Building energy consumption prediction and control system based on machine learning
Through the building energy consumption prediction and management system based on machine learning, combined with the CFD model and graph neural network, the problem of insufficient adaptability and real-time performance of the building complex energy consumption prediction model is solved, and high-precision prediction of building energy consumption is achieved and intelligent management.
Patent Information
- Application Number
- CN202411252868.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-09
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-09-09
AI Technical Summary
The prior art is difficult to accurately simulate and capture the air duct effect in building complexes and its impact on energy consumption, resulting in poor adaptability and insufficient real-time performance of the energy consumption prediction model.
Adopting a building energy consumption prediction and control system based on machine learning, including data integration processing module, simulation parameter extraction module, deep feature analysis module, key variable network module and energy consumption prediction model module, extracted through calculation of CFD model and time-frequency analysis Microenvironment changes parameters, use graph neural network to build a key variable interaction network, and perform hyperparameter optimization and cross-validation of machine learning models.
It realizes high-precision prediction of building energy consumption, can effectively respond to rapidly changing environmental conditions, and improves the intelligence level of building complex energy consumption management.
Smart Images

Figure CN118761859B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of building energy consumption management, and more specifically, to a building energy consumption prediction and control system based on machine learning. Background Art
[0002] In modern urban planning and architectural design, energy management of building complexes has become an important research area, especially in the pursuit of sustainable development and energy conservation and consumption reduction. The energy consumption of buildings is affected by many factors, including the design of the building, the materials used, the location of the building, and its interaction with the surrounding environment. Among them, the wind duct effect in the building complex has a significant impact on energy consumption. This effect is caused by the relative position and spatial layout between buildings, resulting in changes in wind speed and direction in specific areas of the building complex, which in turn affects the thermal environment and the ventilation needs of the building.
[0003] When dealing with the wind duct effect in a building complex, existing technologies usually have difficulty accurately simulating and capturing complex airflow changes and their impact on building energy consumption. Traditional computational fluid dynamics (CFD) models are often unable to accurately predict microenvironmental parameters when faced with large-scale and high-dimensional data, resulting in a lack of accuracy in energy consumption models. In addition, existing methods are deficient in feature extraction and dynamic interaction analysis between variables, and fail to fully utilize potential deep-level information. This results in poor adaptability of energy consumption prediction models, making it difficult to cope with rapidly changing environmental conditions, and having obvious deficiencies in real-time and accuracy.
[0004] In order to solve the above problems, a technical solution is now provided. Summary of the invention
[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a building energy consumption prediction and control system based on machine learning to solve the problems raised in the above-mentioned background technology.
[0006] To achieve the above object, the present invention provides the following technical solutions:
[0007] Building energy consumption prediction and control system based on machine learning, including data integration and processing module, simulation parameter extraction module, deep feature analysis module, key variable network module and energy consumption prediction model module;
[0008] Data integration and processing module: collects the location, design features and surrounding environment data of the building complex, and performs cleaning and normalization processing;
[0009] Simulation parameter extraction module: Use the computational CFD model to simulate the wind duct effect in the building complex, and extract microenvironment change parameters from the CFD model data through time-frequency analysis. The microenvironment change parameters include local turbulence intensity and pressure fluctuations;
[0010] Deep feature analysis module: multi-dimensional scaling and principal component analysis are used to extract deep features of the extracted microenvironment change parameters to identify key variables that affect energy consumption;
[0011] Key variable network module: Use graph neural network to build an interaction network based on key variables that affect energy consumption, and predict how key variables that affect energy consumption jointly affect the energy consumption of buildings;
[0012] Energy consumption prediction model module: Develop energy consumption prediction models and use machine learning methods to optimize model hyperparameters and cross-validate.
[0013] In a preferred embodiment, the location, design features and surrounding environment data of the building complex are collected and cleaned and standardized, specifically:
[0014] Collect multi-dimensional data related to the building complex from multiple data sources; data sources include but are not limited to architectural design drawings and structural documents, geographic information system data, environmental monitoring data, and building usage data;
[0015] The collected data is preprocessed, which includes: using outlier detection algorithms to identify and process abnormal data points; converting the collected heterogeneous data into a unified format; normalizing data of different scales; and integrating data from different sources.
[0016] In a preferred embodiment, a computational CFD model is used to simulate the wind duct effect in a building complex, and microenvironment change parameters are extracted from the CFD model data through time-frequency analysis. The microenvironment change parameters include local turbulence intensity and pressure fluctuations, specifically:
[0017] Based on the design characteristics of the building complex and environmental data, build the CFD model, set the turbulence model and boundary conditions, and calibrate the initial conditions;
[0018] Adopting adaptive mesh generation algorithm, the building complex is divided into fine meshes and meshes are encrypted in key areas;
[0019] Use high-performance computing resources to run CFD simulations and generate high-frequency fluid dynamics data, including velocity field, pressure field, and vorticity;
[0020] Through short-time Fourier transform and wavelet transform, time-frequency analysis is performed on high-frequency fluid dynamics data to extract local turbulence intensity and pressure fluctuation parameters;
[0021] The extracted local turbulence intensity and pressure fluctuation parameters are statistically analyzed and integrated to generate a data set of microenvironment change parameters.
[0022] In a preferred embodiment, multidimensional scaling and principal component analysis are used to perform deep feature extraction on the extracted microenvironment change parameters to identify key variables affecting energy consumption, specifically:
[0023] The extracted microenvironment change parameters were standardized, and the Z-score standardization method was used to eliminate the dimensional differences between different parameters;
[0024] Recursive feature elimination was used to preliminarily screen the standardized data and remove redundant and low-relevance features;
[0025] Multidimensional scaling is applied to project the filtered high-dimensional data into a low-dimensional space, preserving the nonlinear distance relationship between parameters;
[0026] Based on the data processed by multidimensional scaling, principal component analysis is performed to further reduce the dimension and extract the main components in the data, identifying the key variables that have the greatest impact on energy consumption;
[0027] The key variables extracted through multidimensional scaling and principal component analysis were interpreted, and the specific effects of the key variables on energy consumption were verified using correlation analysis.
[0028] In a preferred embodiment, a graph neural network is used to construct an interaction network based on key variables affecting energy consumption to predict how key variables affecting energy consumption jointly affect the energy consumption of a building, specifically:
[0029] The identified key variables are defined as nodes in the graph neural network, and each node represents an independent factor affecting energy consumption;
[0030] Use correlation analysis and mutual information calculation to determine the relationship between key variables and construct edges in graph neural networks;
[0031] Initialize the constructed graph structure and set the initial node feature vector and edge weight;
[0032] Use graph convolutional networks to train graph structures, propagate node features layer by layer, update node feature vectors and edge weights, and capture interactions between variables.
[0033] Through the trained graph convolutional network model, it predicts how key variables work together on building energy consumption and outputs the predicted results of building energy consumption.
[0034] In a preferred embodiment, an energy consumption prediction model is developed, and a machine learning method is used to optimize the hyperparameters and cross-validate the model, specifically:
[0035] Selecting suitable machine learning algorithms for building energy consumption prediction;
[0036] Perform feature engineering on input data to extract and construct new features to enhance the model’s sensitivity and predictive ability to changes in building energy consumption;
[0037] Train the model on the training data set, use the initial hyperparameter combination, verify the model performance through cross-validation, and evaluate its generalization ability;
[0038] Bayesian optimization is used to optimize the model's hyperparameters and determine the optimal parameter combination;
[0039] The optimized model performance is evaluated through the test data set, the mean square error of key indicators is calculated, and then the final model is deployed to the actual energy consumption management system.
[0040] In a preferred embodiment, the key indicator mean square error is calculated by using a test data set to perform a final evaluation on the model and calculating the performance indicator mean square error of the model on the test data set: ;in, is the mean square error, Indicates the actual energy consumption, is the energy consumption value predicted by the model, is the number of test samples, is the test sample number.
[0041] Technical effects and advantages of the building energy consumption prediction and control system based on machine learning of the present invention:
[0042] 1. Through the data integration and processing module, the building and environmental data are comprehensively collected and standardized to ensure the quality and consistency of the input data, providing a solid foundation for subsequent analysis. Secondly, the simulation parameter extraction module uses advanced computational CFD models and time-frequency analysis methods to accurately capture the wind duct effect in the building complex and its microscopic impact on energy consumption, ensuring the high-precision extraction of microenvironment change parameters.
[0043] 2. Through the deep feature analysis module, the present invention can effectively identify and extract key variables that affect energy consumption, and adopt multidimensional scaling and principal component analysis technology to fully explore the deep features in the data, thereby improving the accuracy and relevance of feature extraction. The key variable network module introduces graph neural networks to construct a dynamic interaction network between key variables, realizes the modeling and analysis of complex interactive relationships between variables, and provides deep insights into energy consumption forecasting. Finally, the energy consumption prediction model module uses machine learning methods to perform hyperparameter optimization and cross-validation to ensure the high adaptability and real-time performance of the model, making energy consumption forecasts more accurate, able to effectively respond to rapidly changing environmental conditions, and improving the level of intelligent energy consumption management in building complexes. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] Figure 1It is a structural schematic diagram of the building energy consumption prediction and control system based on machine learning of the present invention. DETAILED DESCRIPTION
[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Example
[0046] Figure 1 A structural schematic diagram of the building energy consumption prediction and control system based on machine learning of the present invention is given. The building energy consumption prediction and control system based on machine learning includes a data integration and processing module, a simulation parameter extraction module, a deep feature analysis module, a key variable network module and an energy consumption prediction model module.
[0047] Data integration and processing module: collects the location, design features and surrounding environment data of the building complex, and performs cleaning and normalization processing.
[0048] Simulation parameter extraction module: Use the computational CFD model to simulate the wind duct effect in the building complex, and extract the microenvironment change parameters from the CFD model data through time-frequency analysis. The microenvironment change parameters include local turbulence intensity and pressure fluctuations.
[0049] Deep feature analysis module: Multidimensional scaling and principal component analysis are used to perform deep feature extraction on the extracted microenvironment change parameters to identify key variables that affect energy consumption.
[0050] Key variable network module: Use graph neural network to build an interaction network based on key variables that affect energy consumption, and predict how key variables that affect energy consumption jointly affect the energy consumption of buildings.
[0051] Energy consumption prediction model module: Develop energy consumption prediction models and use machine learning methods to optimize model hyperparameters and cross-validate.
[0052] Collect the location, design features and surrounding environment data of the building complex, and clean and standardize them, specifically:
[0053] Collect multi-dimensional data related to the building complex from multiple data sources. Data sources include but are not limited to:
[0054] Architectural design drawings and structural documents: Capture design features such as building height, area, volume, window area, material type, number of floors, etc.
[0055] Geographic Information System (GIS) data: including the building’s geographic location, orientation, surrounding terrain, and relative positions to neighboring buildings.
[0056] Environmental monitoring data: Obtain environmental parameters around the building, such as average annual temperature, humidity, wind speed, wind direction, solar radiation intensity, etc. These data can be obtained through satellite data, weather station sensors, or field sensor networks.
[0057] Building usage data: such as the nature of the building’s use (commercial, residential, industrial, etc.), daily usage patterns, personnel flow, etc.
[0058] After completing data collection, the collected data is preprocessed to ensure data quality and consistency. Preprocessing includes the following specific steps:
[0059] Use outlier detection algorithms (such as the Z-score method based on statistical distribution or the isolation forest algorithm based on machine learning) to identify and process abnormal data points. For missing data, interpolation, mean filling, or multiple imputation based on collaborative data (MICE) are used to ensure data integrity.
[0060] Convert the collected heterogeneous data into a unified format, for example, convert geographic location data from different sources into a unified coordinate system (such as WGS84 coordinate system) to ensure the comparability and consistency of the data.
[0061] Normalize data of different scales. For example, the minimum-maximum normalization method is used for the physical characteristics of buildings (such as height and area) to normalize the data to the range of [0,1]. For environmental data (such as temperature and humidity), Z-score normalization is used according to their distribution characteristics to eliminate the impact of dimension on the analysis.
[0062] Integrate data from different sources. Use spatial data fusion technology to align GIS data with building design data, and associate environmental monitoring data with building locations and design features to build a complete multidimensional data set. Use multi-source data fusion algorithms, such as weighted average or Bayesian reasoning-based fusion methods, to ensure interoperability between data.
[0063] The computational CFD model is used to simulate the wind duct effect in the building complex, and the microenvironment change parameters are extracted from the CFD model data through time-frequency analysis. The microenvironment change parameters include local turbulence intensity and pressure fluctuations, specifically:
[0064] Based on the design characteristics of the building complex and environmental data, the CFD model is constructed, the turbulence model and boundary conditions are set, and the initial conditions are calibrated:
[0065] A three-dimensional computational CFD model is established based on the specific design features of the building complex (such as building height, volume, orientation, etc.) and surrounding environmental data (such as terrain, wind speed, and wind direction). When selecting a turbulence model, the geometric complexity of the building and the fluid flow characteristics are taken into account, and turbulence phenomena of different scales are accurately simulated using models such as the k-ε model or large eddy simulation (LES). The boundary conditions are set based on the actual geographical location and meteorological data to ensure that the settings of the fluid inlet and outlet are consistent with the actual environment. By calibrating the initial conditions (such as adjusting the wind speed and direction, and setting the ambient temperature), it is ensured that the simulation results can accurately reflect the actual aerodynamic characteristics of the building complex. This meticulous model construction and initial condition setting avoids the simulation errors that may occur in conventional CFD models and ensures the high accuracy and applicability of the model.
[0066] Adopting adaptive mesh generation algorithm, the building complex is divided into fine meshes and meshes are encrypted in key areas:
[0067] In the constructed CFD model, adaptive mesh generation algorithms (such as Dynamic Mesh Refinement) are applied to finely mesh different parts of the building. In particular, local mesh encryption technology is used to further refine the mesh units at the edges of the building, key areas of the wind duct, and areas with high turbulence intensity. This mesh generation method combines the characteristics of fluid flow to achieve higher resolution simulation of key areas, avoiding the shortcomings of conventional meshing that may ignore details. At the same time, by dynamically adjusting the mesh during the simulation process, the computational efficiency and accuracy of the simulation are maintained. This innovative meshing strategy ensures that high-quality fluid dynamics data can be obtained even with limited resources.
[0068] Use high-performance computing resources to run CFD simulations and generate high-frequency fluid dynamics data, including velocity fields, pressure fields, and vorticity:
[0069] After the meshing is completed, high-performance computing (HPC) resources are used to run CFD simulations to generate high-frequency fluid dynamics data. These data include core variables such as velocity field, pressure field, and vorticity. Time-StepRefinement technology is used during simulation to ensure that the time resolution of the data can reflect the rapid changes in turbulence. The data acquisition frequency is set to be much higher than the sampling frequency of traditional CFD simulations to ensure that small-scale eddies and pressure fluctuations are captured. This high-precision, high-frequency data generation method breaks through the limitations of conventional CFD simulations in time and space resolution, and provides rich data support for subsequent time-frequency analysis.
[0070] Through short-time Fourier transform and wavelet transform, time-frequency analysis is performed on high-frequency fluid dynamics data to extract local turbulence intensity and pressure fluctuation parameters:
[0071] Time-frequency analysis is performed on the high-frequency fluid dynamics data generated by CFD simulation. First, the short-time Fourier transform (STFT) is used to perform a preliminary spectrum analysis on the flow field data to identify the main frequency components of turbulence and pressure fluctuations. Then, the wavelet transform is used to perform multi-scale decomposition of the data within the key frequency range to extract local turbulence intensity and pressure fluctuation parameters with high time and frequency resolution. This method is different from traditional frequency domain analysis in that it can simultaneously capture the time and frequency information of flow field changes, allowing the dynamic flow field characteristics to be accurately separated and identified.
[0072] This innovative time-frequency analysis technique enables a better understanding of transient phenomena in complex flow fields and their impact on building energy consumption.
[0073] Statistical analysis and integration of the extracted local turbulence intensity and pressure fluctuation parameters are performed to generate a microenvironment change parameter data set:
[0074] The local turbulence intensity and pressure fluctuation parameters extracted from the time-frequency analysis are further statistically analyzed. First, the parameter distribution fitting technique is applied to perform probability distribution modeling on the extracted characteristic parameters to determine their statistical characteristics under different environmental conditions. Then, extreme value theory analysis is used to identify outliers or extreme conditions that may have the greatest impact on building energy consumption. After completing the statistical analysis, these parameters are integrated with the design characteristics and environmental data of the building complex to generate a comprehensive microenvironment change parameter dataset. This dataset not only contains static building and environmental characteristics, but also incorporates dynamic microclimate change information, providing comprehensive and accurate input for subsequent deep feature extraction and energy consumption prediction.
[0075] Multidimensional scaling and principal component analysis were used to extract deep features of the extracted microenvironment change parameters to identify the key variables affecting energy consumption, specifically:
[0076] The extracted microenvironment change parameters were standardized, and the Z-score standardization method was used to eliminate the dimensional differences between different parameters:
[0077] Specifically, the original value of each microenvironment change parameter is set to (in represents different microenvironment change parameters), and its standardized formula is: ;in, is the standardized microenvironment change parameter, express The mean of express By this method, the dimension differences between the parameters of different microenvironment changes are eliminated, so that The mean is 0 and the standard deviation is 1; this process ensures that all microenvironment change parameters are processed on the same scale, effectively avoiding deviations caused by differences in parameter ranges.
[0078] Recursive feature elimination is used to perform preliminary screening on the standardized data to remove redundant and low-relevance features:
[0079] After standardization, the recursive feature elimination (RFE) algorithm is used to perform preliminary screening of the data. RFE first trains the initial model (such as support vector machine, SVM) and calculates the impact of each feature on the model performance. The feature weight is set to (in represents different features), the model recursively removes the least important features according to their weights. The specific process is as follows: In each iteration, the feature with the smallest weight is removed , retrain the model, and evaluate its performance. This step is repeated until the number of remaining features meets the preset conditions. This ensures that the feature set retained in the end has the strongest predictive power while significantly reducing the data dimension. Compared with traditional one-time feature selection methods, RFE dynamically adjusts the feature set in each iteration, ensuring the robustness and efficiency of the model.
[0080] Apply multidimensional scaling to project the filtered high-dimensional data into a low-dimensional space, preserving the nonlinear distance relationship between parameters:
[0081] After the initial screening, the multidimensional scaling (MDS) method is applied to the remaining high-dimensional data to preserve the nonlinear distance relationship between parameters. The goal of MDS is to project the data points in the high-dimensional space into the low-dimensional space, maintaining the distance structure between the original data points. The data points in the high-dimensional space are set to (in represents different data points), MDS optimizes the function To minimize the difference between the Euclidean distance between points in the high-dimensional space and the distance between corresponding points in the low-dimensional space: ;in, Represents the midpoint of a high-dimensional space and The actual distance between Represents the corresponding distance in low-dimensional space. Through iterative optimization, MDS repositions these data points in low-dimensional space so that the distance structure retains nonlinear relationships to the maximum extent. This step is particularly suitable for building microenvironment data with complex interactions, thereby revealing hidden parameter correlations. MDS is able to capture complex nonlinear patterns in the data and provide richer feature representations.
[0082] Based on the data processed by multidimensional scaling, principal component analysis is performed to further reduce the dimension and extract the main components in the data, identifying the key variables that have the greatest impact on energy consumption:
[0083] After multidimensional scaling, principal component analysis (PCA) was performed to further reduce the data dimension and extract the main components. The data matrix was set to , whose covariance matrix is Through eigenvalue decomposition, the eigenvalues of the covariance matrix are obtained and the corresponding eigenvector (in Represents different principal components), PCA projects the data onto the principal component directions defined by these eigenvectors: ;in, Indicates In this way, PCA extracts the principal components that explain the most data variance, thereby reducing the data dimension and identifying the key variables that have the greatest impact on energy consumption. The application of PCA combined with MDS can more effectively remove redundant information while ensuring the capture of complex influencing factors of building energy consumption.
[0084] The key variables extracted by multidimensional scaling and principal component analysis were explained, and the specific impact of key variables on energy consumption was verified using correlation analysis:
[0085] Correlation analysis was used to explain and verify the extracted key variables. Specifically, the The Pearson correlation coefficient between the key variables and the building energy consumption is obtained. Through correlation analysis, the degree of influence of these key variables on energy consumption can be quantified, and the predictive ability of these variables can be further verified through regression analysis. In order to ensure the robustness of the results, a cross-validation technique is used to divide the data set into multiple subsets, and training and validation are performed on different subsets. This process ensures that the identified key variables are not only statistically significant, but also have stable predictive ability in practical applications.
[0086] A graph neural network is used to build an interaction network based on key variables that affect energy consumption, and predict how key variables that affect energy consumption jointly affect the energy consumption of buildings, specifically:
[0087] The identified key variables are defined as nodes in the graph neural network, and each node represents an independent factor affecting energy consumption:
[0088] In the graph neural network, the key variables identified from the deep feature analysis are considered as nodes Each node represents a factor that independently affects energy consumption, such as "temperature" and "humidity". This step provides a basis for building complex system interactions by abstracting each influencing factor into a network node.
[0089] Use correlation analysis and mutual information calculation to determine the relationship between key variables and construct edges in graph neural networks:
[0090] According to the correlation and mutual information between variables, the connection between nodes is established to form edges. Here, the edge weight It is obtained based on correlation analysis, and higher correlations produce larger weights. This step provides a data-driven basis for the structure of the network by accurately calculating the statistical relationship between nodes.
[0091] Initialize the constructed graph structure and set the initial node feature vector and edge weight:
[0092] Initialize the graph structure and assign an initial feature vector to each node and the edge weights The initial features of a node can be obtained from its statistical properties, such as the mean or standard deviation. This initialization process is the preparatory work before model training. The key is to set a suitable starting state for each node.
[0093] Use graph convolutional networks to train graph structures, propagate node features layer by layer, update node feature vectors and edge weights, and capture interactions between variables:
[0094] Apply the graph convolutional network (GCN) model to train the network and update the feature vector of the node through forward propagation and back propagation in an iterative process. Here, Represents the number of training iterations, and each iteration is updated according to the edge weights between nodes and the features of adjacent nodes. This step enables the network to more accurately simulate the complex interactions between nodes by learning and adjusting the feature expressions of nodes.
[0095] Through the trained graph convolutional network model, we predict how key variables work together to affect building energy consumption and output the predicted results of building energy consumption:
[0096] The trained graph convolutional network model is used to predict how key variables work together to affect building energy consumption. The model expresses the final features of each node by combining , and combined with their weighted relationships, calculate the overall energy consumption forecast. This prediction step is the core of the application phase and can provide an intuitive understanding of how key variables jointly affect energy consumption.
[0097] Develop an energy consumption prediction model and use machine learning methods to optimize the model's hyperparameters and cross-validate, specifically:
[0098] Select a machine learning algorithm suitable for building energy consumption prediction, ensuring that the model adapts to nonlinear characteristics:
[0099] Choose a machine learning algorithm suitable for building energy consumption prediction to adapt to the nonlinear relationships that may exist in the data. Applicable algorithms include Random Forest (RF) or Gradient Boosting Machine (GBM). Random Forest reduces the risk of overfitting by building multiple decision trees and averaging the prediction results of each tree. Gradient Boosting Machine improves the accuracy of the model by gradually training multiple weak learners (usually decision trees). When choosing a machine learning algorithm, preliminary experiments should be conducted based on the data characteristics (such as nonlinearity and high dimensionality of variables) to determine which algorithm can better handle the complex interaction effects in building energy consumption data.
[0100] Perform feature engineering on input data to extract and construct new features to enhance the model’s sensitivity and predictive ability to changes in building energy consumption:
[0101] In the data preprocessing stage, feature engineering is performed on the input data to extract and construct new features that are sensitive to changes in building energy consumption.
[0102] Specifically, we obtain the original feature set and construct new features such as quadratic terms and interaction terms to form a new feature set. , using methods such as principal component analysis (PCA) or factor analysis to further reduce feature dimensions and remove redundant features. This feature engineering process not only enhances the model's ability to capture key influencing factors, but also reduces model complexity and improves its generalization ability. When selecting features, algorithms such as recursive feature elimination (RFE) are used to ensure that the model only uses the most representative variables for prediction.
[0103] in, Represents the feature index in the new feature set.
[0104] Train the model on the training data set, use the initial hyperparameter combination, verify the model performance through cross-validation, and evaluate its generalization ability:
[0105] After the dataset is prepared, the model is trained using the initially selected hyperparameter combination. The training dataset is set to , containing the input features and target output Select a cross-validation method, such as f-fold cross-validation, to divide the training data into multiple subsets. Each time, select one of the subsets as the validation set, and use the remaining subsets minus one to train the model. When training the model, for the initial hyperparameter combination , adjust the weights in each iteration to minimize the loss function , such as mean square error (MSE). Through cross-validation, the generalization ability of the model can be evaluated to ensure its consistent performance on different data sets.
[0106] Bayesian optimization is used to optimize the model's hyperparameters and determine the optimal parameter combination:
[0107] To further improve the model performance, Bayesian optimization is used to optimize the model's hyperparameters. Assume that the initial hyperparameter combination is , the optimization goal is to find the optimal hyperparameter combination , so that the verification error Minimum. Bayesian optimization builds a proxy model (such as a Gaussian process) to predict new hyperparameter combinations based on past evaluation results. This proxy model effectively searches for optimal parameters by balancing exploration (finding new areas) and exploitation (refining existing optimal areas). Compared with traditional grid search, Bayesian optimization is more efficient, especially when the parameter space is large, and can significantly reduce computing time and cost.
[0108] The optimized model performance is evaluated through the test data set, the mean square error of key indicators is calculated, and then the final model is deployed to the actual energy consumption management system:
[0109] After optimizing the hyperparameters After confirmation, use the test data set Perform a final evaluation of the model and calculate the performance indicators of the model on the test set, especially the mean square error:
[0110] ;in, is the mean square error, Indicates the actual energy consumption, is the energy consumption value predicted by the model, is the number of test samples, is the test sample number. By evaluating the performance of the model on the test set, we can ensure that the generalization performance of the model is good. After the evaluation is completed, the optimized model is deployed to the actual building energy management system. The deployment process includes online monitoring and periodic updates of the model to ensure the accuracy and real-time performance of energy consumption prediction.
[0111] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.
[0112] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or may be transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium may be a solid-state hard disk.
[0113] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.
[0114] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and modules described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0115] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.
[0116] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, and may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0117] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.
[0118] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.
[0119] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
[0120] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. Building energy consumption prediction and control system based on machine learning, characterized by: It includes data integration and processing module, simulation parameter extraction module, deep feature analysis module, key variable network module and energy consumption prediction model module; Data integration and processing module: collects the location, design features and surrounding environment data of the building complex, and performs cleaning and normalization processing; Simulation parameter extraction module: Use the computational CFD model to simulate the wind duct effect in the building complex, and extract microenvironment change parameters from the CFD model data through time-frequency analysis. The microenvironment change parameters include local turbulence intensity and pressure fluctuations; Deep feature analysis module: multi-dimensional scaling and principal component analysis are used to extract deep features of the extracted microenvironment change parameters to identify key variables that affect energy consumption; Key variable network module: Use graph neural network to build an interaction network based on key variables that affect energy consumption, and predict how key variables that affect energy consumption jointly affect the energy consumption of buildings; Energy consumption prediction model module: Develop energy consumption prediction models and use machine learning methods to optimize model hyperparameters and cross-validate; A graph neural network is used to build an interaction network based on key variables that affect energy consumption, and predict how key variables that affect energy consumption jointly affect the energy consumption of buildings, specifically: The identified key variables are defined as nodes in the graph neural network, and each node represents an independent factor affecting energy consumption; Use correlation analysis and mutual information calculation to determine the relationship between key variables and construct edges in graph neural networks; Initialize the constructed graph structure and set the initial node feature vector and edge weight; Use graph convolutional networks to train graph structures, propagate node features layer by layer, update node feature vectors and edge weights, and capture interactions between variables. Through the trained graph convolutional network model, predict how key variables work together on building energy consumption and output the predicted results of building energy consumption; Develop an energy consumption prediction model and use machine learning methods to optimize the model's hyperparameters and cross-validate, specifically: Selecting suitable machine learning algorithms for building energy consumption prediction; Perform feature engineering on input data to extract and construct new features to enhance the model’s sensitivity and predictive ability to changes in building energy consumption; Train the model on the training data set, use the initial hyperparameter combination, verify the model performance through cross-validation, and evaluate its generalization ability; Bayesian optimization is used to optimize the model's hyperparameters and determine the optimal parameter combination; The optimized model performance is evaluated through the test data set, the mean square error of key indicators is calculated, and then the final model is deployed to the actual energy consumption management system.
2. The machine learning-based building energy consumption prediction and control system according to claim 1, characterized in that: Collect the location, design features and surrounding environment data of the building complex, and clean and standardize them, specifically: Collect multi-dimensional data related to the building complex from multiple data sources; data sources include but are not limited to architectural design drawings and structural documents, geographic information system data, environmental monitoring data, and building usage data; The collected data is preprocessed, which includes: using outlier detection algorithms to identify and process abnormal data points; converting the collected heterogeneous data into a unified format; normalizing data of different scales; and integrating data from different sources.
3. The building energy consumption prediction and control system based on machine learning according to claim 2 is characterized in that: The computational CFD model is used to simulate the wind duct effect in the building complex, and the microenvironment change parameters are extracted from the CFD model data through time-frequency analysis. The microenvironment change parameters include local turbulence intensity and pressure fluctuations, specifically: Based on the design characteristics of the building complex and environmental data, build the CFD model, set the turbulence model and boundary conditions, and calibrate the initial conditions; Adopting adaptive mesh generation algorithm, the building complex is divided into fine meshes and meshes are encrypted in key areas; Use high-performance computing resources to run CFD simulations and generate high-frequency fluid dynamics data, including velocity field, pressure field, and vorticity; Through short-time Fourier transform and wavelet transform, time-frequency analysis is performed on high-frequency fluid dynamics data to extract local turbulence intensity and pressure fluctuation parameters; The extracted local turbulence intensity and pressure fluctuation parameters are statistically analyzed and integrated to generate a data set of microenvironment change parameters.
4. The machine learning-based building energy consumption prediction and control system according to claim 3 is characterized in that: Multidimensional scaling and principal component analysis were used to extract deep features of the extracted microenvironment change parameters to identify the key variables affecting energy consumption, specifically: The extracted microenvironment change parameters were standardized, and the Z-score standardization method was used to eliminate the dimensional differences between different parameters; Recursive feature elimination was used to preliminarily screen the standardized data and remove redundant and low-relevance features; Multidimensional scaling is applied to project the filtered high-dimensional data into a low-dimensional space, preserving the nonlinear distance relationship between parameters; Based on the data processed by multidimensional scaling, principal component analysis is performed to further reduce the dimension and extract the main components in the data, identifying the key variables that have the greatest impact on energy consumption; The key variables extracted through multidimensional scaling and principal component analysis were interpreted, and the specific effects of the key variables on energy consumption were verified using correlation analysis.
5. The machine learning-based building energy consumption prediction and control system according to claim 4, characterized in that: The key indicator mean square error is calculated as follows: Use the test data set to perform a final evaluation of the model and calculate the mean square error of the model's performance indicator on the test set: ;in, is the mean square error, Indicates the actual energy consumption, is the energy consumption value predicted by the model, is the number of test samples, is the test sample number.
