A tropical zero-carbon building integrated system design parameter optimization method

Abstract

The application relates to the technical field of building energy saving and green building, and discloses a tropical zero-carbon building integrated system design parameter optimization method. The method constructs a three-dimensional analysis framework of input correlation-output uncertainty-global sensitivity, accurately identifies a dominant parameter set and a key interaction item by high-fidelity simulation modeling, Latin hypercube sampling, Gaussian process proxy model construction and verification, combination of Pearson correlation analysis, Monte Carlo propagation and Sobel sensitivity calculation, and finally solves an optimal low-carbon parameter combination under engineering constraints by using a sequential quadratic programming algorithm. The system comprises twelve functional units of meteorological acquisition, parameter definition, simulation modeling and the like, and outputs a parameter list which can be directly used for enclosure construction, unit selection and photovoltaic arrangement. The application improves the scientificity and implementability of parameterized design of the tropical zero-carbon building, can reduce annual net carbon emissions, and provides intelligent decision support for the building double-carbon target.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of building energy saving and green building, and particularly relates to a design parameter optimization method for a tropical zero-carbon building integrated system. BACKGROUND

[0002] With the intensification of global warming and the full implementation of the double carbon strategy, the zero-carbon transformation in the building field has become a core issue for realizing sustainable development. In the tropical region, the building cooling load is high due to the high temperature and humidity and strong solar radiation throughout the year, and the energy consumption intensity is much higher than that in the temperate and cold regions, making the zero-carbon goal face severe challenges in this climate zone.

[0003] As a key means to realize zero-carbon operation of buildings, the building integrated system covers multiple subsystems such as the thermal performance of the building envelope, the operation strategy of heating, ventilation and air conditioning, the energy consumption mode of lighting and equipment, and the configuration of renewable energy production, etc. The parameters among the subsystems are highly coupled and nonlinearly interacted, resulting in an exponential increase in the complexity of the design space. Traditional building performance optimization methods mostly rely on experience judgment or local parameter adjustment, and lack quantitative analysis capability for the overall behavior of the system, making it difficult to support precise decision-making for tropical zero-carbon buildings under multiple constraints.

[0004] Among them, for the design parameter optimization of the tropical zero-carbon building integrated system, existing research generally focuses on single-dimensional analysis paths, such as only screening key variables through parameter sensitivity sorting, or only evaluating the output fluctuation range through uncertainty propagation. Although such methods can provide certain reference at the local level, they cannot construct a systematic cognitive framework from the three dimensions of input correlation, output distribution characteristics and global sensitivity, resulting in the neglect of the synergistic effect and conflict mechanism implied among parameters, and the optimization results are prone to local optimum or misleading conclusions.

[0005] Especially in complex simulation scenarios without explicit objective functions, most practices directly call commercial software for "black box" parameter scanning, neither establishing a proxy model accuracy verification mechanism, nor lacking cross-validation of the robustness of sensitivity results, resulting in serious lack of engineering credibility and physical interpretability of the final recommended scheme.

[0006] The existing technology has not formed a multi-dimensional parameter diagnosis system for the characteristics of tropical zero-carbon buildings, and has not effectively identified the key parameters and their interaction paths that truly dominate carbon emissions in the energy saving link (such as the thermal insulation performance of the building envelope), the energy production link (such as the layout efficiency of photovoltaic components), and the energy consumption link (such as the regulation of equipment usage rate). At the same time, for key technical nodes such as parameter sampling independence guarantee, output probability distribution characterization, proxy model optimization standard, and sensitivity index verification process, there is also a lack of standardized and closed-loop processing logic.

[0007] This series of methodological defects makes the current tropical zero-carbon building design still stay in the rough trial and error stage, which cannot meet the precise optimization demand of high performance, high reliability and high adaptability. It is urgent to build a systematic analysis method that integrates input decoupling, output quantification, global sensitivity tracing and mechanism-driven optimization to break through the "black box" dilemma of complex parameter systems and provide solid support for the scientific design and intelligent control of tropical zero-carbon buildings. SUMMARY

[0008] A tropical zero-carbon building integrated system design parameter optimization method, comprising the following steps:

[0009] Step S1: Obtain the hourly temperature, humidity, solar radiation intensity and wind speed data of a typical meteorological year in tropical regions as external environmental driving variables;

[0010] Step S2: Define the set of design parameters to be optimized of the building integrated system;

[0011] Step S3: Build a dynamic simulation model of building life cycle carbon emissions, which includes an energy consumption module in the building operation phase, a photovoltaic power generation offset module and a dynamic response module of the power grid carbon factor, and outputs the annual net carbon emissions;

[0012] Step S4: Use the Latin hypercube sampling method, i.e. LHS method, to generate covering samples within the parameter permission range and complete the annual simulation;

[0013] Step S5: Based on the input and output data of all samples, train a Gaussian process regression surrogate model, use anisotropic radial basis kernel, i.e. ARD-RBF, for verification analysis, and set the accuracy threshold for subsequent analysis;

[0014] Step S6: Calculate the Pearson linear correlation coefficient between each pair of input parameters r , form a correlation matrix, identify parameter pairs with strong positive or negative correlation, and mark them as potential coupling interference terms;

[0015] Step S7: Take the surrogate model as a fast response function, calculate the probability density distribution of the annual net carbon emissions, and extract its mean, standard deviation and uncertainty interval representing the output performance;

[0016] Step S8: Apply the Sobol global sensitivity analysis method to calculate the first-order sensitivity index of each single parameter and the second-order interaction sensitivity index of any two parameter combinations;

[0017] According to the descending order of the first-order sensitivity index, select the dominant parameter set and filter the key interaction terms;

[0018] Step S9: For the dominant parameter set and key interaction term, combined with building code limits, material market supply capacity and construction process constraints, the engineering feasible adjustment interval of each parameter is determined;

[0019] Step S10: Output the final optimized parameter combination and its corresponding expected annual net carbon emission estimate and confidence interval as the basis for tropical zero-carbon building design decisions.

[0020] Preferably, the typical meteorological year hourly temperature, humidity, solar radiation intensity and wind speed data in tropical regions are obtained as follows:

[0021] From the standard meteorological year data set published by the National Meteorological Data Center, each time step corresponds to 1 h of continuous observation value, temperature data records air dry bulb temperature and wet bulb temperature in ℃, humidity data is provided in the form of relative humidity, solar radiation intensity is divided into three components of normal direct radiation, scattered radiation and ground reflected radiation, unit is W / m 2 , wind speed data records the horizontal wind speed at 10 m height, unit is m / s, and all meteorological parameters are quality controlled and interpolated to ensure no missing values and continuous and complete time series.

[0022] Preferably, the building operation stage energy consumption module uses energy balance equation to describe the dynamic relationship between building envelope heat transfer, solar heat gain, internal heat and air conditioning refrigeration capacity;

[0023] The building envelope heat transfer calculation is based on the steady-state heat transfer formula combined with non-steady-state correction term, considering the influence of wall heat storage effect and internal and external surface heat transfer coefficient with wind speed change;

[0024] Solar heat gain calculation distinguishes between direct radiation and scattered radiation that enters the room through glass, and attenuates with shading coefficient;

[0025] Internal heat includes personnel heat dissipation, lighting heat dissipation and equipment heat dissipation, among which personnel and lighting heat dissipation are calculated by multiplying fixed power density by use time, and equipment heat dissipation is dynamically adjusted according to load curve shape parameters;

[0026] Air conditioning refrigeration capacity is derived from indoor cooling load demand, and cooling load is equal to the sum of building envelope heat transfer, solar heat gain and internal heat minus the heat taken away by natural ventilation.

[0027] Preferably, the Gaussian process regression proxy model is trained, including:

[0028] ;

[0029] Wherein, is the signal standard deviation, is the The length scale of the input parameters, The total number of parameters, .

[0030] Preferably, the specific steps of the LHS method are: uniformly dividing the value range of each parameter into sub-intervals, ensuring that there is only one sample point in each sub-interval, independently dividing the intervals for the eight parameters, and then recombining the sample index of each dimension by random arrangement.

[0031] Preferably, each group of the generated sample combination contains: the heat transfer coefficient of the outer wall X1, the roof reflectivity X2, the shading coefficient of the outer window X3, the energy efficiency ratio of the air conditioning system X4, the lighting power density X5, the load curve shape parameter X6, the installation angle of the photovoltaic panel X7, and the conversion efficiency of the inverter X8, and their specific values.

[0032] After each group of samples is input into the dynamic simulation model, the model automatically loads the corresponding meteorological data, calculates the building cooling load, air conditioning energy consumption, lighting energy consumption, equipment energy consumption, photovoltaic power generation, and electricity purchase amount hour by hour, and finally accumulates to obtain the total annual net carbon emission amount;

[0033] In the simulation process, a multi-thread parallel computing mechanism is enabled, the samples are distributed to multiple computing cores for synchronous execution, and all simulation results are stored in a structured database to form a complete sample data set containing the input parameter vector and the output carbon emission value.

[0034] Preferably, the calculation formula of the Pearson linear correlation coefficient is:

[0035] ;

[0036] wherein, and are the values of the two parameters in the group of samples, and are the sample means.

[0037] Preferably, the parameter set includes the heat transfer coefficient of the outer wall X1, the roof reflectivity X2, the shading coefficient of the outer window X3, the energy efficiency ratio of the air conditioning system X4, the lighting power density X5, the load curve shape parameter X6, the installation angle of the photovoltaic panel X7, and the conversion efficiency of the inverter X8.

[0038] Preferably, the output final optimized parameter combination and its corresponding expected annual net carbon emission estimate and confidence interval are obtained, the expected annual net carbon emission estimate is taken from the predicted value of the optimized parameter combination on the proxy model, the confidence interval is calculated by the posterior variance of the proxy model, and is expressed as the estimate plus or minus twice the standard deviation.

[0039] Preferably, the sensitivity index of the Sober global sensitivity analysis is defined using variance decomposition, specifically as follows:

[0040] First-order sensitivity index:

[0041] ;

[0042] Second-order interaction sensitivity index:

[0043] ;

[0044] Total effect index:

[0045] ;

[0046] in, Net carbon emissions for the year; For the first One design parameter; Indicates except The set of all other parameters; For conditional expectation operators; For variance operators; for ease of comparison, the first-order exponent is normalized to the present value. The second-order and total effects follow the original definition and are not directly added to the first-order effects for variance closure verification.

[0047] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0048] 1. The method for optimizing design parameters of integrated tropical zero-carbon building systems provided by this invention is the first to construct a three-dimensional comprehensive analysis framework covering input correlation, output uncertainty and global sensitivity, breaking through the limitations of traditional single-dimensional parameter research and realizing a panoramic mechanism analysis of complex coupled parameter systems.

[0049] 2. By introducing a Gaussian process regression surrogate model and setting up a strict accuracy verification mechanism, the problem of lack of transparency and interpretability in black-box analysis of commercial software is effectively solved, ensuring the scientific reliability of the analysis results.

[0050] 3. In response to the unique high temperature and humidity environment and strong coupling characteristics of multiple systems in tropical zero-carbon buildings, this invention systematically identifies the set of dominant parameters and their key interaction terms that have a decisive impact on carbon emissions. For example, it finds that there is a negative interaction effect between roof reflectivity and air conditioning energy efficiency ratio, that is, increasing reflectivity can reduce air conditioning load and thus amplify the emission reduction benefits of high-efficiency equipment. Such insights provide a clear direction for precise targeted optimization.

[0051] 4. The optimization process integrates engineering constraints and algorithm optimization, generating parameter configuration schemes that are both theoretically optimal and engineering practical. Real-world cases show that compared to conventional design schemes, the method can reduce the annual net carbon emissions of tropical zero-carbon buildings, while avoiding the risk of excessive reliance on expensive materials or unfeasible technical paths.

[0052] 5. The invention provides a standardized, quantitative, and intelligent technical support tool for tropical zero-carbon building design from concept to performance, improving design efficiency and outcome quality, and promoting substantial progress in the building sector's dual-carbon goals. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 is a schematic diagram of the overall technical scheme architecture of the integrated system design parameter optimization method for tropical zero-carbon buildings proposed by the invention;

[0054] Figure 2 is a schematic diagram of the core principle framework of the three-dimensional comprehensive analysis framework that integrates input correlation, output uncertainty, and global sensitivity in the invention;

[0055] Figure 3 is a logic flow framework diagram of the building carbon emission dynamic simulation and Gaussian process proxy model construction in the invention;

[0056] Figure 4 is a parameter space exploration and performance uncertainty quantification process framework diagram based on Latin hypercube sampling and Monte Carlo propagation in the invention;

[0057] Figure 5 is a logic framework diagram of Sobol global sensitivity analysis and dominant parameter screening in the invention. DETAILED DESCRIPTION

[0058] REFERENCES Figures 1 to 5 Obtain typical meteorological year hourly temperature, humidity, solar radiation intensity, and wind speed data in tropical regions as external environmental driving variables. The data comes from the standard meteorological year dataset published by the National Meteorological Data Center, covering 8760 time steps throughout the year, with each time step corresponding to 1h of continuous observations.

[0059] Temperature data is recorded in °C, recording air dry bulb temperature and wet bulb temperature; humidity data is provided in the form of relative humidity %; solar radiation intensity is divided into three components: normal direct radiation, scattered radiation, and ground reflected radiation, with units of W / m 2 ; wind speed data records the horizontal wind speed at a height of 10m from the ground, with units of m / s.

[0060] All meteorological parameters were quality controlled and interpolated to ensure no missing values and continuous time series. The meteorological data were used as the boundary conditions of building energy simulation, which were directly input into the subsequent dynamic simulation model to drive the building envelope heat transfer calculation, solar heat gain calculation, natural ventilation potential assessment, and photovoltaic power generation prediction module.

[0061] The set of design parameters to be optimized for building integrated systems was defined, including the heat transfer coefficient of the outer wall, the roof reflectivity, the shading coefficient of the outer window, the energy efficiency ratio of the air conditioning system, the lighting power density, the device use load curve shape parameter, the installation angle of the photovoltaic panel, and the inverter conversion efficiency. The heat transfer coefficient of the outer wall represents the ability of the wall material and structure to hinder heat transfer, and its value range is set to 0.4-0.8 W / (m 2 ·K) according to the current energy-saving design standard and the physical limit of high-performance insulation materials.

[0062] The roof reflectivity is defined as the proportion of solar shortwave radiation reflected by the roof surface, with a value range of 0.2-0.9, corresponding to the spectral characteristics of dark asphalt roof to high-reflective white paint roof. The shading coefficient of the outer window reflects the ability of the window system to inhibit the entry of solar radiation into the room, considering the transmittance of the glass itself and the effect of the external shading component, and is set to 0.2-0.6. The energy efficiency ratio of the air conditioning system refers to the ratio of refrigerating capacity to input power, which is set to 3.0-5.0 according to the current market performance grade of main equipment.

[0063] The lighting power density is divided by function room type, with 10 W / m 2 for office areas, 15 W / m 2 for commercial spaces, and 5 W / m 2 for residential spaces, allowing for floating adjustment within ±20%. The device use load curve shape parameter controls the peak load occurrence period and duration by introducing a scaling factor, with a value range of 0.5-1.5, to simulate the electricity consumption characteristics under different user behavior patterns. The installation angle of the photovoltaic panel is adjusted within ±15° according to the latitude of the building location to maximize the annual solar radiation reception.

[0064] The inverter conversion efficiency represents the energy loss in the process of converting direct current to alternating current, which is set to 92% to 98% according to the current technology level of the photovoltaic industry. The above eight parameters constitute a complete set of design parameters to be optimized, each with a clear physical meaning and engineering adjustability, which together determine the overall carbon emission performance of building integrated systems.

[0065] A dynamic simulation model of building life cycle carbon emissions was constructed, which includes building operation phase energy consumption module, photovoltaic power generation offset module, and grid carbon factor dynamic response module, with the output of annual net carbon emissions.

[0066] The building operation phase energy consumption module uses an energy balance equation to describe the dynamic relationship between the heat transfer of the envelope, solar heat gain, internal heat generation, and air conditioning refrigeration capacity. The time step is set to 1 hour. The envelope heat transfer calculation is based on the steady-state heat transfer formula combined with a non-steady-state correction term, considering the influence of wall heat storage effect and the change of internal and external surface heat transfer coefficient with wind speed.

[0067] The solar heat gain calculation distinguishes between direct radiation and scattered radiation that enters the room through the glass, and combines the shading coefficient for attenuation processing. Internal heat generation includes personnel heat dissipation, lighting heat dissipation, and equipment heat dissipation. Personnel and lighting heat dissipation are calculated by multiplying fixed power density by usage time, and equipment heat dissipation is dynamically adjusted according to load curve shape parameters.

[0068] The photovoltaic power generation deduction module calculates the annual power generation and converts it into carbon emission reduction based on photovoltaic panel area, conversion efficiency, inclination angle, and local solar radiation. The photovoltaic panel conversion efficiency is set to 20%, the area is determined according to the available projection area of the roof, and the power generation is calculated using the inclined surface radiation reception model combined with the inverter conversion efficiency for correction. The grid carbon factor dynamic response module calculates the indirect carbon emissions during operation by multiplying the annual average carbon emission intensity of the regional power grid by the electricity purchase quantity. The carbon emission intensity is set to 0.61 kgCO2e / kWh (based on the average value of Hainan Province in recent years, which can be adjusted according to the region and year).

[0069] Latin hypercube sampling method is used to generate not less than 1000 parameter sample combinations within the preset parameter value range. Each sample is input into the dynamic simulation model to perform a complete annual simulation, and the corresponding annual net carbon emissions output value is obtained. Latin hypercube sampling uniformly divides the value range of each parameter into N sub-intervals, ensuring that there is only one sample point in each sub-interval, thereby achieving full coverage and uniform distribution of the parameter space.

[0070] The sampling process first independently divides the eight parameters into intervals, and then randomly rearranges the sample index of each dimension to avoid artificial correlation between parameters. The 1000 generated sample combinations each contain specific values of external wall heat transfer coefficient, roof reflectivity, external window shading coefficient, air conditioning system energy efficiency ratio, lighting power density, load curve shape parameter, photovoltaic panel installation inclination angle, and inverter conversion efficiency. After each sample is input into the dynamic simulation model, the model automatically loads the corresponding meteorological data, calculates the building cooling load, air conditioning energy consumption, lighting energy consumption, equipment energy consumption, photovoltaic power generation, and electricity purchase quantity hour by hour, and finally accumulates the total annual net carbon emissions.

[0071] Multi-thread parallel computing mechanism is enabled during simulation to shorten the total time consumption. All simulation results are stored in a structured database to form a complete sample dataset containing input parameter vectors and output carbon emission values, providing basic data support for subsequent proxy model training.

[0072] Based on the input and output data of all samples, a Gaussian process regression proxy model is trained. The kernel function is selected as a radial basis function, and the hyperparameters are automatically optimized by maximum likelihood estimation. The model prediction accuracy needs to meet the ten-fold cross-validation RMSE < 5% and R 2 > 0.95. The Gaussian process regression model assumes that the target function obeys the Gaussian process prior distribution, and its covariance structure is defined by the kernel function. In this embodiment, the radial basis function is selected as the kernel function, and its mathematical expression is:

[0073] ;

[0074] wherein, is the signal standard deviation, is the length scale of the th input parameter, is the total number of parameters (in this embodiment, the total number of input parameters is ). During model training, the hyperparameters are automatically optimized by maximizing the marginal likelihood function , so that the model can adaptively learn the scale difference of the influence of different parameters on the output.

[0075] After training, the ten-fold cross-validation is used to evaluate the generalization ability of the model: the sample data is randomly divided into 10 mutually exclusive subsets, and each time 9 subsets are selected as the training set and the remaining 1 subset is selected as the validation set. After repeating 10 times, the root mean square value of the prediction error of all validation sets is calculated. At the same time, the coefficient of determination R 2 measures the proportion of the explained variance of the model.

[0076] If the cross-validation RMSE exceeds 5% or R 2 is less than 0.95, the hyperparameters are reinitialized and the number of training iterations is increased until the accuracy requirements are met. The model finally retains 10% of the original simulation model samples as an independent test set to calculate the relative error distribution between the predicted value and the true value, ensuring that the absolute value of the relative error of more than 95% of the samples does not exceed 8%. The proxy model that meets the accuracy requirements will replace the original simulation model for subsequent large-scale sensitivity analysis and optimization calculation, reducing the computational cost.

[0077] The Pearson linear correlation coefficient between each pair of input parameters is calculated to form a correlation matrix, identifying parameter pairs with strong positive or negative correlation, which are marked as potential coupled interference terms. The calculation formula of the Pearson linear correlation coefficient is:

[0078] ;

[0079] in, and The two parameters are respectively in the first... The values ​​in the group sample, and The mean of the sample is used. Based on 1000 sets of sample data, the eight parameters are combined pairwise to form a symmetric correlation matrix. The diagonal elements of the matrix are always 1, indicating that the parameter is perfectly correlated with itself.

[0080] Calculate the Pearson linear correlation coefficients between each pair of input parameters to form a correlation matrix, which is used for sampling quality checks. Under LHS independent sampling and random rearrangement in this method, the input-input correlations are mostly satisfied. To gain a deeper understanding of the underlying mechanisms, input-output correlations were calculated to identify potential coupling effects: for example, roof reflectivity was significantly negatively correlated with annual net carbon emissions, and the energy efficiency ratio of air conditioning systems was significantly negatively correlated with annual net carbon emissions; these findings corroborated subsequent Sobol first-order / second-order results.

[0081] Using the surrogate model as a fast response function, 100,000 Monte Carlo random samples were performed to statistically analyze the probability density distribution of annual net carbon emissions, extracting its mean, standard deviation, and uncertainty interval characterizing the output performance. The Monte Carlo sampling randomly generated 100,000 parameter combinations within a uniform distribution across eight preset parameter values. Each combination was input into the surrogate model, and the predicted annual net carbon emissions were returned immediately. Because the surrogate model's computation time is extremely low, the 100,000 samples could be completed within minutes.

[0082] After collecting all output results, the central tendency and dispersion of the output values ​​are observed through probability density distribution histograms. The arithmetic mean of the output sequence is calculated as the expected performance index, and the standard deviation characterizes the performance fluctuation. Furthermore, the 5% and 95% quantiles on the cumulative distribution function are extracted to define the confidence intervals of the performance values.

[0083] The Sober global sensitivity analysis method is applied to calculate the first-order sensitivity index of each individual parameter and the second-order interaction sensitivity index of any combination of two parameters. The sensitivity index calculation is based on the principle of variance decomposition, with a sampling scale of no less than 200,000 times. Sober sensitivity analysis quantifies the contribution ratio of each parameter to the total output variance through variance decomposition.

[0084] First-order sensitivity index Defined as fixed parameters only The percentage decrease in output variance reflects the parameter Independent main effects:

[0085] ;

[0086] Second-order interaction sensitivity index defined as fixing parameters simultaneously compared to fixing parameters separately compared to fixing parameters separately or the proportion of additional variance reduction, reflecting the interaction strength between parameters compared to fixing parameters separately compared to fixing parameters separately

[0087] ;

[0088] Total effect index for measuring the total contribution of parameters including all their interactions:

[0089] ;

[0090] where, is the annual net carbon emission; is the i-th design parameter of step S2; represents the set of all parameters except ; is the conditional expectation operator; is the variance operator.

[0091] In this embodiment, 200,000 sets of parameter samples are generated using the quasi-Monte Carlo sampling strategy, and the corresponding output values are quickly calculated on the verified proxy model. By constructing a specific sample matrix, the conditional variance and the unconditional variance are estimated, and then the first-order sensitivity index, the second-order interaction index and the total effect index are solved. For easy comparison, the set of design parameters to be optimized is numbered as follows: X1: external wall heat transfer coefficient; X2: roof reflectivity; X3: external window shading coefficient; X4: air conditioning system energy efficiency ratio; X5: lighting power density; X6: device usage load curve shape parameter; X7: photovoltaic panel installation inclination; X8: inverter conversion efficiency.

[0092] The first-order sensitivity index calculation results are as follows: roof reflectivity , air conditioning system energy efficiency ratio , external window shading coefficient , photovoltaic panel installation inclination , inverter conversion efficiency , external wall heat transfer coefficient , lighting power density , device usage load curve shape parameter .

[0093] The significant items of the second-order interaction sensitivity are:​ (Roof reflectivity x Air conditioner EER), (External window shading coefficient x Equipment load curve shape parameter), (Roof reflectivity x External window shading coefficient). The absolute values of the second-order interaction sensitivity indices of the rest of the parameter combinations are all less than 0.01. The sum of the absolute values of all the second-order interaction sensitivity indices is about 0.24.

[0094] The total effect index is basically consistent with the first-order ranking, and satisfies the property: Roof reflectivity , Air conditioner EER , External window shading coefficient , Photovoltaic panel installation inclination ; the of inverter conversion efficiency, external wall heat transfer coefficient, lighting power density, and equipment load curve shape parameter is 0.04 (X8), 0.02 (X1), 0.01 (X5), and 0.01 (X6), respectively.

[0095] According to the descending order of the first-order sensitivity indices, and according to the cumulative contribution degree of the cumulative contribution degree in turn, the dominant factors are screened: roof reflectivity 0.32 (cumulative 0.32), air conditioner EER 0.28 (cumulative 0.60), external window shading coefficient 0.19 (cumulative 0.79), photovoltaic panel installation inclination 0.13 (cumulative 0.92), inverter conversion efficiency 0.04 (cumulative 0.96), external wall heat transfer coefficient 0.02 (cumulative 0.98), lighting power density 0.01 (cumulative 0.99), and equipment load curve shape parameter 0.01 (cumulative 1.00). The cumulative value first exceeds 80% in the first four items, so the dominant parameter set is: roof reflectivity (X2), air conditioner EER (X4), external window shading coefficient (X3), and photovoltaic panel installation inclination (X7). The key interaction items are screened according to the standard of ||>0.05, and there are three groups . The above interaction items are not only significant in value, but also have clear physical meaning: one embodies the synergistic effect of passive cooling (high-reflectivity roof) and active refrigeration (high-EER air conditioner); the second reflects the difference in the adaptation of shading measures to different use time periods and load intensity scenarios; the third describes the coupling mechanism of the roof and the external window jointly affecting the heat gain of the enclosure. The dominant parameter set and the key interaction items jointly constitute the core object of subsequent optimization, and the rest of the parameters can be temporarily fixed as the benchmark value to simplify the optimization dimension due to their low contribution degree.

[0096] The implementation suggestions for key interactions are as follows: the linkage adjustment of roof reflectivity and air conditioning system energy efficiency ratio should be implemented synchronously to avoid resource mismatching such as "highly reflective roof + low energy efficiency air conditioner"; the linkage of external window shading coefficient and device usage load curve shape parameters needs to be combined with user usage habits and operation period investigation to match the shading strategy with the peak and valley period of load; the combination of roof reflectivity and external window shading coefficient should reduce the peak value of summer enclosure heat gain under the condition of meeting natural lighting.

[0097] All parameter adjustments must not violate the mandatory provisions of the current specifications, including "General Code for Building Energy Conservation and Renewable Energy Utilization" GB55015-2021, "Standard for Energy Saving Design of Public Buildings" GB50189-2023, "Standard for Building Solar Energy Application Technology" GB / T50411-2019, etc. For example, the heat transfer coefficient of external wall must meet the upper limit requirement (such as not greater than 0.40 W / (m2·K)) to ensure the insulation performance. The establishment of the above engineering feasible interval and collaborative control points ensures that the subsequent optimization results have practical landing nature and avoid the theoretical optimal solution from deviating from the engineering constraints.

[0098] Within the engineering feasible adjustment interval, the sequential quadratic programming algorithm is used to search for the optimal combination of parameters that minimizes the annual net carbon emissions. During the optimization process, the proxy model is called in real time to evaluate the objective function, and the convergence criterion is that the relative change amplitude of the objective function is less than 1‰ for 50 consecutive iterations.

[0099] The sequential quadratic programming algorithm is a high-efficiency numerical method suitable for nonlinear constraint optimization problems. Its core idea is to construct a quadratic approximation sub-problem of the original problem and solve it at each iteration, gradually approaching the global optimal solution. In this embodiment, the annual net carbon emissions are set as the objective function, and the roof reflectivity, air conditioning system energy efficiency ratio, and external window shading coefficient are the three dominant parameters to be optimized. The constraint conditions include: roof reflectivity , air conditioning system energy efficiency ratio , and external window shading coefficient .

[0100] The algorithm initialization point is selected from the top 5% sample points in the Latin hypercube sampling, i.e., the 50 groups of parameter combinations with the lowest carbon emissions, and five points are randomly selected as the initial guess value to start the optimization process in parallel. In each iteration, the proxy model replaces the original simulation model to calculate the objective function value and its gradient, and the gradient is approximated by the finite difference method; the Hessian matrix is updated using the BFGS formula of quasi-Newton method to avoid the high cost of directly calculating the second-order derivative.

[0101] The optimization process records the parameter values, objective function values and constraint violation levels of each iteration, and finally outputs the optimal parameter combination after convergence. The final optimized parameter combination and its corresponding expected annual net carbon emission estimate and confidence interval are output as the basis for tropical zero-carbon building design decisions. The expected annual net carbon emission estimate is obtained from the predicted value of the optimized parameter combination on the proxy model. The confidence interval is calculated from the posterior variance of the proxy model and is expressed as the estimate plus or minus twice the standard deviation. The proxy model not only outputs the mean value when predicting, but also provides the standard deviation of the posterior distribution, reflecting the prediction uncertainty.

[0102] The local sensitivity gradient of each dominant parameter near the optimization solution is also output. The gradient information guides subsequent fine-tuning and tolerance design, such as prioritizing investment in energy efficiency improvement when the budget is limited, as it brings the highest emission reduction benefit per unit improvement. All output results are presented in a structured report format, including parameter recommendations, performance predictions, uncertainty intervals, sensitivity gradients and implementation recommendations, directly serving architects, engineers and project decision-makers.

[0103] Comparison results of the embodiments: The present invention uniformly uses the annual net carbon as the caliber, which is determined by the annual purchased electricity and the average emission factor of the power grid, and the photovoltaic power generation is not separately counted as carbon offset. In the case of an office building in Sanya City, Hainan Province, the optimal parameter combination obtained by the present method reduces the annual net carbon emission from 134.2tCO2e to 102.5tCO2e (-26%), the annual electricity purchase from 220MWh to 168MWh (-24%), the peak cooling load from 350kW to 300kW (-14%), and the photovoltaic annual power generation from 95MWh to 98MWh (+3.2%); the proxy model has a ten-fold cross-validation RMSE<5%, R 2 >0.95, and the optimization convergence satisfies the relative change amplitude of the objective function <1‰ for 50 consecutive iterations.

[0104] It should be noted that in this paper, relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the term "includes" "contains" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment.

[0105] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary of the principles and application of the present application. Numerous modifications and adaptions can be effected without departing from the spirit and scope of the present application, which is not limited to the exact construction and arrangement described. It is intended, therefore, to cover all modifications and adaptions that fall within the scope of the claims and their equivalents.

Claims

1. A method for optimizing design parameters of a tropical zero-carbon building integrated system, characterized in that, The method comprises the following steps: Step S1: Obtain the hourly temperature, humidity, solar radiation intensity and wind speed data of a typical meteorological year in a tropical region as external environmental driving variables; Step S2: Define a set of design parameters to be optimized of the building integrated system; Step S3: Build a dynamic simulation model of carbon emissions in the whole life cycle of the building, wherein the model comprises an energy consumption module in the operation stage of the building, a photovoltaic power generation offset module and a dynamic response module of the carbon factor of the power grid, and the output is the annual net carbon emissions; Step S4: Use the Latin hypercube sampling method, i.e. the LHS method, to generate a covering sample within the permissible range of parameters and complete the whole year simulation; Step S5: Based on the input and output data of all samples, train a Gaussian process regression surrogate model, use an anisotropic radial basis kernel, i.e. ARD-RBF, for verification analysis, and set an accuracy threshold for subsequent analysis; Step S6: Calculate the Pearson linear correlation coefficient between each pair of input parameters r , form a correlation matrix, identify the pairs of parameters that have strong positive or strong negative correlation, and mark them as potential coupled interference terms; Step S7: Take the surrogate model as a fast response function, statistically analyze the probability density distribution of the annual net carbon emissions, extract the mean value, standard deviation and uncertainty interval representing the output performance; Step S8: Apply the Sobol global sensitivity analysis method to calculate the first-order sensitivity index of each single parameter and the second-order interaction sensitivity index of any two parameter combinations; According to the descending order of the first-order sensitivity index, all parameters are arranged in descending order, and a dominant parameter set is selected, and key interaction items are screened; Step S9: For the dominant parameter set and the key interaction items, combined with the building specification limit, material market supply capacity and construction process constraint conditions, the engineering feasible adjustment interval of each parameter is determined; Step S10: Output the final optimized parameter combination, the expected annual net carbon emissions estimate and the confidence interval thereof as the basis for decision-making of the tropical zero-carbon building design.

2. The method of claim 1, wherein, The hourly temperature, humidity, solar radiation intensity and wind speed data of a typical meteorological year in a tropical region are obtained, and specifically: The standard meteorological year dataset is obtained from the National Meteorological Data Center, and each time step corresponds to 1 h of continuous observation value. The temperature data records the air dry-bulb temperature and the wet-bulb temperature in units of ℃. The humidity data is provided in the form of relative humidity. The solar radiation intensity is divided into three components: normal direct radiation, scattered radiation, and ground reflected radiation, with units of W / m 2 . The wind speed data records the horizontal wind speed at a height of 10 m from the ground, with units of m / s. All meteorological parameters are quality-controlled and interpolated to ensure that there are no missing values and that the time series is continuous and complete.

3. The method of claim 1, wherein, The energy consumption module in the operation stage of the building uses an energy balance equation to describe the dynamic relationship between the heat transfer of the envelope structure, solar heat gain, internal heat generation and air conditioning refrigeration capacity; The heat transfer calculation of the envelope structure is based on the steady-state heat transfer formula combined with a non-steady-state correction term, considering the influence of wall heat storage effect and internal and external surface heat transfer coefficient changes with wind speed; The solar heat gain calculation distinguishes between direct radiation and scattered radiation that penetrates through the glass into the room, and attenuates the calculation combined with the shading coefficient; The internal heat generation includes personnel heat dissipation, lighting heat dissipation and equipment heat dissipation, wherein the personnel and lighting heat dissipation is calculated by multiplying the fixed power density by the use time, and the equipment heat dissipation is dynamically adjusted according to the load curve shape parameters; The air conditioning refrigeration capacity is obtained by inversely calculating the indoor cooling load demand, and the cooling load is equal to the sum of the heat transfer of the envelope structure, the solar heat gain and the internal heat generation minus the heat removed by natural ventilation.

4. The method of claim 1, wherein, The Gaussian process regression surrogate model is trained, including: ； wherein, is the signal standard deviation, is the number of parameters, is the length scale of the input parameter, is the total number of parameters, .

5. The method of claim 1, wherein, The specific steps of the LHS method are: uniformly dividing the value interval of each parameter into sub-intervals, ensuring that there is and only one sample point in each sub-interval, independently dividing the intervals for the eight parameters, and then recombining the sample indexes of each dimension by random arrangement.

6. The method of claim 1, wherein, Each combination of the generated samples contains: external wall heat transfer coefficient X1, roof reflectivity X2, external window shading coefficient X3, air conditioning system energy efficiency ratio X4, lighting power density X5, load curve shape parameter X6, photovoltaic panel installation inclination X7 and inverter conversion efficiency X8 and their specific values; After inputting each sample into the dynamic simulation model, the model automatically loads the corresponding meteorological data, calculates the building cooling load, air conditioning energy consumption, lighting energy consumption, equipment energy consumption, photovoltaic power generation and electricity purchase quantity hour by hour, and finally accumulates the total annual net carbon emission. During the simulation process, a multi-thread parallel computing mechanism is enabled, the samples are allocated to multiple computing cores for synchronous execution, and all simulation results are stored in a structured database to form a complete sample dataset containing input parameter vectors and output carbon emission values.

7. The method of claim 1, wherein, Pearson linear correlation coefficient The formula for calculating the Pearson linear correlation coefficient is: ； wherein and are the values of the two parameters in the first group of samples, and are the sample means.

8. The method of claim 1, wherein, The parameter set covers the heat transfer coefficient of the outer wall X1, the roof reflectivity X2, the shading coefficient of the outer window X3, the energy efficiency ratio of the air conditioning system X4, the lighting power density X5, the load curve shape parameter X6, the installation angle of the photovoltaic panel X7, and the inverter conversion efficiency X8.

9. The method of claim 1, wherein, The output final optimized parameter combination and its corresponding expected annual net carbon emission estimate and confidence interval are obtained. The expected annual net carbon emission estimate is obtained from the predicted value of the optimized parameter combination on the surrogate model, and the confidence interval is calculated by the posterior variance of the surrogate model, expressed as the estimate plus or minus twice the standard deviation.

10. The method of claim 1, wherein, The sensitivity index of the Sobol global sensitivity analysis is defined by variance decomposition, specifically: First-order sensitivity index: ; Second-order interaction sensitivity index: ； Total effect index: ； in, Net carbon emissions for the year; For the first One design parameter; Indicates except The set of all other parameters; For conditional expectation operators; For variance operators; for ease of comparison, the first-order exponent is normalized to the present value. The second-order and total effects follow the original definition and are not directly added to the first-order effects for variance closure verification.

Images