Knowledge-guided building performance optimization method and system adapted to extreme working conditions

Abstract

The application discloses a knowledge-guided building performance optimization method and system suitable for extreme working conditions, which decouples building energy consumption into two independent optimization targets of extreme refrigeration energy consumption in summer and extreme heating energy consumption in winter, firstly constructs a double-target data set through parameterized modeling and extreme climate directional sampling, and trains a proxy model as an energy consumption prediction engine; secondly, prior knowledge is embedded in the process of executing a genetic algorithm for multi-objective optimization, and a Pareto optimal solution set is obtained; finally, the Pareto optimal solution set is deconstructed reversely to generate elastic control intervals of each building form control parameter and parameter interlinkage compensation rules. Through the seasonal directional double-target collaborative optimization, the genetic algorithm reconstruction driven by prior knowledge and the reverse form control and dynamic compensation mechanism, the efficiency and engineering applicability of building energy consumption optimization under extreme climate working conditions are effectively improved, and a reliable form control scheme is provided for energy-saving design of public buildings in hot summer and cold winter areas.

Description

Technical Field

[0001] This invention relates to the field of building energy-saving design technology, specifically to a knowledge-guided method and system for optimizing building performance under extreme conditions. Background Technology

[0002] To reduce energy consumption in public buildings during operation and achieve green building design goals, existing technologies typically combine building performance simulation with optimization algorithms. Typical existing technical solutions mainly include the following:

[0003] (1) Architectural design method based on empirical rules: relying on architects' empirical rules to conduct preliminary design of morphological parameters such as building shape and window-to-wall ratio, or determining parameters based on design standards for a single season (such as summer heat protection or winter insulation), lacking quantitative multi-objective collaborative design methods.

[0004] (2) Data-driven surrogate model prediction method: Using machine learning models (such as neural networks, random forests, etc.) as black box surrogate models to fit the mapping relationship between building form parameters and energy consumption indicators, so as to replace physical simulation software for rapid energy consumption assessment.

[0005] (3) Parametric optimization method based on heuristic algorithm: In the parametric design platform, heuristic algorithms (such as particle swarm algorithm, simulated annealing algorithm, genetic algorithm, etc.) are combined to search for building form parameters in order to find a scheme with lower annual energy consumption.

[0006] The methods described above each have their own applicable scenarios in practical engineering applications. However, when these conventional techniques are used to optimize building forms under extreme climatic conditions, their internal search mechanisms still rely on mathematical randomness, making it difficult to balance optimization efficiency and engineering applicability. Summary of the Invention

[0007] The purpose of this application is to provide a knowledge-guided building performance optimization method and system that can adapt to extreme working conditions in summer and winter, while taking into account both optimization efficiency and engineering applicability.

[0008] The technical solution provided in this application is: Firstly, this application provides a knowledge-guided building performance optimization method adapted to extreme working conditions, including: Step S1: Determine the building form control parameters, extract summer extreme heat wave meteorological data and winter extreme cold wave meteorological data, and obtain multiple sets of building form control parameter combinations and their corresponding summer extreme cooling energy consumption and winter extreme heating energy consumption through parametric modeling and sampling, and construct a dual-objective dataset. Step S2: Based on the dual-objective dataset, train a proxy model to establish a nonlinear mapping relationship between the building form control parameters and the extreme cooling energy consumption in summer and the extreme heating energy consumption in winter. The proxy model is used to output the corresponding predicted values ​​of extreme cooling energy consumption in summer and extreme heating energy consumption in winter for any combination of input form parameters. Step S3: Using the surrogate model as the fitness evaluation engine, and minimizing the extreme summer cooling energy consumption and the extreme winter heating energy consumption as the dual optimization objectives, a genetic algorithm is used to search for the Pareto optimal solution set; prior knowledge is embedded during the execution of the genetic algorithm, and the prior knowledge includes at least one of the following: prior knowledge based on the empirical distribution of building thermal performance for population initialization, prior knowledge based on energy consumption-morphological attribution mapping for directed mutation, and prior knowledge based on building energy-saving design specifications for feasible region projection; Step S4: Dimensionality reduction and sensitivity analysis are performed on the Pareto optimal solution set to obtain the contribution information of each building form control parameter to the energy consumption boundary, and the elastic control range of each building form control parameter and the linkage compensation rules between parameters are generated as the output of the building form control information.

[0009] In one possible implementation, in step S2, the proxy model employs a deep neural network, and its training process includes: Introducing an adaptive weighted loss function Perform parameter optimization, among which The adaptive weighted loss function value measures the overall level of weighted error between the surrogate model's predicted values ​​and the true values, and serves as the objective function for optimizing model parameters. The total number of training samples, For the first The adaptive weight coefficients for each sample are adjusted adaptively based on the sample residuals. Samples with larger residuals are assigned lower weights to reduce gradient interference from local outlier simulation samples during model training; samples with smaller residuals are assigned higher weights to enhance the contribution of mainstream distribution samples to model accuracy. For the first The actual energy consumption value of each sample For the first The model predicts energy consumption values ​​for each sample.

[0010] In one possible implementation, K-fold cross-validation and Bayesian optimization algorithms can be combined to automatically optimize the number of network layers, learning rate, and number of hidden layer nodes of the deep neural network globally.

[0011] In one possible implementation, step S3, the prior for population initialization based on the empirical distribution of building thermal performance, specifically includes: This scheme breaks away from the standard genetic algorithm's globally uniform distribution initialization method and constructs a multivariate Gaussian distribution model based on building thermal priors: ; This formula represents: the initial population of individuals probability distribution Obey As the mean vector, with It is a multivariate Gaussian distribution (multivariate normal distribution) of the covariance matrix.

[0012] in, The vector represents the initial population of individuals, with each dimension corresponding to a building form control parameter; the mean vector... The combination of building form control parameters corresponding to the empirically optimal solution; covariance matrix It is a diagonal matrix, and its diagonal elements control the exploration step size of the corresponding building form control parameters; Population sampling is performed using the aforementioned multivariate Gaussian distribution model to generate an initial population, which is then concentrated within a solution space region with optimization potential.

[0013] In one possible implementation, step S3, specifically the prior for directional variation based on the energy consumption-morphological attribution mapping, includes: Constructing an energy consumption anomaly-morphological attribution prior mapping matrix The mapping matrix records the direction and intensity of the influence between each building form control parameter and each energy consumption target; During the mutation phase, if the predicted extreme winter heating energy consumption or the predicted extreme summer cooling energy consumption of the current individual is worse than the population mean, then according to the mapping matrix... Identify the dominant morphological parameter that triggers the energy consumption degradation state, and assign the dominant morphological parameter a higher probability of variation than other parameters; The gradient direction of the dominant morphological parameter along the variable asynchronous length is adjusted with respect to the energy consumption target, and the variation formula is as follows: ; in, Individual building form control parameter values; For the mutated first Individual building form control parameter values; This is the step size control value, used to control the adjustment range of a single variation; Energy consumption objective function For the The partial derivative of a building form control parameter represents the direction and sensitivity of the effect of the parameter change on energy consumption. The sign function takes +1 when the partial derivative is positive, -1 when it is negative, and 0 when it is zero. It is used to ensure that the mutation always proceeds in the direction of decreasing energy consumption.

[0014] Among them, the energy consumption objective function , The objective function for extreme cooling energy consumption in summer; The objective function is the energy consumption for extreme winter heating.

[0015] If the predicted extreme winter heating energy consumption of a current individual is worse than the population mean, then The partial derivative is ; If the predicted extreme summer cooling energy consumption of a current individual is worse than the population mean, then The partial derivative is .

[0016] In one possible implementation, step S3, the prior for projecting the feasible region based on building energy efficiency design specifications, specifically includes: After executing the crossover and / or mutation operators of the genetic algorithm, a nonlinear constraint discriminator based on building energy efficiency design specifications is invoked to determine whether the current individual lies within the strictly feasible closed set defined by the specifications. Inside; If the current individual If the location is outside the closed set of the feasible region, then a forced projection operation is performed: ; in, For infeasible individuals In the feasible region closed set The projection point on the graph serves as a legitimate alternative for the individual to participate in subsequent non-dominated sorting; This is the vector of infeasible individuals generated after crossover and mutation operations, with each dimension corresponding to the control parameters of each building form. This is a strictly feasible closed set as defined in the building energy efficiency design code; For the feasible closed set Any feasible individual vector within; for and The Euclidean distance (L2 norm) between them is used to measure the degree of deviation between the infeasible solution and each compliant point within the closed set of the feasible region; For minimization operators, represents the closed set in the feasible region. Find the compliance point that minimizes the Euclidean distance. The projection operation will exclude infeasible individuals. Mapped to the feasible region closed set At the compliance boundary point closest to its Euclidean distance, ensure that all individuals participating in subsequent non-dominated sorting meet engineering specification requirements.

[0017] Will Projected onto the feasible region closed set At the nearest compliance boundary, the projection result is used as the legitimate individual participating in the subsequent non-dominated ranking.

[0018] In one possible implementation, step S4, which involves performing dimensionality reduction and sensitivity analysis on the Pareto optimal solution set to obtain the contribution information of each building form control parameter to the energy consumption boundary, specifically includes: Using partial correlation analysis, the partial correlation coefficients between each building form control parameter in the Pareto optimal solution set and the extreme cooling energy consumption in summer and the extreme heating energy consumption in winter are calculated. The absolute value of the partial correlation coefficient represents the independent contribution of each building form control parameter to the corresponding energy consumption target—the closer the absolute value is to 1, the stronger the independent influence of the parameter on energy consumption; the closer the absolute value is to 0, the more insensitive the parameter is. The sign of the partial correlation coefficient represents the direction of the parameter's influence on the energy consumption target—a positive sign indicates that the parameter value is positively correlated with energy consumption (the larger the parameter, the higher the energy consumption), and a negative sign indicates that the parameter value is negatively correlated with energy consumption (the larger the parameter, the lower the energy consumption). Based on the partial correlation coefficients of various building form control parameters, horizontal comparisons and sign comparisons are performed on each parameter (comparing the absolute values ​​of parameters and the signs of energy consumption targets). A contribution matrix representing the distribution of collaborative weights among parameters is constructed: parameters with larger absolute values ​​of partial correlation coefficients are assigned higher weights in the contribution matrix, representing their dominant control over the energy consumption boundary; when the signs of the partial correlation coefficients of the same parameter for summer cooling energy consumption and winter heating energy consumption are opposite, it indicates that there is a physical conflict between the two seasons, and the parameter is assigned priority in the collaborative weights in the contribution matrix (that is, which parameter is fundamentally conflicting in terms of physical demand in the two seasons, is the focus of the entire optimization problem, and is given priority in the dynamic compensation mechanism).

[0019] Preferably, to further enhance the comprehensiveness of the sensitivity analysis, the partial correlation coefficients can be normalized to weights. This normalized weight reflects the proportion of each building form control parameter's contribution to the energy consumption target, serving as a global sensitivity indicator. Represents the control parameter of the j-th building form. energy consumption objective function The normalized weight, with a value range of [0,1], represents the parameter's contribution to the energy consumption objective function among all n parameters. The higher the weight of the parameter in the energy consumption objective function, the greater its contribution percentage. The stronger the dominant control role; Represents the control parameter of the j-th building form. With energy consumption objective function The partial correlation coefficient, after controlling for (eliminating) the influence of all other building form control parameters, is in the range [−1, 1]: a positive value indicates a positive correlation (the larger the parameter value, the higher the energy consumption), and a negative value indicates a negative correlation (the larger the parameter value, the lower the energy consumption). A contribution matrix is ​​constructed based on normalized weights.

[0020] In one possible implementation, step S4, where the inversion generates the flexible control range of various building form control parameters, specifically includes: For the frontier of the Pareto optimal solution set, an energy consumption tolerance threshold is set, and Pareto solutions with energy consumption degradation less than or equal to the energy consumption tolerance threshold are selected to form a qualified solution subset. Based on the value distribution of each building form control parameter in the qualified solution subset, the confidence interval of each building form control parameter is mapped and inverted, and the upper and lower limits of the confidence interval are used as the elastic control boundaries of each building form control parameter.

[0021] In one possible implementation, in step S4, the inter-parameter linkage compensation rule is generated in the following manner: Construct the Jacobian matrix of each building form control parameter with respect to the energy consumption target within the Pareto optimal solution set: ; in Let be a Jacobian matrix, and its element in the i-th row and j-th column be... Let represent the partial derivative of the j-th building form control parameter with respect to the i-th energy consumption target; The objective function for extreme cooling energy consumption in summer; The objective function for extreme winter heating energy consumption; There are n building form control parameters, where n is the total number of building form control parameters; When any building form control parameter When the deviation from its optimal threshold boundary causes a poorly-conditioned shift in the objective function, the compensation gradient vector is calculated based on the analytical solution of the Jacobian matrix: ; in, For the Jacobian matrix The generalized inverse matrix (Moore-Penrose pseudo-inverse). This is the bias vector of the objective function caused by parameter deviation. To compensate for the gradient vector, Theoretical compensation amounts are included for each building form control parameter; according to Determine at least one compensation parameter and its compensation amount that require coordinated reverse adjustment in order to maintain energy consumption within the Pareto optimal frontier range within the elastic control range.

[0022] In one possible implementation, in step S1, the building form control parameters include macroscopic geometric attributes, spatial composition attributes, and interface characteristics; the macroscopic geometric attributes include building volume, building height, and shape coefficient; the spatial composition attributes include floor area ratio and building shape index; and the interface characteristics include window-to-wall ratio for each orientation and building orientation angle. The parametric modeling and sampling specifically involves generating multiple sets of building form control parameter combinations through Latin hypercube sampling, and using an energy consumption simulation engine to calculate the extreme summer cooling energy consumption and extreme winter heating energy consumption corresponding to each set of building form control parameter combinations.

[0023] Secondly, this application provides a knowledge-guided building performance optimization system adapted to extreme working conditions, comprising: The dual-objective dataset acquisition module is used to determine building form control parameters, extract extreme summer heat wave meteorological data and extreme winter cold wave meteorological data, and obtain multiple sets of building form control parameter combinations and their corresponding extreme summer cooling energy consumption and extreme winter heating energy consumption through parametric modeling and sampling, thereby constructing a dual-objective dataset. The proxy model construction module is used to train a proxy model based on the dual-objective dataset to establish a nonlinear mapping relationship between the building form control parameters and the extreme cooling energy consumption in summer and the extreme heating energy consumption in winter. The proxy model is used to output the corresponding predicted values ​​of extreme cooling energy consumption in summer and extreme heating energy consumption in winter for any combination of input form parameters. A prior knowledge-driven multi-objective optimization module is used to use the surrogate model as a fitness evaluation engine, with the minimization of extreme summer cooling energy consumption and extreme winter heating energy consumption as dual optimization objectives, and to search for Pareto optimal solutions using a genetic algorithm. Prior knowledge is embedded during the execution of the genetic algorithm, and the prior knowledge includes at least one of the following: prior knowledge for population initialization based on the empirical distribution of building thermal performance, prior knowledge for directional mutation based on energy consumption-morphological attribution mapping, and prior knowledge for feasible region projection based on building energy-saving design specifications. The reverse deconstruction and control information generation module is used to perform dimensionality reduction and sensitivity analysis on the Pareto optimal solution set, obtain the contribution information of each building form control parameter to the energy consumption boundary, and generate the elastic control range of each building form control parameter and the linkage compensation rules between parameters as the output of building form control information.

[0024] The specific implementation of the second aspect of this application can refer to the implementation of the first aspect, and will not be elaborated here.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] (1) Break through the “annual compromise” and achieve real extreme climate adaptation.

[0027] This invention decouples building energy consumption in hot-summer and cold-winter regions into two independent optimization objectives: extreme cooling energy consumption in summer and extreme heating energy consumption in winter, which are mutually interdependent. Through a seasonally oriented dual-objective collaborative optimization architecture, it directly addresses the physical contradiction between "heat gain" and "heat protection" in summer and winter. The resulting optimized building form exhibits significantly better real-world energy consumption performance and climate resilience when facing extreme high temperatures and extreme cold waves, avoiding the performance imbalance between summer and winter caused by existing technologies that use "total annual energy consumption" as a single objective.

[0028] (2) A leapfrog improvement in algorithm performance and engineering compliance.

[0029] This invention reconstructs the core mechanism of the standard genetic algorithm by embedding prior knowledge of buildings, specifically including: preference seeding initialization based on the probability distribution of building thermal experience, directional mutation operation based on physical causal partial derivative mapping, and forced mathematical projection of the feasible region based on the closed set of engineering energy-saving specifications. Thus, the optimization algorithm starts with an empirical baseline, the mutation process can be targeted and corrected, and the output results have regulatory barriers. This not only saves huge amounts of unnecessary computational overhead but also eliminates the distorted solutions that traditional algorithms often output, which are "floating on theory and unable to be implemented," ensuring that the optimization results are 100% compliant with building design specifications and achieving highly efficient search for compliant solutions.

[0030] (3) Upgrade "forward optimization" to "reverse morphological control and dynamic compensation" which has great practical value.

[0031] This invention delves into the coupling mechanism of various morphological control parameters within the Pareto optimal solution set. Using partial correlation analysis and sensitivity indices, it inversely extracts the elastic confidence intervals of each architectural morphological control parameter, thus defining a safe "morphological design boundary (threshold band)" for architects. Furthermore, when any design parameter is forced to exceed the preset optimal threshold boundary due to architectural function or aesthetic requirements, the system can instantly calculate the set of compensation parameters requiring coordinated reverse adjustment and their precise compensation gradient based on a multi-parameter dynamic linkage compensation mechanism derived from the Jacobian matrix. This mechanism grants architects ample creative freedom while ensuring that the overall energy consumption of the building remains within an acceptable range, achieving a deep integration of architectural aesthetic design and engineering performance rationality. Attached Figure Description

[0032] Figure 1 This is a flowchart of a method in one embodiment of this application; Figure 2 This is a flowchart of the kernel of a prior knowledge-driven heuristic algorithm in one embodiment of this application; Figure 3 This is a logic diagram of the dynamic reverse control and compensation mechanism for building form control parameters in one embodiment of this application. Detailed Implementation

[0033] To enable those skilled in the art to better understand the present application, the technical solution of the present application will be further described in detail below with reference to the embodiments and accompanying drawings.

[0034] Example 1: like Figure 1 As shown, this application provides a knowledge-guided building performance optimization method adapted to extreme working conditions. Its core lies in: abandoning the linear logic of "total annual energy consumption" and proposing "seasonal targeted" optimization; using a high-precision surrogate model as an accelerated evaluation engine; deeply modifying the standard genetic algorithm; and ultimately achieving intelligent reverse control and dynamic compensation of key building form indicators.

[0035] Method and steps: Step 1: Obtain a dual-objective dataset of building form control parameters and extreme climate energy consumption.

[0036] This step is achieved through parametric modeling of building form control and targeted sampling for extreme climates, including: Determine key building form control parameters, including: macro geometric attributes (building volume, height, shape coefficient), spatial composition attributes (floor area ratio, building shape index, BSI), and interface characteristics (window-to-wall ratio, WWR for each orientation, building orientation angle).

[0037] Extract real meteorological data of extreme summer heat wave weeks and extreme winter cold wave weeks in a typical year.

[0038] Multiple sets of morphological parameters are generated through Latin hypercube sampling, and the extreme summer cooling energy consumption corresponding to each set of morphological parameters is independently calculated using an energy consumption simulation engine. ) and extreme heating energy consumption in winter ( ), and construct a dataset for the model to learn from.

[0039] Step 2: Construct and train a dual-objective energy consumption prediction proxy model.

[0040] This step involves building and training a high-precision, accelerated evaluation surrogate model, specifically including: Constructing nonlinear mapping relationships ,in It is a high-dimensional architectural form feature vector. This represents the energy consumption response matrix under extreme weather conditions. To overcome the computational bottleneck of excessively long traditional physical simulations, this invention employs a deep neural network to establish a surrogate model. During the training phase, to prevent overfitting and improve robustness to extreme values, an adaptive weighted loss function is introduced to reduce gradient interference from local abnormal simulation samples. Simultaneously, K-fold cross-validation and Bayesian optimization algorithms are combined for global automatic optimization of core hyperparameters such as the number of network layers, learning rate, and number of hidden layer nodes. This ensures that the surrogate model possesses millisecond-level inference capabilities while strictly controlling the generalization error within engineering-allowable thresholds, thus providing a low-latency, high-fidelity dual-objective fitness evaluation engine for subsequent large-scale multi-objective heuristic searches.

[0041] Step 3: Heuristic multi-objective optimization based on prior knowledge embedding.

[0042] Standard genetic algorithms (such as NSGA-II) often get bogged down in inefficient random Markov chain walks due to a lack of domain knowledge guidance. This method reconstructs its kernel structure using both mathematical and physical principles, constructing a domain-specific heuristic search framework: First, establish the two extreme opposing objective functions: construct a multi-objective optimization problem. The constraints are ,in This is the morphological parameter compliance domain.

[0043] Secondly, prior knowledge is embedded: "Experience Seeding" Preferential Initialization (Prior Embedding 1): Breaking the Global Uniform Initialization of Standard Genetic Algorithms Construct a multivariate Gaussian distribution model based on building thermal priors. , where the mean vector Anchored to empirically sound solutions (e.g., empirical values ​​for "south-facing window-to-wall ratio, compact form factor"), the covariance matrix The system controls the exploration step size of the search domain. It forces the initial population to be "seeded" using this probability density function, enabling the algorithm to converge to a solution space with high optimization potential in the early stages of iteration.

[0044] Knowledge-driven "directed mutation" operator (prior embedding 2): Completely abandons the purely random perturbation mutation mode of traditional algorithms. The system has a built-in "energy consumption anomaly-morphological attribution" prior mapping matrix. During the mutation phase, the algorithm continuously evaluates the skewness of the objective function of the current individual; if a certain offspring... Significantly inferior to the population mean, mutation operator According to For specific characteristic locations that cause this adverse condition (such as the ratio of south-facing window wall to wall), This assigns an exponentially amplified mutation probability and applies a step size adjustment along the gradient optimization direction, i.e. This "targeted" heuristic leap leads to an exponential increase in convergence efficiency.

[0045] Standardized feasible region forced projection (prior embedding 3): After the crossover and mutation operators of the genetic algorithm are executed, a nonlinear constraint discriminator based on building energy efficiency design specifications is cascaded. This addresses infeasible solution vectors that exceed boundaries or violate physical principles. The algorithm does not adopt a simple discard strategy, but performs a mathematically forced projection operation. ,in This step involves creating a strictly feasible closed set according to the building energy efficiency design code. It forcibly pulls free parameters back to the nearest compliance boundary, ensuring that the Pareto offspring participating in the non-dominated sorting and output are 100% legally valid at the engineering level. The building energy efficiency design code can be the "Standard for Energy Efficiency Design of Public Buildings" (GB 50189-2015). It can be defined according to the various mandatory clauses in the "Standard for Energy-Saving Design of Public Buildings".

[0046] Step 4: Perform reverse deconstruction on the Pareto optimal solution set to generate building form control information: This step involves reverse deconstruction of the architectural form and dynamic compensation control, specifically including: Pareto solution set feature dimensionality reduction and sensitivity analysis: Extract the Pareto optimal frontier solution set that tends to converge in step three. Using partial correlation analysis and global sensitivity index, quantify the independent contribution and collaborative weight matrix of each combination of high-dimensional morphological parameters in the solution set to the winter / summer optimal energy consumption boundary, and complete the black-box deconstruction from performance results to geometric morphology.

[0047] Form control safety threshold zone delineation: overcoming the limitations of outputting a single "optimal determined value" to users. For the Pareto front, the algorithm maps and inverses the confidence intervals of multidimensional parameters that satisfy the preset energy consumption tolerance. In the form of a "threshold zone," it delineates the flexible engineering control boundaries of various parameters (such as form factor, window ratio, and width-to-depth ratio), giving architects room for creative design.

[0048] Multi-dimensional feature linkage and dynamic compensation matrix generation: This involves in-depth exploration of the nonlinear compensation manifold between parameters within the Pareto solution set. By constructing the partial derivative matrix of morphological parameters, quantified dynamic linkage guidelines are output. Specifically: when a certain architectural form control parameter (set as a feature) is affected by aesthetic or functional requirements of the building form... When the system is forced to exceed the preset optimal threshold boundary, causing a poorly-conditioned shift in the objective function, it will automatically calculate and output the set of compensation parameters (such as features) that need to be adjusted in the opposite direction based on the analytical solution of the Jacobian matrix. ) and its precise compensation gradient (such as synchronous reduction of body coefficient) (Value). This mechanism ensures that the overall energy consumption performance remains within the Pareto optimal frontier while achieving the ultimate fusion of architectural aesthetics and engineering performance rationality.

[0049] Example 2: This embodiment provides a knowledge-guided building performance optimization system adapted to extreme working conditions, including: a dual-objective dataset acquisition module, a proxy model construction module, a prior knowledge-driven multi-objective optimization module, and a reverse deconstruction and control information generation module.

[0050] The dual-objective dataset acquisition module is used to determine building form control parameters, extract extreme summer heat wave meteorological data and extreme winter cold wave meteorological data, and obtain multiple sets of building form control parameter combinations and their corresponding extreme summer cooling energy consumption and extreme winter heating energy consumption through parametric modeling and sampling, thus constructing a dual-objective dataset. This module can achieve dual-mode sampling of extreme climate patterns and energy consumption.

[0051] The surrogate model construction module is used to train a surrogate model based on a dual-objective dataset to establish a nonlinear mapping relationship between building morphological control parameters and extreme summer cooling energy consumption and extreme winter heating energy consumption, respectively. The surrogate model outputs predicted values ​​for extreme summer cooling energy consumption and extreme winter heating energy consumption for any combination of input morphological parameters. This module utilizes sampled data to train an efficient machine learning surrogate model, replacing time-consuming energy consumption simulation software, and achieving second-level dual-objective energy consumption assessment for any input morphological parameters.

[0052] A prior knowledge-driven multi-objective optimization module uses a surrogate model as the fitness evaluation engine and minimizes both extreme summer cooling energy consumption and extreme winter heating energy consumption as dual optimization objectives. It employs a genetic algorithm to search for the Pareto optimal solution set, embedding prior knowledge during the execution of the genetic algorithm. This prior knowledge includes at least one of the following: prior knowledge based on empirical distribution of building thermal performance for population initialization; prior knowledge based on energy consumption-morphological attribution mapping for directed mutation; and prior knowledge based on building energy-saving design codes for feasible region projection. Accordingly, this module can be divided into an "experience seeding" population initialization subunit, a knowledge-driven "directed mutation" subunit, and a code-based "feasible region projection" subunit, used to search for the dual-objective Pareto front of summer cooling energy consumption and winter heating energy consumption in parallel, efficiently, and compliantly.

[0053] The reverse deconstruction and control information generation module is used to perform dimensionality reduction and sensitivity analysis on the Pareto optimal solution set, obtain the contribution information of each building form control parameter to the energy consumption boundary, and generate the elastic control range of each building form control parameter, as well as the linkage compensation rules between parameters, as the output of building form control information.

[0054] This application can clearly separate the energy consumption contradiction under extreme winter and summer climates; by deeply embedding the building physics mechanism and design specification prior knowledge into the non-dominated sorting genetic algorithm (NSGA-II), the search logic of the algorithm is reconstructed from the bottom layer to greatly improve efficiency; and finally, it realizes the leap from a single "energy consumption target solution" to "dynamic reverse control of building form control parameters".

[0055] Obviously, those skilled in the art should understand that the various units or steps of this application described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device, or fabricating them separately as individual integrated circuit modules, or fabricating multiple modules or steps into a single integrated circuit module. Thus, this application is not limited to any particular combination of hardware and software.

[0056] The above are merely preferred embodiments of this application and are not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A knowledge-guided building performance optimization method adapted to extreme working conditions, characterized in that, include: Step S1: Determine the building form control parameters, extract summer extreme heat wave meteorological data and winter extreme cold wave meteorological data, and obtain multiple sets of building form control parameter combinations and their corresponding summer extreme cooling energy consumption and winter extreme heating energy consumption through parametric modeling and sampling, and construct a dual-objective dataset. Step S2: Based on the dual-objective dataset, train a proxy model to establish a nonlinear mapping relationship between the building form control parameters and the extreme cooling energy consumption in summer and the extreme heating energy consumption in winter. The proxy model is used to output the corresponding predicted values ​​of extreme cooling energy consumption in summer and extreme heating energy consumption in winter for any combination of input form parameters. Step S3: Using the surrogate model as the fitness evaluation engine, and minimizing the extreme cooling energy consumption in summer and the extreme heating energy consumption in winter as the dual optimization objectives, a genetic algorithm is used to search for the Pareto optimal solution set. Prior knowledge is embedded during the execution of the genetic algorithm. The prior knowledge includes: prior knowledge for population initialization based on the empirical distribution of building thermal performance, prior knowledge for directional mutation based on energy consumption-morphological attribution mapping, and prior knowledge for feasible domain projection based on building energy-saving design specifications. Step S4: Dimensionality reduction and sensitivity analysis are performed on the Pareto optimal solution set to obtain the contribution information of each building form control parameter to the energy consumption boundary, and the elastic control range of each building form control parameter and the linkage compensation rules between parameters are generated as the output of the building form control information.

2. The method according to claim 1, characterized in that, In step S2, the proxy model employs a deep neural network, and its training process includes: Introducing an adaptive weighted loss function Perform parameter optimization, among which The value of the adaptive weighted loss function. The total number of training samples, For the first The adaptive weighting coefficients for each sample are adaptively adjusted based on the corresponding sample residuals. For the first The actual energy consumption value of each sample For the first The model predicts energy consumption values ​​for each sample.

3. The method according to claim 1, characterized in that, In step S3, the prior for population initialization based on empirical distribution of building thermal performance specifically includes: Construct a multivariate Gaussian distribution model based on building thermal priors: ； This formula represents: the initial population of individuals probability distribution Obey As the mean vector, with Let be a multivariate Gaussian distribution of the covariance matrix, where The vector represents the initial population of individuals, with each dimension corresponding to a building form control parameter; the mean vector... The combination of building form control parameters corresponding to the empirically optimal solution; covariance matrix It is a diagonal matrix, and its diagonal elements control the exploration step size of the corresponding building form control parameters; Population sampling is performed using the aforementioned multivariate Gaussian distribution model to generate an initial population, which is then concentrated within a solution space region with optimization potential.

4. The method according to claim 1, characterized in that, In step S3, the prior for directional variation based on the energy consumption-morphological attribution mapping specifically includes: Constructing an energy consumption anomaly-morphological attribution prior mapping matrix The mapping matrix records the direction and intensity of the influence between each building form control parameter and each energy consumption target; During the mutation phase, if the predicted extreme winter heating energy consumption or the predicted extreme summer cooling energy consumption of the current individual is worse than the population mean, then according to the mapping matrix... Identify the dominant morphological parameter that triggers the energy consumption degradation state, and assign the dominant morphological parameter a higher probability of variation than other parameters; The gradient direction of the dominant morphological parameter along the variable asynchronous length is adjusted with respect to the energy consumption target, and the variation formula is as follows: ； in For the current individual's first Individual building form control parameter values; For the mutated first Individual building form control parameter values; This is the step size control quantity; Energy consumption objective function For the Partial derivatives of each architectural form control parameter; It is a symbolic function.

5. The method according to claim 1, characterized in that, In step S3, the prior knowledge of projecting the feasible region based on building energy-saving design specifications specifically includes: After executing the crossover and / or mutation operators of the genetic algorithm, a nonlinear constraint discriminator based on building energy efficiency design specifications is invoked to determine whether the current individual lies within the strictly feasible closed set defined by the specifications. Inside; If the current individual If the location is outside the closed set of the feasible region, then a forced projection operation is performed: ； in, For infeasible individuals In the feasible region closed set Projection points on; This is the vector of infeasible individuals generated after crossover and mutation operations, with each dimension corresponding to the control parameters of each building form. This is a strictly feasible closed set as defined in the building energy efficiency design code; For the feasible closed set Any feasible individual vector within; for and The Euclidean distance between them; For minimization operators, represents the closed set in the feasible region. Find the compliance point that minimizes the Euclidean distance. The projection operation will exclude infeasible individuals. Mapped to the feasible region closed set At the compliance boundary point closest to its Euclidean distance, ensure that all individuals participating in subsequent non-dominated sorting meet the engineering specification requirements; Will Projected onto the feasible region closed set At the nearest compliance boundary, the projection result is used as the legitimate individual participating in the subsequent non-dominated ranking.

6. The method according to claim 1, characterized in that, In step S4, the dimensionality reduction and sensitivity analysis of the Pareto optimal solution set to obtain the contribution information of each building form control parameter to the energy consumption boundary specifically includes: Using partial correlation analysis, the partial correlation coefficients between each building form control parameter in the Pareto optimal solution set and the extreme cooling energy consumption in summer and the extreme heating energy consumption in winter are calculated. Based on the partial correlation coefficient, the control parameters of each building form are compared horizontally and their signs are compared to construct a contribution matrix that represents the distribution of collaborative weights among the parameters: the parameter with the larger absolute value of the partial correlation coefficient is given a higher weight in the contribution matrix; when the partial correlation coefficients of the same parameter for summer cooling energy consumption and winter heating energy consumption have opposite signs, they are given a priority of collaborative weights in the contribution matrix.

7. The method according to claim 1, characterized in that, In step S4, the inversion generation of the elastic control ranges for various building form control parameters specifically includes: For the frontier of the Pareto optimal solution set, an energy consumption tolerance threshold is set, and Pareto solutions with energy consumption degradation less than or equal to the energy consumption tolerance threshold are selected to form a qualified solution subset. Based on the value distribution of each building form control parameter in the qualified solution subset, the confidence interval of each building form control parameter is mapped and inverted, and the upper and lower limits of the confidence interval are used as the elastic control boundaries of each building form control parameter.

8. The method according to claim 1, characterized in that, In step S4, the parameter linkage compensation rule is generated in the following manner: Construct the Jacobian matrix of each building form control parameter with respect to the energy consumption target within the Pareto optimal solution set: ； in Let be a Jacobian matrix, and its element in the i-th row and j-th column be... Let represent the partial derivative of the j-th building form control parameter with respect to the i-th energy consumption target; The objective function for extreme cooling energy consumption in summer; The objective function for extreme winter heating energy consumption; There are n building form control parameters, where n is the total number of building form control parameters; When any building form control parameter When the deviation from its optimal threshold boundary causes a poorly-conditioned shift in the objective function, the compensation gradient vector is calculated based on the analytical solution of the Jacobian matrix: ； in, For the Jacobian matrix The generalized inverse matrix, This is the bias vector of the objective function caused by parameter deviation. To compensate for the gradient vector, Theoretical compensation amounts are included for each building form control parameter; according to Determine at least one compensation parameter and its compensation amount that require coordinated reverse adjustment in order to maintain energy consumption within the Pareto optimal frontier range within the elastic control range.

9. The method according to claim 1, characterized in that, In step S1, the building form control parameters include macroscopic geometric attributes, spatial composition attributes, and interface characteristics; the macroscopic geometric attributes include building volume, building height, and shape coefficient; the spatial composition attributes include floor area ratio and building shape index; and the interface characteristics include window-to-wall ratio for each orientation and building orientation angle. The parametric modeling and sampling specifically involves generating multiple sets of building form control parameter combinations through Latin hypercube sampling, and using an energy consumption simulation engine to calculate the extreme summer cooling energy consumption and extreme winter heating energy consumption corresponding to each set of building form control parameter combinations.

10. A knowledge-guided building performance optimization system adapted to extreme working conditions, characterized in that, include: The dual-objective dataset acquisition module is used to determine building form control parameters, extract extreme summer heat wave meteorological data and extreme winter cold wave meteorological data, and obtain multiple sets of building form control parameter combinations and their corresponding extreme summer cooling energy consumption and extreme winter heating energy consumption through parametric modeling and sampling, thereby constructing a dual-objective dataset. The proxy model construction module is used to train a proxy model based on the dual-objective dataset to establish a nonlinear mapping relationship between the building form control parameters and the extreme cooling energy consumption in summer and the extreme heating energy consumption in winter. The proxy model is used to output the corresponding predicted values ​​of extreme cooling energy consumption in summer and extreme heating energy consumption in winter for any combination of input form parameters. A prior knowledge-driven multi-objective optimization module is used to use the surrogate model as a fitness evaluation engine, with minimizing the extreme summer cooling energy consumption and the extreme winter heating energy consumption as dual optimization objectives, and to use a genetic algorithm to search for Pareto optimal solutions. Prior knowledge is embedded during the execution of the genetic algorithm. The prior knowledge includes: prior knowledge for population initialization based on the empirical distribution of building thermal performance, prior knowledge for directional mutation based on energy consumption-morphological attribution mapping, and prior knowledge for feasible domain projection based on building energy-saving design specifications. The reverse deconstruction and control information generation module is used to perform dimensionality reduction and sensitivity analysis on the Pareto optimal solution set, obtain the contribution information of each building form control parameter to the energy consumption boundary, and generate the elastic control range of each building form control parameter and the linkage compensation rules between parameters as the output of building form control information.

Images