GAN and RF-NSGA integration algorithm-based gymnasium wind-solar-energy collaborative optimization design method

Abstract

The invention relates to the technical field of wind, light and storage collaboration, in particular to a gymnasium wind, light and energy collaborative optimization design method based on a GAN and RF-NSGA integration algorithm. The method comprises the following steps that skylight design parameters are determined, wherein the skylight design parameters comprise the area ratio of skylights to roofs, the number of the skylights, the length-width ratio of the skylights and the change rate of the skylights; based on the skylight design parameters, building performance simulation data are obtained through building performance simulation, and the building performance simulation data comprise the effective sunlight utilization rate, the thermal comfort degree percentage and the photovoltaic generating capacity; and carrying out generative data enhancement on the building performance simulation data to generate an expansion performance data set. According to the invention, the problem of tradeoff and tradeoff between lighting and power generation caused by traditional single-target optimization can be fundamentally solved, and roof resource utilization rate maximization and building operation carbon emission minimization are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of wind-solar-storage coordination technology, and particularly relates to a gymnasium wind-solar-energy coordination optimization design method based on a GAN and RF-NSGA integrated algorithm. BACKGROUND

[0002] As an important part of the gymnasium building, the design of the roof not only affects natural lighting, thermal comfort and ventilation performance, but also has great potential for integrating building photovoltaic systems (BIPV). By reasonably designing the roof skylight system, the indoor light environment quality and natural ventilation performance can be effectively improved, and the energy consumption of air conditioning and lighting can be reduced; at the same time, arranging the BIPV system in the remaining area of the roof helps to realize the on-site power generation of renewable energy and reduce the carbon emissions of building operation.

[0003] However, there is a clear spatial conflict and performance coupling relationship between the skylight design and the arrangement of the BIPV system.

[0004] On the one hand, the increase in the number of skylights and the decentralization of the layout help to improve the uniformity of lighting and thermal comfort, but they occupy the effective area of the roof, reduce the arrangement density of photovoltaic panels, and cause the power generation to decrease; on the other hand, the efficient arrangement of the BIPV system requires the roof to have a large continuous available area, and it is extremely sensitive to shading, and the existence of skylights may cause the skylight diffusion effect (SDE) between photovoltaic arrays, further reducing the power generation efficiency of the system.

[0005] At present, the research on the gymnasium roof system mainly focuses on the optimization of a single performance target. For example, some research focuses on improving lighting and thermal comfort through skylight optimization, and others focus on the power generation efficiency and arrangement strategy of the roof photovoltaic system. However, few studies include lighting, thermal comfort and photovoltaic power generation in a unified multi-objective optimization framework, and lack a systematic parameter control and performance trade-off mechanism. SUMMARY

[0006] Therefore, it is necessary to provide a gymnasium wind-solar-energy coordination optimization design method based on a GAN and RF-NSGA integrated algorithm to solve at least one of the above technical problems.

[0007] To achieve the above-mentioned purpose, a gymnasium wind-solar-energy coordination optimization design method based on a GAN and RF-NSGA integrated algorithm includes the following steps: Step S1: determining the skylight design parameters, the skylight design parameters including the skylight-to-roof area ratio, the number of skylights, the skylight length-width ratio and the skylight change rate; Step S2: Based on the skylight design parameters, building performance simulation data is obtained through building performance simulation, including effective daylight utilization, thermal comfort percentage and photovoltaic power generation; Step S3: The building performance simulation data is generated by generative data augmentation, and an expanded performance data set is generated; Step S4: Based on the expanded performance data set, a prediction relationship between the skylight design parameters and the effective daylight utilization, thermal comfort percentage and photovoltaic power generation is established through regression analysis; Step S5: Based on the prediction relationship, the effective daylight utilization, thermal comfort percentage and photovoltaic power generation are optimized by a multi-objective optimization method, and a first Pareto optimal solution set is generated; Step S6: The optimal skylight design parameter range is determined based on the first Pareto optimal solution set.

[0008] The beneficial effects of the present application are: On the one hand, GAN is used to generate a high-fidelity expanded data set by high-dimensional interpolation of a small number of real simulation samples in the latent space, to completely cover the skylight-photovoltaic coupling performance surface in a data-driven manner, significantly relieve the simulation resource pressure caused by the roof scheme combination explosion, provide sufficient and diverse training samples for subsequent agent modeling, and thus improve the modeling accuracy and robustness.

[0009] On the other hand, random forest regression is used to quickly fit and quantify the feature importance of high-dimensional nonlinear mapping, and an interpretable agent model is established between the skylight geometric parameters and the daylight utilization, thermal comfort and photovoltaic power generation, which provides parameter sensitivity information while ensuring prediction accuracy, and provides a stable, differentiable and low-computing-cost evaluation engine for multi-objective optimization.

[0010] On the other hand, RF-NSGA dynamically adjusts the search pressure in the target space through a reference point mechanism, balances the convergence and uniformity of the Pareto front, and outputs a balanced solution set of high daylighting, high thermal comfort and high power generation at one time, so that designers can directly lock the optimal balance area under the conflict between skylight and photovoltaic panels without repeated trial calculations, fundamentally solving the problem of trade-off between daylighting and power generation caused by traditional single-objective optimization, and realizing the comprehensive benefits of maximizing roof resource utilization, minimizing building operation carbon emissions and shortening design cycle. BRIEF DESCRIPTION OF DRAWINGS

[0011] Other features, objects and advantages of the present application will become more apparent through reading the detailed description made with reference to the following drawings: Fig. 1 A step flowchart of a gym wind and light energy collaborative optimization design method based on the integrated algorithm of GAN and RF-NSGA of an embodiment is shown.

[0012] Fig. 2 A detailed step flow diagram of step S6 of an embodiment is shown.

[0013] Fig. 3 A stadium roof diagram of an embodiment is shown. DETAILED DESCRIPTION

[0014] The technical method of the present application will be described clearly and completely below in conjunction with the drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0015] In addition, the drawings are only schematic illustrations of the present application and are not necessarily drawn to scale. Identical reference numerals in the drawings represent identical or similar parts, and thus repeated descriptions thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities, which do not necessarily have to correspond to physically or logically independent entities. The functional entities can be implemented in the form of software, or in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0016] It should be understood that although the terms "first", "second" and the like can be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, without departing from the scope of the example embodiments, a first element can be referred to as a second element, and similarly a second element can be referred to as a first element. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0017] To achieve the above-mentioned purpose, please refer to Figs. 1 to 3 The present application provides a stadium view and wind energy collaborative optimization design method based on GAN and RF-NSGA integrated algorithm, comprising the following steps: Step S1: determining the skylight design parameters, the skylight design parameters including the skylight to roof area ratio, the number of skylights, the skylight length-width ratio and the skylight change rate; Step S2: based on the skylight design parameters, obtaining building performance simulation data through building performance simulation, the building performance simulation data including effective daylight utilization rate, thermal comfort percentage and photovoltaic power generation capacity; Step S3: generating augmented data set by generating data augmentation on building performance simulation data; Step S4: Based on the extended performance dataset, a predictive relationship between the skylight design parameters and the effective daylight utilization rate, the percentage of thermal comfort, and the photovoltaic power generation amount is established by regression analysis; Step S5: Based on the predictive relationship, the effective daylight utilization rate, the percentage of thermal comfort, and the photovoltaic power generation amount are synergistically optimized by a multi-objective optimization method to generate a first Pareto optimal solution set; Step S6: The optimal skylight design parameter range is determined based on the first Pareto optimal solution set.

[0018] It should be noted that in step S2, the effective daylight utilization rate is calculated by daylight simulation, the percentage of thermal comfort is calculated by thermal comfort simulation, and the photovoltaic power generation amount is calculated by photovoltaic power generation simulation.

[0019] The specific operation is: based on the skylight design parameters (area ratio, number, length-width ratio, and change rate) and the architectural features (orientation, roof structure, and use mode) of the target gymnasium, a three-dimensional model of the gymnasium is generated by building information modeling (BIM) software; the specific physical parameters are set by using Energy Plus or Design Builder software in conjunction with the BIM model, including solar radiation, optical properties of glass and frame materials, indoor thermal load, etc., thereby providing basic data for building performance simulation. Daylight simulation calculates the illuminance distribution on the working surface of each room throughout the year, and calculates the area ratio that meets the illuminance standard (e.g. 300 lux), to obtain the effective daylight utilization rate; thermal comfort simulation is based on the ISO 7730 standard, combined with indoor temperature, humidity, air flow speed, etc. to calculate the predicted mean vote (PMV) value, and to calculate the time ratio that meets the thermal comfort standard (PMV between -0.5 and +0.5); photovoltaic power generation simulation combines the installation angle of the photovoltaic panel, the radiation amount of the inclined surface, and the system efficiency to calculate the total annual power generation.

[0020] Especially important is that before step S1, it also includes: Collecting the architectural feature parameters of the target gymnasium, including the building orientation, roof form, use function, and climate zone; In the embodiments of the present application, please refer to Fig. 3 Before establishing any skylight optimization model, it is necessary to first digitize and map the target gymnasium by laser scanning in conjunction with the BIM model to obtain four categories of architectural feature parameters, including the building orientation, roof form, use function, and climate zone; the building orientation is recorded as 0° south, clockwise to 360°, the roof form is coded as four categories of double slope, single slope, dome, and flat roof, the use function is divided into four enumerated values of competition, training, performance, and comprehensive, and the climate zone directly quotes the thermal division number of GB 50178-93 standard.

[0021] In an implementation manner of the embodiment of the application, it is assumed that a target gymnasium is located at somewhere in the northern hemisphere, a BIM model of the target gymnasium shows that the building is oriented to south by 15 degrees, the roof is a double-slope structure, the main use function is competition, and the climate zone belongs to a cold region (division number 2); after on-site measurement and cross verification with drawings, the generated feature vector is [15, 1, 0, 2], wherein the 1 in the second position represents the double-slope roof coding.

[0022] The building type feature library is established based on the building feature parameters, and the target gymnasium is matched with typical building types in the feature library. In the embodiment of the application, 300 completed gymnasiums at home and abroad are pre-collected and cleaned up by a cloud server, 8-bit feature vectors are clustered by using a K-means++ algorithm, 12 typical building prototypes are obtained, the centroid vector, sample quantity and skylight design parameter statistical value of each prototype are saved, and a building type feature library is formed; when the feature vector of the target gymnasium is uploaded, the Euclidean distance between the feature vector and 12 centroids is calculated, and a type with the smallest distance and less than a set threshold value 0.15 is selected as a matching result.

[0023] In an implementation manner of the embodiment of the application, the Euclidean distance between the feature vector [15, 1, 0, 2] of the target gymnasium and the centroid [18, 1, 0, 2] of the 7th prototype is 3, which is lower than the threshold value 0.15 (the distance after Z-score normalization), and therefore the target gymnasium is classified into the 7th type; the 7th type includes 32 double-slope roof competition gymnasiums, the average value of the historical skylight and roof area ratio is 8.5%, the average value of the skylight quantity is 24, the average value of the length-width ratio is 1.6, and the average value of the skylight change rate is 1.

[0024] According to the matching result, the initial value range of the skylight design parameter is adjusted, and a skylight design parameter benchmark range suitable for a specific building type is formed. In the embodiment of the application, after the matching is completed, the parameter statistical quartile value saved by the 7th prototype is called, and the initial value interval of the skylight design parameter is automatically generated by using a mean ± 0.5 times standard deviation strategy: the area ratio range is 6%-11%, the skylight quantity range is 18-30, the length-width ratio range is 1.2-2.0, and the change rate range is 0.6-1.2; the interval is narrowed by 42% compared with the industry general specification, which not only ensures that the initial solution of the genetic algorithm falls in a high-potential area, but also avoids non-feasible solutions caused by blind search.

[0025] In one implementation manner of the embodiment of the application, if the project party proposes special requirements of the Winter Olympics event venue, the system allows secondary manual correction on the locked interval: the lower limit of the area ratio is increased by 1% and the upper limit is reduced by 1% in the XML, so as to reserve a larger continuous roof area for arranging BIPV; the new interval after correction is 7%-10%, and the remaining parameters remain unchanged, the version is marked as 7-Winter sub-class and stored in the feature library to realize knowledge self-accumulation.

[0026] Based on the building type feature library, a skylight design parameter recommendation value table of different types of gymnasiums is established to provide initial parameter setting basis for step S1.

[0027] In the embodiment of the application, based on the statistical results of 12 types of prototypes, a skylight design parameter recommendation value table of gymnasiums is automatically generated, and the table structure includes 10 columns of building type ID, typical orientation, roof form, use function, climate zone, recommended area ratio, recommended number, recommended length-width ratio, recommended change rate and confidence. The table is stored in the local edge computing node in the form of SQLite database, supports SQL query, and the designer can input the project number before step S1 to return the corresponding recommended value in seconds.

[0028] In one implementation manner of the embodiment of the application, the designer inputs the project number BJ-2025-01 in the front-end interface, the background retrieves SQLite to obtain the 7th record, and returns the recommended values: area ratio 8.5%, number 24, length-width ratio 1.6, change rate 1.0, and confidence 92%; the interface synchronously displays the three-dimensional effect diagram of the same case, and the designer confirms the recommended values to be automatically filled into the initial population individuals of step S1.

[0029] Preferably, step S3 comprises: The generator of the preset generative adversarial network extracts features from the building performance simulation data, wherein the generator adopts a U-Net architecture, and includes a preset number of down-sampling layers and a preset number of up-sampling layers; In the embodiment of the application, the generative adversarial network is selected as the data enhancement engine, the generator thereof adopts a U-Net architecture, and each of the down-sampling layers and the up-sampling layers is provided with 4 levels; each level of the down-sampling layer is multiplied by the number of channels and halved in spatial size through 3*3 convolution+BN+Leaky ReLU, until the size of the latent vector is compressed to 8*8*512; the corresponding up-sampling layer uses a skip connection, and after the channel dimension of the encoded features and the decoded features at the same level is spliced, the features are gradually restored to the original input size 64*64*3 through 3*3 transpose convolution+BN+ReLU.

[0030] In one implementation manner of the embodiment of the present application, the input is a three-channel effective daylight utilization-thermal comfort-photovoltaic power generation combined heat map of 64*64 pixels, the batch size is set to 32, the generator is first compressed to 8*8*512 through 4 times of downsampling, then the latent space is cascaded with a 128-dimensional Gaussian noise, and the spatial resolution is restored through 4 times of upsampling and jump connection; in the training stage, the Adam optimizer is adopted, the generator learning rate is , the discriminator learning rate is , the loss function is a Wasserstein loss with gradient penalty, and after 200 epochs of training, the SSIM of the generated image is greater than or equal to 0.92, so as to meet the subsequent synthesis requirements.

[0031] The synthetic performance data is generated in the latent space, and the noise vector dimension and batch size are set in the synthesis process. In the embodiment of the present application, after the generator completes encoding, the 512*8*8 tensor is flattened into a 32768-dimensional latent feature, and then spliced with a 128-dimensional random noise to form a 32896-dimensional mixed vector; the vector is mapped back to the decoding entrance of 8*8*256 through two fully connected layers (node number 4096->2048), and a 64*64*3 synthetic performance heat map is output through the upsampling network, so as to realize end-to-end generation of latent vector-performance distribution; the noise vector dimension and batch size are fixed in the configuration file.

[0032] In one implementation manner of the embodiment of the present application, the noise vector dimension is set to 128, and the batch size is 64; before each training iteration, a 64*128-dimensional noise matrix is sampled from N(0, 1), and then spliced with the latent features corresponding to 64 real samples to be sent to the decoding network to generate 64 synthetic performance maps; in order to avoid mode collapse, the FID index is used to monitor the distribution offset every 20 epochs, and when the FID is greater than 30, the generator learning rate is automatically reduced by 10%, until the FID is stable below 25, so as to ensure that the synthetic data and the real data are highly overlapped.

[0033] The synthetic performance data and the building performance simulation data are merged according to a preset ratio to form an initial expanded data set. In the embodiment of the present application, the synthetic data qualified through FID screening and the building performance simulation data are mixed in a ratio of 3:7 to form an initial expanded data set; before merging, the same normalization is performed on the two types of data, that is, the minimum value is 0 and the maximum value is 1, so as to prevent the scale difference from causing model bias; after merging, the total amount of the data set is expanded to 1.43 times of the original real sample.

[0034] In one implementation manner of the embodiment of the present application, if the original real samples are 1000, 429 synthetic data verified by FID are selected to splice, and 1429 initial expansion sets are obtained; SQLite records the source label of each sample (0 = real, 1 = synthetic), and the prediction errors of the real and synthetic subsets are counted respectively in subsequent cross-validation, so as to ensure that the synthetic data will not introduce significant deviation, and if the prediction error is greater than 5%, the GAN retraining is triggered.

[0035] The Jensen-Shannon divergence of the synthetic performance data and the building performance simulation data is calculated, and when the Jensen-Shannon divergence exceeds a preset divergence threshold, the corresponding synthetic data sample is removed, so as to determine the expansion performance data set.

[0036] In the embodiment of the present application, for each synthetic sample in the initial expansion data set, the Jensen-Shannon divergence (JSD) between it and the real sample set is calculated: first, the histogram distribution is estimated by flattening the 64x64x3 vector, and the number of buckets is 512, then the JSD value of the real distribution P and the synthetic distribution Q is calculated; if the JSD of a single synthetic sample is greater than 0.025, it is determined that the distribution drift is too large and is automatically removed; the remaining synthetic samples are screened in a loop until the average JSD of the remaining synthetic samples is less than or equal to 0.020, so as to lock the final expansion performance data set.

[0037] In one implementation manner of the embodiment of the present application, multi-thread parallel computing JSD is enabled, and each thread processes 50 synthetic samples; in this round, 87 high-diversity samples are removed, and the remaining 342 synthetic data and 1000 real data jointly constitute an expansion performance data set of 1342, the average JSD of which is reduced to 0.018, and the determination coefficient of the subsequent random forest training is 0.91 is improved to 0.94, and the verification screening mechanism is effective.

[0038] Preferably, the regression analysis in step S4 includes: Based on the expansion performance data set, a random forest algorithm is used to establish a mapping relationship between the design parameters of the window and the preset performance indicators, the number of decision trees is set to H, the maximum tree depth is set to F, and the determination coefficient and the root mean square error are calculated as evaluation indexes, wherein the value range of H is 50 to 200, and the value range of F is 5 to 15. In the embodiment of the present application, the 1342 expansion performance data sets obtained in step S3 are divided into a training set and a test set according to 8:2, the input variable is the 4-dimensional window design parameter (area ratio, number, length-width ratio, change rate), and the output variable is the 3-dimensional performance indicator (effective daylight utilization rate, thermal comfort percentage, and photovoltaic power generation capacity); a multi-output regression model is constructed by using the RandomForest Regressor of scikit-learn, and three-dimensional indicators are predicted simultaneously.

[0039] In one implementation of this invention, the number of decision trees is set to H=150, the maximum depth F=12, the minimum number of leaf node samples=5, and the maximum number of features=sqrt(4)=2, and bootstrap sampling is used; after training, the model's coefficient of determination on the training set is... =0.94, test set =0.91, Root Mean Square Error (RMSE) = (2.1%, 1.8%, 0.05 kWh / All meet the preset accuracy threshold. ≥0.90 and RMSE≤ (3%, 3%, 0.1kWh / Therefore, the mapping relationship is deemed to meet the standard and is locked as the prediction relationship for subsequent optimization.

[0040] When the coefficient of determination reaches or exceeds the preset accuracy threshold and the root mean square error is lower than the preset error threshold, the mapping relationship is considered to meet the preset requirements. Select the mapping relationship that meets the preset requirements as the prediction relationship.

[0041] In this embodiment of the invention, the training set is further divided into 5 parts, and 4 parts are used for training and 1 part for validation in turn, recording the results of each fold. With RMSE; only when 50% average ≥0.90 and average RMSE ≤ (3%, 3%, 0.1kWh / Only when the condition is met is the model deemed robust; otherwise, H and F are automatically adjusted using a grid search, with H step size of 25 and F step size of 2, until the condition is met.

[0042] In one implementation of this invention, the initial combination H=100 and F=8 is a 5-fold average. =0.88 was not met, so grid search was automatically initiated; after 3 rounds of iteration, when H=150 and F=12, the 50% average was achieved. It rose to 0.905, and the average RMSE decreased to (2.3%, 2.0%, 0.048kWh / If the dual threshold requirements are met, the search stops; this set of hyperparameters is written to config.json as the sole predictive relation model for subsequent multi-objective optimization stages.

[0043] Preferably, the multi-objective optimization method in step S5 includes: A multi-objective optimization problem is constructed based on the predictive relationship, where the optimization objectives include maximizing the effective solar utilization rate, maximizing the thermal comfort percentage, and maximizing photovoltaic power generation. In the embodiment of the present application, the random forest agent model obtained in step S4 is taken as a fast evaluator to construct a three-objective optimization problem of maximizing effective daylight utilization, maximizing the percentage of thermal comfort and maximizing photovoltaic power generation; the decision variable is the 4-dimensional window design parameter, and boundary constraints (area ratio 6%-11%, number 18-30, length-width ratio 1.2-2.0, change rate 0.6-1.2) are applied to the input layer of the agent model to ensure that the individuals generated by the genetic algorithm fall within the physically feasible domain.

[0044] In one implementation manner of the embodiment of the present application, the three objectives are written in vector form , and the negative sign indicates that maximizing is converted into minimizing fitness; the NSGA-III framework of the pymoo library is called, real number coding is performed on the decision variable , and the agent model is called once to return three-dimensional target values, and the time consumption of single evaluation is about 2 ms, which is more than 4000 times faster than the traditional Energy Plus calling, so that large-scale population iteration becomes possible.

[0045] The solution parameters of the multi-objective optimization problem are set, and the solution parameters include the population size, the iteration number and the number of reference points; In the embodiment of the present application, the population size N=120, the iteration number T=150 and the number of reference points p=12 according to the Das-Dennis rule are preset according to the problem complexity, and the 3-dimensional combination is generated, and a total of C(12+3-1, 3-1)=91 reference points are generated, which are uniformly distributed on the unit hyperplane Each reference point corresponds to a preferred direction to guide the algorithm to form a uniform grid in the target space.

[0046] In one implementation manner of the embodiment of the present application, if it is monitored during running that the front distribution is too sparse, the system can dynamically insert additional reference points: when the minimum neighborhood distance is greater than 1.5 times the average distance, one reference point is added at the center of the neighborhood, and the total number does not exceed 150.

[0047] The non-dominated sorting genetic algorithm based on reference point distribution is used to solve the multi-objective optimization problem; In the embodiment of the present application, the algorithm process follows the standard NSGA-III: first, the initial parent is generated, the child is obtained through simulated binary crossover (SBX, =30) and polynomial mutation ( =20), the is combined, and the non-dominated sorting is performed, and then the next generation is selected according to the reference point association and niche count ; the whole process uses the agent model to replace the real simulation.

[0048] In one implementation of the embodiment of the application, the crossover probability =0.9, the mutation probability =1 / 4=0.25, and rebound repair is performed on the genes exceeding the boundary after mutation, that is, if the area ratio is greater than 11%, 11%-(x-11%) is taken, so that the search space is continuous; after 150 generations of evolution, 87 of the 91 reference points are occupied by at least one individual, with an occupation rate of 95.6%, indicating that the algorithm successfully drives the population to cover all the preferred directions.

[0049] In the solving optimization process, a preset reference point adjustment mechanism is introduced, and the positions and quantities of the reference points are dynamically adjusted according to the distribution density of the solution set in the target space; In the embodiment of the application, the reference point health degree is checked once every 30 generations: the number of associated individuals of each reference point is counted, if the number of associated individuals of a reference point is zero for two generations in succession, it is determined that the reference point deviates from the front, and then the reference point is moved 5% distance in the direction of the centroid of the target space and is normalized to the unit hyperplane again.

[0050] In one implementation of the embodiment of the application, at the 90th generation, it is found that reference points 17 and 42 have no associated individuals, after migration of the system, at the 91st generation, three new individuals fall into the neighborhood of 17, and the association is successful; before and after migration, the front hyper-volume (HV) index increases from 0.847 to 0.851.

[0051] The distribution uniformity and convergence index of the intermediate solution set of the optimization process in the calculation solving are calculated, and when the distribution uniformity and convergence index are both lower than the preset standard, the optimization strategy is automatically adjusted; In the embodiment of the application, IGD (Inverted Generational Distance) and Spacing double indexes are selected as the process monitoring: the IGD and Spacing are calculated once every 10 generations, if the IGD is greater than 0.015 and the Spacing is greater than 0.020, it is considered that the uniformity and convergence are both lower than the preset standard, and the system automatically activates the intensive search mode: the crossover distribution index η is reduced from 30 to 15 within the next 20 generations, the search range is improved, and the algorithm is prompted to jump out of the local.

[0052] In one implementation of the embodiment of the application, at the 100th generation, the IGD is 0.018 and the Spacing is 0.023, which triggers the intensive mode; the system automatically reduces η and introduces a large mutation probability of 0.4, at the 120th generation, the IGD is reduced to 0.012 and the Spacing is reduced to 0.018, which meets the double thresholds, and then the normal parameters are restored; in the case, the feedback closed loop ensures that the whole evolution process is always in a high-quality exploration state.

[0053] After preset number of iterations of optimization, a first Pareto optimal solution set containing preset number of non-dominated solutions is output, wherein each solution in the first Pareto optimal solution set represents a sunroof design scheme and a corresponding performance index value.

[0054] In the embodiment of the present application, when the iteration count reaches 150 generations and the HV increment of two consecutive generations is less than 0.1%, the evolution is terminated, the individuals most adjacent to the 91 reference points are screened from the last generation of non-dominated layer Level 0, and a first Pareto optimal solution set containing 87 non-dominated solutions is formed; each solution corresponds to a set of sunroof parameters (area ratio, number, length-width ratio, and change rate) and three-objective performance values predicted by the proxy model.

[0055] Preferably, step S6 comprises: Step S61: Extracting the sunroof-to-roof area ratio, the number of sunroofs, the length-width ratio of sunroofs, and the change rate of sunroofs from the first Pareto optimal solution set respectively, and calculating the mean, standard deviation, and quantile thereof to obtain the distribution characteristics of each sunroof design parameter; In the embodiment of the present application, the sunroof design variables are analyzed from the 87 non-dominated solutions obtained in step S5, and four one-dimensional arrays of area ratio, number, length-width ratio, and change rate are established respectively; the mean , standard deviation , and [10%, 25%, 50%, 75%, 90%] quantile of each set of data are calculated using Numpy to form a five-number summary table for describing the actual distribution pattern of each parameter on the Pareto frontier.

[0056] In one implementation manner of the embodiment of the present application, the mean of the area ratio array is 8.7%, the standard deviation is 1.2%, the 50% quantile is 8.5%, and the 75% quantile is 9.6%; the mean of the number array is 24.3, the standard deviation is 2.1; the mean of the length-width ratio is 1.58, the standard deviation is 0.18; the mean of the change rate is 17.8°, and the standard deviation is 1.4°; all the coefficients of variation are less than 0.15, indicating that the parameter distribution in the frontier solution set is concentrated, and the preferred interval can be selected based on ± .

[0057] Step S62: Determining the preferred interval of each sunroof design parameter based on the distribution characteristics; In the embodiment of the present application, the - , + interval is taken as the preliminary selection interval, and the intersection with the 10%-90% quantile interval is calculated to ensure that the preferred interval covers both the statistical main body and the extreme good solution; if the lower limit of the intersection is lower than the initial boundary in step S1, the boundary value is taken as the hard constraint, and finally the preferred interval of area ratio 7.5%-10.0%, number 22-27, length-width ratio 1.4-1.8, and change rate 0.9-1.1 is obtained.

[0058] In one implementation of the embodiment of the application, if a certain parameter - is less than the initial lower limit, 25% quantile is used instead: for example, the aspect ratio - =1.40, which is higher than the boundary 1.2, remains unchanged.

[0059] Step S63: The preferred interval is corrected according to the preset engineering practical constraints to determine the corrected preferred interval, and the correction factors include structural safety constraints and construction feasibility constraints; In the embodiment of the application, the structural safety and construction feasibility are quantified as calculable constraints: 1) Structural safety - area ratio ≤10% to ensure that the roof continuous purlin span ≤4.2m; 2) Construction feasibility - single window width ≤1.8m and weight ≤200kg, so that the number ≥22, the single window area should not be >2.5 , and the upper limit of the aspect ratio is inversely calculated to be 1.8; only when both conditions are met can the corrected preferred interval be entered, otherwise it is cut according to the boundary.

[0060] In one implementation of the embodiment of the application, the original preferred area ratio upper limit 10.0% is equal to the structural limit value, which remains unchanged; the upper limit of the number 27 corresponds to the average area of a single window 2.4 , which meets the hoisting requirements after rechecking; the aspect ratio 1.8 is compared with the window type library to be 1.68m×3.02m, and the weight 190kg<200kg, which is qualified.

[0061] Step S64: Adjust the parameter range based on the corrected preferred interval to obtain the optimal skylight design parameter range.

[0062] In the embodiment of the application, the corrected interval is packaged into an optimal parameter range table, which contains the fields {variable, lower limit, upper limit, unit, source}.

[0063] Preferably, after step S5, there is also a performance balance factor calculation step: Based on the first Pareto optimal solution set, the change rate ratio values between the effective daylight utilization rate, the thermal comfort percentage and the photovoltaic power generation capacity are calculated; Wherein, the change rate ratio value is calculated by the performance index difference ratio of adjacent solutions on the Pareto frontier; The geometric mean of the three change rate ratio values is defined as the performance balance factor.

[0064] In the embodiment of the application, for the 87 non-dominated solutions obtained in step S5, the adjacent solution change rate ratio values on two targets are calculated in turn along the Pareto frontier after normalizing the target space: define , , where is the difference of adjacent solutions in the ith dimension objective; the geometric mean of the three ratios obtained for each pair of adjacent solutions is taken as the local performance balance factor of the edge, and finally the global performance balance factor is obtained by averaging the performance balance factors of all the edges .

[0065] In an implementation manner of the embodiment of the present application, taking the sorted solutions 21 and 22 as an example, the normalized objective difference is = 0.018, = 0.012, = 0.009, then = 1.50, = 1.33, = 0.75, and the geometric mean is ; after traversing all the 86 edges, the global = 1.23, The closer to 1 indicates that the front balance is more balanced.

[0066] Preferably, the application of the performance balance factor after step S6 includes: Based on the performance balance factor, the influence degree of the area ratio of the skylight and the roof, the number of the skylight, the length-width ratio of the skylight and the change rate of the skylight on the performance balance factor is calculated respectively, wherein the influence degree is quantified by a sensitivity analysis method; In the embodiment of the present application, the global performance balance factor is the response value, Sobol sensitivity analysis based on variance decomposition is performed on the 4-dimensional skylight design parameters: N = 1024 groups of samples are generated by using Saltelli sampling in the optimal parameter range locked in step S6, and a random forest proxy model is called to quickly calculate the corresponding , and then estimate the first-order and total-order sensitivity indices.

[0067] In an implementation manner of the embodiment of the present application, the calculation result shows that the total-order index of the area ratio is 0.72, the number is 0.18, the length-width ratio is 0.06, and the change rate is 0.9; setting the influence threshold value to 0.15, the area ratio and the number are determined as the key skylight design parameters, and the subsequent simplified optimization is only expanded around these two dimensions, and the decision space can be reduced from 4 dimensions to 2 dimensions.

[0068] When the influence degree of a certain skylight design parameter on the performance balance factor exceeds the preset influence threshold value, it is defined as a key skylight design parameter; A simplified optimization model is established based on the key skylight design parameter; In this embodiment of the invention, the selected key parameters (area ratio, quantity) are used as new decision vectors. The aspect ratio and rate of change are kept fixed at the median of the optimal interval in step S6, which is 1.6 and 1.0, respectively. A simplified surrogate model is constructed: the random forest is retrained with only 2-dimensional input, and the number of trees is reduced to 100 to further accelerate the process. The optimization objective is still set to maximize... , , However, the constraint degenerates from a 4-dimensional hyperrectangle to a 2-dimensional plane, reducing the time for a single evaluation from 2ms to 0.3ms.

[0069] A predetermined number of solutions are randomly selected from the first Pareto optimal solution set as the initial solution set; In this embodiment of the invention, 20 solutions are randomly selected from the 87 solutions in the first Pareto optimal solution set as the initial solution set. The selection process uses a fixed random seed to ensure reproducibility. After selection, only the key parameter values ​​are retained, and the aspect ratio and rate of change dimensions are removed to form an initial matrix of 20×2, which is used for subsequent rapid re-optimization and screening in 2D space.

[0070] The simplified optimization model is used to quickly screen the skylight design schemes in the initial scheme set. During the screening process, the scheme evaluation time is set to not exceed 10% of the evaluation time of the original method, and a list of the preferred schemes after screening is output.

[0071] In this embodiment of the invention, a simplified surrogate model is used as the evaluator to perform a 2D greedy local search on the initial solution set: eight neighborhood solutions are generated in each generation, and those no worse than the parent's are selected. The solution is moved to the next generation, and the process is terminated if there is no improvement after 5 consecutive generations; the total number of calls is limited to ≤200, and the total evaluation time is about 0.06s, which is only 8.6% of the original 4D NSGA-III evaluation time (≈0.7s), meeting the preset requirement of no more than 10%.

[0072] Preferably, the process further includes the following after step S5: A photovoltaic panel layout scheme is generated based on the photovoltaic power generation simulation results in building performance simulation. In this embodiment of the invention, after completing the photovoltaic power generation module for building performance simulation, the usable area of ​​the roof and its radiation distribution are extracted, and the irradiance ≥1200kWh / The continuous surface region is defined as the photovoltaic panel layout area. Using a rectangular layout algorithm, a single photovoltaic panel of 1.65m×0.99m is used as a unit. The panels are arranged row by row along the roof slope direction within the layout area, with a row spacing of 0.05m and a panel edge distance of 0.3m from the skylight outline. An initial photovoltaic panel layout scheme without overlap is automatically generated and the center coordinates of each panel are output.

[0073] Based on the skylight design scheme and photovoltaic panel layout scheme, calculate the Euclidean distance from each skylight edge point to the nearest photovoltaic panel center; In the embodiment of the present application, the skylight polygon is uniformly discretized into boundary point clouds with a side length of 0.1 m, and the three-dimensional Euclidean distance from each boundary point to the center of all photovoltaic panels is calculated . =‖ - ‖2, and min( ) is taken as the minimum distance of the point; after traversing all the boundary points, a distance array D is formed, and the length is equal to the total number of boundary points.

[0074] In an implementation manner of the embodiment of the present application, the perimeter of a certain skylight contour is 28 m, and after discretization, there are 280 points, and the coordinates of a point are (12.4, 7.2, 13.1) m, and the nearest panel center (13.1, 7.8, 13.4) m is calculated , = 0.92 m; the minimum distance of all 280 points ranges from 0.31 m to 1.05 m, and the average value is 0.62 m.

[0075] The Euclidean distance is taken as the reciprocal, and the weighted sum is obtained according to the skylight area, to obtain a skylight diffusion effect index; In the embodiment of the present application, the reciprocal of the array D is taken element by element to obtain a local diffusion coefficient 1 / , and the weighted sum is obtained by taking the skylight area A as the weight, to define the skylight diffusion effect index SDE= , the unit ; the larger the SDE is, the more serious the interference of the skylight on the photovoltaic array is.

[0076] According to the preset threshold range and the skylight diffusion effect index, the severity level of the spatial conflict is determined; In the embodiment of the present application, the preset SDE threshold interval is: ≤0.8 is mild, >0.8 and ≤1.4 is moderate, and >1.4 is severe; after reading the SDE of each skylight, a grade label is automatically assigned, and the Pareto solution set metadata is written, so that the designer can quickly identify high-conflict schemes without manual comparison of each picture.

[0077] According to the severity level of the spatial conflict, a hierarchical conflict resolution strategy is formulated.

[0078] In the embodiment of the present application, differentiated strategies are formulated for three levels of conflicts: the original scheme is maintained for mild conflicts; for moderate conflicts, the photovoltaic panels within 0.3 m on both sides of the skylight are moved outward by 0.2 m and are reconnected; and for severe conflicts, the panels are first moved outward, then the skylight area is reduced by 5%, and the input of the agent model is updated synchronously, and the performance is re-evaluated to ensure that the resolved scheme is still near the Pareto frontier.

[0079] Preferably, after step S6, it further comprises: Obtaining the building use mode of the target stadium, the building use mode being any one of a competition mode, a training mode and a daily mode; In the embodiment of the present application, the electronic questionnaire is issued by the owner end applet, the stadium operator selects one of the competition / training / daily three options, and the system writes the reply into the project table after receiving the reply; if no feedback is received within 3 days, the daily mode is defaulted and a reminder email is sent, ensuring that the subsequent threshold adjustment has a clear basis.

[0080] In one implementation manner of the embodiment of the present application, the target stadium is determined to be the 2025 Winter Olympics test stadium, the operator selects the competition mode, encodes the label as 2 and stores it in the mode field of SQLite, and triggers the threshold adjustment sub-process to prepare for subsequent regeneration of the Pareto frontier.

[0081] Collecting the local climate conditions of the target stadium; In the embodiment of the present application, the hourly typical meteorological year data of the target stadium site is automatically downloaded by latitude and longitude using Meteonorm API, the average outdoor temperature, the scattering radiation ratio and the average wind speed in winter (November-March) are extracted, and they are used as the cold-high scattering-low wind climate label; the label together with the use mode determines the correction coefficient of the satisfaction threshold.

[0082] According to the local climate conditions, the satisfaction threshold of the effective daylight utilization rate and the thermal comfort percentage is adjusted, wherein the satisfaction threshold of the competition mode is increased by a preset percentage than that of the daily mode; In the embodiment of the present application, the preset daily mode reference threshold is: effective daylight utilization rate ≥ 60%, thermal comfort percentage ≥ 70%; the competition mode is increased by 10% on the basis, i.e. ≥ 66% and ≥ 77%; after reading the mode code, the threshold is automatically written into the multi-objective optimization constraint layer to replace the original constraint value, so that the frontier of the re-search moves in the direction of high lighting and high comfort.

[0083] A second Pareto optimal solution set is regenerated based on the adjusted satisfaction threshold.

[0084] In the embodiment of the present application, NSGA-III is called with the same population size N=120 and the number of generations T=150, but the constraint boundary is replaced by the adjusted threshold, and the surrogate model remains unchanged; after evolution, the second Pareto optimal solution set is output, wherein all solutions meet the higher lighting and comfort requirements of the competition mode+local climate, so as to provide the owner with the preferred skylight-photovoltaic scheme during the event period.

[0085] Preferably, after step S6, it further comprises: Latin Hypercube method of scipy-stats is used to generate N verification schemes in the optimal range of the window design parameters, wherein N is an integer between 50 and 200; In the embodiment of the application, the Latin Hypercube method of scipy-stats is used to generate N=100 verification schemes in the optimal range of the window design parameters (area ratio 7.5%-10.0%, number 22-27, length-width ratio 1.4-1.8, and change rate 0.9-1.1); before sampling, the four-dimensional variables are normalized to [0, 1], and after sampling, the boundary is linearly transformed, so that each dimension only appears once in the 100 equal interval, and a uniform and well-space-filled verification data set is obtained.

[0086] In one implementation of the embodiment of the application, the random seed is set to 42, the generation time is 0.8s, and a 100x4 array is output; the first sample is decoded as an area ratio of 7.6%, a number of 23, a length-width ratio of 1.45, and a change rate of 1.06, and the last sample is decoded as an area ratio of 9.8%, a number of 26, a length-width ratio of 1.76, and a change rate of 1.08, the coverage range is consistent with the boundary, and there is no out-of-bound phenomenon, which can be directly sent to the building performance simulation.

[0087] The effective daylight utilization, the percentage of thermal comfort, and the photovoltaic power generation of each verification scheme are calculated by building performance simulation. In the embodiment of the application, the above-mentioned 100 schemes are simulated in batches by the automatic process of Rhino / GH+Energy Plus: the indoor illumination, PMV, and photovoltaic power generation are output for 8760h per year, and the effective daylight utilization, the percentage of thermal comfort, and the annual power generation are obtained after post-processing.

[0088] The standard deviation and the coefficient of variation of the effective daylight utilization, the percentage of thermal comfort, and the photovoltaic power generation of each verification scheme are calculated respectively. In the embodiment of the application, the standard deviation and the coefficient of variation of the three performance indicators of 100 groups are calculated respectively and the coefficient of variation CV= / ; wherein is the average of each index, and CV is dimensionless and can be used for horizontal comparison of fluctuation amplitude; the threshold values of CV are set to daylight ≤5%, comfort ≤6%, and power generation ≤4% as high robustness criteria, and if all three conditions are met, the scheme is marked as a high robustness scheme, otherwise, it is listed as a normal scheme.

[0089] The high robustness scheme in the N verification schemes is determined according to the standard deviation and the coefficient of variation. In the embodiment of the present application, the CV normalized scores are weighted and summed: Score = 0.4 x CV_sunlight + 0.4 x CV_comfort + 0.2 x CV_power, the lower the score, the higher the robustness; the top 30% (i.e. 30) of the schemes are selected as high-robustness schemes.

[0090] Based on the high-robustness schemes, the optimal skylight design parameter range is divided into a core recommended area and an extended applicable area.

[0091] In the embodiment of the present application, the minimum-maximum envelope of the 4-dimensional parameters of the high-robustness schemes is taken to form the core recommended area; and the envelope of the remaining 70 ordinary schemes is subtracted from the core area to obtain the extended applicable area.

[0092] Especially important is that after step S6, it further includes: Based on the monitoring performance data collected during the actual building operation, the actual effective daylight utilization rate, the actual thermal comfort percentage and the actual photovoltaic power generation capacity are calculated; In the embodiment of the present application, after the roof is put into operation, the illuminance, indoor temperature and photovoltaic direct current metering signals are continuously collected through the BAS system from the first year: the illuminance is recorded by 16 roof light sensors every minute, and the daily average value is taken; the thermal comfort is calculated PMV according to ISO 7730, and the whole point output is taken; the photovoltaic power generation capacity is uploaded daily power generation capacity through the inverter RS485.

[0093] The monitoring performance data is compared and analyzed with the expected performance indicators corresponding to the optimal skylight design parameter range, and a performance deviation index is calculated; In the embodiment of the present application, the normalized deviation is defined as , wherein is the expected value, is the actual and expected difference; the preset deviation threshold = 5%, if > The model drift is determined, the agent model needs to be retrained and the optimal parameter range is updated.

[0094] In one implementation manner of the embodiment of the present application, the above-mentioned monitoring performance data of January 2026 = 4.2%, which does not exceed the standard; but by July 2026, the dust accumulation and the shadow of the newly-built tower around cause to rise to 5.8%, which exceeds the threshold, and the system automatically starts the recalibration process and calls steps S4-S6 for online updating.

[0095] When the performance deviation index exceeds the preset deviation threshold, steps S4 to S6 are re-executed based on the monitoring performance data, and the optimal skylight design parameter range is updated; In the embodiment of the present application, 12-month monitoring data is added to the original expanded performance data set as a new real sample to form an enhanced training set; the GAN-RF-NSGA process is kept unchanged, a new proxy model is re-established, and a new Pareto front is evolved, and finally the corrected optimal window parameter range is output.

[0096] A parameter range optimization record database is established to store key parameters and performance indicators of the previous parameter range optimization process.

[0097] In the embodiment of the present application, range_update.db is established, and the table structure includes {version ID, update time, trigger reason, old range JSON, new range JSON, old , new , value}; each time the re-optimization is completed, a record is written.

[0098] Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting, the scope of the present application is defined by the appended claims rather than the above description, and therefore all changes falling within the meaning and scope of the equivalent elements of the application file are intended to be included in the present application.

[0099] The above description is only a specific embodiment of the present application, enabling those skilled in the art to understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for coordinated optimization design of wind and solar energy in stadiums based on a fusion algorithm of GAN and RF-NSGA, characterized in that, Includes the following steps: Step S1: Determine the skylight design parameters, including the skylight-to-roof area ratio, the number of skylights, the skylight length-to-width ratio, and the skylight variation rate; Step S2: Based on the skylight design parameters, obtain building performance simulation data through building performance simulation. The building performance simulation data includes effective solar utilization rate, thermal comfort percentage, and photovoltaic power generation. Step S3: Perform generative data augmentation on the building performance simulation data to generate an expanded performance dataset; Step S4: Based on the extended performance dataset, establish the predictive relationship between skylight design parameters and effective solar utilization rate, thermal comfort percentage and photovoltaic power generation through regression analysis; Step S5: Based on the predicted relationship, the effective solar utilization rate, thermal comfort percentage and photovoltaic power generation are synergistically optimized using a multi-objective optimization method to generate the first Pareto optimal solution set; Step S6: Determine the optimal range of sunroof design parameters based on the first Pareto optimal solution set.

2. The method for coordinated optimization design of wind and solar energy in stadiums based on the integrated algorithm of GAN and RF-NSGA as described in claim 1, characterized in that, Step S3 includes: Feature extraction is performed on building performance simulation data using a pre-defined generative adversarial network generator, which adopts a U-Net architecture and includes a pre-defined number of downsampling layers and a pre-defined number of upsampling layers. Generate synthetic performance data in the latent space, and set the noise vector dimension and batch size during the synthesis process; The synthetic performance data and the building performance simulation data are merged according to a preset ratio to form an initial expanded dataset; The Jensen-Shannon divergence between the synthetic performance data and the building performance simulation data is calculated. When the Jensen-Shannon divergence exceeds a preset divergence threshold, the corresponding synthetic data sample is removed, thereby determining the expanded performance dataset.

3. The method for coordinated optimization design of wind and solar energy in stadiums based on the integrated algorithm of GAN and RF-NSGA as described in claim 1, characterized in that, The regression analysis in step S4 includes: Based on the extended performance dataset, the random forest algorithm is used to establish the mapping relationship between the window design parameters and the preset performance indicators. The number of decision trees is set to H, the maximum tree depth is F, and the coefficient of determination and root mean square error are calculated as evaluation indicators. The value of H ranges from 50 to 200, and the value of F ranges from 5 to 15. When the coefficient of determination reaches or exceeds the preset accuracy threshold and the root mean square error is lower than the preset error threshold, the mapping relationship is considered to meet the preset requirements. Select the mapping relationship that meets the preset requirements as the prediction relationship.

4. The method for coordinated optimization design of wind and solar energy in a stadium based on the integrated algorithm of GAN and RF-NSGA as described in claim 1, characterized in that, The multi-objective optimization methods in step S5 include: A multi-objective optimization problem is constructed based on the predictive relationship, where the optimization objectives include maximizing the effective solar utilization rate, maximizing the thermal comfort percentage, and maximizing photovoltaic power generation. Set the solution parameters for the multi-objective optimization problem. The solution parameters include population size, number of iterations, and number of reference points. A non-dominated sorting genetic algorithm based on the distribution of reference points is used to solve the multi-objective optimization problem. In the optimization process of solving the problem, a preset reference point adjustment mechanism is introduced to dynamically adjust the position and number of reference points according to the distribution density of the solution set in the target space. The algorithm calculates the uniformity of distribution and convergence index of intermediate solution set during the optimization process. When both the uniformity of distribution and convergence index are lower than the preset standard, the optimization strategy is automatically adjusted. After a preset number of iterations, the output is a first Pareto optimal solution set containing a preset number of non-dominated solutions, where each solution in the first Pareto optimal solution set represents a sunroof design scheme and its corresponding performance index value.

5. The method for coordinated optimization design of wind and solar energy in a stadium based on the integrated algorithm of GAN and RF-NSGA as described in claim 1, characterized in that, Step S6 includes: Step S61: Extract the ratio of skylight to roof area, number of skylights, length-to-width ratio of skylights, and skylight change rate from the first Pareto optimal solution set, and calculate their mean, standard deviation, and quantiles to obtain the distribution characteristics of each skylight design parameter; Step S62: Determine the optimal range of design parameters for each skylight based on distribution characteristics; Step S63: Modify the preferred range according to the preset actual constraints of the project, and determine the modified preferred range. The modification factors include structural safety constraints and construction feasibility constraints. Step S64: Adjust the parameter range based on the modified preferred range to obtain the optimal sunroof design parameter range.

6. The method for coordinated optimization design of wind and solar energy in a stadium based on the integrated algorithm of GAN and RF-NSGA as described in claim 1, characterized in that, Step S5 is followed by a step to calculate the performance balance factor: Based on the first Pareto optimal solution set, calculate the ratio of the rate of change between each pair of effective solar utilization, thermal comfort percentage, and photovoltaic power generation; The ratio of the rate of change is calculated by the ratio of the performance index differences between adjacent solutions on the Pareto front; The geometric mean of the ratios of the three rates of change is defined as the performance balance factor.

7. The method for coordinated optimization design of wind and solar energy in stadiums based on the integrated algorithm of GAN and RF-NSGA as described in claim 1, characterized in that, The application of the performance balance factor after step S6 includes: Based on the performance balance factor, the influence of the ratio of skylight to roof area, the number of skylights, the length-to-width ratio of skylights, and the rate of change of skylights on the performance balance factor were calculated, and the degree of influence was quantified by the sensitivity analysis method. When the influence of a certain sunroof design parameter on the performance balance factor exceeds the preset influence threshold, it is defined as a key sunroof design parameter. A simplified optimization model was established based on key sunroof design parameters; A predetermined number of solutions are randomly selected from the first Pareto optimal solution set as the initial solution set; The simplified optimization model is used to quickly screen the skylight design schemes in the initial scheme set. During the screening process, the scheme evaluation time is set to not exceed 10% of the evaluation time of the original method, and a list of the preferred schemes after screening is output.

8. The method for coordinated optimization design of wind and solar energy in a stadium based on the integrated algorithm of GAN and RF-NSGA as described in claim 1, characterized in that, The process after step S5 also includes: A photovoltaic panel layout scheme is generated based on the photovoltaic power generation simulation results in building performance simulation. Based on the skylight design scheme and photovoltaic panel layout scheme, calculate the Euclidean distance from each skylight edge point to the nearest photovoltaic panel center; The reciprocal of the Euclidean distance is taken, and the sum is weighted according to the skylight area to obtain the skylight diffusion effect index; The severity level of spatial conflict is determined based on a preset threshold range and the skylight diffusion effect index. Develop tiered conflict resolution strategies based on the severity of spatial conflicts.

9. The method for coordinated optimization design of wind and solar energy in stadiums based on the integrated algorithm of GAN and RF-NSGA as described in claim 1, characterized in that, The process after step S6 also includes: Obtain the building usage mode of the target stadium. The building usage mode can be any one of the following: competition mode, training mode, and daily mode. Collect data on the climate conditions of the target stadium's location; The satisfaction thresholds for effective sunlight utilization and thermal comfort percentage are adjusted according to local climate conditions, with the satisfaction threshold for competition mode being increased by a preset percentage compared to daily mode. The second Pareto optimal solution set is regenerated based on the adjusted satisfaction threshold.

10. The method for coordinated optimization design of wind and solar energy in a stadium based on the integrated algorithm of GAN and RF-NSGA as described in claim 1, characterized in that, The process after step S6 also includes: Within the optimal sunroof design parameter range, N verification schemes are generated using the Latin hypercube sampling method, where N is an integer between 50 and 200; The effective solar utilization rate, thermal comfort percentage, and photovoltaic power generation of each verification scheme were calculated through building performance simulation. Calculate the standard deviation and coefficient of variation of the effective solar utilization rate, thermal comfort percentage, and photovoltaic power generation for each verification scheme; The most robust scheme among N validation schemes is determined based on the standard deviation and coefficient of variation; Based on a highly robust approach, the optimal sunroof design parameter range is divided into a core recommended area and an extended applicable area.

Images