Building design scheme generation method and system based on energy saving rate target driving, and medium

Abstract

This invention discloses a method, system, and medium for generating building design schemes based on energy efficiency targets, relating to the fields of building energy conservation and building design technology. It combines the semantic generation capability of a large language model for building energy conservation with the predictive capability of a deep neural network prediction model to realize a reverse optimization design method for building design schemes based on energy efficiency targets. Under a given energy efficiency target, it automatically generates multiple sets of feasible building energy conservation design parameter schemes, realizes semantic understanding and performance prediction closed-loop reasoning, and improves the efficiency of generating building energy conservation design schemes.

Description

Technical Field

[0001] This invention relates to the fields of building energy conservation and building design technology, specifically to a method, system, and medium for generating building design schemes driven by energy-saving rate targets. Background Technology

[0002] With the continuous rise in the proportion of building energy consumption in total social energy consumption, building energy-saving design has become an important research direction in the building design stage. Existing building energy-saving designs typically employ a forward optimization method, using design parameters such as building orientation, window-to-wall ratio, thermal performance of the building envelope, and photovoltaic system configuration as input variables. Energy consumption is then calculated through energy consumption simulation to determine the building's energy consumption or energy saving rate, and multiple rounds of parameter adjustments are made to achieve the goal of minimizing energy consumption. When owners or regulations specify clear energy-saving targets (such as 30% or 50% energy saving rates), designers struggle to quickly obtain the parameter combinations that meet these targets, making it difficult to quantify and implement energy-saving goals in the early design stages.

[0003] Furthermore, in recent years, Large Language Models (LLMs) have demonstrated powerful semantic expression and knowledge reasoning capabilities in areas such as graphics generation, text understanding, and design assistance, sparking exploration of their applications in intelligent design within the architectural community. However, existing research indicates that LLMs can only generate semantic design descriptions and cannot directly generate parametric architectural design schemes that meet specific requirements. Summary of the Invention

[0004] To address the shortcomings of existing forward optimization methods for building energy efficiency design, which struggle to quickly obtain parameter combinations that meet specific energy efficiency targets, or can only generate semantic design descriptions, making it difficult to quantify and implement energy efficiency targets in the early design stages, this invention aims to provide a method, system, and medium for generating building design schemes based on energy efficiency rate targets. This method combines the semantic generation capabilities of a large-scale building energy efficiency language model with the predictive capabilities of a deep neural network prediction model to achieve a reverse optimization design method for building design schemes driven by energy efficiency rate targets. Given an energy efficiency rate target, this method automatically generates multiple sets of feasible building energy efficiency design parameter schemes, enabling semantic understanding and performance prediction closed-loop reasoning, thereby improving the efficiency of building energy efficiency design scheme generation.

[0005] The above-mentioned technical objective of the present invention is achieved through the following technical solution:

[0006] This solution provides a method for generating building design schemes based on energy efficiency targets. The method includes:

[0007] Construct a multidimensional dataset containing architectural design parameters, energy consumption characteristics, and energy-saving design measures of architectural design schemes, and preprocess the multidimensional dataset;

[0008] A building design model driven by energy-saving rate targets is trained based on a preprocessed multidimensional dataset. The building design model is obtained by fusing a deep neural network prediction model with a building energy-saving big data language model. After semantically understanding the building design constraints and energy-saving rate targets, the building energy-saving big data language model generates multiple candidate building design schemes with different focuses. The deep neural network prediction model predicts the energy-saving rate of each candidate building design scheme. The building energy-saving big data language model updates the building design parameters of the candidate building design schemes according to the energy-saving rate target achievement, and finally obtains multiple candidate building design schemes that achieve the energy-saving rate target.

[0009] Output candidate building design schemes that meet the energy-saving rate standards, along with design scheme descriptions.

[0010] A further optimized solution is that the preprocessing includes the following methods:

[0011] The energy-saving design measures of the building design scheme are parametrically described to obtain the energy-saving design feature matrix, and the building design parameters of the building design scheme are standardized and vectorized to obtain the design parameter feature matrix.

[0012] Based on the energy-saving design feature matrix and the design parameter feature matrix, the Latin hypercube sampling method is used to expand the sample.

[0013] The energy efficiency of each building design scheme is calculated based on its energy consumption characteristics.

[0014] A further optimization is that the architecture of the architectural design model includes:

[0015] The representation vectors of the output layer of the deep neural network prediction model are embedded as semantic feature inputs into the intermediate layer structure of the building energy conservation big language model;

[0016] The deep neural network prediction model is trained based on the Adam optimizer and an adaptive learning rate decay strategy, and the contribution of each building design parameter to the building design parameter energy saving rate is evaluated based on the SHAP value method.

[0017] A further optimization scheme is that the building energy conservation language model uses a cross-modal feature fusion method to jointly encode the representation vector of the output layer of the deep neural network prediction model, the semantic understanding of building design constraints, and the semantic understanding of energy conservation rate targets. This enables the mapping of the patterns of changes in building design parameters and energy conservation rate to the language space, and the reverse mapping of energy conservation rate targets and building design constraints to the parameter space.

[0018] A further optimization scheme is proposed, in which the training method for the building energy conservation large language model includes:

[0019] Build a pre-trained language model;

[0020] A knowledge corpus for building energy conservation was constructed based on textual data in the field of building energy conservation, and a pre-trained language model was trained based on the knowledge corpus for building energy conservation, enabling the pre-trained language model to learn the semantic structure and professional knowledge in the field of building energy conservation.

[0021] A further optimization scheme is as follows: the building energy conservation big data model updates the building design parameters of the candidate building design schemes according to the energy conservation rate target, including the following method: if the energy conservation rate of the current candidate building design scheme reaches the energy conservation rate target, then save the current candidate building design scheme; otherwise, repeat the following process until the energy conservation rate of the current candidate building design scheme reaches the energy conservation rate target: after adjusting the building design parameters of the candidate building design schemes, the deep neural network prediction model re-predicts the energy conservation rate of the current group of candidate building design schemes.

[0022] A further optimization scheme is that the building design model compresses high-dimensional reasoning features into a lightweight model based on the knowledge distillation method, while retaining high-order semantic features of the relationship between energy-saving parameters and energy-saving rate during the distillation process.

[0023] A further optimized approach is that the design scheme specification includes the energy-saving mechanism and physical explanation of the candidate building design scheme, as well as one or more optional design semantic descriptions.

[0024] This solution also provides a building design scheme generation system based on energy efficiency target, used to implement the above-mentioned building design scheme generation method based on energy efficiency target. The system includes:

[0025] The preprocessing module is used to construct a multidimensional dataset containing architectural design parameters, energy consumption characteristics, and energy-saving design measures of architectural design schemes, and to preprocess the multidimensional dataset.

[0026] The model building module is used to train an energy-saving rate target-driven building design model based on a preprocessed multidimensional dataset. The building design model is obtained by fusing a deep neural network prediction model and a building energy-saving big data language model. After semantically understanding the building design constraints and energy-saving rate targets, the building energy-saving big data language model generates multiple candidate building design schemes with different focuses. The deep neural network prediction model predicts the energy-saving rate of each candidate building design scheme. The building energy-saving big data language model updates the building design parameters of the candidate building design schemes according to the energy-saving rate target achievement, and finally obtains multiple candidate building design schemes that achieve the energy-saving rate target.

[0027] The output module is used to output candidate building design schemes and design scheme descriptions that meet the energy-saving rate standards.

[0028] This solution also provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, can realize the above-described method for generating building design schemes based on energy efficiency targets.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] 1. This invention provides a method, system, and medium for generating building design schemes based on energy efficiency targets. It combines the semantic generation capability of a large language model for building energy efficiency with the predictive capability of a deep neural network prediction model to realize a reverse optimization design method for building design schemes based on energy efficiency targets. Under a given energy efficiency target, it automatically generates multiple sets of feasible building energy efficiency design parameter schemes, realizes semantic understanding and performance prediction closed-loop reasoning, and improves the efficiency of generating building energy efficiency design schemes.

[0031] 2. This invention provides a method, system, and medium for generating building design schemes based on energy-saving rate targets. In the generation stage, a diverse sampling strategy is adopted to enable the building energy-saving big language model to generate a variety of energy-saving design schemes with differences.

[0032] 3. This invention provides a method, system, and medium for generating building design schemes based on energy efficiency targets. The final output is not a single optimal solution, but multiple candidate building design schemes with different focuses, which allows designers to choose according to cost, standards, and owner preferences. Attached Figure Description

[0033] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0034] Figure 1 A schematic diagram of a method for generating building design schemes based on energy efficiency targets;

[0035] Figure 2 Generate system structure diagrams for building design schemes driven by energy efficiency targets. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0037] In view of this, the present solution provides the following embodiments to solve the above-mentioned technical problems.

[0038] Example 1

[0039] This embodiment provides a method for generating building design schemes based on energy efficiency targets, such as... Figure 1 As shown, the method includes:

[0040] Step 1: Construct a multidimensional dataset containing architectural design parameters, energy consumption characteristics, and energy-saving design measures of architectural design schemes, and preprocess the multidimensional dataset;

[0041] Specific architectural design parameters include building orientation, window-to-wall ratio, heat transfer coefficient of external windows, shading method, thickness of external wall insulation layer, roof reflectivity, photovoltaic module arrangement and capacity ratio, etc.

[0042] In step one, the preprocessing includes the following methods:

[0043] The energy-saving design measures of the building design scheme are parametrically described to obtain the energy-saving design feature matrix, and the building design parameters of the building design scheme are standardized and vectorized to obtain the design parameter feature matrix.

[0044] Based on the energy-saving design feature matrix and the design parameter feature matrix, the Latin hypercube sampling method is used to expand the sample.

[0045] The energy efficiency of each building design scheme is calculated based on its energy consumption characteristics.

[0046] Specifically, the energy efficiency of the building design scheme Calculate according to the following formula:

[0047] ;

[0048] in, Indicates the energy consumption of the building design scheme; This indicates that the baseline energy consumption value is determined based on current energy efficiency standards or reference building energy consumption.

[0049] Step 2: Based on the preprocessed multidimensional dataset, a building design model driven by the energy-saving rate target is trained. The building design model is obtained by fusing a deep neural network prediction model and a building energy-saving big data language model. After semantically understanding the building design constraints and energy-saving rate target, the building energy-saving big data language model generates multiple candidate building design schemes with different focuses. The deep neural network prediction model predicts the energy-saving rate of each candidate building design scheme. The building energy-saving big data language model updates the building design parameters of the candidate building design schemes according to the energy-saving rate target achievement, and finally obtains multiple candidate building design schemes that achieve the energy-saving rate target.

[0050] In step two, the architecture of the architectural design model includes:

[0051] The representation vector of the output layer of the deep neural network prediction model is embedded as the semantic feature input into the intermediate layer structure of the building energy conservation big language model. The deep neural network prediction model is used to learn the nonlinear mapping relationship between building design parameters and energy saving rate, and the building energy conservation big language model is used to understand design semantics and generate candidate building design schemes.

[0052] The specific deep neural network prediction model consists of an input layer, several hidden layers, and an output layer. The input dimension of the input layer corresponds to the number of building design parameters (such as 10 to 20 features, including building orientation, window-to-wall ratio, external window heat transfer coefficient, shading type, external wall insulation thickness, roof reflectivity, photovoltaic layout ratio, and other feature variables).

[0053] The hidden layers contain 4 to 6 fully connected network layers, with the number of neurons in each layer decreasing from large to small (e.g., 256→128→64→32); Batch Normalization and Dropout mechanisms are used to prevent overfitting.

[0054] The output layer is a single-node structure, corresponding to the predicted energy saving rate. The activation function of the deep neural network prediction model is Sigmoid to ensure that the output range is between 0 and 1 (i.e., 0–100% energy saving rate).

[0055] The mean squared error (MSE) loss function is used as the optimization objective during model training.

[0056] The deep neural network prediction model is trained using the Adam optimizer and an adaptive learning rate decay strategy. The contribution of each building design parameter to the energy-saving rate is evaluated using the SHAP value method, generating a feature importance matrix to provide interpretable weights for subsequent back-reasoning of energy-saving design schemes. The initial learning rate of the Adam optimizer is set to 0.001, and an early stopping mechanism is used to prevent overfitting. Training and validation data are divided in a 7:3 ratio, with 300-500 training epochs. The accuracy of the deep neural network prediction model is evaluated using R², RMSE, and CVRMSE metrics.

[0057] After the model training is completed, the contribution of each building design parameter to the energy saving rate prediction result is calculated by the SHAP value method, and the feature importance matrix of the building design parameters is obtained, thus providing interpretable weight information for subsequent reverse generation of energy-saving design schemes.

[0058] In step two, the building energy conservation language model uses a cross-modal feature fusion method to jointly encode the representation vector of the deep neural network prediction model output layer, the semantic understanding of building design constraints, and the semantic understanding of energy conservation rate targets. This enables the mapping of the patterns of changes in building design parameters and energy conservation rate to the language space (such as understanding the logical relationship between semantic instructions such as "increase the thermal resistance of the building envelope", "increase photovoltaic coverage", and "optimize orientation" and energy consumption features), and the reverse mapping of energy conservation rate targets and building design constraints to the parameter space.

[0059] In step two, the training method for the building energy conservation big language model includes:

[0060] S21, Build pre-trained language models; such as the Transformer architecture model.

[0061] S22. A knowledge corpus for building energy conservation is constructed based on textual materials in the field of building energy conservation. A pre-trained language model is then trained based on this corpus, enabling the model to learn the semantic structure and professional knowledge within the building energy conservation field. Specifically, a large amount of textual materials, including building energy conservation design specifications, papers, technical guidelines, and design drawing descriptions, are collected to construct the building energy conservation corpus. Through data cleaning, word segmentation, semantic annotation, and knowledge graph construction, the knowledge corpus for building energy conservation is formed.

[0062] In step two, the building energy conservation big data model updates the building design parameters of the candidate building design schemes according to the energy conservation rate target, including the following method: if the energy conservation rate of the current candidate building design scheme reaches the energy conservation rate target, then save the current candidate building design scheme; otherwise, repeat the following process until the energy conservation rate of the current candidate building design scheme reaches the energy conservation rate target: after adjusting the building design parameters of the candidate building design schemes, the deep neural network prediction model re-predicts the energy conservation rate of the current group of candidate building design schemes.

[0063] The architectural design model compresses high-dimensional reasoning features into a lightweight model based on the knowledge distillation method, preserving high-order semantic features of the relationship between energy-saving parameters and energy-saving rates during the distillation process. This enables the generated model to possess the following capabilities: ① Identify the logical constraints between input energy-saving targets and building types; automatically avoid infeasible designs (such as parameter combinations that exceed shape coefficients or violate regulations); and balance multiple objective conditions such as energy-saving rate, comfort, and economy at the semantic level.

[0064] After training the architectural design model, fine-tuning can be performed. Inputs include natural language descriptions (e.g., "We aim for a 35% energy saving rate and maximize photovoltaic use") and corresponding building parameters and energy saving rate labels. Supervised learning optimizes the model's output, enabling it to generate design parameters that match the energy-saving goals under natural language conditions. This fine-tuning process can employ lightweight training strategies such as LoRA (Low-Rank Adaptation) to ensure efficient adaptation of the large-scale architectural energy-saving language model within limited computational resources.

[0065] Specifically, in order to generate energy-saving schemes with different design focuses, design strategy labels or weight parameters are introduced as control conditions at the input end of the building energy conservation big language model, such as "passive energy saving priority", "photovoltaic utilization priority", "comprehensive balance optimization", etc., so that the corresponding type of energy-saving measures are tended to be strengthened when generating design parameters.

[0066] Step 3: Output the candidate building design schemes and design scheme descriptions that meet the energy saving rate standards.

[0067] The design specification includes the energy-saving mechanism and physical explanation of the candidate building design scheme, as well as one or more optional design semantic descriptions.

[0068] This solution provides a building design model that, when given an energy efficiency target and several design constraints (such as "in hot-summer, cold-winter regions, office buildings, energy efficiency ≥ 30%, rooftop photovoltaics available"), outputs: multiple feasible candidate building design schemes, the energy-saving mechanism and physical explanation for each scheme, and optional design semantic descriptions (such as "increasing the rooftop photovoltaic ratio and reducing the window-to-wall ratio can effectively improve the energy efficiency"). It enables the reverse generation of energy-saving design parameters and technology combinations that meet the constraints, starting from the energy efficiency target.

[0069] Specifically, the user inputs their desired energy saving rate target. The scope (e.g., ≥30%) and constraints (building height, shape coefficient, economic cost ceiling, etc.) are defined. Based on the building energy conservation language model, a series of feasible candidate building design schemes (combinations of building energy conservation design parameters) are generated according to semantic understanding. The generated content includes energy conservation parameters of the building envelope (heat transfer coefficients of exterior walls, windows, roof, and ground, insulation material thickness, etc.); passive design strategies (shading methods, building orientation adjustment, natural ventilation enhancement measures, etc.); active energy conservation measures (photovoltaic layout ratio, photovoltaic module type, cold and heat source efficiency optimization, etc.); specific parameter values ​​and combinations; and the design logic and energy conservation rationale of the design scheme (e.g., "increasing the roof reflectivity can reduce solar radiation heat gain").

[0070] This solution also embeds the trained deep neural network prediction model into the inference process of the building energy efficiency big data model. After the building energy efficiency big data model generates candidate building design schemes, it automatically inputs structured parameters into the deep neural network prediction model to predict the energy saving rate and returns the prediction results in real time. When the predicted energy saving rate meets or is close to the target energy saving rate, the candidate building design scheme is retained and output. When the predicted energy saving rate is insufficient, the building energy efficiency big data model automatically adjusts relevant design parameters, such as reducing the window-to-wall ratio, increasing the insulation thickness, and increasing the photovoltaic coverage. Through several rounds of generation-prediction-adjustment, candidate building design schemes that meet the energy saving rate target can be obtained.

[0071] To avoid getting stuck in a single solution or local optima, this invention employs a diversity sampling strategy during the generation phase, prompting the building energy efficiency big data model to generate multiple energy-saving design schemes with distinct characteristics. Each sampled scheme is used to predict energy efficiency rates through a deep neural network prediction model. The system automatically selects schemes that meet the energy efficiency targets, forming the final candidate building design scheme set. This candidate design scheme set typically includes multiple optimization directions, such as designs focusing on strengthening the building envelope, passive strategies, and photovoltaic integration. This diverse scheme set allows users to choose the optimal scheme based on the actual needs and implementation conditions of the project. The final output scheme is not a single optimal solution, but rather 3–5 energy-saving design schemes with significantly different focuses, allowing designers to decide which type of scheme to adopt based on cost, regulations, and owner preferences. Each design scheme includes key design parameters (window-to-wall ratio, U-value, shading type, photovoltaic area, etc.); predicted energy efficiency rates; a design orientation description ("biased to passive energy saving," "suitable for large-area rooftop photovoltaics," "suitable for high-rise office buildings," etc.); and a brief design basis or precautions.

[0072] Example 2

[0073] This embodiment provides a building design scheme generation system driven by energy efficiency targets, such as... Figure 2 As shown, the system for generating building design schemes based on energy efficiency targets as described in Example 1 includes:

[0074] The preprocessing module is used to construct a multidimensional dataset containing architectural design parameters, energy consumption characteristics, and energy-saving design measures of architectural design schemes, and to preprocess the multidimensional dataset.

[0075] The model building module is used to train an energy-saving rate target-driven building design model based on a preprocessed multidimensional dataset. The building design model is obtained by fusing a deep neural network prediction model and a building energy-saving big data language model. After semantically understanding the building design constraints and energy-saving rate targets, the building energy-saving big data language model generates multiple candidate building design schemes with different focuses. The deep neural network prediction model predicts the energy-saving rate of each candidate building design scheme. The building energy-saving big data language model updates the building design parameters of the candidate building design schemes according to the energy-saving rate target achievement, and finally obtains multiple candidate building design schemes that achieve the energy-saving rate target.

[0076] The output module is used to output candidate building design schemes and design scheme descriptions that meet the energy-saving rate standards.

[0077] Example 3

[0078] This embodiment provides a computer-readable medium storing a computer program, characterized in that the computer program, when executed by a processor, can implement the building design scheme generation method based on energy efficiency target as described in Embodiment 1; specifically, it performs the following steps:

[0079] Step 1: Construct a multidimensional dataset containing architectural design parameters, energy consumption characteristics, and energy-saving design measures of architectural design schemes, and preprocess the multidimensional dataset;

[0080] Step 2: Based on the preprocessed multidimensional dataset, a building design model driven by the energy-saving rate target is trained. The building design model is obtained by fusing a deep neural network prediction model and a building energy-saving big data language model. After semantically understanding the building design constraints and energy-saving rate target, the building energy-saving big data language model generates multiple candidate building design schemes with different focuses. The deep neural network prediction model predicts the energy-saving rate of each candidate building design scheme. The building energy-saving big data language model updates the building design parameters of the candidate building design schemes according to the energy-saving rate target achievement, and finally obtains multiple candidate building design schemes that achieve the energy-saving rate target.

[0081] Step 3: Output the candidate building design schemes and design scheme descriptions that meet the energy saving rate standards.

[0082] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for generating building design schemes based on energy-saving rate targets, characterized in that, The method includes: Construct a multidimensional dataset containing architectural design parameters, energy consumption characteristics, and energy-saving design measures of architectural design schemes, and preprocess the multidimensional dataset; A building design model driven by energy-saving rate targets is trained based on a preprocessed multidimensional dataset. The building design model is obtained by fusing a deep neural network prediction model with a building energy-saving big data language model. After semantically understanding the building design constraints and energy-saving rate targets, the building energy-saving big data language model generates multiple candidate building design schemes with different focuses. The deep neural network prediction model predicts the energy-saving rate of each candidate building design scheme. The building energy-saving big data language model updates the building design parameters of the candidate building design schemes according to the energy-saving rate target achievement, and finally obtains multiple candidate building design schemes that achieve the energy-saving rate target. Output candidate building design schemes that meet the energy-saving rate standards, along with design scheme descriptions.

2. The method for generating building design schemes based on energy-saving rate targets according to claim 1, characterized in that, The preprocessing includes the following methods: The energy-saving design measures of the building design scheme are parametrically described to obtain the energy-saving design feature matrix, and the building design parameters of the building design scheme are standardized and vectorized to obtain the design parameter feature matrix. Based on the energy-saving design feature matrix and the design parameter feature matrix, the Latin hypercube sampling method is used to expand the sample. The energy efficiency of each building design scheme is calculated based on its energy consumption characteristics.

3. The method for generating building design schemes based on energy-saving rate targets according to claim 2, characterized in that, The architecture of the architectural design model includes: The representation vectors of the output layer of the deep neural network prediction model are embedded as semantic feature inputs into the intermediate layer structure of the building energy conservation big language model; The deep neural network prediction model is trained based on the Adam optimizer and an adaptive learning rate decay strategy, and the contribution of each building design parameter to the building design parameter energy saving rate is evaluated based on the SHAP value method.

4. The method for generating building design schemes based on energy-saving rate targets according to claim 3, characterized in that, The aforementioned building energy conservation language model uses a cross-modal feature fusion method to jointly encode the representation vector of the output layer of the deep neural network prediction model, the semantic understanding of building design constraints, and the semantic understanding of energy conservation rate targets. This enables the mapping of the patterns of changes in building design parameters and energy conservation rate to the language space, and the reverse mapping of energy conservation rate targets and building design constraints to the parameter space.

5. The method for generating building design schemes based on energy-saving rate targets according to claim 4, characterized in that, The training method for the building energy efficiency big language model includes: Build a pre-trained language model; A knowledge corpus for building energy conservation was constructed based on textual data in the field of building energy conservation, and a pre-trained language model was trained based on the knowledge corpus for building energy conservation, enabling the pre-trained language model to learn the semantic structure and professional knowledge in the field of building energy conservation.

6. The method for generating building design schemes based on energy-saving rate targets according to claim 1, characterized in that, The building energy conservation big data model updates the building design parameters of candidate building design schemes based on the energy conservation rate target, including the following method: if the energy conservation rate of the current candidate building design scheme reaches the energy conservation rate target, then save the current candidate building design scheme; otherwise The following process is repeated until the energy saving rate of the current candidate building design scheme reaches the energy saving rate target: After adjusting the building design parameters of the candidate building design scheme, the energy saving rate of the current group of candidate building design schemes is re-predicted by the deep neural network prediction model.

7. The method for generating building design schemes based on energy-saving rate targets according to claim 1, characterized in that, The architectural design model compresses high-dimensional reasoning features into a lightweight model based on the knowledge distillation method, while retaining high-order semantic features of the relationship between energy-saving parameters and energy-saving rates during the distillation process.

8. The method for generating building design schemes based on energy efficiency targets according to claim 1, characterized in that, The design specification includes the energy-saving mechanism and physical explanation of the candidate building design scheme, as well as one or more optional design semantic descriptions.

9. A building design scheme generation system driven by energy-saving rate targets, characterized in that, The system is used to implement the building design scheme generation method based on energy efficiency target as described in any one of claims 1-8, the system comprising: The preprocessing module is used to construct a multidimensional dataset containing architectural design parameters, energy consumption characteristics, and energy-saving design measures of architectural design schemes, and to preprocess the multidimensional dataset. The model building module is used to train an energy-saving rate target-driven building design model based on a preprocessed multidimensional dataset. The building design model is obtained by fusing a deep neural network prediction model and a building energy-saving big data language model. After semantically understanding the building design constraints and energy-saving rate targets, the building energy-saving big data language model generates multiple candidate building design schemes with different focuses. The deep neural network prediction model predicts the energy-saving rate of each candidate building design scheme. The building energy-saving big data language model updates the building design parameters of the candidate building design schemes according to the energy-saving rate target achievement, and finally obtains multiple candidate building design schemes that achieve the energy-saving rate target. The output module is used to output candidate building design schemes and design scheme descriptions that meet the energy-saving rate standards.

10. A computer-readable medium having a computer program stored thereon, characterized in that, The computer program, when executed by a processor, can implement the building design scheme generation method based on energy efficiency target as described in any one of claims 1-8.

Images