levels for each factor is set to h, according to and Select an orthogonal experimental table and determine the number of experiments based on the orthogonal experimental table. ;

[0113] Specifically, the orthogonal experimental setup was configured with 4 factors, and each factor had 4 levels. The L-axis was selected.256 (4) 4 The orthogonal experimental setup is used, with a total of 256 experiments for each optimization objective. Orthogonal experiments are conducted for each of the five optimization objectives.

[0114] S22, determine the value of each factor for each experimental sample based on the uniform distribution of the level quantity and the value range of the factor;

[0115] Specifically, the number of levels and the range of values ​​for each factor are evenly distributed to determine the level. , , , The values ​​of each experimental sample;

[0116] S23, implement parametric geometric and physical modeling of buildings through programming on the Grasshopper platform;

[0117] S24, using the Butterfly plugin on the Grasshopper platform, calculates each optimization objective value according to orthogonal experiments;

[0118] Specifically, using the Butterfly plugin on the Grasshopper platform, the optimization objective for all samples is calculated according to orthogonal experiments. , , , , The programming part is shown in Figure 3;

[0119] S25, The completed dataset, which includes: ( ), referring to the first The first optimization variable is the... The values ​​for this experiment are determined by the orthogonal experimental setup. ( (Refers to the first) The first positive goal The calculated values ​​for this experiment; ( (Refers to the first) The first negative target The calculated values ​​for this experiment.

[0120] Specifically, the completed dataset is shown in Table 1, and the optimization target values ​​in the dataset are shown in Figure 4.

[0121] Table 1. Dataset

[0122] Sample number surface

[0123] Variable descriptions in the table:

[0124] ( ), referring to the first The first optimization variable is the... The values ​​for this experiment are determined by the orthogonal experimental setup and the range of values ​​for the optimized variables.

[0125] ( (Refers to the first) The first positive goal The calculated values ​​for this experiment;

[0126] ( (Refers to the first) The first negative target The calculated values ​​for this experiment.

[0127] S3, The optimization objective is weighted using the entropy method to obtain the entropy weights of the optimization objective; specifically, the optimization objective is weighted using the entropy method. , , , , The specific steps for assigning weights are as follows:

[0128] S31, Use the dataset from step S2 as the sample dataset for the entropy method;

[0129] Specifically, the dataset in Table 1 will be used as the sample dataset for the entropy method;

[0130] S32, using formula (1) to target the positive direction , , Data standardization is performed, and equation (2) is used for negative targets. , Data standardization is performed to obtain a standardized sample dataset that has eliminated differences in the units of measurement for each indicator.

[0131] (1)

[0132] (2)

[0133] In the formula, ( (Refers to the first) The first positive goal The standard value of this experiment;

[0134] ( (Refers to the first) The first negative target The standard value for this experiment;

[0135] S33, In order to avoid some data having low values ​​after standardization, equations (3) and (4) are used to perform translation processing on the positive and negative targets respectively;

[0136] (3)

[0137] (4)

[0138] In the formula:

[0139] H represents the magnitude of the index shift, with a value of 0.01.

[0140] , ), referring to the The first goal The translation value of the experimental results

[0141] , , refers to the The first goal The translation value of the experimental results;

[0142] S34, use equations (5) and (6) to normalize the positive and negative targets respectively;

[0143] (5)

[0144] (6)

[0145] In the formula:

[0146] , refers to the The first positive goal Normalized values ​​for this experiment;

[0147] , refers to the The first negative target Normalized values ​​for this experiment;

[0148] S35, use equations (7) and (8) to calculate the information entropy of the optimized target for the positive and negative targets respectively;

[0149] (7)

[0150] (8)

[0151] In the formula:

[0152] Refers to the first Information entropy of a positive target

[0153] (Refers to the first) The information entropy of a negative target;

[0154] S36, use equations (9) and (10) to calculate the difference coefficients of the optimized target for the positive and negative targets respectively;

[0155] (9)

[0156] (10)

[0157] (Refers to the first) The difference coefficient of a positive objective

[0158] (Refers to the first) The difference coefficient of each negative target;

[0159] S37, use equations (11) and (12) to calculate the normalized entropy weights of the optimization target for the positive and negative targets respectively;

[0160] (11)

[0161] (12)

[0162] (Refers to the first) The normalized entropy weights of a positive objective.

[0163] (Refers to the first) Normalized entropy weights for each negative objective;

[0164] In this embodiment, the normalized weights of the optimization objective are calculated according to steps S31 to S37. =0.1168, = 0.0464, =0.3054, =0.2657, =0.2657.

[0165] S4. Using the Python platform and the Tensorflow toolkit, an optimized target prediction model based on a Deep Feedforward Neural Network (DFNN) is established and trained using the dataset obtained in step S2. The specific steps are as follows:

[0166] S41. Configure the environment: Import the following libraries into the Python platform: pandas for data processing, numpy for numerical computation, sk-learn for machine learning tools, tensorflow for building neural networks, and matplotlib for plotting.

[0167] S42, Read in the dataset from Table 1, specifying the features in Table 1 using input_features. , , , The columns represent the input data, and the output_targets column specifies the data in Table 1. , , , , The columns represent the output data;

[0168] S43 randomly partitions the dataset imported in S42, with 80% used as the training set and 20% as the test set. The random_state=42 instruction ensures that the partitioning results are the same every time the code is run, thus guaranteeing the reproducibility of the experiment.

[0169] S44, the optimization parameter values ​​in the dataset are standardized to a distribution with "mean 0, standard deviation 1" using the scaler_x command to eliminate the influence of dimensions, and the optimization target values ​​in the dataset are standardized to a distribution with "mean 0, standard deviation 1" using the scaler_y command to eliminate the influence of dimensions.

[0170] S45. Create a Deep Feedforward Neural Network (DFNN) using SK-Learn and Tensorflow toolkits, setting the input layer dimension to 4 and the output layer dimension to 5.

[0171] Neurons are created using the Dense(128,256,128,64,32,activation='relu') instruction, forming fully connected layers with the ReLU activation function. There are a total of five fully connected (Dense) layers. The first two layers increase the number of neurons (128-256) to enhance network balance, while subsequent layers reduce the number of neurons (128-64-32) to compress feature dimensions and avoid redundant information interference.

[0172] The Dropout(0.3) command is used to randomly drop 30% of neurons during training to prevent the model from over-relying on certain neurons and causing overfitting.

[0173] Setting the learning rate to 0.001 using `optimizer=Adam(learning_rate=0.001)` is suitable for moderately complex neural networks, balancing convergence speed and stability. The mean squared error (MSE) is set as the loss function using `loss=MeanSquaredError(name='loss')`. The mean absolute error (MAE) is set as the overfitting metric using `metrics=[MeanAbsoluteError(name='mae')]`. The model is trained 500 times using `epochs=500`. The batch size is set to 16 samples per parameter update.

[0174] S46. Input the dataset processed in step S44 into the optimized target prediction model for training. Using the validation_split=0.2 instruction, 20% of the training set is divided into a validation set to monitor whether the model is overfitting during training. At the same time, an early stopping mechanism is set. If the mean squared error (MSE) of this validation set does not improve for 50 consecutive rounds, training is stopped early to avoid invalid computation and prevent overfitting.

[0175] Save the trained model structure and weights as the courtyard_ventilation_model.keras file, save the processing functions used for standardization as the scaler_x.pkl file and the scaler_y.pkl file respectively, and save the prediction script as the predict.py file.

[0176] S5. Based on the optimized target prediction model trained in step S4 and the optimized target entropy weights obtained in step S3, the NSGA-II algorithm and Pareto algorithm are used to generate the ventilation opening design scheme. The specific steps are as follows:

[0177] S51, Set the optimization variable range as follows: and The value range is set to [1.6m, 6.3m]. and The value range is set to [0.25m, 0.95m]. The above data range is standardized using the scaler_x command to make it consistent with the format used when training the model.

[0178] S52, using the pymoo.core.problem.ElementwiseProblem code to configure multi-objective optimization problems, , , Defined as a positive optimization objective, , Defined as a negative optimization objective; , , , Defined as the input optimization variable; the courtyard_ventilation_model.keras file generated in step S46 is used as the prediction tool for the optimization objective; the entropy weights of the input optimization objective are as follows: =0.1168, =0.0464, =0.3054, =0.2657, =0.2657, the weight vector is named (target_weights), and the optimization target weights for the multi-objective optimization problem are set by the directive optim_problem.target_weights = target_weights;

[0179] S53, run the algorithm = NSGA2 instruction, set the population size to 500, iterate 120 times, set the crossover rate to 90%, the mutation rate to 20%, and set the number of Pareto non-dominated solution sets to 20;

[0180] S54, calculate the overall priority weight of the scheme by using the Pareto non-dominated solution set according to equation (13).

[0181] (13)

[0182] In the formula:

[0183] ( ) is the comprehensive priority weight value of the j-th solution in the Pareto non-dominated solution set;

[0184] ( The value is the standardized value of the i-th positive objective value of the j-th solution in the Pareto non-dominated solution set, obtained by applying scaler_y.

[0185] ( The value is the standardized value of the i-th negative objective value of the j-th solution in the Pareto non-dominated solution set using scaler_y.

[0186] S55 uses the scheme corresponding to the maximum comprehensive priority weight value in the Pareto non-dominated solution set as the multi-objective optimization design scheme for year-round natural ventilation in the scenario of adding a roof to the courtyard of a residential building. The following solutions are obtained: the width of the east-side ventilation opening C1 is 1.6m, the height of the east-side ventilation opening C2 is 0.95m, the width of the west-side ventilation opening C3 is 1.6m, and the height of the west-side ventilation opening C4 is 0.25m. The design schemes are shown in Figures 5-1 to 5-3. The corresponding natural ventilation optimization objective values ​​are: spring ventilation optimization objective... The optimization target for summer ventilation conditions is 1.028 m / s. The target for autumn ventilation optimization is 1.478 m / s. The optimization target for summer wind shelter conditions is 0.981 m / s. The target for winter wind protection is 0.201 m / s. It is 0.144 m / s;

[0187] This example demonstrates the effect analysis:

[0188] 1) Comparative analysis of the running time of the technical solution of this invention and conventional technical solutions

[0189] This paper implements multi-objective optimization of natural ventilation in this example using the Butterfly, Ladybug, and Wallecei plugins on the Grasshopper platform in Rhino. As a comparative technique, this is referred to as the conventional technique. Partial code of the conventional technique's Grasshopper implementation is shown in Figure 6. The performance of the conventional technique's multi-objective optimization programming on the Grasshopper platform is shown in Figure 7. Except for using average weight optimization for the optimization objectives, the other settings of the conventional technique are the same as in the implementation case of this invention. Running the conventional technique on a computer with an i5-12400 processor and 32GB of RAM, it took 11 hours and 1 minute. Using the technique of this invention, running on the same computer, completing model training and multi-objective optimization took 51 minutes and 16 seconds. Compared to the conventional technique, the technique of this invention reduces the running time by approximately 92%. Therefore, this invention provides a time-cost-acceptable technical solution for this type of multi-objective optimization, offering a more cost-effective solution for multi-objective optimization research and practice in buildings.

[0190] 2) Impact analysis of optimizing target weights

[0191] To explore the impact of optimization objective weights on the multi-objective optimization design scheme for year-round natural ventilation, the optimization objective weights were averaged and the technical solution of this invention was used to perform multi-objective optimization for year-round natural ventilation in this example. The resulting opening schemes are shown in Figures 8-1 and 8-2: the width of the east ventilation opening of C1 is 6.29m, the height of the east ventilation opening of C2 is 0.95m, the width of the west ventilation opening of C3 is 1.62m, and the height of the west ventilation opening of C4 is 0.25m; the optimization objective for spring ventilation... The optimization target for summer ventilation conditions is 1.029 m / s. The target for autumn ventilation is 1.69 m / s. The optimization target for summer wind shelter conditions is 0.953 m / s. The target for winter wind protection is 0.178 m / s. The value is 0.191 m / s. The Pareto non-dominated solution set and optimal solution using the present invention and optimized objective entropy weight are shown in Figure 9, and the Pareto non-dominated solution set and optimal solution using the present invention and optimized objective average weight are shown in Figure 10.

[0192] It is evident from both the optimization results and the optimization process that the optimization target weight allocation scheme has a significant impact on the ventilation design in this case.

[0193] 3) This invention provides a flexible optimization target weight configuration scheme.

[0194] Conventional technical solutions default to assigning average weights to the optimization target, without offering diverse configuration options. In this invention, the entropy weight method is used, which is an objective weighting method based on statistical data. However, in practical applications, subjective weights are a crucial factor. For example, in cold regions, the importance of shelter from the wind in winter outweighs ventilation in summer and transitional seasons for residential buildings; conversely, in hot-summer-cold-winter regions, ventilation during the transitional season is more important than shelter from the wind in winter. When the subjective weights of the optimization target have a clear bias, this technical solution makes appropriate adjustments, omitting the entropy weight assignment step and directly assigning subjective weights to the optimization target. This allows for flexible configuration of the optimization target weights and integrates these subjective weights into the Pato non-dominated solution set comprehensive sorting, resulting in a more optimized opening solution.

[0195] Example 2

[0196] This renovation plan proposes to install a skylight at the top of the courtyard of the building in the case study. The courtyard roof will be constructed of 1mm thick structural steel on the top and bottom, with a 50mm thick rock wool board sandwiched in between. The thermal conductivity of the steel is 58.2. The thermal conductivity of rock wool board is 0.044. The thermal resistance of the opaque portion of the top cover is 1.136. The inner surface reflectivity is 0.5. The skylight uses a double-glazed construction with a heat transfer coefficient of 2.8. The visible light transmittance is 0.8, and the solar heat gain coefficient (SHGC) is 0.4. The length and width of the skylight are important factors affecting the natural lighting and cooling / heating energy consumption of the building. This invention provides an optimized design method for the above-mentioned skylight lighting and cooling / heating energy consumption, achieving a balance between cooling / heating energy consumption and lighting optimization goals, and seeking the optimal skylight design scheme. The preliminary skylight design scheme of this case is shown in Figure 2-7. The two skylight sloping roof sections are divided into five equal parts in the north-south direction and four equal parts in the east-west direction. The center of the rectangle formed by their intersection is the center of the skylight, which is also the center of the openable window area. The length and width of each skylight are equal. The shaded area is the optimized skylight design area. The minimum window size (opening size is the length in the east-west direction × the length in the north-south direction) is 160mm × 330mm, and the maximum window size is 850mm × 1150mm. This invention seeks the optimal window opening scheme between the minimum and maximum window opening. The window opening scheme is represented by the following optimization variable: the length of the east-west side of the skylight. The length of the north-south side of the skylight The range between the minimum and maximum window opening is the optimal range for the skylight.

[0197] This embodiment provides a skylight design method for adding a roof over the courtyard of a residential building. Referring to the schematic diagram of the technical solution of the present invention in Figure 1, the specific steps of this embodiment are as follows:

[0198] S1, Set lighting optimization goals Annual cooling energy consumption optimization target Annual thermal energy consumption optimization target Glare optimization target ;

[0199] The optimization indicators set in this embodiment The annual average effective daylight intensity (UDI) is measured at 0.8m intervals on a horizontal plane 0.8m above the ground in the courtyard. 300～3000lux Useful Daylight Illuminance (unit: %) The annual cooling energy consumption per unit area within the courtyard is expressed in kWh / m². 2 , The annual heating energy consumption per unit area within the courtyard is expressed in kWh / m². 2 , The average annual glare probability (DGP) is measured at 0.8m intervals on a horizontal plane 0.8m above the ground in the courtyard.

[0200] Natural lighting optimization goals The objective is to maximize the value, so the calculation method is to use the Grasshopper platform and the Ladybug plugin to obtain meteorological parameters of Hangzhou City, and set the effective light illuminance range of the measuring point to be 300 lux to 3000 lux.

[0201] Annual cooling energy consumption optimization target The negative objective is that the smaller the value, the better. The calculation method is to use the Grasshopper platform and the ladybug plugin to obtain meteorological parameters of Hangzhou. The air conditioner is set to turn on when the room temperature exceeds 24°C.

[0202] Annual thermal energy consumption optimization target The negative objective is that the smaller the value, the better. The calculation method is to use the grasshopper platform and the ladybug plugin to obtain meteorological parameters of Hangzhou. The air conditioner is set to turn on when the room temperature is below 17°C.

[0203] Annual glare optimization target The negative objective is that the smaller the value, the better. The calculation method is to use the Grasshopper platform and the Ladybug plugin to obtain meteorological parameters of Hangzhou City. The surface reflectivity of the interior walls is set to 0.7 and the surface reflectivity of the top of the courtyard is set to 0.5.

[0204] Set the following two optimization variables: the length of the east-west side of the sunroof. The length of the north-south side of the skylight All units are meters (m). The value range is set to [0.16m, 0.85m]. The value range is set to [0.33m, 1.15m].

[0205] S2, Perform orthogonal experiments on the optimization objectives in step S1 respectively, and use the ventilation plugin of the Grasshopper platform to calculate and analyze the optimization objectives and construct a dataset;

[0206] In this embodiment S21, the number of factors in the orthogonal experiment is set to 2, and the number of levels for each factor is set to 23. L is selected. 529 (twenty three 2 The orthogonal experimental setup was used, with a total of 529 experiments for each optimization objective. A total of 2116 experiments were conducted for each of the four optimization objectives, resulting in the dataset shown in Table 2. This setup includes the following sub-steps:

[0207] S22, determined by evenly distributing the number of levels and the range of values ​​for factors. , The values ​​of each experimental sample;

[0208] S23, Implement parametric geometric and physical modeling of the case building using programming on the Grasshopper platform;

[0209] S24. Using the HoneyBee plugin on the Grasshopper platform, the optimization objective for all samples is calculated according to an orthogonal experiment. , , The programming portion is shown in Figure 11;

[0210] S25, The completed dataset is shown in Table 2, and the optimization objective values ​​in the dataset are shown in Figure 12.

[0211] Table 2. Dataset

[0212] Sample number surface

[0213] Variable descriptions in the table:

[0214] ( ), referring to the first The first optimization variable is the... The values ​​for this experiment are determined by the orthogonal experimental setup and the range of values ​​for the optimized variables.

[0215] ( (Refers to the first) The first positive goal The calculated values ​​for this experiment;

[0216] ( (Refers to the first) The first negative target The calculated values ​​for this experiment.

[0217] S3, Optimization objective based on entropy method , , , The process of assigning weights includes the following sub-steps:

[0218] S31, Use the dataset in Table 2 as the sample dataset for the entropy method;

[0219] S32, Using Equation (1) for the positive target Data standardization is performed, and equation (2) is used for negative targets. , , Data standardization is performed to obtain a standardized sample dataset that has eliminated differences in the units of measurement for each indicator.

[0220] (1)

[0221] (2)

[0222] In the formula, ( (Refers to the first) The first positive goal The standard value of this experiment;

[0223] ( (Refers to the first) The first negative target The standard value for this experiment;

[0224] S33, In order to avoid some data having low values ​​after standardization, equations (3) and (4) are used to perform translation processing on the positive and negative targets respectively;

[0225] (3)

[0226] (4)

[0227] In the formula:

[0228] H represents the magnitude of the index shift, with a value of 0.01.

[0229] , ), referring to the The first positive goal The translation value of the experimental results

[0230] , , refers to the The first negative target The translation value of the experimental results;

[0231] S34, use equations (5) and (6) to normalize the positive and negative targets respectively;

[0232] (5)

[0233] (6)

[0234] In the formula:

[0235] , refers to the The first positive goal Normalized values ​​for this experiment;

[0236] , refers to the The first negative target Normalized values ​​for this experiment;

[0237] S35 uses equations (7) and (8) to calculate the information entropy of the optimized target for both the positive and negative targets, respectively;

[0238] (7)

[0239] (8)

[0240] In the formula:

[0241] (Refers to the first) Information entropy of a positive target

[0242] (Refers to the first) The information entropy of a negative target;

[0243] S36, use equations (9) and (10) to calculate the difference coefficients of the optimized target for the positive and negative targets respectively;

[0244] (9)

[0245] (10)

[0246] (Refers to the first) The difference coefficient of a positive objective

[0247] (Refers to the first) The difference coefficient of each negative target;

[0248] S37, use equations (11) and (12) to calculate the normalized entropy weights of the optimization target for the positive and negative targets respectively;

[0249] (11)

[0250] (12)

[0251] (Refers to the first) The normalized entropy weights of a positive objective.

[0252] ,3) refers to the first Normalized entropy weights for each negative objective;

[0253] In this embodiment, the normalized weights of the optimization objective are calculated according to steps S31 to S37. =0.297, = 0.227, =0.136, =0.340.

[0254] S4, using the Python platform and the Tensorflow toolkit, and the dataset constructed in S2, trains and optimizes the target prediction model based on a deep feedforward neural network (DFNN), including the following sub-steps.

[0255] S41, Configure the environment: Import the following libraries into the Python platform: pandas for data processing, numpy for numerical computation, sk-learn for machine learning tools, tensorflow for building neural networks, and matplotlib for plotting.

[0256] S42, Read in the dataset from Table 1, specifying the features in Table 1 using input_features. , The columns represent the input data, and the output_targets column specifies the data in Table 1. , , , The columns represent the output data;

[0257] S43, randomly splits the dataset imported in S42, with 80% used as the training set and 20% as the test set. The random_state=42 instruction ensures that the splitting results are the same every time the code is run, thus guaranteeing the reproducibility of the experiment.

[0258] S44, the optimization parameter values ​​in the dataset are standardized to a distribution with "mean 0, standard deviation 1" using the scaler_x command to eliminate the influence of dimensions, and the optimization target values ​​in the dataset are standardized to a distribution with "mean 0, standard deviation 1" using the scaler_y command to eliminate the influence of dimensions.

[0259] S45, create a Deep Feedforward Neural Network (DFNN) using SK-Learn and Tensorflow toolkits, setting the input layer dimension to 4 and the output layer dimension to 5;

[0260] Neurons are created using the Dense(128,256,128,64,32,activation='relu') instruction, forming fully connected layers with the ReLU activation function. There are a total of five fully connected (Dense) layers. The first two layers increase the number of neurons (128-256) to enhance network balance, while subsequent layers reduce the number of neurons (128-64-32) to compress feature dimensions and avoid redundant information interference.

[0261] The Dropout(0.3) command is used to randomly drop 30% of neurons during training to prevent the model from over-relying on certain neurons and causing overfitting.

[0262] S46 sets the learning rate to 0.001 using `optimizer=Adam(learning_rate=0.001)`, which is suitable for moderately complex neural networks, balancing convergence speed and stability. It sets the mean squared error (MSE) as the loss function using `loss=MeanSquaredError(name='loss')`, sets the mean absolute error (MAE) as the overfitting metric using `metrics=[MeanAbsoluteError(name='mae')]`, sets the training epochs to 500, and sets the batch size to 16 samples per parameter update.

[0263] S47 uses the validation_split=0.2 command to further divide the training set into 20% as a validation set, which is used to monitor whether the model is overfitting during training. At the same time, an early stopping mechanism is set. If the mean squared error (MSE) of this validation set does not improve for 50 consecutive rounds, training will be stopped early to avoid invalid calculations and prevent overfitting.

[0264] S48. Save the trained model structure and weights as the courtyard_energy_light_model.keras file, save the processing functions used for standardization as the scaler_X_energy_light.pkl file and the scaler_y_energy_light.pkl file respectively, and save the prediction script as the predict_energy_light.py file.

[0265] S5. Based on the completed optimized target prediction model and optimized target entropy weights, the sunroof design scheme is generated using the NSGA-II algorithm and the Pareto algorithm, including the following sub-steps.

[0266] S51, Set the optimization variable range as follows: The value range is set to [0.16m, 0.85m]. The value range is set to [0.33m, 1.15m]. The above data range is standardized using the scaler_X_energy_light.pkl command to make it consistent with the format used when training the model.

[0267] S52, using the pymoo.core.problem.ElementwiseProblem code to configure multi-objective optimization problems, Defined as a positive optimization objective, , , Defined as a negative optimization objective; , Defined as the input optimization variable; the courtyard_energy_light_model.keras file generated in step S49 is used as the prediction tool for the optimization objective; the entropy weights of the input optimization objective are as follows: =0.297, =0.227, =0.136, =0.34, the weight vector is named (target_weights), and the optimization target weights for the multi-objective optimization problem are set by the command optim_problem.target_weights = target_weights;

[0268] S53, run the algorithm = NSGA2 command, set the population size to 500, iterate 120 times, set the crossover rate to 80%, the mutation rate to 20%, and set the number of solutions in the Pareto non-dominated solution set to 20;

[0269] S54, calculate the overall priority weight of the solutions in the Pareto non-dominated solution set according to equation (13).

[0270] (13)

[0271] In the formula:

[0272] ( ) is the comprehensive priority weight value of the j-th solution in the Pareto non-dominated solution set;

[0273] ( The value is the standardized value of the i-th positive objective value of the j-th solution in the Pareto non-dominated solution set, obtained by applying scaler_y.

[0274] ( The value is the standardized value of the i-th negative objective value of the j-th solution in the Pareto non-dominated solution set using scaler_y.

[0275] S55, the scheme corresponding to the maximum comprehensive priority weight value in the Pareto non-dominated solution set is used as the multi-objective optimization skylight design scheme for light environment and building energy consumption in the scenario of adding a roof to the roof of a residential courtyard, resulting in the roof skylight design scheme of this embodiment: the side length of the skylight in the east-west direction. The length is 0.59m, and the side length of the skylight in the north-south direction is... The height is 0.16 meters. The skylight design scheme is shown in Figure 13; the corresponding optimization target value is: annual natural light. The annual cooling energy consumption is 68.842%. The value is: 55.565 kWh / m³ 2 Annual heat energy consumption It is 87.845 kWh / m³ 2 Annual glare probability It is 0.317.

[0276] This example demonstrates the effect analysis:

[0277] 1) Comparative analysis of the running time of the technical solution of this invention and conventional technical solutions

[0278] This paper describes a multi-objective optimization method for year-round lighting environment and building energy consumption in Rhino using the Ladybug and Wallecei plugins on the Grasshopper platform. This method serves as a comparative technique, referred to as the conventional approach. Partial code of the conventional approach's Grasshopper implementation is shown in Figure 14. The performance of the conventional approach's multi-objective optimization programming on the Grasshopper platform is shown in Figure 15. Except for using average weight optimization for the optimization objectives, the other settings of the conventional approach are identical to those in this invention's implementation. Running the conventional approach on a computer with an i5-12400 processor and 32GB of RAM took 4 hours and 53 minutes. Using the present invention's approach on the same computer, model training and multi-objective optimization were completed in 42 minutes and 16 seconds. Compared to the conventional approach, the present invention reduces runtime by approximately 85%. Therefore, this invention provides a time-cost-acceptable solution for such multi-objective optimization implementations, offering a more cost-effective solution for multi-objective optimization research and practice in buildings.

[0279] 2) Impact analysis of optimizing target weights

[0280] To explore the impact of optimization objective weights on the multi-objective optimization of skylight design for light environment and building energy consumption, the optimization objective weights were averaged and the technical solution of this invention was used to perform multi-objective optimization of light environment and building energy consumption throughout the year in this example. The resulting skylight design is shown in Figure 16: the side length of the skylight in the east-west direction. The length of the skylight's north-south side is 0.16m. The value is 1.01m. The corresponding optimization target value is: annual natural daylight. The annual cooling energy consumption is 67.839%. It is: 55.538 kWh / m³2 Annual heat energy consumption It is 88.047 kWh / m³ 2 Annual glare probability It is 0.335.

[0281] Figure 17 shows the Pareto solution set and optimal solution using the present invention and optimized target entropy weight, and Figure 18 shows the Pareto solution set and optimal solution using the present invention and optimized target average weight.

[0282] It is evident from both the optimization results and the optimization process that the optimization target weight allocation scheme has an impact on the sunroof design in this case.

[0283] 3) This invention provides a flexible optimization target weight configuration scheme.

[0284] Conventional technical solutions default to using average weights for the optimization target, without offering diverse configuration options. In this invention, the optimization target entropy weight method is employed, a method based on objective statistical data. However, in practical applications, subjective weights are a crucial factor that cannot be ignored. For example, when the atrium space serves as a display area for artworks or merchandise, the subjective weight of the atrium space's glare index may be greater. When the subjective weight of the optimization target exhibits a clear subjective bias, this technical solution allows for flexible adjustment: by omitting the process of assigning entropy weights to the optimization target and directly allocating subjective weights to the core target, the optimization target weights can accurately adapt to actual needs. This subjective weight is then incorporated into the comprehensive ranking of the Pareto non-dominated solution set, resulting in a more practical atrium roof and skylight design.

Claims

1. A design method for ventilation openings or skylights in the scenario of roof renovation and addition of residential courtyards. The specific steps are as follows: S1, set annual natural ventilation volume optimization targets or light environment and building energy consumption optimization targets according to the seasons. The optimization targets include positive and negative targets, and optimization variables and their value ranges are set. S2, perform orthogonal experiments on the optimization targets in step S1 respectively, calculate the optimization targets using the ventilation plugin of the Grasshopper platform, and construct a dataset. S3, assign weights to the optimization targets based on the entropy method to obtain the optimization target entropy weights. S4, establish an optimization target prediction model based on a deep feedforward neural network and train it using the dataset obtained in step S2. S5, based on the optimization target prediction model trained in step S4 and the optimization target entropy weights obtained in step S3, generate ventilation opening design schemes or skylight design schemes using the NSGA-II algorithm and the Pareto algorithm.

2. The method for designing ventilation openings or skylights in the scenario of adding a roof to the courtyard of a residential building, as described in claim 1, is characterized in that: In step S1, the total number of objectives to be optimized is f, where a is the number of positive objectives and b is the number of negative objectives, and f = a + b. 、 、 Indicates a positive goal, using 、 、 This indicates a negative objective. A positive objective means the larger the target value, the better, while a negative objective means the smaller the target value, the better. Set optimization variables 、 、 The number of optimization variables is d. The optimization variables refer to the design parameters of ventilation openings that affect natural ventilation or the design parameters of skylights that affect the light environment and building energy consumption.

3. The method for designing ventilation openings or skylights in the scenario of adding a roof to the courtyard of a residential building, as described in claim 2, is characterized in that: The specific steps for constructing the dataset in step S2 are as follows: S21, set the number of factors in the orthogonal experiment to be... The number of levels for each factor is set to h, according to and Select an orthogonal experimental table and determine the number of experiments based on the orthogonal experimental table. S22, Determine the values ​​of each factor and each experimental sample based on the uniform distribution of the number of levels and the range of values ​​of the factors; S23, Implement parametric geometric modeling and physical modeling of the building by programming on the Grasshopper platform; S24, Calculate the optimization target values ​​according to the orthogonal experiment using the corresponding plugins on the Grasshopper platform. S25, The completed dataset, which includes: （ ), referring to the first The first optimization variable is the... The values ​​for this experiment are determined by the orthogonal experimental setup. （ (Refers to the first) The first positive goal The calculated values ​​for this experiment; （ (Refers to the first) The first negative target The calculated values ​​for this experiment.

4. The method for designing ventilation openings or skylights in the scenario of adding a roof to the courtyard of a residential building, as described in claim 3, is characterized in that: The specific steps of assigning weights to all optimization indicators based on the entropy method in step S3 are as follows: S31, use the dataset in step S2 as the sample dataset for the entropy method; S32, use equation (1) to weight the positive target 、 、 Data standardization is performed, and equation (2) is used for negative targets. 、 、 、 Data standardization is performed to obtain a standardized dataset that eliminates differences in measurement units; (1) (2) In the formula, （ (Refers to the first) The first positive goal The standard value of this experiment; （ (Refers to the first) The first negative target The standard value of this experiment; S33, in order to avoid some data having low values ​​after standardization, the positive and negative targets are translated using equations (3) and (4) respectively; （3） (4) In the formula, H is the magnitude of the index shift, which is 0.01; （ (Refers to the first) The first positive goal The translation value of this experiment; （ (Refers to the first) The first negative target The translation value of this experiment; S34, normalization is performed on the positive and negative targets using equations (5) and (6) respectively; （5） (6) Where: （ (Refers to the first) The first positive goal Normalized values ​​for this experiment; （ (Refers to the first) The first negative target The normalized value of the experiment; S35, use equations (7) and (8) to calculate the information entropy of the optimized target for the positive and negative targets respectively; （7） (8) Where: (Refers to the first) Information entropy of a positive target (Refers to the first) Information entropy of negative targets; S36, use equations (9) and (10) to calculate the difference coefficient of the optimized target for positive and negative targets respectively; （9） （10） (Refers to the first) The difference coefficient of a positive objective (Refers to the first) The difference coefficients of the negative targets; S37, the normalized entropy weights of the optimization targets are calculated for the positive and negative targets respectively using equations (11) and (12); （11） （12） (Refers to the first) The normalized entropy weights of a positive objective. (Refers to the first) Normalized entropy weights for each negative objective.

5. The method for designing ventilation openings or skylights in the scenario of adding a roof to the courtyard of a residential building, as described in claim 1, is characterized in that: In step S4, the establishment and training of the target prediction model were performed using the Python platform and the Tensorflow toolkit.

6. The method for designing ventilation openings or skylights in the scenario of adding a roof to the courtyard of a residential building, as described in claim 5, is characterized in that: The specific steps for establishing and training the optimization target prediction model in step S4 are as follows: S41, Configure the environment: Import the relevant libraries on the Python platform; S42, Read and import the dataset from step S2, and... 、 、 、 The column is the input data, 、 、 、 、 、 The output data is listed below; S43, the dataset is randomly divided, with 80% used as the training set and 20% as the test set; S44, the optimization parameter values ​​in the dataset are standardized to a distribution with a mean of 0 and a standard deviation of 1, and the optimization target values ​​in the dataset are also standardized to a distribution with a mean of 0 and a standard deviation of 1; S45, an optimization target prediction model based on a deep feedforward neural network is created, with the input layer dimension set to d and the output layer dimension set to f; neurons are created, and fully connected layers using the ReLU activation function are used, for a total of five fully connected layers. The first two layers increase the number of neurons to expand the network's balancing ability, and the subsequent layers reduce the number of neurons to compress the feature dimension; Set the learning rate to 0.001; set the mean squared error as the loss function; set the mean absolute error as the overfitting evaluation index; set the training epochs of the model; set the number of samples used for each parameter update; S46, input the dataset processed in step S44 into the optimized target prediction model for training, and divide 20% of the training set as the validation set, while setting an early stopping mechanism.

7. The method for designing ventilation openings or skylights in the scenario of adding a roof to the courtyard of a residential building, as described in claim 6, is characterized in that: The libraries in step S41 include: pandas for data processing, numpy for numerical computation, sklearn for machine learning tools, tensorflow for building neural networks, and matplotlib for plotting.

8. The method for designing ventilation openings or skylights in the scenario of adding a roof to the courtyard of a residential building, as described in claim 6, is characterized in that: Step S45 also includes setting up the random discarding of 30% of neurons during training.

9. The method for designing ventilation openings or skylights in the scenario of adding a roof to the courtyard of a residential building, as described in claim 1, is characterized in that: The specific steps for generating the ventilation outlet design scheme in step S5 are as follows: S51, set the range of optimization variables and standardize the range of optimization variables to make it consistent with the format used when training the model; S52, configure the multi-objective optimization problem, and... 、 、 、 Defined as a positive goal, 、 、 、 Defined as a negative target, 、 、 、 Define the optimization variables; call the optimization objective prediction model in step S4; input the entropy weights of the optimization objective. and ), set the optimization objective weights for the multi-objective optimization problem; S53, set the population size, number of iterations, crossover rate, and mutation rate, and set the number of schemes in the Pareto non-dominated solution set to q; S54, calculate the comprehensive priority weight of the schemes in the Pareto non-dominated solution set according to equation (13). In equation (13), （ ) is the comprehensive priority weight value of the j-th solution in the Pareto non-dominated solution set; ( The value is the standardized value of the i-th positive objective value of the j-th solution in the Pareto non-dominated solution set, obtained by using scaler_y. ( ) is the value after standardizing the i-th negative objective value of the j-th scheme in the Pareto non-dominated solution set using scaler_y; S55, the scheme corresponding to the maximum comprehensive priority weight value in the Pareto non-dominated solution set is taken as the multi-objective optimization ventilation opening design scheme for year-round natural ventilation in the scenario of adding a roof to the courtyard of a residential building, or the multi-objective optimization skylight design scheme for lighting and energy consumption in the scenario of adding a roof to the courtyard of a residential building.

Images