Multi-convolutional neural network building energy consumption prediction method based on attention mechanism and gating unit

By introducing attention mechanism and gated units into building energy consumption prediction, combined with local mean decomposition and genetic algorithm optimization, the problems of low energy consumption prediction accuracy and limited scenarios in the existing technology are solved, and the problem of low energy consumption prediction accuracy and limited scenarios are achieved. Higher prediction accuracy and robustness.

CN118981965AInactive Publication Date: 2024-11-19SHANDONG HECHENG CONSTR CO LTD

Patent Information

Application Number
CN202411144469.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-20
Publication Date
2024-11-19
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing building energy consumption prediction methods have low accuracy when considering random factors such as weather and human activities, high mathematical model dependence, and limited application scenarios, making it difficult to accurately predict building energy consumption in complex scenarios.

Method used

Using a multi-convolutional neural network method based on attention mechanism and gated unit, the external environment feature data is recorded in real time by deploying sensors, and the timing data is processed using local mean decomposition, building energy consumption prediction model is designed, and the model structure is optimized through genetic algorithms. to improve prediction accuracy.

Benefits of technology

Improve the accuracy and robustness of building energy consumption prediction, can better capture long-term dependence and local features in sequence data, automatically learn and pay attention to important features, reduce the error between the predicted value and the real value, and improve the model generalization ability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118981965A_ABST
    Figure CN118981965A_ABST
Patent Text Reader

Abstract

The invention provides a multi-convolutional neural network building energy consumption prediction method based on an attention mechanism and a gating unit, and aims to solve the problems of insufficient prediction precision, limited application scenarios, high mathematical model dependency and the like in the prior art. According to the technology, firstly, external environment characteristics (including temperature, humidity, wind speed, atmospheric pressure and solar intensity) are collected by deploying sensors, building energy consumption data of the day are recorded, and an initial data set is constructed; then, the collected time series data is processed by using local mean decomposition (LMD) to further extract data features. A multi-convolutional neural network building energy consumption prediction model based on a gating unit and an attention mechanism is designed to achieve high-precision building energy consumption prediction, finally, the model is optimized through a genetic algorithm, the optimized building energy consumption model is actually deployed in a building, energy consumption prediction is conducted according to data collected in real time, and the energy consumption prediction accuracy is improved. Therefore, effective distribution of energy is realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of building energy consumption, and in particular relates to a multi-convolutional neural network building energy consumption prediction method based on an attention mechanism and a gating unit. Background Art

[0002] With the increase in the proportion of new buildings, the total energy consumption and carbon emissions of the entire construction process have been rising year by year, posing severe challenges to the environment. The report pointed out that in 2019, the carbon emissions of the national building operation stage reached 2.13 billion tons of carbon dioxide, accounting for 21.6% of the national carbon emissions. At the same time, the total energy consumption of the entire construction process also reached 2.233 billion tons of standard coal, of which the energy consumption in the operation stage accounted for 21.2% of the total national energy consumption. This shows that China's current environmental situation is still severe. The excessive use of fossil energy such as coal and oil has aggravated the problems of resource scarcity and environmental pollution. Therefore, the construction industry faces huge challenges in energy conservation and emission reduction. Accurate building energy consumption prediction technology can not only provide data support for managers, enabling them to formulate more reasonable energy conservation and energy allocation plans, but also help respond to the national advocacy of energy conservation and emission reduction. According to relevant research, by relying on accurate prediction technology to optimize the operation process of the building energy system, energy efficiency can be improved by about 30%, which significantly improves the efficiency of building energy use and is of great significance to promoting green buildings and sustainable development.

[0003] In order to improve the accuracy of building energy consumption prediction, the present invention aims to propose a multi-convolutional neural network building energy consumption prediction method based on attention mechanism and gating unit. Existing building energy consumption prediction methods mainly include the following three methods: engineering method, statistical method and traditional machine learning method.

[0004] Engineering method: The engineering method is also called the physical method or "white box" method. It relies on the basic principles of heat transfer and deeply analyzes the energy consumption of each component of the building. This method relies on highly accurate physical formulas for calculations to ensure the accuracy and reliability of the analysis. When building a prediction model based on engineering methods, it is usually necessary to collect and integrate rich input information, which covers meteorological data, building design details parameters, and residential use patterns. Comprehensive and detailed information input is the key to ensuring the accuracy of the prediction results, but weather factors are random, and predictions based on fixed physical formulas will not be able to take random factors into account.

[0005] Statistical method: Statistical method, also known as regression method, is a technique that uses classical statistical principles to explore the intrinsic relationship between input variables and output results. In the field of building energy consumption prediction, commonly used basic statistical methods include autoregression model (AR), differential integrated moving average autoregression model (ARIMA) and multivariate linear regression model (MLR). However, when statistical methods are used for nonlinear fitting, their fitting accuracy is low, but when predicting building energy consumption, it often involves the fitting of nonlinear relationships, so this will lead to a decrease in the accuracy of the prediction.

[0006] Traditional machine learning methods: Compared with engineering methods and statistical methods, traditional machine learning methods have shown higher superiority in fitting nonlinear inputs and have achieved better performance in prediction tasks. However, these methods are generally based on shallow structure design models, that is, they only perform simple linear or nonlinear transformations on input features. When faced with a large and growing amount of training data, these shallow models may encounter bottlenecks in insufficient computing power, making it difficult to deeply mine the potential high-level information in the data. This limitation is particularly significant in complex scenarios, limiting their performance in improving prediction accuracy.

[0007] The main shortcomings of existing building energy consumption prediction technologies include: 1. Low prediction accuracy: When applying pure engineering methods to the prediction of time-varying building energy consumption, it is impossible to fully consider the impact of random factors such as weather and human activities on building energy consumption, thus affecting the prediction accuracy of building energy consumption.

[0008] 2. High dependence on mathematical models: Existing building energy consumption prediction methods usually rely on accurate mathematical models. However, in actual building energy consumption prediction, these models may be difficult to obtain or insufficiently accurate, thus affecting the accuracy of building energy consumption prediction.

[0009] 3. Limited application scenarios: The existing building energy consumption prediction methods take into account limited influencing factors during prediction, but the actual building energy consumption is a complex scenario that requires consideration of the impact of multiple parameters on building energy consumption. However, the existing methods have limited prediction capabilities in such complex scenarios.

[0010] Therefore, in view of the shortcomings of the above-mentioned scheme in actual production and implementation, it has been revised and improved. At the same time, in the spirit and concept of seeking excellence, with the assistance of professional knowledge and experience, and after many ingenuity and experiments, the present invention was created. A multi-convolutional neural network building energy consumption prediction method based on attention mechanism and gating unit is provided to solve the above-mentioned problems. Summary of the invention

[0011] The present invention proposes a multi-convolutional neural network building energy consumption prediction method based on attention mechanism and gating unit, which solves the problems in the prior art.

[0012] The technical solution of the present invention is implemented as follows: a multi-convolutional neural network building energy consumption prediction method based on attention mechanism and gating unit, the method comprising: S1. Initial data set construction: The external environment characteristics (including outdoor temperature) are fully considered in the prediction of building energy consumption. , outdoor humidity , outdoor wind speed , atmospheric pressure , solar intensity ) for building energy consumption The impact of building characteristics is clearly defined. Sensors are deployed on the basis of building characteristic data to record external environment characteristic data in real time and record corresponding building energy consumption data. S2, initial data set time series data processing: Based on the initial data set construction process of S1, a data processing module is set up. This module is based on the local mean decomposition (LMD) operation to further extract the characteristics of the initial data, and also plays a role in eliminating noise, providing more sufficient data support for subsequent building energy consumption prediction; S3. Design of building energy consumption prediction model: A building energy consumption prediction model based on gated units and convolutional neural networks is designed, and an attention mechanism module is added to the building energy consumption prediction model to further improve the prediction accuracy of the model; S4. Overall optimization of building energy consumption model based on genetic algorithm: Establish an optimization method for building energy consumption prediction model, and further optimize and adjust the network structure of the building energy consumption prediction model to further reduce the error between the predicted value and the true value; S5. Actual deployment of the building energy consumption prediction model. This model is deployed in the building. The external environment data collected by various sensors is combined with the building's own characteristics to predict the building's energy consumption. Based on the predicted data, energy allocation can be carried out in advance for the building.

[0013] As a preferred implementation, the process of constructing the initial data set of the building energy consumption prediction model includes: S11. Initial data collection: Deploy high-precision sensors around the building to collect characteristic data related to building energy consumption, including deploying temperature sensors to obtain outdoor temperature data , deploy humidity sensors to obtain outdoor humidity data , deploy wind speed sensors to obtain outdoor wind speed data , deploy atmospheric pressure sensors to obtain outdoor air pressure data , deploy solar radiation sensors to obtain outdoor solar intensity data , while recording the building’s energy consumption data ; S12, Time series data division: for daily outdoor temperature data , 48 evenly sampled samples are taken, i.e. the sampling period is set to half an hour, to obtain daily outdoor temperature data for: ;in, For the The specific value of building energy consumption at each sampling time point, and ; S13, data processing method based on normalization: Since the dimensions of different data are different, in order to eliminate the dimension differences in different data, the maximum and minimum normalization methods are used to transform all sampled data into intervals. The specific formula for normalization is: ; in, Used to represent the time series data collected during the building energy consumption prediction process, that is, the outdoor temperature , outdoor humidity , outdoor wind speed , atmospheric pressure , solar intensity , so formula (1) is rewritten as: ; in, After normalized data processing The building energy consumption data at each sampling time point, and ; Similarly, outdoor humidity , outdoor wind speed , atmospheric pressure , solar intensity Conduct and outdoor temperature Using the same sampling and normalization operations, the corresponding time series data is: ; ; ; ; ;in , , , and They are the first The outdoor temperature, outdoor humidity, outdoor wind speed, outdoor atmospheric pressure and outdoor solar intensity values ​​at each sampling time point, and ; S14, based on the time series data partitioning method of S12 and the data normalization method of S13, collects data related to building energy consumption prediction within one year (i.e., outdoor temperature , outdoor humidity 1. Outdoor wind speed 1. Atmospheric pressure , solar intensity ), a total of 365 sets of data were obtained for subsequent model training: ; ; ; ; ; and the outdoor temperature , outdoor humidity , outdoor wind speed 1. Atmospheric pressure , solar intensity , the data in is used as the model input data, and the building energy consumption data of the day is collected. Since a total of 365 days of data are collected, the normalized building energy consumption data of the day is recorded as and , then the building energy consumption data can be obtained : As a preferred embodiment, the operation process of the time series data processing module in the initial data set includes: S21, first determine the decomposed input signal, and define the secondary input signal as , and then extract the data to be decomposed The local extreme point of , and based on the extracted local extreme points , find the average value of adjacent local extreme points : ; in, Indicates input signal Middle Local extreme points, Indicates input signal Middle local extreme points; these average points Connect them with a straight line to get the local mean function ; S22, then solve the envelope function value of the adjacent local extreme point : ; Use the sliding average method to convert two adjacent envelope function values Connect them together and get the envelope estimation function ; S23, the local mean function From the original data Separate it from the data to get a new data : ; Reuse Divide by the envelope estimate function , we can get the frequency modulation function : ; S24, then judge The envelope estimation function of the function is always 1. If this condition is not met, the iterative process S21-S23 is repeated until The envelope estimation function of the function satisfies this condition; S25, multiply all envelope estimation functions generated in the above-mentioned entire iterative process to obtain an envelope signal : ; S26, the obtained envelope signal and frequency modulation function Multiply to get the initial data The first Component; S27, from the initial data The lieutenant general will get The components are separated and the newly obtained data As a new raw data, repeat S21-S26 to get the next Components, since the characteristic information of the data is mainly concentrated in the first few Among the components, the number of decompositions selected in this technology is 4, and the initial data is finally obtained Decomposed into 4 Components, and these four components are spliced ​​together to obtain data with new feature information without noise information. ; S28, the outdoor temperature , outdoor humidity , outdoor wind speed 1. Atmospheric pressure , solar intensity Each set of time series data in is used as the input time series signal of S21, and the signal decomposition and splicing steps of S21-S27 are repeated to obtain data with noise removed and new feature information. , then the corresponding standard input feature data for: .

[0014] As a preferred implementation, the operation process of the building energy consumption prediction model based on the door control unit includes: S31. Five different external environmental characteristic data will affect the building energy consumption prediction value. In the constructed building energy consumption prediction model, an attention mechanism module needs to be designed. Using this attention mechanism module, the optimal weight set of each external environmental characteristic in the building energy consumption prediction is calculated. , Indicates outdoor temperature , outdoor humidity , outdoor wind speed , atmospheric pressure , solar intensity The optimal weights corresponding to each environmental feature are then input into the feature data With the best weight set Multiply to get the weighted input feature data : ; S32, obtain weighted input feature data After that, the data is sent to the multi-layer convolutional layer in the building energy consumption prediction model designed by the present invention to further extract the weighted input feature data. Detailed features : ; in, Table convolution layer processing function; S33, the detailed features of the convolutional layer output After the gated unit layer, we get In order to prevent overfitting, The output is sent to the Dropout layer, and finally a regularization layer is added to further adjust the output of the Dropout layer to obtain the predicted value of building energy consumption. : ; ; in, represents the gated unit layer processing function, Represent the Dropout layer processing function and the regularization layer processing function respectively; S34, in the gate control unit training, first initialize and reset the gate weight , update the gate weights , output gate weight Then, each data point in each group of data processed by the multi-layer convolutional neural network is sent to the gate control unit for training, and the gate weights are continuously updated and reset. , update the gate weights , output gate weight , after training, the final reset gate weight is obtained , update the gate weights , output gate weight .

[0015] As a preferred implementation, the building energy consumption prediction model optimization method based on genetic algorithm includes: S41. Use MAE loss function to measure the prediction effect of building energy consumption model; ;in, Indicates the actual building energy consumption value. represents the building energy consumption value predicted by this model, Indicates the number of predicted values, ; S42, Random Generation Initial population , each population represents the number of attention layers in a model And the number of convolutional layers and number of gate control unit layers The combination of solutions, namely: Among them, it means in Number of attention layers in a population And the number of convolutional layers and number of gate control unit layers The specific number of ; S43, the number of attention layers determined for each population in S42 And the number of convolutional layers and number of gate control unit layers The combination scheme is trained and the value, and this The value is taken as the fitness of this population; S44. Select the initial population The population with the lowest fitness is the best population ; S45. Initialize the population based on the determined optimal population Population update operation: ; in, is the update speed weight, , Indicates taking an interval A random number, represents the updated population, ; S46, repeating the iterative steps S43-S45, when the maximum number of iterations is reached, obtaining the optimal population after multiple iterations : ;in, They represent the number of attention layers, convolutional layers, and gated unit layers in the best combination solution respectively; S47, based on the number of attention layers, convolution layers, and gating unit layers in the best combination solution determined in S46, determine them as the number of attention layers, convolution layers, and gating unit layers in the building energy consumption prediction model, so as to obtain the network structure of the building energy consumption prediction model with the best prediction effect. Based on the optimal network result, the model is trained using the standard feature data set after LMD processing to obtain the final trained building energy consumption prediction model.

[0016] As a preferred implementation, the actual deployment of the building energy consumption prediction model includes: S51. High-precision sensors are arranged outside the building for which building energy consumption forecasting is required, with a sampling period of 30 minutes to obtain external environmental data of the building, including temperature sensors to obtain outdoor temperature data. , deploy humidity sensors to obtain outdoor humidity data , deploy wind speed sensors to obtain outdoor wind speed data , deploy atmospheric pressure sensors to obtain outdoor air pressure data , deploy solar radiation sensors to obtain outdoor solar intensity data ; S52, sending the sampled data to a normalization module for processing to eliminate the differences in dimensions between the data, and then sending the normalized data to a data processing module based on LMD to further extract data features and eliminate the influence of noise; S53, the data processed by LMD is sent to the building energy consumption prediction model trained based on the data set, and the predicted amount of building energy consumption is obtained ; S54 obtains the building energy consumption forecast Allocate the energy required by the building in advance to achieve further energy optimization allocation.

[0017] After adopting the above technical solution, the beneficial effects of the present invention are: 1. Improve data reliability: Deploy sensors to record data in real time, ensuring the accuracy and real-time nature of the data, and providing reliable data support for subsequent energy consumption prediction. Local mean decomposition (LMD) is used to process the collected data, further extracting the deep features of the data and eliminating noise, improving the quality and expression of the data.

[0018] 2. Improve the accuracy of building energy consumption prediction: The building energy consumption prediction model designed by the present invention based on the gated unit and the convolutional neural network effectively captures the long-term dependencies and local features in the sequence data by combining the advantages of both, thereby improving the prediction performance of the model. At the same time, the attention mechanism module is introduced to enable the model to automatically learn and focus on more important input features, thereby further improving the prediction accuracy.

[0019] 3. Improved robustness of the building energy consumption prediction process: The present invention uses a genetic algorithm to optimize the network structure of the model, which can automatically adjust the network structure and reduce the error between the predicted value and the true value, so as to improve the robustness and generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0021] Figure 1 This is a schematic diagram of data collection in the present invention; Figure 2 This is a flow chart of LMD operation of the present invention; Figure 3 It is an overall schematic diagram of the building energy consumption prediction model of the present invention; Figure 4 This is the actual deployment diagram of the building energy consumption prediction model of the present invention. DETAILED DESCRIPTION

[0022] 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

[0023] A multi-convolutional neural network building energy consumption prediction method based on attention mechanism and gated unit, the method comprising: S1, initial data set construction: fully consider the external environment characteristics (including outdoor temperature) in the prediction of building energy consumption , outdoor humidity , outdoor wind speed , atmospheric pressure , solar intensity ) for building energy consumption The impact of building characteristics is clearly defined. Sensors are deployed on the basis of building characteristic data to record external environment characteristic data in real time and record corresponding building energy consumption data. S2, initial data set time series data processing: Based on the initial data set construction process of S1, a data processing module is set up. This module is based on the local mean decomposition (LMD) operation to further extract the characteristics of the initial data, and also plays a role in eliminating noise, providing more sufficient data support for subsequent building energy consumption prediction; S3. Design of building energy consumption prediction model: A building energy consumption prediction model based on gated units and convolutional neural networks is designed, and an attention mechanism module is added to the building energy consumption prediction model to further improve the prediction accuracy of the model; S4. Overall optimization of building energy consumption model based on genetic algorithm: Establish an optimization method for building energy consumption prediction model, and further optimize and adjust the network structure of the building energy consumption prediction model to further reduce the error between the predicted value and the true value; S5. Actual deployment of the building energy consumption prediction model. This model is deployed in the building. The external environment data collected by various sensors is combined with the building's own characteristics to predict the building's energy consumption. Based on the predicted data, energy allocation can be carried out in advance for the building.

[0024] like Figure 1 As shown in the figure, the initial data set construction process of the building energy consumption prediction model includes: S11. Initial data collection: Deploy high-precision sensors around the building to collect characteristic data related to building energy consumption, including deploying temperature sensors to obtain outdoor temperature data , deploy humidity sensors to obtain outdoor humidity data , deploy wind speed sensors to obtain outdoor wind speed data , deploy atmospheric pressure sensors to obtain outdoor air pressure data , deploy solar radiation sensors to obtain outdoor solar intensity data , while recording the building’s energy consumption data ; S12, Time series data division: for daily outdoor temperature data , 48 evenly sampled samples are taken, i.e. the sampling period is set to half an hour, to obtain daily outdoor temperature data for: ;in, For the The specific value of building energy consumption at each sampling time point, and ; S13, data processing method based on normalization: Since the dimensions of different data are different, in order to eliminate the dimension differences in different data, the maximum and minimum normalization methods are used to transform all sampled data into intervals. The specific formula for normalization is: ;in, Used to represent the time series data collected during the building energy consumption prediction process, that is, the outdoor temperature , outdoor humidity , outdoor wind speed , atmospheric pressure , solar intensity , so formula (1) is rewritten as: ;in, After normalized data processing The building energy consumption data at each sampling time point, and ; Similarly, outdoor humidity , outdoor wind speed , atmospheric pressure , solar intensity Conduct and outdoor temperature Using the same sampling and normalization operations, the corresponding time series data is: ; ; ; ; ;in , , , and They are the first The outdoor temperature, outdoor humidity, outdoor wind speed, outdoor atmospheric pressure and outdoor solar intensity values ​​at each sampling time point, and ; S14, based on the time series data partitioning method of S12 and the data normalization method of S13, collects data related to building energy consumption prediction within one year (i.e., outdoor temperature , outdoor humidity 1. Outdoor wind speed 1. Atmospheric pressure , solar intensity ), a total of 365 sets of data were obtained for subsequent model training: ; ; ; ; ; and the outdoor temperature , outdoor humidity , outdoor wind speed 1. Atmospheric pressure , solar intensity , the data in is used as the model input data, and the building energy consumption data of the day is collected. Since a total of 365 days of data are collected, the normalized building energy consumption data of the day is recorded as and , then the building energy consumption data can be obtained : .

[0025] like Figure 2 As shown, the operation process of the time series data processing module in the initial data set includes: S21, first determine the decomposed input signal, and define the secondary input signal as , and then extract the data to be decomposed The local extreme point of , and based on the extracted local extreme points , find the average value of adjacent local extreme points : ;in, Indicates input signal Middle Local extreme points, Indicates input signal Middle local extreme points; these average points Connect them with a straight line to get the local mean function ; S22, then solve the envelope function value of the adjacent local extreme point ; Use the sliding average method to convert two adjacent envelope function values Connect them together and get the envelope estimation function ; S23, the local mean function From the original data Separate it from the data to get a new data : ; Reuse Divide by the envelope estimate function , we can get the frequency modulation function : ; S24, then judge The envelope estimation function of the function is always 1. If this condition is not met, the iterative process S21-S23 is repeated until The envelope estimation function of the function satisfies this condition; S25, multiply all envelope estimation functions generated in the above-mentioned entire iterative process to obtain an envelope signal : ; S26, the obtained envelope signal and frequency modulation function Multiply to get the initial data The first Quantity; S27, from the initial data The lieutenant general will get The components are separated and the newly obtained data As a new raw data, repeat S21-S26 to get the next Components, since the characteristic information of the data is mainly concentrated in the first few Among the components, the number of decompositions selected in this technology is 4, and the initial data is finally obtained Decomposed into 4 Components, and these four components are spliced ​​together to obtain data with new feature information without noise information. ; S28, the outdoor temperature , outdoor humidity , outdoor wind speed 1. Atmospheric pressure , solar intensity Each set of time series data in is used as the input time series signal of S21, and the signal decomposition and splicing steps of S21-S27 are repeated to obtain data with noise removed and new feature information. , then the corresponding standard input feature data for: .like Figure 3 As shown in the figure, the operation process of the building energy consumption prediction model based on the gate control unit includes: The present invention predicts building energy consumption based on time series data; for time series data, the gating unit can better capture the connection between time series data, and the convolutional neural network can perform deeper feature extraction. Therefore, the present invention designs a building energy consumption prediction model based on the gating unit and the convolutional neural network, and designs an attention mechanism module to further improve the prediction effect of the model.

[0026] S31. Five different external environmental characteristic data will affect the building energy consumption prediction value. In the constructed building energy consumption prediction model, an attention mechanism module needs to be designed. Using this attention mechanism module, the optimal weight set of each external environmental characteristic in the building energy consumption prediction is calculated. , Indicates outdoor temperature , outdoor humidity , outdoor wind speed , atmospheric pressure , solar intensity The optimal weights corresponding to each environmental feature are then input into the feature data With the best weight set Multiply to get the weighted input feature data : ; S32, obtain weighted input feature data After that, the data is sent to the multi-layer convolutional layer in the building energy consumption prediction model designed by the present invention to further extract the weighted input feature data. Detailed features : ;in, Represents the convolution layer processing function; S33, the detailed features of the convolutional layer output After the gated unit layer, we get In order to prevent overfitting, The output is sent to the Dropout layer, and finally a regularization layer is added to further adjust the output of the Dropout layer to obtain the predicted value of building energy consumption. : ; ;in, represents the gated unit layer processing function, Represent the Dropout layer processing function and the regularization layer processing function respectively; S34, in the gate control unit training, first initialize and reset the gate weight , update the gate weights , output gate weight Then, each data point in each group of data processed by the multi-layer convolutional neural network is sent to the gate control unit for training, and the gate weights are continuously updated and reset. , update the gate weights , output gate weight , after training, the final reset gate weight is obtained , update the gate weights , output gate weight The optimization methods of building energy consumption prediction model based on genetic algorithm include: S41. Use MAE loss function to measure the prediction effect of building energy consumption model; ;in, Indicates the actual building energy consumption value. represents the building energy consumption value predicted by this model, Indicates the number of predicted values, ; S42, Random Generation Initial population , each population represents the number of attention layers in a model And the number of convolutional layers and number of gate control unit layers The combination of solutions, namely: Among them, it means in Number of attention layers in a population And the number of convolutional layers and number of gate control unit layers The specific number of ; S43, the number of attention layers determined for each population in S42 And the number of convolutional layers and number of gate control unit layers The combination scheme is trained and the value, and this The value is taken as the fitness of this population; S44. Select the initial population The population with the lowest fitness is the best population ; S45. Initialize the population based on the determined optimal population Population update operation: ;in, is the update speed weight, , Indicates taking an interval A random number, represents the updated population, ; S46, repeating the iterative steps S43-S45, when the maximum number of iterations is reached, obtaining the optimal population after multiple iterations : ;in, They represent the number of attention layers, convolutional layers, and gated unit layers in the best combination solution respectively; S47, based on the number of attention layers, convolution layers, and gating unit layers in the best combination solution determined in S46, determine them as the number of attention layers, convolution layers, and gating unit layers in the building energy consumption prediction model, so as to obtain the network structure of the building energy consumption prediction model with the best prediction effect. Based on the optimal network result, the model is trained using the standard feature data set after LMD processing to obtain the final trained building energy consumption prediction model.

[0027] like Figure 4 As shown in Figure 2, the actual deployment of the building energy consumption prediction model includes: S51. High-precision sensors are arranged outside the building for which building energy consumption forecasting is required, with a sampling period of 30 minutes to obtain external environmental data of the building, including temperature sensors to obtain outdoor temperature data. , deploy humidity sensors to obtain outdoor humidity data , deploy wind speed sensors to obtain outdoor wind speed data , deploy atmospheric pressure sensors to obtain outdoor air pressure data , deploy solar radiation sensors to obtain outdoor solar intensity data ; S53, the data processed by LMD is sent to the building energy consumption prediction model trained based on the data set, and the predicted amount of building energy consumption is obtained ; S54 obtains the building energy consumption forecast Allocate the energy required by the building in advance to achieve further energy optimization allocation.

[0028] This technology deploys sensors to collect external environmental characteristics (including temperature, humidity, wind speed, atmospheric pressure, solar intensity) and record the building energy consumption data of the day to construct an initial data set, and then uses LMD to process the collected initial time series data to further extract data features, and then designs a multi-convolutional neural network building energy consumption prediction model based on gating units and attention mechanisms to achieve high-precision building energy consumption prediction. Finally, the model is optimized through genetic algorithms, and this building energy consumption model is actually deployed in the building, and energy consumption prediction is performed based on real-time collected data to achieve effective energy allocation.

[0029] In the description of the present invention, it should be understood that the terms "longitudinal", "lateral", "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In the description of the present invention, unless otherwise specified and limited, it should be noted that the terms "installed", "connected" and "connected" should be understood in a broad sense, for example, it can be a mechanical connection or an electrical connection, or it can be the internal communication of two elements, it can be a direct connection, or it can be an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0030] The above are only preferred embodiments of the present invention and are 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. A multi-convolutional neural network building energy consumption prediction method based on attention mechanism and gating unit, characterized in that: The method includes: S1. Initial data set construction: fully consider the external environmental characteristics (including outdoor temperature) in the prediction of building energy consumption. , outdoor humidity , outdoor wind speed , atmospheric pressure , solar intensity ) for building energy consumption Influence, on the basis of clarifying the building characteristic data, sensors are deployed to record the external environment characteristic data in real time and record the corresponding building energy consumption data; S2, initial data set time series data processing: based on the initial data set construction process of S1, a data processing module is established. This module is based on the local mean decomposition (LMD) operation to further extract the characteristics of the initial data, and also plays a role in eliminating noise, providing more sufficient data support for subsequent building energy consumption prediction; S3, building energy consumption prediction model design: a building energy consumption prediction model based on the gating unit and convolutional neural network, and an attention mechanism module is added to the building energy consumption prediction model to further improve the prediction accuracy of the model; S4. Overall optimization of building energy consumption model based on genetic algorithm: Establish an optimization method for building energy consumption prediction model, and further optimize and adjust the network structure of the building energy consumption prediction model to further reduce the error between the predicted value and the true value; S5. Actual deployment of the building energy consumption prediction model. This model is deployed in the building. The external environmental data collected by various sensors is used to predict the building energy consumption. Based on the predicted data, energy allocation can be carried out in advance for the building.

2. According to claim 1, a multi-convolutional neural network building energy consumption prediction method based on attention mechanism and gating unit is characterized in that: The process of constructing the initial data set of the building energy consumption prediction model includes: S11. Initial data collection: deploying high-precision sensors around the building to collect characteristic data related to building energy consumption, including deploying temperature sensors to obtain outdoor temperature data , deploy humidity sensors to obtain outdoor humidity data , deploy wind speed sensors to obtain outdoor wind speed data , deploy atmospheric pressure sensors to obtain outdoor air pressure data , deploy solar radiation sensors to obtain outdoor solar intensity data , while recording the building’s energy consumption data ; S12, Time series data division: for daily outdoor temperature data , 48 evenly sampled samples are taken, i.e. the sampling period is set to half an hour, to obtain daily outdoor temperature data for: ; in, For the The specific value of building energy consumption at each sampling time point, and ; S13, data processing method based on normalization: Since the dimensions of different data are different, in order to eliminate the dimension differences in different data, the maximum and minimum normalization methods are used to transform all sampled data into intervals. The specific formula for normalization is: ; in, Used to represent the time series data collected during the building energy consumption prediction process, that is, the outdoor temperature , outdoor humidity , outdoor wind speed , atmospheric pressure , solar intensity , so formula (1) is rewritten as: ; in, After normalized data processing The building energy consumption data at each sampling time point, and ; Similarly, outdoor humidity , outdoor wind speed , atmospheric pressure , solar intensity Conduct and outdoor temperature Using the same sampling and normalization operations, the corresponding time series data is: ; ; ; ; ;in , , , and They are the first The outdoor temperature, outdoor humidity, outdoor wind speed, outdoor atmospheric pressure and outdoor solar intensity values ​​at each sampling time point, and ; S14, based on the time series data partitioning method of S12 and the data normalization method of S13, collect data related to building energy consumption prediction within one year (i.e. outdoor temperature , outdoor humidity 1. Outdoor wind speed 1. Atmospheric pressure , solar intensity ), a total of 365 sets of data were obtained for subsequent model training: ; ; ; ; ; and the outdoor temperature , outdoor humidity , outdoor wind speed 1. Atmospheric pressure , solar intensity , the data in is used as the model input data, and the building energy consumption data of the day is collected. Since a total of 365 days of data are collected, the normalized building energy consumption data of the day is recorded as and , then the building energy consumption data can be obtained : .

3. According to claim 1, a multi-convolutional neural network building energy consumption prediction method based on attention mechanism and gating unit is characterized in that: The operation process of the time series data processing module in the initial data set includes: S21, first determine the decomposed input signal, and define the secondary input signal as , and then extract the data to be decomposed The local extreme point of , and based on the extracted local extreme points , find the average value of adjacent local extreme points : ;in, Indicates input signal Middle Local extreme points, Indicates input signal Middle Local extreme points; These average points Connect them with a straight line to get the local mean function ; S22, then solve the envelope function value of the adjacent local extreme point : ; Use the sliding average method to convert two adjacent envelope function values Connect them together and get the envelope estimation function ; S23, the local mean function From the original data Separate it from the data to get a new data : ; Reuse Divide by the envelope estimate function , we can get the frequency modulation function : ; S24, then judge The envelope estimation function of the number is always 1. If this condition is not met, the iterative process S21-S23 is repeated until The envelope estimation function of the function satisfies this condition; S25, multiply all envelope estimation functions generated in the above-mentioned entire iterative process to obtain an envelope signal : ; S26, the obtained envelope signal and frequency modulation function Multiply to get the initial data The first Quantity; S27, from the initial data The lieutenant general will get The components are separated and the newly obtained data As a new raw data, repeat S21-S26 to get the next Components, since the characteristic information of the data is mainly concentrated in the first few Among the components, the number of decompositions selected in this technology is 4, and the initial data is finally obtained Decomposed into 4 Components, and these four components are spliced ​​together to obtain data with new feature information without noise information. ; S28, the outdoor temperature , outdoor humidity , outdoor wind speed 1. Atmospheric pressure , solar intensity Each set of time series data in is used as the input time series signal of S21, and the signal decomposition and splicing steps of S21-S27 are repeated to obtain data with noise removed and new feature information. , then the corresponding standard input feature data for: .

4. According to claim 1, a multi-convolutional neural network building energy consumption prediction method based on attention mechanism and gating unit is characterized in that: The operation process of the building energy consumption prediction model based on the gate control unit includes: S31. Five different external environmental characteristic data will affect the building energy consumption prediction value. In the constructed building energy consumption prediction model, an attention mechanism module needs to be designed. Using this attention mechanism module, the optimal weight set of each external environmental characteristic in the building energy consumption prediction is calculated. , Indicates outdoor temperature , outdoor humidity , outdoor wind speed , atmospheric pressure , solar intensity The optimal weights corresponding to each environmental feature are then input into the feature data With the best weight set Multiply to get the weighted input feature data : ; S32, obtain weighted input feature data After that, the data is sent to the multi-layer convolutional layer in the building energy consumption prediction model designed by the present invention to further extract the weighted input feature data. Detailed features : ; in, Represents the convolution layer processing function; S33, the detailed features of the convolutional layer output After the gated unit layer, we get In order to prevent overfitting, The output is sent to the Dropout layer, and finally a regularization layer is added to further adjust the output of the Dropout layer to obtain the predicted value of building energy consumption. : ; ;in, represents the gated unit layer processing function, Represent the Dropout layer processing function and the regularization layer processing function respectively; S34, in the gate control unit training, first initialize and reset the gate weight , update the gate weights , output gate weight Then, each data point in each group of data processed by the multi-layer convolutional neural network is sent to the gate control unit for training, and the gate weights are continuously updated and reset. , update the gate weights , output gate weight , after training, the final reset gate weight is obtained , update the gate weights , output gate weight .

5. According to claim 1, a multi-convolutional neural network building energy consumption prediction method based on attention mechanism and gating unit is characterized in that: The optimization methods of building energy consumption prediction model based on genetic algorithm include: S41. Use MAE loss function to measure the prediction effect of building energy consumption model; ; in, Indicates the actual building energy consumption value. represents the building energy consumption value predicted by this model, Indicates the number of predicted values, ; S42, Random Generation Initial population , each population represents the number of attention layers in a model And the number of convolutional layers and number of gate control unit layers The combination of solutions, namely: Among them, it means in Number of attention layers in a population And the number of convolutional layers and number of gate control unit layers The specific number of ; S43, the number of attention layers determined for each population in S42 And the number of convolutional layers and number of gate control unit layers The combination scheme is trained and the value, and this The value is taken as the fitness of this population; S44. Select the initial population The population with the lowest fitness is the best population ; S45. Initialize the population based on the determined optimal population Population update operation: ; in, is the update speed weight, , Indicates taking an interval A random number, represents the updated population, ; S46, repeating the iterative steps S43-S45, when the maximum number of iterations is reached, obtaining the optimal population after multiple iterations : ;in, They represent the number of attention layers, convolutional layers, and gated unit layers in the best combination solution respectively; S47, based on the number of attention layers, convolution layers, and gating unit layers in the best combination solution determined in S46, determine them as the number of attention layers, convolution layers, and gating unit layers in the building energy consumption prediction model, so as to obtain the network structure of the building energy consumption prediction model with the best prediction effect. Based on the optimal network result, the model is trained using the standard feature data set after LMD processing to obtain the final trained building energy consumption prediction model.

6. According to claim 1, a method for predicting building energy consumption based on a multi-convolutional neural network with an attention mechanism and a gating unit is characterized in that: The actual deployment of the building energy consumption prediction model includes: S51. High-precision sensors are arranged outside the building for which building energy consumption forecasting is required, with a sampling period of 30 minutes to obtain external environmental data of the building, including temperature sensors to obtain outdoor temperature data. , deploy humidity sensors to obtain outdoor humidity data , deploy wind speed sensors to obtain outdoor wind speed data , deploy atmospheric pressure sensors to obtain outdoor air pressure data , deploy solar radiation sensors to obtain outdoor solar intensity data ; S52, sending the sampled data to a normalization module for processing to eliminate the differences in dimensions between the data, and then sending the normalized data to a data processing module based on LMD to further extract data features and eliminate the influence of noise; S53, the data processed by LMD is sent to the building energy consumption prediction model trained based on the data set, and the predicted amount of building energy consumption is obtained ; S54 obtains the building energy consumption forecast Allocate the energy required by the building in advance to achieve further energy optimization allocation.

Citation Information

Patent Citations

  • Process industrial energy consumption prediction method based on GA-AM-GRU

    CN116881683A

  • Building energy consumption rapid prediction method

    CN118114816A

  • Energy-saving control method for heating, ventilating and air conditioning of building

    CN118347106A

  • Magnetic suspension molecular pump real-time control method and rotor rotating speed prediction model building method

    CN118446090A

Cited By

  • Method for predicting energy-saving effect of building envelope

    CN120632636A

  • A method for predicting energy-saving effect of a building envelope

    CN120632636B

  • Building energy consumption prediction method of convolutional bidirectional gating cycle unit neural network based on efficient channel attention mechanism optimization

    CN120822857A

  • Building air conditioning load forecasting method fusing physical simulation and graph neural network

    CN122927616A