A load simulation calculation method and system for building energy efficiency

Abstract

The present invention relates to the field of building energy conservation and intelligent building technology, and in particular to a load simulation calculation method and system for building energy efficiency, comprising: collecting multi-source data and constructing a multidimensional data set; deconstructing building load data into three types of heterogeneous modes: sequence-like, image-like, and video-like; calculating a dynamic correction coefficient based on the thermal storage effect of the enclosure structure; processing the three types of heterogeneous modes through a BiGRU network, an STNN network, and a 3DCNN network, respectively, and adopting Stacking ensemble learning to fuse and output the initial load; optimizing and outputting the final load prediction value through a multi-feature recurrent neural network; the system integrates multi-source data collection, multi-modal feature processing, physical correction, hybrid prediction, and timing optimization modules, and deploys the system on a local edge node through a knowledge distillation compression model. The present invention integrates physical models and data-driven methods, effectively solving the problems of insufficient data processing, simple feature construction, shallow model fusion, and insufficient computing architecture and timeliness in the prior art.

Description

Technical Field

[0001] The present invention relates to the technical field of building energy conservation and intelligent building technology, and in particular to a load simulation calculation method and system for building energy efficiency. Background Art

[0002] In the early days, building load prediction mainly relied on traditional physical models, calculating theoretical loads through building envelope parameters and meteorological data. Although such methods have physical interpretability, they have poor adaptability to complex building forms and user behaviors, and require a large number of precise input parameters, resulting in large errors in practical applications.

[0003] With the development of Internet of Things (IoT) technology and machine learning, data-driven approaches have become mainstream. Modeling the correlation between historical load data and meteorological characteristics has enhanced forecasting flexibility. Some studies have attempted to combine physical models with data-driven models, but these efforts have only achieved synergy through simple overlay or parameter adjustments, failing to delve deeper into the inherent connections between building physics and data characteristics.

[0004] At present, building load forecasting technology faces the following core bottlenecks: existing methods are mostly based on one-dimensional time series data modeling, ignoring the spatial distribution and spatiotemporal dynamic evolution of loads, resulting in insufficient prediction accuracy for load mutations or spatial coupling scenarios; physical models are difficult to adapt to parameter uncertainties, data models lack physical interpretability, and traditional hybrid methods do not implement explicit guidance of physical constraints on data modeling; cloud-based centralized computing architectures have data transmission delays, and complex deep learning models are difficult to deploy lightweight on local edge nodes, making it difficult to meet the real-time regulation needs of building energy systems; there is a lack of effective processing mechanism for abnormal load data, and simple elimination strategies are often adopted, resulting in limited model generalization capabilities.

[0005] Chinese invention patent CN115659459A discloses a method and system for calculating the dynamic heat load of solar-heated buildings. To address the issue of inaccurate calculations of heat load fluctuations in solar-heated buildings, this method, based on the Z-transfer function, is proposed. However, this invention lacks multidimensional feature engineering for load data and cannot account for spatial heterogeneity and dynamic fluctuations in loads.

[0006] Therefore, the present invention discloses a load simulation calculation method and system for building energy efficiency. Summary of the Invention

[0007] The purpose of the present invention is to solve the problems in the prior art of single data modality, insufficient coordination between physical models and data drivers, poor real-time performance, and weak robustness under extreme working conditions, and to propose a load simulation calculation method and system for building energy efficiency.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: a load simulation calculation method for building energy efficiency, comprising:

[0009] Step S1, multi-source data preprocessing, obtains building physical parameters, meteorological data and building load data through the Internet of Things sensor network, and uses the isolation forest algorithm to remove outliers and construct a multidimensional data set;

[0010] Step S2, multimodal feature construction, deconstructing the building load data into three heterogeneous modalities: sequence-like, image-like, and video-like;

[0011] Step S3: Dynamic physical correction, calculate the dynamic correction coefficient based on the heat storage effect of the enclosure structure , quantify the real-time impact of wall heat storage and release on load;

[0012] Step S4: Hybrid model prediction, which processes the three types of heterogeneous modalities through BiGRU network, STNN network and 3DCNN network respectively, and uses Stacking ensemble learning to fuse and output the initial load;

[0013] Step S5, timing optimization output, initial load, dynamic correction coefficient , building load data series and meteorological data series are input into the multivariate feature recurrent neural network, and the final load prediction value is output to achieve collaborative optimization of physical model and data-driven.

[0014] A load simulation calculation system for building energy efficiency, comprising:

[0015] Multi-source data acquisition module, multi-modal feature processing module, physical correction module, hybrid prediction module, timing optimization module and edge computing module;

[0016] The multi-source data acquisition module is used to obtain building physical parameters, meteorological data and building load data;

[0017] The multimodal feature processing module is used to construct sequence-like, image-like, and video-like features;

[0018] The physical correction module is used to calculate the thickness of the violent fluctuation layer of the enclosure structure and the dynamic correction coefficient ;

[0019] The hybrid prediction module includes BiGRU, STNN, 3DCNN sub-models and Stacking integrator;

[0020] The timing optimization module runs a multivariate feature recurrent neural network and outputs a final load forecast value;

[0021] The edge computing module is deployed on the local edge node through knowledge distillation to compress the multivariate feature recurrent neural network, so as to improve the real-time prediction capability of the local edge node.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] The present invention captures temporal dependencies through sequence-like modalities, models spatial distribution through image-like modalities, and mines spatiotemporal coupling features through video-like modalities. The present invention can reduce load prediction errors and improve the prediction accuracy of spatial load mutation scenarios.

[0024] The present invention uses dynamic physical correction to calculate the dynamic correction coefficient based on the heat storage effect of the enclosure structure, quantifies the impact of wall heat storage and release on the load, and achieves a deep integration of the physical model and the data-driven model. The present invention can make up for the defects of the traditional method that the physical model is difficult to adapt to parameter uncertainty and the data-driven model lacks physical interpretability.

[0025] The bottom layer of the present invention uses BiGRU, STNN, and 3DCNN to independently process three types of heterogeneous modalities and extract differentiated features. The top layer uses Stacking Ridge Regression to integrate and output the initial load, which is then integrated with LSTM and historical sequences for timing optimization. The present invention can improve the generalization ability of the model and reduce the fluctuation of prediction errors across building types.

[0026] The present invention compresses a multivariate feature recurrent neural network through knowledge distillation and deploys it on local edge nodes. The present invention can reduce model inference delay and realize real-time linkage between load forecasting and energy regulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. 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 any creative work.

[0028] Figure 1 A schematic diagram of a method flow for load simulation calculation of building energy efficiency provided by an embodiment of the present invention;

[0029] Figure 2 A schematic diagram of a dynamic correction coefficient calculation process for a load simulation calculation method for building energy efficiency provided by an embodiment of the present invention;

[0030] Figure 3 A schematic diagram of a hybrid model prediction process of a load simulation calculation method for building energy efficiency provided by an embodiment of the present invention;

[0031] Figure 4 A schematic diagram of the system architecture of a load simulation calculation system for building energy efficiency provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0032] To further illustrate the technical means and effectiveness of the present invention in achieving its intended objectives, the following, in conjunction with the accompanying drawings and preferred embodiments, describes in detail a load simulation calculation method and system for building energy efficiency, including its specific implementation, structure, features, and effectiveness. In the following description, references to "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.

[0033] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0034] The following examples are for illustrative purposes only and are not intended to limit the scope of the present invention.

[0035] The following describes in detail a load simulation calculation method and system for building energy efficiency provided by the present invention with reference to the accompanying drawings.

[0036] Example

[0037] Please refer to Figure 1 It is a flowchart of the method provided by an embodiment of the present invention.

[0038] In an embodiment of the present invention, a load simulation calculation method for building energy efficiency is provided, the method comprising:

[0039] Step S1, multi-source data preprocessing, obtains building physical parameters, meteorological data and building load data through the Internet of Things sensor network, and uses the isolation forest algorithm to remove outliers and construct a multidimensional data set;

[0040] Wherein, step S1 further includes the following sub-steps:

[0041] S1-1, building physical parameters include thermal conductivity, material density, material specific heat capacity, envelope heat transfer coefficient and envelope area;

[0042] S1-2, meteorological data including outdoor temperature and indoor set temperature;

[0043] S1-3, building load data includes historical total load data and user unit load data;

[0044] S1-4, abnormal data points are detected and eliminated by the isolation forest algorithm, and missing values ​​are filled by linear interpolation;

[0045] S1-5, partition and standardize the cleaned data by feature type, including:

[0046] Z-score standardization is used for meteorological data and building load data;

[0047] Min-Max normalization is used for building physical parameters;

[0048] S1-6, construct a multidimensional dataset containing building load data, meteorological data and time series features. The time series features include sine or cosine coding of hours, week type and holiday flag.

[0049] It should be noted that meteorological data is collected using temperature and humidity sensors; historical total load is collected through smart meters; user unit load is collected through household meters; all sensor data is collected synchronously at 1-hour intervals, and the timestamps are aligned to the ISO 8601 standard format.

[0050] Building physical parameters, thermal conductivity, density and specific heat capacity are collected through material testing instruments.

[0051] The heat transfer coefficient of the enclosure structure is calculated using the following formula: , where R represents the total thermal resistance and K represents the heat transfer coefficient of the enclosure structure.

[0052] The area F of the enclosure structure is obtained by obtaining point cloud data through a 3D laser scanner and processed by BIM software.

[0053] The abnormal data detection of the isolation forest algorithm sets the abnormal value ratio э=1%, and identifies anomalies by calculating the average path length of the data points in the random forest; if the path length of a point , it is determined as an outlier and removed; among them, represents the mean, Represents standard deviation.

[0054] In the missing value processing, for data points with a missing duration of ≤2 hours in the continuous time series, linear interpolation is used to fill in the missing data; if the missing duration exceeds 2 hours, the mean value of the previous and next 24 hours is used to fill in the missing data.

[0055] Time features capture periodicity through sine / cosine encoding, day of week / holiday marks distinguish load patterns, and all data are aligned at hourly granularity to construct a multidimensional dataset.

[0056] Step S2, multimodal feature construction, deconstructing the building load data into three heterogeneous modalities: sequence-like, image-like, and video-like;

[0057] Wherein, in step S2, the following sub-steps are also included:

[0058] S2-1, quasi-sequence mode construction, the historical total load data, meteorological data and time series features are spliced ​​into a one-dimensional feature vector according to the time axis, which is used as the input of the BiGRU network to generate the first prediction value reflecting the time series evolution law ;The vector dimension is T×m, where T is the time step and m is the feature dimension;

[0059] S2-2, image-like modality construction, including:

[0060] Map the user unit load value to spatial pixel points, and each pixel value represents the load size of the corresponding unit;

[0061] Construct a spatial matrix representing the load distribution of the building's functional areas, where cells in the same functional area are spatially adjacent in the matrix; the functional areas include office areas, commercial areas, and public areas;

[0062] Based on the statistical correlation analysis of load time series data, the spatial arrangement of units in the matrix is ​​optimized so that units with high correlation are close to each other;

[0063] Perform Min-Max normalization on the matrix so that the element value range is [0,1];

[0064] S2-3, video-like modality construction, stacks the image-like modality matrices of nine consecutive equally spaced moments into a three-dimensional tensor according to the time dimension to capture the spatiotemporal dynamic changes of the load.

[0065] It should be noted that the feature splicing method splices the historical total load data, meteorological data and time series features into a T×6-dimensional feature vector along the time axis; the time step T is 24, covering the complete daily cycle load changes.

[0066] Pixel mapping rules, including:

[0067] Divide the building floor plan into an N×N grid (N is dynamically adjusted by the total number of cells, preferably 11×11);

[0068] Units in the same functional area are four-connected and adjacent in the matrix;

[0069] Optimize permutations based on Spearman's correlation coefficient.

[0070] For tensor construction, we take the image-like matrices for 9 consecutive hours (from time t-8 to time t) and stack them into a [11, 11, 9] three-dimensional tensor. The number of channels input to the 3DCNN is set to 9.

[0071] The three types of modalities synchronize data based on the hour to ensure timing consistency.

[0072] Please refer to Figure 2 It is a schematic diagram of the dynamic correction coefficient calculation process provided by an embodiment of the present invention.

[0073] Step S3: Dynamic physical correction, calculate the dynamic correction coefficient based on the heat storage effect of the enclosure structure , quantify the real-time impact of wall heat storage and release on load;

[0074] Wherein, in step S3, the following sub-steps are also included:

[0075] S3-1, based on Fourier heat conduction theory, calculate the thickness of the violent fluctuation layer of the enclosure structure δ. The specific formula is: , where α represents the thermal conductivity of the wall material; T represents the temperature fluctuation period, which is 24 hours; It represents the thickness of the fluctuation layer and is used to define the effective heat storage area of ​​the wall; π represents pi, which is a mathematical constant;

[0076] The calculation formula of the thermal conductivity coefficient α of the wall material is: ,in, represents thermal conductivity; Indicates the material density; Indicates the specific heat capacity of the material;

[0077] S3-2, calculate the heat storage of the violent fluctuation layer at time τ , the specific formula is: , where ρ represents the material density; c represents the material specific heat capacity; represents the thickness of the wave layer; F represents the area of ​​the enclosure structure; Indicates the temperature difference on both sides of the wave layer; It represents the heat storage of the violent fluctuation layer at time τ;

[0078] The outdoor temperature at time τ and indoor set temperature The absolute value of the difference between the two is calculated;

[0079] S3-3, calculate the heat transfer of the inner surface of the enclosure structure at time τ , the specific formula is: , where K represents the heat transfer coefficient of the enclosure structure; F represents the area of ​​the enclosure structure; Indicates the indoor set temperature; is the outdoor temperature; It represents the heat transfer on the inner surface of the enclosure structure at time τ;

[0080] S3-4, calculate the dynamic correction coefficient , the specific formula is: ,in, It represents the heat transfer on the inner surface of the enclosure structure at time τ; It represents the heat storage of the violent fluctuation layer at time τ; Indicates the dynamic correction factor.

[0081] It should be noted that based on Fourier heat conduction theory, the thickness of the fluctuation layer δ is calculated by the formula, where the thermal conductivity α is determined by the material's thermal physical parameters (λ, ρ, c), and the temperature fluctuation period T is taken as 24 hours to reflect day and night changes.

[0082] Heat storage of violent fluctuation layer at time τ Calculated by material heat capacity, fluctuation layer volume and indoor and outdoor temperature difference, where the temperature difference The absolute value of the indoor set temperature and the real-time outdoor temperature is taken to reflect the driving effect of the ambient temperature difference on heat storage.

[0083] Heat transfer on the inner surface of the enclosure structure at time τ Based on the heat transfer coefficient K of the envelope structure and the indoor and outdoor temperature difference, the contribution of direct heat transfer from the wall to the load is characterized.

[0084] Dynamic correction factor It is the ratio of the heat transfer amount to the heat storage amount on the inner surface, quantifying the real-time impact of the heat storage effect. The value range is 0-1 and is used to directly input the subsequent timing optimization module to correct the load forecast.

[0085] Please refer to Figure 3 It is a schematic diagram of the hybrid model prediction process provided by an embodiment of the present invention.

[0086] Step S4: Hybrid model prediction, which processes the three types of heterogeneous modalities through BiGRU network, STNN network and 3DCNN network respectively, and uses Stacking ensemble learning to fuse and output the initial load;

[0087] Wherein, in step S4, the following sub-steps are also included:

[0088] S4-1, sequence-like modal processing, processes the sequence-like modal features through the BiGRU network and outputs the first prediction value ;

[0089] S4-2, image-like modality processing, processes image-like modality features through the STNN network and outputs the second prediction value ;

[0090] S4-3, video-like modality processing, processes the video-like modality features through the 3DCNN network and outputs the third prediction value ;

[0091] S4-4, using the Stacking ensemble learning method to fuse the first prediction value , the second predicted value and the third predicted value , generating the initial load ;

[0092] The meta-learner of ensemble learning is the ridge regression model, which is trained through 5-fold cross-validation. The ensemble formula is: ,in, represents the integrated mapping function based on the ridge regression model; Indicates the initial load.

[0093] It should be noted that BiGRU stands for bidirectional gated recurrent unit, which captures the bidirectional temporal dependency of sequence-like modalities. The BiGRU network consists of two hidden layers, each with 64 neurons, an activation function of tanh, an input dimension of T×m (T=24 hours, m=6), and outputs the first predicted value. .

[0094] STNN stands for spatiotemporal neural network, which extracts the spatial distribution features of image-like modalities. The STNN network contains three convolutional layers with kernel sizes of 5×5, 3×3, and 3×3, respectively. The input is an 11×11 functional area load matrix and the output is the second predicted value. .

[0095] 3DCNN stands for three-dimensional convolutional neural network, which captures the spatiotemporal dynamic characteristics of video-like modalities. The 3DCNN network consists of three layers of 3D convolutional layers, with convolution kernel sizes of 5×5×3, 3×3×3, and 3×3×3, respectively, and outputs the third prediction value. .

[0096] Stacking ensemble learning is used to fuse the three types of modal prediction values, and the ridge regression model is used as the meta-learner with the regularization parameter λ=1.0. The initial load is generated through 5-fold cross-validation training. , achieving complementary enhancement of temporal, spatial and spatiotemporal features.

[0097] The input and output dimensions of each modal network match the load forecasting target, the training data is divided into 7:3, and the model performance is evaluated by overall error and robustness to ensure the generalization ability of the initial load forecast.

[0098] Step S5, timing optimization output, initial load, dynamic correction coefficient , building load data series and meteorological data series are input into the multivariate feature recurrent neural network, which outputs the final load forecast value to achieve collaborative optimization of physical model and data-driven;

[0099] Wherein, in step S5, the following sub-steps are also included:

[0100] S5-1 uses a multi-feature recurrent neural network to process input data, where the LSTM memory unit update formula is: ,in Indicates the current state of the memory unit; Represents the cell state at the previous moment τ-1, that is, historical memory information; 、 are the outputs of the forget gate and input gate respectively; Indicates the dynamic correction coefficient; Indicates the concatenation of the previous hidden state, current input, and dynamic correction coefficient; represents the weight matrix; represents the bias term; Represents element-wise multiplication; tanh represents the hyperbolic tangent activation function;

[0101] S5-2, calculate the final load forecast value through the output layer, the specific formula is: ,in, Indicates the current LSTM hidden state; represents the output layer weight matrix; represents the bias vector, Indicates the final load forecast value.

[0102] It should be noted that the LSTM memory unit combines the dynamic correction coefficient through the forget gate and input gate , optionally retaining historical load patterns and current physical correction information. As an external control signal, it participates in gating calculations through splicing operations, allowing the model to quickly adjust memory weights when the load changes suddenly, thereby improving prediction robustness.

[0103] The output layer converts the LSTM hidden state into Mapping to the final predicted value , which integrates the timing characteristics (initial load), physical correction ( ) and environmental variables (meteorological data). A weighted fusion of multi-source information is achieved through linear transformation, ensuring that the prediction results are consistent with data trends and follow the laws of thermodynamics.

[0104] The input data are all standardized, the initial load, building load data series and meteorological data are normalized by Z-score, and the dynamic correction coefficient Maintaining the range of [0,1], all sequences are aligned at the hour level to ensure the temporal consistency of LSTM input.

[0105] The model was trained using a weighted loss function to balance overall error, robustness, and relative error. It was iteratively trained using the Adam optimizer with a learning rate of 0.001 to ensure the model captured the details of load fluctuations while maintaining generalization capabilities.

[0106] Please refer to Figure 4 It is a schematic diagram of the system architecture provided by an embodiment of the present invention.

[0107] A load simulation calculation system for building energy efficiency, comprising:

[0108] Multi-source data acquisition module, multi-modal feature processing module, physical correction module, hybrid prediction module, timing optimization module and edge computing module;

[0109] Multi-source data acquisition module for acquiring building physical parameters, meteorological data and building load data;

[0110] Multimodal feature processing module, used to construct sequence-like, image-like, and video-like features;

[0111] Physical correction module, used to calculate the thickness of the violent fluctuation layer of the enclosure structure and the dynamic correction coefficient ;

[0112] Hybrid prediction module, including BiGRU, STNN, 3DCNN sub-models and Stacking integrator;

[0113] The timing optimization module runs a multi-feature recurrent neural network and outputs the final load forecast value;

[0114] The edge computing module compresses the multi-feature recurrent neural network through knowledge distillation and is deployed on local edge nodes to improve the real-time prediction capabilities of local edge nodes.

[0115] It should be noted that the multi-source data acquisition module obtains data through BIM models, IoT sensors and smart meters, and outputs a multidimensional data set after outlier elimination, missing value filling and standardization.

[0116] The multimodal feature processing module deconstructs the data into sequence-like (temporal), image-like (spatial), and video-like (spatiotemporal) features, which are input into the BiGRU, STNN, and 3DCNN sub-models respectively.

[0117] The physical correction module calculates the thickness of the fluctuating layer and the dynamic correction coefficient based on Fourier heat conduction theory to quantify the heat storage effect of the enclosure structure.

[0118] The hybrid prediction module fuses the three-modal prediction values ​​through stacking ensemble learning to generate the initial load.

[0119] The timing optimization module uses the LSTM network to combine the dynamic correction coefficient and timing data to output the final load forecast value.

[0120] The edge computing module compresses the model through knowledge distillation and is deployed on local edge nodes to improve the real-time prediction capabilities of local edge nodes and support dynamic regulation of building energy systems.

[0121] For the load simulation calculation method and system for building energy efficiency in this embodiment, after the operation is completed, each module and equipment is shut down one by one according to the prescribed procedures to ensure safe power off, and the monitoring data and operation records during the operation are organized and saved to provide a basis for subsequent clinical research and quality control.

[0122] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the scope of protection of the present application.

Claims

1. A load simulation calculation method for building energy efficiency, characterized in that: The method includes: Step S1, multi-source data preprocessing, obtains building physical parameters, meteorological data and building load data through the Internet of Things sensor network, and uses the isolation forest algorithm to remove outliers and construct a multidimensional data set; Step S2, multimodal feature construction, deconstructing the building load data into three heterogeneous modalities: sequence-like, image-like, and video-like; Step S3: Dynamic physical correction, calculate the dynamic correction coefficient based on the heat storage effect of the enclosure structure , quantify the real-time impact of wall heat storage and release on load; Step S4: Hybrid model prediction, which processes the three types of heterogeneous modalities through BiGRU network, STNN network and 3DCNN network respectively, and uses Stacking ensemble learning to fuse and output the initial load; Step S5, timing optimization output, initial load, dynamic correction coefficient , building load data series and meteorological data series are input into the multivariate feature recurrent neural network, and the final load prediction value is output to achieve collaborative optimization of physical model and data-driven.

2. The load simulation calculation method for building energy efficiency according to claim 1, characterized in that: Wherein, step S1 further includes the following sub-steps: S1-1, the building physical parameters include thermal conductivity, material density, material specific heat capacity, enclosure heat transfer coefficient and enclosure area; S1-2, the meteorological data includes outdoor temperature and indoor set temperature; S1-3, the building load data includes historical total load data and user unit load data; S1-4, abnormal data points are detected and eliminated by the isolation forest algorithm, and missing values ​​are filled by linear interpolation; S1-5, partition and standardize the cleaned data by feature type, including: The meteorological data and building load data are standardized using Z-score; Min-Max normalization is applied to the building physical parameters; S1-6, constructing a multidimensional dataset including building load data, meteorological data and time series features, wherein the time series features include sine or cosine codes of hours, week types and holiday flags.

3. The load simulation calculation method for building energy efficiency according to claim 1, characterized in that: Wherein, in step S2, the following sub-steps are also included: S2-1, quasi-sequence mode construction, the historical total load data, meteorological data and time series features are spliced ​​into a one-dimensional feature vector according to the time axis, which is used as the input of the BiGRU network to generate the first prediction value reflecting the time series evolution law ; The vector dimension is T×m, where T is the time step and m is the feature dimension; S2-2, image-like modality construction, including: Map the user unit load value to spatial pixel points, and each pixel value represents the load size of the corresponding unit; Constructing a spatial matrix representing load distribution of functional areas of a building, where cells in the same functional area are spatially adjacent in the matrix; the functional areas include office areas, commercial areas, and public areas; Based on the statistical correlation analysis of load time series data, the spatial arrangement of units in the matrix is ​​optimized so that units with high correlation are close to each other; Perform Min-Max normalization on the matrix so that the element value range is [0, 1]; S2-3, video-like modality construction, stacks the image-like modality matrices of nine consecutive equally spaced moments into a three-dimensional tensor according to the time dimension to capture the spatiotemporal dynamic changes of the load.

4. The load simulation calculation method for building energy efficiency according to claim 1, characterized in that: Wherein, in step S3, the following sub-steps are also included: S3-1, based on Fourier heat conduction theory, calculate the thickness of the violent fluctuation layer of the enclosure structure δ. The specific formula is: , where α represents the thermal conductivity of the wall material; T represents the temperature fluctuation period, which is 24 hours; It represents the thickness of the fluctuation layer and is used to define the effective heat storage area of ​​the wall; π represents pi, which is a mathematical constant; The calculation formula of the thermal conductivity coefficient α of the wall material is: ,in, represents thermal conductivity; Indicates the material density; Indicates the specific heat capacity of the material; S3-2, calculate the heat storage of the violent fluctuation layer at time τ , the specific formula is: , where ρ represents the material density; c represents the material specific heat capacity; represents the thickness of the wave layer; F represents the area of ​​the enclosure structure; Indicates the temperature difference on both sides of the wave layer; It represents the heat storage of the violent fluctuation layer at time τ; described The outdoor temperature at time τ and indoor set temperature The absolute value of the difference between the two is calculated; S3-3, calculate the heat transfer of the inner surface of the enclosure structure at time τ , the specific formula is: , where K represents the heat transfer coefficient of the enclosure structure; F represents the area of ​​the enclosure structure; Indicates the indoor set temperature; is the outdoor temperature; It represents the heat transfer on the inner surface of the enclosure structure at time τ; S3-4, calculate the dynamic correction coefficient , the specific formula is: ,in, It represents the heat transfer on the inner surface of the enclosure structure at time τ; It represents the heat storage of the violent fluctuation layer at time τ; Indicates the dynamic correction factor.

5. The load simulation calculation method for building energy efficiency according to claim 1 is characterized in that : Wherein, in step S4, the following sub-steps are also included: S4-1, sequence-like modal processing, processes the sequence-like modal features through the BiGRU network and outputs the first prediction value ; S4-2, image-like modality processing, processes image-like modality features through the STNN network and outputs the second prediction value ; S4-3, video-like modality processing, processes the video-like modality features through the 3DCNN network and outputs the third prediction value ; S4-4, using the Stacking ensemble learning method to fuse the first prediction value , the second predicted value and the third predicted value , generating the initial load ; The meta-learner of the ensemble learning is a ridge regression model, which is trained through 5-fold cross-validation. The ensemble formula is: ,in, represents the integrated mapping function based on the ridge regression model; Indicates the initial load.

6. The load simulation calculation method for building energy efficiency according to claim 1, characterized in that: Wherein, in step S5, the following sub-steps are also included: S5-1 uses a multi-feature recurrent neural network to process input data, where the LSTM memory unit update formula is: ,in Indicates the current state of the memory unit; Represents the cell state at the previous moment τ-1, that is, historical memory information; 、 are the outputs of the forget gate and input gate respectively; Indicates the dynamic correction coefficient; Indicates the concatenation of the previous hidden state, current input, and dynamic correction coefficient; represents the weight matrix; represents the bias term; Represents element-wise multiplication; tanh represents the hyperbolic tangent activation function; S5-2, calculate the final load forecast value through the output layer, the specific formula is: ,in, Indicates the current LSTM hidden state; represents the output layer weight matrix; represents the bias vector, Indicates the final load forecast value.

7. A load simulation calculation system for building energy efficiency, used to implement a load simulation calculation method for building energy efficiency according to any one of claims 1 to 6, characterized in that: include: Multi-source data acquisition module, multi-modal feature processing module, physical correction module, hybrid prediction module, timing optimization module and edge computing module; The multi-source data acquisition module is used to obtain building physical parameters, meteorological data and building load data; The multimodal feature processing module is used to construct sequence-like, image-like, and video-like features; The physical correction module is used to calculate the thickness of the violent fluctuation layer of the enclosure structure and the dynamic correction coefficient ; The hybrid prediction module includes BiGRU, STNN, 3DCNN sub-models and Stacking integrator; The timing optimization module runs a multivariate feature recurrent neural network and outputs a final load forecast value; The edge computing module is deployed on the local edge node through knowledge distillation to compress the multivariate feature recurrent neural network, so as to improve the real-time prediction capability of the local edge node.

Images