Energy performance prediction method and equipment based on urban residential building, and medium

Abstract

The invention discloses an energy performance prediction method and device based on urban residential buildings and a medium, belongs to the technical field of energy efficiency prediction, and is used for solving the technical problems that traditional physical modeling still has certain limitation on building energy consumption prediction, the calculation complexity is high, used data parameters are limited, and overall energy consumption prediction is not accurate enough. The method comprises the following steps: carrying out collection processing on building stock data, meteorological data and auxiliary data on a building group in a city range to obtain building key data; generating a building reference model based on the building key data; performing energy consumption simulation processing on the building group through the building reference model to obtain a multi-dimensional energy efficiency feature data set; performing energy efficiency performance prediction evaluation processing on the building energy efficiency data by using a non-separation model, a terminal demand isolation model and a set isolation model, and determining an optimal energy efficiency performance prediction model; and performing energy efficiency analysis on the building group to obtain a building energy efficiency prediction result.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of energy efficiency prediction, and in particular to an energy performance prediction method, device and medium based on urban residential buildings. BACKGROUND

[0002] According to global data, building operation accounts for 30% of global energy consumption and 27% of greenhouse gas emissions (8% of which is direct emission and 19% of which is indirect emission). To address this challenge, the European Union has set the goal of improving building energy efficiency and decarbonization by 2030 and 2050 through the Building Energy Efficiency Directive (EPBD). However, existing building databases only cover 30-50% of the total amount, and the data quality is uneven, which seriously restricts the formulation of energy-saving policies.

[0003] Although traditional physical modeling methods (such as EnergyPlus) are accurate, they require detailed building feature data and have high computational complexity. In contrast, data-driven methods can achieve energy consumption prediction under limited information conditions and reveal the complex relationship between building features and energy consumption through machine learning algorithms. However, most current research focuses only on single building energy consumption prediction, and the parameters used are limited. SUMMARY

[0004] The embodiments of the present application provide an energy performance prediction method, device and medium based on urban residential buildings, which are used to solve the following technical problems: traditional physical modeling has certain limitations for building energy consumption prediction, has high computational complexity, uses limited data parameters, and the overall energy consumption prediction is not accurate enough.

[0005] The embodiments of the present application adopt the following technical solutions:

[0006] On the one hand, the embodiments of the present application provide an energy performance prediction method based on urban residential buildings, comprising: collecting and processing building stock data, meteorological data and auxiliary data of a building group in a city range to obtain building key data; generating a building benchmark model based on the building key data; performing energy consumption simulation processing on the building group through the building benchmark model to obtain a multi-dimensional energy efficiency feature data set; performing energy efficiency performance prediction and evaluation processing on building energy efficiency data based on the multi-dimensional energy efficiency feature data set, determining the best energy efficiency performance prediction model, and performing energy efficiency analysis on the building group through the best energy efficiency performance prediction model to obtain building energy efficiency prediction results.

[0007] The embodiments of the present application efficiently predict the energy performance of urban buildings through machine learning technology, effectively solving the bottleneck problem of high complexity and large data demand of traditional physical models. Parameterized simulation and Latin hypercube sampling technology are used to generate a large number of synthetic data sets of buildings, significantly improving the coverage and practical value of the model. In terms of methodology, integrated learning and terminal demand separation technology are also integrated, and advanced algorithms such as XGBoost and LightGBM are used to significantly improve the prediction accuracy. At the same time, the SHAP explainability analysis method is used to accurately identify key features such as air exchange rate and wall U value. The interdisciplinary integration of building physical simulation, parameterized modeling and machine learning technology is also realized, and a whole-process analysis framework from data collection to decision support is constructed, providing a systematic solution for urban building energy management, which has important theoretical value and practical significance.

[0008] In a feasible implementation, the building group in the city range is collected and processed for building stock data, meteorological data and auxiliary data to obtain building key data, specifically including: through the ABULA professional database, the geometric characteristics of the building group in the city range are collected to obtain building geometric characteristics; wherein the geometric characteristics at least include: building envelope, building shape and building floor number; through population census and energy efficiency certificate, the non-geometric characteristics of the building group in the city range are collected to obtain building non-geometric characteristics; wherein the non-geometric characteristics at least include: energy use mode and energy equipment load; the building geometric characteristics and the building non-geometric characteristics are combined to obtain the building stock data; the building stock data, the meteorological data and the auxiliary data are collected for key data multi-source collection to obtain the building key data; wherein the meteorological data includes: typical meteorological year data, future climate scenario data and EnergyPlus weather file covering the whole world; the auxiliary data at least includes: population census data and energy policy report data.

[0009] In a feasible implementation, based on the building key data, a building benchmark model is generated, specifically including: through the national building group database, a reference building standard model is extracted; through the building key data, the reference building standard model is simulated for baseline model parameters, and based on the simulated building prototype covering a series of parameter combinations, the building benchmark model applied to the building group is determined.

[0010] In an embodiment, the building group is simulated for energy consumption by the building benchmark model to obtain a multi-dimensional energy efficiency feature dataset, which includes: the initial building performance data obtained by uniformly distributing the building energy efficiency influencing features in the building benchmark model in the relevant parameter space by a Latin hypercube sampling method; wherein the building energy efficiency influencing features include building physical features, use features, and heat system parameters; the standard building performance data obtained by simulating the initial building performance data for complex energy consumption under multi-parameter coordination according to the integrated EnergyPlus physical engine and DesignBuilder template library; wherein the standard building performance data includes building energy consumption performance indicators and building comfort indicators; and the multi-dimensional energy efficiency feature dataset obtained by collecting the multi-dimensional energy efficiency features in the standard building performance data.

[0011] In an embodiment, before determining the optimal energy efficiency performance prediction model by predicting and evaluating the energy efficiency performance of the building energy efficiency data according to the multi-dimensional energy efficiency feature dataset for non-separation models, terminal demand isolation models, and set isolation models, the method further includes: training and constructing the prediction model for the energy efficiency data in the multi-dimensional energy efficiency feature dataset by a preset multi-type machine learning algorithm to obtain the non-separation model for predicting energy efficiency performance; wherein the multi-type machine learning algorithm includes at least linear regression, neural network, decision tree, random forest, and gradient boosting; calculating the energy use intensity of the terminal demand data in the multi-dimensional energy efficiency feature dataset for energy use types by a preset regression model to obtain the energy use intensity indicators for each energy use type; and constructing the terminal demand isolation model based on the energy use intensity indicators; wherein the terminal demand data includes at least heating energy use data, cooling energy use data, equipment energy use data, lighting energy use data, and hot water energy use data; and constructing the set isolation model by model prediction training of the terminal demand data output by the terminal demand isolation model for energy efficiency performance by stacking technology and voting technology in the set learning algorithm.

[0012] In an implementable embodiment, according to the multi-dimensional energy efficiency feature dataset, building energy efficiency data is subjected to energy efficiency performance prediction evaluation processing on non-separation models, terminal demand isolation models and set isolation models, and the optimal energy efficiency performance prediction model is determined, specifically comprising: through a preset SHAP technology, the prediction accuracy contribution of each terminal demand feature in the non-separation model, the terminal demand isolation model and the set isolation model is calculated to obtain a prediction feature contribution degree; wherein the terminal demand feature is derived from the terminal demand data; through the prediction feature contribution degree, the importance of the prediction influence degree of each terminal demand feature is divided to obtain a key driving factor feature related to building energy efficiency; based on an integrated multi-type machine learning model, the key driving factor feature is subjected to energy efficiency performance prediction training related to the building group to generate the optimal energy efficiency performance prediction model.

[0013] In an implementable embodiment, based on an integrated multi-type machine learning model, the key driving factor feature is subjected to energy efficiency performance prediction training related to the building group to generate the optimal energy efficiency performance prediction model, specifically comprising: based on a preset performance index, the energy efficiency performance of the non-separation model, the terminal demand isolation model and the set isolation model is subjected to prediction calculation to obtain an energy use intensity prediction value of each machine learning model; the energy use intensity prediction value of each machine learning model is converted into an energy performance certification level; according to the energy performance certification level, each machine learning model is subjected to energy efficiency performance prediction evaluation processing to determine the prediction accuracy of each machine learning model; based on the sorting of the prediction accuracy, the prediction results of each machine learning model are combined with corresponding weight values to obtain an integrated multi-type machine learning model.

[0014] In an implementable embodiment, through the optimal energy efficiency performance prediction model, the building group is subjected to energy efficiency analysis to obtain a building energy efficiency prediction result, specifically comprising: inputting current building key data of the building group into the optimal energy efficiency performance prediction model; through the optimal energy efficiency performance prediction model, the current building key data is subjected to energy efficiency analysis processing related to building energy performance to obtain the building energy efficiency prediction result.

[0015] In a second aspect, the embodiments of the present application also provide an energy performance prediction device based on urban residential buildings, the device comprising: at least one processor; and a memory in communication connection with the at least one processor; wherein the memory stores instructions executable by the at least one processor, so that the at least one processor can execute the energy performance prediction method based on urban residential buildings according to any of the embodiments described above.

[0016] In a third aspect, the embodiments of the present application further provide a non-volatile computer storage medium, which is a non-volatile computer readable storage medium, and stores at least one program, each of which includes instructions, which, when executed by a terminal, causes the terminal to perform the energy performance prediction method for urban residential buildings according to any one of the above embodiments.

[0017] The present application provides an energy performance prediction method, device and medium for urban residential buildings, which has the following beneficial technical effects compared with the prior art:

[0018] The embodiments of the present application efficiently predict the energy performance of urban buildings through machine learning technology, effectively solving the bottleneck problem of high computational complexity and large data demand of traditional physical models. By using parameterized simulation and Latin hypercube sampling technology, a large number of building synthetic data sets are generated, which significantly improves the coverage and practical value of the model. It also integrates ensemble learning and terminal demand separation technology, and uses advanced algorithms such as XGBoost and LightGBM to greatly improve the prediction accuracy. At the same time, the SHAP explainability analysis method is used to accurately identify key features such as air exchange rate and wall U value. It also realizes the cross-disciplinary integration of building physical simulation, parameterized modeling and machine learning technology, and builds a whole-process analysis framework from data collection to decision support, providing a systematic solution for urban building energy management, which has important theoretical value and practical significance. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiment or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments described in the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor. In the drawings:

[0020] Figure 1 A flowchart of an energy performance prediction method for urban residential buildings is provided for the embodiments of the present application;

[0021] Figure 2 A flowchart of predicting energy efficiency of urban buildings using machine learning is provided for the embodiments of the present application;

[0022] Figure 3 A terminal use demand isolation modeling flowchart for predicting energy use intensity using machine learning is provided for the embodiments of the present application;

[0023] Figure 4A set machine learning modeling schematic diagram for enhancing the prediction performance of a machine learning model is provided for the embodiments of the present application.

[0024] Figure 5 A structural schematic diagram of an energy performance prediction device based on urban residential buildings is provided for the embodiments of the present application. DETAILED DESCRIPTION

[0025] In order for those skilled in the art to better understand the technical solutions in the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor should fall within the scope of protection of the present application.

[0026] It should be noted that the present application combines parameterized simulation and machine learning technology: first, a million-level building synthetic data set is generated through Latin hypercube sampling (note: a method for approximately random sampling from a multi-dimensional parameter distribution, belonging to stratified sampling technology, commonly used in computer experiments or Monte Carlo integration, etc.), solving the data shortage problem of urban scale modeling; second, the terminal demand separation method is used to model the heating, lighting and other terminal energy respectively and then aggregate, which significantly improves the prediction accuracy (RMSE is reduced to 6.48); at the same time, XGBoost and other integrated algorithms are used to improve the overall prediction accuracy from 76% to 91%. The present application also identifies key features such as air change rate and wall U value through SHAP analysis, and develops a policy evaluation tool to quantitatively analyze the effect of energy saving reconstruction (such as only 27-34% of buildings can reach high energy efficiency through reconstruction). In addition, by integrating RCP4.5 climate scenario data, the prediction of building energy consumption changes under future climate conditions is realized, providing a more comprehensive scientific basis for policy making.

[0027] The embodiments of the present application provide a method for predicting the energy performance of urban residential buildings, as shown in Figure 1 The method for predicting the energy performance of urban residential buildings specifically includes steps S101-S105:

[0028] S101, collecting and processing the building stock data, meteorological data and auxiliary data of the building group in the city to obtain the key data of the building.

[0029] Specifically, first, the geometric characteristics of the building group in the city are collected through the ABULA professional database to obtain the geometric characteristics of the building. The geometric characteristics at least include: building envelope, building shape and building floor number.

[0030] Further, the non-geometric features of the building group in the city range are collected through the population census and the energy efficiency certificate to obtain building non-geometric features. The non-geometric features at least include: energy use mode and energy equipment load.

[0031] Further, the building stock data is obtained by combining the building geometric features and the building non-geometric features.

[0032] Further, the building key data is obtained by performing key data multi-source collection on the building stock data, the meteorological data and the auxiliary data. The meteorological data includes: typical meteorological year data, future climate scenario data and EnergyPlus weather file covering the whole world. The auxiliary data at least includes: population census data and energy policy report data.

[0033] In one embodiment, Figure 2 A building energy efficiency prediction process diagram using machine learning is provided for the embodiments of the present application, as shown in Figure 2 As shown, a multi-source data fusion method is used to construct a building energy performance prediction model. The key data includes: 1) building stock data, covering geometric features (envelope, shape, number of layers, etc.) and non-geometric features (use mode, equipment load, etc.), wherein the geometric data comes from professional databases such as TABULA, and the non-geometric data is obtained through population census and energy efficiency certificate; 2) meteorological data, including typical meteorological year (TMY) and EnergyPlus weather file (EPW) covering 3034 locations worldwide, and integrating future climate scenario data (2030 / 2050) from Meteonorm and other sources; 3) auxiliary data, such as population census and energy policy report, providing social and economic dimension support for the model.

[0034] As a feasible implementation, the current building energy data collection faces two major challenges: large-scale acquisition of non-geometric building data and standardized integration of multi-source heterogeneous data. The building prototype method is used to solve the problem of missing non-geometric data, and the data fusion technology is used to realize the collaborative analysis of meteorological data, building features and policy targets. This multi-dimensional data system not only supports energy consumption prediction from single building to city scale, but also provides scientific basis for evaluating the energy saving effect under different climate scenarios, and finally realizes the organic combination of historical data and future prediction.

[0035] S102, generating a building benchmark model based on the building key data.

[0036] Specifically, the reference building standard model is extracted through the national building group database.

[0037] Furthermore, the key building data is used to simulate the parameters of the baseline model of the reference building standard model, and based on the simulated building prototype covering a series of parameter combinations, the building benchmark model applied to the building complex is determined.

[0038] As a feasible implementation method, Figure 2 As shown, several buildings within a city often share similar characteristics and can be categorized as building prototypes. In urban building energy simulations, a building prototype, referred to as a reference building, is a representative model that captures the typological characteristics and performance of a specific category or cluster of buildings within a larger building complex. Parametric simulation frameworks use each building prototype as a baseline model. This data can be obtained from established national building complex databases, such as TABULA or the EPC database. Building prototypes, or reference buildings, serve as standardized models that simplify the simulation process by providing an analytical baseline or template. They are typically developed based on existing data collection, statistical analysis, and analytical results. However, simulating any building prototype requires both geometric and non-geometric data for each baseline model. These building prototypes serve as a starting point for parametric modeling of different buildings to develop synthetic building complexes. These methods generate representative synthetic datasets covering a range of parameter combinations, facilitating more comprehensive analysis of design options and optimizing energy modeling results.

[0039] S103. Perform energy consumption simulation on the building complex using a building benchmark model to obtain a multi-dimensional energy efficiency characteristic data set.

[0040] Specifically, the Latin hypercube sampling method can be used to uniformly distribute the building energy efficiency impact characteristics in the building benchmark model across the relevant parameter space to obtain initial building performance data. Building energy efficiency impact characteristics include: building physical characteristics, usage characteristics, and thermal system parameters.

[0041] Furthermore, based on the integrated EnergyPlus physics engine and DesignBuilder template library, a complex energy consumption simulation with multi-parameter coordination was conducted on the initial building performance data to obtain standard building performance data. The standard building performance data includes building energy performance indicators and building comfort indicators.

[0042] Furthermore, data collection is performed on the multidimensional energy efficiency characteristics in the standard building performance data to generate a multidimensional energy efficiency characteristic data set.

[0043] In one embodiment, Figure 2As shown, the energy consumption parameter simulation provides an optimal solution, mainly using parameterized simulation technology to solve the data scarcity problem in building energy efficiency modeling. Through the jEPlus parameterized tool combined with the EnergyPlus simulation engine and the DesignBuilder template, a complex energy consumption simulation with multiple parameter coordination is realized. This method can systematically evaluate the impact of design parameters on building energy consumption, comfort and other performance indicators, and is particularly suitable for energy modeling under sparse data set conditions. Key parameters include building envelope characteristics (walls, windows, etc.), internal loads (equipment, personnel, etc.), and HVAC system parameters. These parameter values are determined through literature research on climate adaptability to ensure the regional applicability of the model.

[0044] As a feasible implementation, parameter selection is a decisive factor in simulation accuracy. The present application adopts three categories of 19 key parameters, including building physical characteristics (U value, window-wall ratio, etc.), use characteristics (personnel density, equipment load, etc.), and system parameters (heating set value, air conditioning energy efficiency, etc.), to explore the impact of different design schemes through parameter combination. To overcome the problem of parameter combination explosion, the Latin hypercube sampling (LHS) method is applied. Compared with simple random sampling (SRS), LHS can achieve more uniform sample distribution in the parameter space, ensuring the balance of each parameter combination. This method successfully generates a synthetic data set covering 1 million parameter combinations, providing sufficient training samples for subsequent machine learning modeling.

[0045] As a feasible implementation, the benefits of parameterized simulation are reflected in three aspects: first, through LHS sampling optimization, the data representativeness is guaranteed while the computational load is greatly reduced; second, the integration of the EnergyPlus physical engine and the DesignBuilder template library realizes the standardization of parameterized modeling; finally, the generated synthetic data set not only contains building physical characteristics, but also covers non-geometric parameters such as use behavior, providing multi-dimensional feature support for data-driven modeling. This parameterization method not only solves the data bottleneck of urban-scale building energy efficiency modeling, but also the generated parameter combination library can be directly used for the effect pre-evaluation of energy-saving schemes, providing decision support for building energy-saving renovation.

[0046] S104、According to the multi-dimensional energy efficiency feature data set, the building energy efficiency data is processed for energy efficiency performance prediction and evaluation of non-separation models, terminal demand isolation models, and set isolation models, to determine the optimal energy efficiency performance prediction model.

[0047] It should be noted that, as Figure 2As shown, a machine learning model is needed to estimate building energy efficiency, i.e., building energy performance. Data preprocessing is first performed on the multidimensional energy efficiency feature dataset, which begins with data preprocessing to identify and eliminate inconsistencies in the dataset before the data is used for further analysis and model development. Data splitting is then performed, i.e., the preprocessed multidimensional energy efficiency feature dataset is split into two subsets to ensure optimal training of the model: one is a training dataset for training the model, and the other is a testing dataset for evaluating the performance of the trained model. Two standard techniques for data splitting are random data splitting and cross-validation.

[0048] wherein random data splitting is a straightforward method that randomly splits the data into a training dataset and a testing dataset, usually with a split ratio of 80%-20%. However, this method can lead to uneven distribution of data, and incorrect selection of the training dataset and the testing dataset can adversely affect the performance of the machine learning model. On the other hand, cross-validation is a more complex method that is usually used to strike a balance between the minimum bias and variance of the trained model. The present application uses a k-fold cross-validation algorithm for data splitting to prevent overfitting or underfitting of the model.

[0049] Specifically, in combination with the pre-set multi-type machine learning algorithm, the energy efficiency data in the multidimensional energy efficiency feature dataset is trained and constructed with a prediction model to obtain a non-separation model for predicting energy efficiency performance. The multi-type machine learning algorithm at least includes: linear regression, neural network, decision tree, random forest, and gradient boosting.

[0050] As a feasible implementation, in a single model method (also referred to as a "non-separation" method), the present application compares and analyzes various machine learning algorithms, evaluates their prediction accuracy, ease of use, and applicability to building energy efficiency modeling. In recent years, machine learning models have received extensive attention in the field of data-driven modeling. The most commonly used models include linear regression (LR), neural network (NN), decision tree (DT), random forest (RF), K-nearest neighbor (KNN), gradient boosting (GB), and support vector regression (SVR). Some popular implementations of gradient boosting include XGBoost (extreme gradient boosting), histogram-based gradient boosting (HGB), and LGBM (light gradient boosting machine). Due to the widespread application and success of these algorithms in previous studies, they have shown excellent performance in energy prediction and forecasting, especially in energy modeling. By evaluating the effectiveness of these models, the present application aims to find the most effective method for predicting building energy efficiency using machine learning technology.

[0051] Further, through a preset regression model, the terminal demand data related to the energy use type in the multi-dimensional energy efficiency feature data set is used to calculate the energy use intensity, and the energy use intensity index under each energy use type is obtained; and based on the energy use intensity index, a terminal demand isolation model is constructed. The terminal demand data at least includes heating energy use data, refrigeration energy use data, equipment energy use data, lighting energy use data, and hot water energy use data.

[0052] As a feasible implementation, Figure 3 A terminal use demand isolation modeling process for predicting energy use intensity using machine learning is provided for the embodiments of the present application, as shown in Figure 3 As shown, the terminal demand separation method uses different machine learning models to predict each terminal demand. This strategy is different from the traditional method of using a single machine learning model. This modification aims to achieve higher prediction performance (see the figure description for details Figure 3 ). The workflow, i.e., terminal demand data includes developing different regression machine learning models for each terminal demand such as heating, refrigeration, lighting, and hot water. The prediction results of these end uses will be aggregated into mandsto calculate the final energy performance of the building and measured in energy use intensity (EUI). The predicted value of each terminal demand is multiplied by the corresponding primary energy factor. Then the resulting heating, refrigeration, equipment, lighting, and hot water values are aggregated and the photovoltaic power generation is deducted from them to calculate the total energy consumption of the building. Then the cumulative total is divided by the building area to calculate the energy use intensity (EUI), which is an indicator of building energy efficiency, as follows:

[0053] Finally, the EUI is divided into energy performance certificate (EPC) labels or grades. Among them, E 加热 , E 制冷 , E 照明 , E 设备 , E 热水 , and E 光伏 represent the energy consumption (or E 光伏 power generation) of each relevant category, in units of kilowatt-hours per year (kWh / year).

[0054] PEF 加热 , PEF 制冷 , PEF 照明 , PEF 设备 , PEF 热水 , and PEF 光伏 are the primary energy factors (PEFS) of the respective categories. The total floor area of the building,

[0055] Further, by using the stacking technique and voting technique in the ensemble learning algorithm, the terminal demand data output by the terminal demand isolation model is trained for model prediction related to energy efficiency performance, thereby constructing an ensemble isolation model.

[0056] As a feasible implementation, Figure 4 An ensemble machine learning modeling schematic diagram for enhancing the prediction performance of a machine learning model is provided for the embodiments of the present application, as shown in Figure 4 The flow further implements an ensemble machine learning method to test multiple learning algorithms and obtain better prediction performance. Ensemble techniques are commonly used in machine learning to improve model accuracy by reducing overfitting and improving generalization. By taking advantage of the complementary strengths of multiple models, ensemble learning can provide more stable predictions and allow for detailed analysis of each model. Compared with traditional methods using a single model, ensemble learning can improve accuracy. There are two main ensemble learning techniques, which differ mainly in model types, data sampling, and decision functions. Therefore, ensemble learning techniques can be divided into stacking techniques and voting techniques.

[0057] As a feasible implementation, as shown in Figure 4 Stacking, also known as stacked generalization, is used to reduce the generalization error of different machine learning models. Through k-fold cross-validation, the final meta-model is composed of the prediction results of “n” machine learning-based models. On the other hand, voting ensemble is one of the most intuitive and easiest to understand methods. Voting ensemble is composed of “n” machine learning models, and the final prediction result is the prediction result with the highest probability of “winning the most votes” or weighted average probability. Generally, ensemble learning techniques use multiple machine learning models with the best prediction performance. The present application uses a semi-learning method based on stacking to predict the demand of each terminal use, thereby improving the accuracy of the model and predicting building energy efficiency. This method combines the predictions of multiple models by training another model to merge its output, which often results in more accurate and robust predictions compared with voting ensemble

[0058] Further, the SHAP technique can be used to calculate the prediction accuracy contribution of each terminal demand feature in the non-isolation model, terminal demand isolation model, and ensemble isolation model, thereby obtaining the prediction feature contribution. The terminal demand features are derived from the terminal demand data.

[0059] Further, by predicting the importance of the prediction impact of each terminal demand feature based on the prediction feature contribution, the key driving factor features related to building energy efficiency are obtained.

[0060] In one embodiment, it is also necessary to find out the importance of each terminal demand feature using the developed machine learning model. Feature importance refers to determining the relevance or contribution of individual features in a machine learning model to accurate prediction. It helps to understand which features have the greatest impact on model prediction. Among them, a common method to calculate feature importance is SHAP (SHapley Additive exPlanations). SHAP values provide a unified measurement method for calculating feature importance. Then the importance of each feature is determined by considering the contribution of each feature value to the prediction of a specific instance, while also considering the interaction between features. By using SHAP values, we can gain insight into which features have the greatest impact on model prediction. This information is very valuable for understanding the underlying relationships in the data and identifying key drivers or factors that affect the target variable. That is, perfectly combining data-driven methods with parametric simulation models to predict the energy performance of buildings in the city.

[0061] Further, based on the preset performance indicators, the energy efficiency performance of the non-separation model, the terminal demand isolation model and the set isolation model is calculated to obtain the energy use intensity prediction value of each machine learning model.

[0062] Further, the energy use intensity prediction value of each machine learning model is converted into an energy performance certification level.

[0063] Further, according to the energy performance certification level, the energy efficiency performance prediction evaluation process of each machine learning model is performed to determine the prediction accuracy of each machine learning model.

[0064] Further, based on the ranking of prediction accuracy, the prediction results of each machine learning model are combined with corresponding weight values to finally obtain an integrated multi-type machine learning model.

[0065] As a feasible implementation, in order to evaluate the effectiveness of the machine learning model, the present application adopts some commonly used performance indicators such as R-square (R 2 ), mean absolute error (MAE) and root mean square error (RMSE). The model with the lowest RMSE value and MAE value and the R 2 value close to 1 is the best model. Among all models, the model with a prediction value of 1 is considered better. Finally, in order to evaluate the accuracy of the model, the predicted value of EUI is converted into an energy performance certification (EPC) label or level (energy performance certification level). In addition, precision and recall are also crucial. Precision is an indicator used for detailed analysis of each class. Precision evaluates the accuracy of positive predictions made by the model, while recall quantifies the ability of the model to detect all positive instances in the dataset.

[0066] Furthermore, based on the integrated multi-type machine learning model, the key driving factor characteristics are used to train the energy efficiency performance prediction of the relevant building complex to generate the optimal energy efficiency performance prediction model.

[0067] S105. Perform energy efficiency analysis on the building complex using the optimal energy efficiency performance prediction model to obtain building energy efficiency prediction results.

[0068] Specifically, the current key building data of the building complex are first input into the optimal energy efficiency performance prediction model.

[0069] Furthermore, the optimal energy efficiency performance prediction model is used to perform energy efficiency analysis on the current building key data under the building energy performance, and finally the building energy efficiency prediction result is obtained.

[0070] In one embodiment, Figure 2 As shown, using the optimal energy performance prediction model, the developed machine learning model will predict the energy performance of an entire building complex. Comprehensive building stock data can help stakeholders analyze city-scale building stocks and successfully implement sustainable energy policies. Furthermore, the developed model can be applied in real-world scenarios, such as implementing and evaluating proposed retrofit measures as part of national-level policy decisions. These measures, often proposed at the national level, aim to improve the energy performance of existing buildings through retrofits and improvements. For example, this might include installing heat pumps or integrating renewable energy systems such as solar panels. The proposed model can assess their impact before implementation and identify potential energy savings. This predictive capability reduces the risk of implementing ineffective or inappropriate measures and ensures optimal resource utilization. It also facilitates fine-tuning these measures to better suit the specific needs and constraints of the building stock.

[0071] In addition, the embodiment of the present application also provides an energy performance prediction device based on urban residential buildings, such as Figure 5 As shown, the energy performance prediction device 500 based on urban residential buildings specifically includes:

[0072] At least one processor 501. And a memory 502 in communication with the at least one processor 501. The memory 502 stores instructions that can be executed by the at least one processor 501, so that the at least one processor 501 can execute:

[0073] Collect and process building inventory data, meteorological data, and auxiliary data for buildings within the city to obtain key building data;

[0074] Generate a building benchmark model based on key building data;

[0075] The building group is simulated for energy consumption by using a building benchmark model to obtain a multi-dimensional energy efficiency characteristic data set.

[0076] According to the multi-dimensional energy efficiency characteristic data set, the building energy efficiency data is subjected to energy efficiency performance prediction and evaluation processing related to a non-separation model, a terminal demand isolation model, and a set isolation model, to determine an optimal energy efficiency performance prediction model.

[0077] The building group is subjected to energy efficiency analysis by using the optimal energy efficiency performance prediction model to obtain building energy efficiency prediction results.

[0078] The embodiments of the present application efficiently predict the energy performance of urban buildings by using machine learning technology, effectively solving the bottleneck problem of high complexity and large data demand of traditional physical models. By using parameterized simulation and Latin hypercube sampling technology, a large number of synthetic data sets of buildings are generated, significantly improving the coverage and practical value of the model. Moreover, by integrating ensemble learning and terminal demand separation technology, the prediction accuracy is greatly improved by using advanced algorithms such as XGBoost and LightGBM. At the same time, the SHAP explainability analysis method is used to accurately identify key features such as air exchange rate and wall U value. The cross-disciplinary integration of building physical simulation, parameterized modeling, and machine learning technology is also achieved, and a full-process analysis framework from data collection to decision support is constructed, providing a systematic solution for urban building energy management, which has important theoretical value and practical significance.

[0079] The embodiments of the present application are described in a progressive manner, and the same or similar parts of each embodiment can be referred to each other. Each embodiment mainly describes the differences from other embodiments. In particular, the device and medium embodiments are described simply because they are basically similar to the method embodiments, and the relevant parts can be referred to the description of the method embodiments.

[0080] The device and medium provided by the embodiments of the present application are one-to-one corresponding to the method, so the device and medium also have similar beneficial technical effects as the method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the device and medium will not be described here.

[0081] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0082] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 means for functionally implementing the steps listed in the flowchart block or blocks.

[0083] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 means for functionally implementing the steps listed in the flowchart block or blocks.

[0084] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 means for functionally implementing the steps listed in the flowchart block or blocks.

[0085] In one typical configuration, the computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0086] The memory can include non-persistent memory and / or volatile memory, such as random access memory (RAM) and / or cache memory, for storing instructions and data used and / or generated by the computing device. The memory can also include non-volatile memory, such as read-only memory (ROM), electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), flash memory, or nonvolatile random access memory (RAM) (e.g., fault tolerant RAM), for storing instructions and data used and / or generated by the computing device. The memory is an example of computer-readable media.

[0087] Computer-readable media includes permanent and non-permanent, movable and non-movable media that can implement information storage by any method or technology. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic tape disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer-readable media does not include transitory media such as modulated data signals and carriers.

[0088] It should also be noted that the terms "comprising", "containing", or any other variant thereof are intended to cover non-exclusive inclusions, so that a process, method, article or apparatus that includes a list of elements not only includes those elements, but also includes other elements not explicitly listed, or inherent to such a process, method, article or apparatus. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or apparatus that includes the element.

[0089] The above only describes the embodiments of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the scope of the specification of the present application.

Claims

1. A method for predicting energy performance of urban residential buildings, characterized in that: The method comprises: Collect and process building inventory data, meteorological data, and auxiliary data for buildings within the city to obtain key building data; generating a building benchmark model based on the key building data; Performing energy consumption simulation on the building complex using the building benchmark model to obtain a multi-dimensional energy efficiency feature data set; Based on the multidimensional energy efficiency feature data set, the building energy efficiency data is subjected to energy efficiency performance prediction and evaluation processing related to the non-separation model, the terminal demand isolation model, and the collective isolation model to determine the optimal energy efficiency performance prediction model; The optimal energy efficiency performance prediction model is used to perform energy efficiency analysis on the building complex to obtain building energy efficiency prediction results.

2. The energy performance prediction method based on urban residential buildings according to claim 1 is characterized in that: Collect and process building stock data, meteorological data, and auxiliary data for buildings within the city to obtain key building data, including: Using the ABULA professional database, geometric features of buildings within the city are collected to obtain building geometric features, wherein the geometric features include at least: building envelope, building shape, and number of floors; Through population census and energy efficiency certificates, non-geometric features of buildings within the city are collected to obtain non-geometric features of buildings; wherein the non-geometric features include at least: energy usage pattern and energy equipment load; Combining the building geometric features and the building non-geometric features to obtain the building stock data; The building stock data, the meteorological data and the auxiliary data are subjected to multi-source key data collection to obtain the building key data; wherein the meteorological data includes: typical meteorological year data, future climate scenario data and EnergyPlus weather files covering the world; the auxiliary data includes at least: census data and energy policy report data.

3. The energy performance prediction method based on urban residential buildings according to claim 1 is characterized in that: Based on the key building data, a building benchmark model is generated, specifically including: Extract reference building standard models through the national building complex database; The reference building standard model is simulated with respect to parameters of the baseline model using the key building data, and the building benchmark model applied to the building complex is determined based on the simulated building prototype covering a series of parameter combinations.

4. The energy performance prediction method based on urban residential buildings according to claim 1 is characterized in that: The energy consumption of the building complex is simulated using the building benchmark model to obtain a multi-dimensional energy efficiency feature dataset, specifically including: Using a Latin hypercube sampling method, the building energy efficiency impact characteristics in the building benchmark model are subjected to a uniform sample distribution process in a related parameter space to obtain initial building performance data; wherein the building energy efficiency impact characteristics include: building physical characteristics, usage characteristics, and thermal system parameters; Based on the integrated EnergyPlus physics engine and DesignBuilder template library, a complex energy consumption simulation under multi-parameter coordination is performed on the initial building performance data to obtain standard building performance data; wherein the standard building performance data includes: building energy consumption performance indicators and building comfort indicators; Data collection is performed on the multidimensional energy efficiency characteristics in the standard building performance data to generate the multidimensional energy efficiency characteristic data set.

5. The energy performance prediction method based on urban residential buildings according to claim 1 is characterized in that: Before performing energy efficiency performance prediction and evaluation processing on the building energy efficiency data using the non-separation model, the terminal demand isolation model, and the collective isolation model based on the multi-dimensional energy efficiency feature data set and determining the optimal energy efficiency performance prediction model, the method further includes: The energy efficiency data in the multi-dimensional energy efficiency feature dataset are trained and constructed using a preset multi-type machine learning algorithm to obtain the non-separable model for predicting energy efficiency performance; wherein the multi-type machine learning algorithm includes at least: linear regression, neural network, decision tree, random forest, and gradient boosting; Using a preset regression model, energy use intensity is calculated for the terminal demand data of the energy use types in the multi-dimensional energy efficiency feature dataset to obtain an energy use intensity index for each energy use type; and based on the energy use intensity index, the terminal demand isolation model is constructed; wherein the terminal demand data includes at least: heating energy use data, cooling energy use data, equipment energy use data, lighting energy use data, and hot water energy use data; The terminal demand data output by the terminal demand isolation model is subjected to model prediction training on energy efficiency performance through the stacking technology and voting technology in the set decomposition learning algorithm to construct the set isolation model.

6. The energy performance prediction method based on urban residential buildings according to claim 5 is characterized in that: Based on the multi-dimensional energy efficiency feature data set, the building energy efficiency data is subjected to energy efficiency performance prediction and evaluation processing related to the non-separation model, the terminal demand isolation model, and the collective isolation model to determine the optimal energy efficiency performance prediction model, specifically including: Using a preset SHAP technique, the contribution of each terminal demand feature in the non-separated model, the terminal demand isolation model, and the collective isolation model is calculated to obtain a prediction feature contribution; wherein the terminal demand feature is derived from the terminal demand data; Based on the predicted feature contribution, each terminal demand feature is divided into the importance of the predicted impact, and the key driving factor characteristics of the building energy efficiency are obtained; Based on an integrated multi-type machine learning model, the key driving factor characteristics are trained on the energy efficiency performance prediction of the building complex to generate the optimal energy efficiency performance prediction model.

7. The energy performance prediction method based on urban residential buildings according to claim 6 is characterized in that: Based on an integrated multi-type machine learning model, the key driving factor characteristics are trained on the energy efficiency performance prediction of the building complex to generate the optimal energy efficiency performance prediction model, specifically including: Based on preset performance indicators, predict and calculate the energy efficiency performance of the non-separation model, the terminal demand isolation model, and the collective isolation model to obtain a predicted value of energy usage intensity for each machine learning model; Convert each machine learning model’s energy use intensity predictions into an energy performance certification rating; According to the energy performance certification level, each machine learning model is evaluated for energy efficiency performance prediction to determine the prediction accuracy of each machine learning model; Based on the ranking of the prediction accuracy, the prediction results of each machine learning model are combined with the data of the corresponding weight values ​​to obtain an integrated multi-type machine learning model.

8. The energy performance prediction method based on urban residential buildings according to claim 1 is characterized in that: The energy efficiency analysis of the building complex is performed using the optimal energy efficiency performance prediction model to obtain building energy efficiency prediction results, specifically including: inputting current key building data of the building complex into the optimal energy efficiency performance prediction model; The optimal energy efficiency performance prediction model is used to perform energy efficiency analysis on the current building key data under the relevant building energy performance to obtain the building energy efficiency prediction result.

9. An energy performance prediction device based on urban residential buildings, characterized in that: The device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, so that the at least one processor can execute the energy performance prediction method based on urban residential buildings according to any one of claims 1-8.

10. A non-volatile computer storage medium, characterized in that The storage medium is a non-volatile computer-readable storage medium, which stores at least one program. Each of the programs includes instructions. When the instructions are executed by the terminal, the terminal executes the energy performance prediction method based on urban residential buildings according to any one of claims 1 to 8.

Images