Method and system for public building seasonal energy consumption prediction and energy saving retrofit analysis

Abstract

The application discloses a kind of public building quarter energy consumption prediction and energy-saving reconstruction analysis method and system, the method includes: obtaining the energy consumption parameters of four quarters of different types of predicted public building, constructs multi-feature dataset and carries out data preprocessing;Energy consumption prediction model is constructed and is trained and tested;Unit area energy consumption and per capita energy consumption are predicted using trained energy consumption prediction model;According to energy consumption parameters, energy consumption prediction model and energy consumption prediction result, based on the method of spearman and tree model coupling analysis, the correlation analysis of feature vector of the energy consumption of each type of building in different seasons is carried out.The application predicts the energy consumption of different types of buildings in each quarter in the future based on the improved convolutional neural network, with high prediction accuracy and fast response speed;The correlation ranking is strong in pertinence, and the energy-saving goal is clear;It can give direct energy-saving reconstruction suggestions to related buildings and timely find abnormal conditions, and provide support for building energy management and energy saving.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of public building energy consumption prediction, and particularly relates to a method and system for public building quarterly energy consumption prediction and energy-saving reconstruction analysis. BACKGROUND

[0002] The building field is one of the three fields of social energy consumption, and the building energy consumption accounts for 46%, 40% and 39% of the total energy consumption in China, the United States and Europe, respectively. In China, the operation energy consumption of public buildings accounts for 40% of the total operation energy consumption of all buildings, and the energy consumption is more than that of urban and rural residential buildings. Therefore, it is necessary to predict the energy consumption of public buildings in the operation stage, which is beneficial to the government's management of public building energy consumption.

[0003] In order to more reasonably evaluate the building energy consumption and eliminate the influence of different influencing factors on the energy consumption, researchers usually establish relevant prediction models to predict the building energy consumption. The data-driven model only needs to input the relevant historical energy consumption or operation data of the building to train the model, and the demand for building structure, environment and other parameters is low, so in recent years, the data-driven model based on building energy consumption prediction and evaluation is more popular among researchers. Most of the previous researches are based on annual energy consumption data model training and prediction, ignoring the influence of seasonality and building type, and it is difficult to realize the energy-saving guidance for the predicted public buildings in stages. Moreover, the prediction model has problems such as unstable accuracy and slow running speed. SUMMARY

[0004] The main purpose of the present application is to overcome the shortcomings and deficiencies of the prior art, and to provide a method and system for public building quarterly energy consumption prediction and energy-saving reconstruction analysis, which is based on an improved convolutional neural network (hereinafter referred to as CNN) to predict the energy consumption of different types of public buildings in different quarters in the future, and to make more effective energy-saving plans for public building managers.

[0005] In order to achieve the above purpose, the technical scheme adopted by the present application is as follows:

[0006] In one aspect of the present application, a method for public building quarterly energy consumption prediction and energy-saving reconstruction analysis is provided, comprising the following steps:

[0007] Obtain the energy consumption parameters of different types of public buildings to be predicted in four quarters, including date parameters, load parameters and energy-saving measures parameters;

[0008] Construct a multi-feature data set according to the obtained energy consumption parameters, and perform data verification and cleaning, data feature selection and data normalization processing;

[0009] The energy consumption prediction model comprises a vector screening layer and a convolutional neural network connected in sequence, and a plurality of convolutional layers and pooling layers are added in the convolutional neural network according to different types of public institutions;

[0010] The energy consumption prediction model is trained and tested by using the preprocessed multi-feature data set;

[0011] The trained energy consumption prediction model is used to predict the energy consumption per unit area and per capita of different types of public buildings in multiple quarters;

[0012] According to the energy consumption parameters, the energy consumption prediction model and the energy consumption prediction results, the characteristic vector correlation analysis of the energy consumption of each type of building in different seasons is performed based on the method of Spearman and tree model coupling analysis.

[0013] As a preferred technical solution, the obtained public buildings to be predicted include educational buildings, administrative buildings, venue buildings and medical buildings; the date parameters include date, year, month and quarter; the load parameters include water consumption per capita, building area, energy consumption number, number of personnel, number of gasoline vehicles, number of diesel vehicles, number of new energy vehicles, power consumption, coal consumption, natural gas consumption, oil consumption, diesel consumption, liquefied petroleum consumption, other energy consumption, solar collector area, solar system installed capacity and geothermal system installed capacity; the energy saving measures data include whether energy saving reconstruction is performed, the time length from energy saving reconstruction to the present and the specific energy saving reconstruction projects performed.

[0014] As a preferred technical solution, the multi-feature data set is labeled based on the energy consumption parameters, 70% of the data of the multi-feature data set is constructed as a training set, and 30% of the data of the multi-feature data set is constructed as a test set;

[0015] In the training phase, the attention weight and the parameters of the convolutional neural network are adjusted by an optimization algorithm;

[0016] The core selects Adam, the batch is 100 / 800, the iteration is 800 times, the initial learning rate is 0.01, and the learning rate reduction factor and period are 0.5 and 400;

[0017] The energy consumption of each type of building in each quarter is predicted respectively, and when the prediction result passes the preset index evaluation, it is considered that the energy consumption prediction model is trained.

[0018] As a preferred technical solution, the data verification and cleaning are specifically as follows:

[0019] The obtained energy consumption load parameters are subjected to data checking and cleaning by using the quartile method, specifically, taking Q3+1.5IQR as the upper threshold value and Q1-1.5IQR as the lower threshold value of the distribution interval; wherein Q3 is the third quartile of the distribution interval, that is, the number at the 75% position after arranging the data sequence from small to large; Q1 is the first quartile of the distribution interval, that is, the number at the 25% position after arranging the data sequence from small to large; IQR is the interquartile range, IQR=Q3-Q1, and the values outside Q1-Q3 are discarded as outliers.

[0020] As a preferred technical solution, the data feature screening is specifically:

[0021] The stepwise linear regression method is used to analyze the correlation between the per capita energy consumption and unit area energy consumption of the public building and each influencing factor in the sequence S=[year, month, quarter, per capita water consumption PWC, building area BA, energy-using number NEU, number of people H, number of gasoline vehicles NGV, number of diesel vehicles NDV, oil consumption GC, other oil consumption OC, diesel consumption DC, liquefied petroleum consumption CLP, other energy consumption OEC, solar collector area SCA, solar system installed capacity SSIC, geothermal system installed capacity GIC, energy saving reconstruction ESR], and the influencing factors with a correlation higher than a preset value are screened out, and the influencing factors with a correlation lower than the preset value are deleted.

[0022] As a preferred technical solution, the data normalization processing is specifically as follows:

[0023]

[0024] wherein X is the data to be processed, X max is the maximum value in the data set, X min is the minimum value in the data set, and X n is the normalized data.

[0025] As a preferred technical solution, the energy consumption prediction model comprises a stepwise regression vector screening layer, a stepwise linear regression layer, a normalization layer, a plurality of convolution layers and pooling layers, a dropout layer and a full connection layer, a regression layer, a reverse normalization layer and an output layer connected in sequence;

[0026] The stepwise regression vector screening layer is used to assign different weights to features of different importance degrees;

[0027] The stepwise linear regression layer is used to perform stepwise regression fitting on each feature vector to obtain the weight of each feature vector, so as to determine the importance of the input element;

[0028] The normalization layer is used to normalize the order of magnitude of the variables in the input feature set.

[0029] The pooling layer is used for dimensionality reduction of data;

[0030] The dropout layer sets a dropout rate to enhance the stability of the model; the fully connected layer is used for matrix multiplication and bias addition operation between the connection weight of the input feature and each neuron, so as to obtain the output result;

[0031] The regression layer is used for calculating the semi-mean square error loss of the regression task, and the output layer of the energy consumption prediction model is returned in the form of a RegressionOutputLayer object;

[0032] The inverse normalization layer is used for restoring the normalized data to the original data;

[0033] The output layer is used for outputting the unit area energy consumption and per capita energy consumption of the public building to be predicted.

[0034] As a preferred technical solution, the method based on the coupling analysis of Spearman and tree model performs feature vector correlation analysis on energy consumption of different seasons of various types of buildings, specifically:

[0035] Based on the Spearman formula, the correlation numerical calculation of the energy consumption parameters of different types of public buildings is performed, and the linear feature vector correlation sequence result is obtained;

[0036] Based on the tree model formula, the correlation numerical calculation of the energy consumption parameters of different types of public buildings is performed, and the nonlinear feature vector correlation sequence result is obtained;

[0037] According to the linear and nonlinear feature vector correlation sequence results, important feature vectors are screened out.

[0038] Another aspect of the present application also provides a system for public building seasonal energy consumption prediction and energy saving reconstruction analysis, which is applied to the above-mentioned method for public building energy consumption prediction and energy saving reconstruction analysis, and comprises a data acquisition and processing module, a data preprocessing module, a model construction and training module, a prediction module and a correlation analysis module.

[0039] The data acquisition and processing module is used for acquiring energy consumption parameters of four seasons of different types of public buildings to be predicted, including date parameters, load parameters and energy saving measure parameters; and a multi-feature data set is constructed according to the acquired energy consumption parameters;

[0040] The data preprocessing module is used for data verification and cleaning, data feature screening and data normalization processing of the multi-feature data set;

[0041] The model construction and training module is used for constructing an energy consumption prediction model, the energy consumption prediction model comprises a vector screening layer and a convolutional neural network connected in sequence, and a plurality of convolutional layers and pooling layers are added in the convolutional neural network according to different types of public institutions; and the energy consumption prediction model is trained and predicted by using a preprocessed multi-feature data set.

[0042] The prediction module is used for predicting the unit area energy consumption and the per capita energy consumption of different types of public buildings in multiple quarters by using the trained energy consumption prediction model.

[0043] The correlation analysis module is used for performing feature vector correlation analysis on the energy consumption of different seasons of each type of building based on the energy consumption parameters, the energy consumption prediction model and the energy consumption prediction result, and a method of coupling analysis of Spearman and tree models.

[0044] Another aspect of the present application also provides a computer readable storage medium and an electronic device, the computer readable storage medium stores a program, and the program is executed by the processor of the electronic device to realize the above-mentioned public building quarter energy consumption prediction and energy saving reconstruction analysis method.

[0045] Compared with the prior art, the present application has the following advantages and beneficial effects:

[0046] (1) The present application realizes the prediction of the unit area energy consumption and the per capita energy consumption of different types of public institution buildings in the field of public institution building energy consumption prediction, and has high prediction accuracy, fast speed and strong pertinence; can timely find abnormal conditions and propose energy saving reconstruction analysis for different types of public institution buildings in stages, and provides support for building energy management and energy saving; the data verification and missing data filling of the present application can be automatically completed, the construction of effective data set is completed, the processing speed is fast and the operation is convenient; since the present application can construct different energy consumption prediction models for different types of public institution buildings, the underfitting and overfitting problems can be better solved, and the prediction accuracy is high. By comparison, the time used by the model of the present application in the prediction process is much less than that of other deep learning models. The present application can construct different prediction models for different seasons of the building and analyze the correlation of each feature vector, so that it can give the public institution building strong pertinence and stage energy saving reconstruction suggestions, and has strong friendliness to the institution. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 is the flowchart of the public building quarter energy consumption prediction method in embodiment one of the present application;

[0048] Figure 2 is the structure diagram of the improved convolutional neural network in embodiment one of the present application;

[0049] Figure 3A flowchart of a public building quarterly energy-saving reconstruction analysis method in the embodiment one of the present application is shown in the figure.

[0050] Figure 4 A structure diagram of a public building quarterly energy consumption prediction and energy-saving reconstruction analysis system in the embodiment one of the present application is shown in the figure.

[0051] Figure 5 A structure diagram of a storage medium in the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0052] In order to make the personnel in the technical field better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0053] It should be pointed out that the following detailed description is exemplary and is intended to provide further description of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as generally understood by those skilled in the art to which the present application belongs.

[0054] It should be noted that the terms used herein are only for the purpose of describing specific embodiments, and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form, and in addition, it should be understood that when the terms "comprise" and / or "include" are used in the specification, they indicate the presence of a feature, step, operation, device, component and / or combination thereof.

[0055] The embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0056] Embodiment one

[0057] As shown in the figure, the embodiment one of the present application discloses a public building quarterly energy consumption prediction method, including the following steps: Figure 1 S1, obtaining the energy consumption parameters of four quarters of different types of public buildings to be predicted, including date parameters, load parameters and energy-saving measures parameters.

[0058]

[0059] ​As one or more implementations, the acquired public building to be predicted at least includes an educational building, an administrative building, a venue building, and a medical building; the date parameter at least includes a date Date, a year year, a month month, and a quarter quarter; the load parameter at least includes a per capita water consumption PWC, a building area BA, a number of energy users NEU, a number of personnel H, a number of gasoline vehicles NGV, a number of diesel vehicles NDV, a number of new energy vehicles NNV, an electricity consumption EC, a coal consumption CC, a natural gas consumption NGC, an oil consumption GC, an other oil consumption OC, a diesel consumption DC, a liquefied petroleum consumption CLP, an other energy consumption OEC, a solar collector area SCA, a solar system installed capacity SSIC, and a geothermal system installed capacity GIC. The energy-saving measure parameter at least includes whether energy-saving reconstruction (what kind of energy-saving reconstruction) ESR is performed, how long after the current energy-saving reconstruction is performed, and specific energy-saving reconstruction projects. The electricity consumption and water consumption information is acquired through a power meter and a water meter, and the other energy consumption information and the energy-saving reconstruction information is acquired through human reporting.

[0060] S2, constructing a multi-feature data set according to the acquired energy consumption parameters, and performing data verification and cleaning, data feature screening, and data normalization processing.

[0061] Further, in the embodiment, the multi-feature data set is constructed according to the acquired energy consumption parameters, and specifically:

[0062] S21, based on the energy consumption parameters, the multi-feature data set is classified by labels, 70% of the data of the multi-feature data set is constructed as a training set, and 30% of the data of the multi-feature data set is constructed as a test set.

[0063] Further, since the acquired related parameters may have abnormal data and missing data, in the embodiment, for the load parameter, the abnormal data is verified to avoid abnormal values, and the missing data is filled using a forward calculation method; for the date data, more dimensional features are extracted; for the energy-saving measure data, the energy consumption change in previous years is combined to judge and correct. Specifically as follows:

[0064] S22, data verification and cleaning: the acquired energy consumption load parameters are verified and cleaned by using the quartile method, specifically: taking Q3+1.5IQR of the distribution interval as the upper threshold, and taking Q1-1.5IQR as the lower threshold; wherein Q3 is the third quartile of the distribution interval, that is, the number at the 75% position after arranging the data sequence from small to large; Q1 is the first quartile of the distribution interval, that is, the number at the 25% position after arranging the data sequence from small to large; IQR is the interquartile range, IQR=Q3-Q1, and the values outside Q1-Q3 are discarded as abnormal values.

[0065] The year, month, and quarter are extracted for subsequent data analysis and processing to provide more accurate and detailed changes and trends in energy consumption.

[0066] Finally, the data after verification is unified in format to ensure the standardization and consistency of the data, with Boolean data represented by "0" for "False" and "1" for "True". The final sequence S contains the following elements:

[0067] S = [year, month, quarter, PWC, BA, NEU, H, NGV, NDV, GC, OC, DC, CLP, OEC, SCA, SSIC, GIC, ESR].

[0068] In particular, for abnormal cases of missing historical data, substitute data is generated based on the same type of energy-using unit's historical data or the target energy-using unit's last period historical data to fill in the missing reported values; for cases of missing data to be verified, the target energy-using unit is verified to fill in the missing monthly reported values.

[0069] S23, data feature screening:

[0070] Since the correlation between the per capita energy consumption and unit area energy consumption of public buildings and the elements in sequence S is different, in order to obtain more accurate prediction results, the per capita energy consumption and unit area energy consumption are predicted respectively. Therefore, for the two kinds of energy intensity, data with higher correlation is selected as the input of the model, which can reduce the difficulty of model training and improve the accuracy of the model. The specific steps include:

[0071] Stepwise linear regression method is used to analyze the correlation between the per capita energy consumption and unit area energy consumption of public buildings and the influencing factors in sequence S, and the influencing factors with correlation higher than the preset value are selected, and the influencing factors with correlation lower than the preset value are deleted.

[0072] The parameters with larger correlation coefficient with per capita energy consumption are S1 = [quarter, PWC, NEU, H, NGV, NDV, GC, OC, DC, CLP, OEC, SCA], and the parameters with larger correlation coefficient with unit area energy consumption are S2 = [quarter, PWC, BA, GC, OC, DC, CLP, OEC, SCA].

[0073] S24, data normalization:

[0074] In order to eliminate the influence of different characteristics between the dimension and the order of magnitude, make the influence of each characteristic on the result more balanced, avoid some characteristics because the value range is too large or too small, thereby reducing the precision of the model, all data are normalized, which can make the contribution of each characteristic more equal, improve the precision of the model, as follows:

[0075]

[0076] Wherein, X is the data to be processed, X max is the maximum value in the data set, X min is the minimum value in the data set, X n is the normalized data.

[0077] Because the convolutional neural network has a data set requirement, it must be 4-dimensional, so the data matrix size is [m*1*1*n], m is the number of public buildings, n is the number of input vectors, and 1 is the dimension filling number.

[0078] S3, constructing an energy consumption prediction model, the energy consumption prediction model comprises a vector screening layer and a convolutional neural network connected in sequence, and a plurality of convolutional layers and pooling layers are added in the convolutional neural network according to different types of public institutions.

[0079] As shown in Figure 2 As one or more embodiments, the present embodiment is improved based on the existing convolutional neural network to obtain an energy consumption prediction model, so as to better cope with complex energy consumption data and improve the prediction performance: first, a stepwise regression vector screening layer is added before the convolutional neural network, which pays more attention to more important features and gives different weights to features of different importance; second, based on different seasonal energy consumption of different types of public buildings, different numbers of convolutional layers and pooling layers are set between the convolutional neural networks to improve the prediction accuracy. Specifically as follows:

[0080] In particular, in the present embodiment, a stepwise linear regression layer can also be constructed according to the feature vector screening step in the preprocessing step, and the obtained data is applied to the features of the input sequence; each feature vector is fitted by stepwise regression to obtain the weight of each feature vector to determine which input elements are most important for the current task.

[0081] A normalization layer is constructed because the orders of magnitude of variables in each feature set are quite different, and normalization is beneficial to the rapid convergence of data-driven models.

[0082] The convolutional layer of the convolutional neural network is constructed, the first group of convolution kernels is 3*1 dimension, and 16 groups of data can be generated.

[0083] The pooling layer of the convolutional neural network is constructed, which can reduce the dimension of data, represent the data with higher level features, and reduce information redundancy and prevent overfitting.

[0084] Based on the different seasonal energy consumption prediction of different types of public buildings, different layers of convolutional and pooling layers are constructed, such as the energy consumption prediction model of administrative buildings in the first quarter needs to construct 4 layers of convolutional and pooling layers to obtain the optimal prediction accuracy.

[0085] The dropout layer and the fully connected layer are constructed, in order to enhance the stability of the model, and at the same time help to alleviate the problem of gradient vanishing and explosion, the dropout rate is set to 0.01. The fully connected layer can perform matrix multiplication and bias addition operation on the connection weight between the input feature and each neuron, so as to obtain the output result.

[0086] The regression layer is constructed, and the semi-mean square error loss of the regression task is calculated, and the regression output layer of the energy consumption prediction model is returned in the form of RegressionOutputLayer object.

[0087] The inverse normalization layer is constructed, which restores the normalized data to the original data, so as to better evaluate the performance of the model.

[0088] Output layer: output the unit area energy consumption and per capita energy consumption of the predicted public building.

[0089] Among them, the vector screening layer is stepwise regression screening, the input dimension is 20, and the output dimension is 12; the convolutional neural network layer is multi-layer, the specific number of layers is related to the building type and season, the input dimension is 4, and the number of CNN units is 4 times the number of layers; the output layer is a 3-layer fully connected layer, and the output dimension is 1.

[0090] S4, using the preprocessed multi-feature data set to train and predict the energy consumption prediction model.

[0091] Further, the training and prediction of the energy consumption prediction model are as follows:

[0092] Model training and adjustment: in the training stage, the attention weight and the parameters of the energy consumption prediction model are adjusted through the optimization algorithm, so that the energy consumption prediction model can better capture the relationship and pattern between the input sequences.

[0093] Hyperparameter selection: the core selects Adam, the batch is 100 / 800, the iteration is 800 times, the initial learning rate is 0.01, and the learning rate reduction factor and period are 0.5 and 400.

[0094] The energy consumption of each type in each quarter is predicted respectively, and when the prediction result passes the preset index evaluation, it is considered that the energy consumption prediction model is trained.

[0095] S5, predicting the unit area energy consumption and the per capita energy consumption of multiple quarters of different types of public buildings by using the trained energy consumption prediction model.

[0096] Embodiment Two

[0097] As shown in Figure 3 , the embodiment two of the present application discloses an energy-saving reconstruction analysis method, comprising the following steps:

[0098] Step one: according to the energy consumption parameters and prediction model of each type of public building and the energy consumption prediction result, based on the method of spireman and tree model coupling analysis, the characteristic vector correlation analysis of different seasonal energy consumption of each type of building is carried out, specifically:

[0099] Based on the spireman formula, the correlation numerical calculation of the energy consumption parameters of different types of public buildings is carried out, and the linear characteristic vector correlation sequence result is obtained;

[0100] Based on the tree model formula, the correlation numerical calculation of the energy consumption parameters of different types of public buildings is carried out, and the non-linear characteristic vector correlation sequence result is obtained;

[0101] According to the linear and nonlinear characteristic vector correlation sequence results, important characteristic vectors are screened out.

[0102] Step two: based on the obtained vector correlation sequence, the energy consumption index issued by the government and the energy consumption prediction result, the energy-saving reconstruction analysis is carried out for different public buildings.

[0103] The detailed steps of the energy consumption prediction method are the same as the building energy consumption prediction method provided in embodiment one, which will not be repeated here.

[0104] Embodiment Three

[0105] As shown in Figure 4 , in another embodiment of the present application, a public building seasonal energy consumption prediction and energy-saving reconstruction analysis system is provided, which comprises a data acquisition and processing module, a data preprocessing module, a model construction and training module, a prediction module and a correlation analysis module;

[0106] The data acquisition and processing module is used to obtain the energy consumption parameters of different types of public buildings to be predicted in four quarters, including date parameters, load parameters and energy-saving measures parameters; a multi-feature data set is constructed according to the obtained energy consumption parameters;

[0107] The data preprocessing module is used for data verification and cleaning, data feature screening and data normalization processing of the multi-feature data set;

[0108] The model construction and training module is configured to construct an energy consumption prediction model, the energy consumption prediction model comprising a vector screening layer and a convolutional neural network connected in sequence, and a plurality of convolutional layers and pooling layers are added in the convolutional neural network according to different types of public institutions; and the preprocessed multi-feature data set is used to train and predict the energy consumption prediction model.

[0109] The prediction module is configured to use the trained energy consumption prediction model to predict the energy consumption per unit area and the energy consumption per capita of the different types of public buildings in multiple quarters.

[0110] The correlation analysis module is configured to perform feature vector correlation analysis on the energy consumption of different types of buildings in different seasons based on the energy consumption parameters, the energy consumption prediction model, and the energy consumption prediction results, and a method of Spearman and tree model coupling analysis.

[0111] It should be noted that the system provided in the above embodiments is only used as an example for the division of the above functional modules, and in actual applications, the above functions can be completed by different functional modules according to needs, that is, the internal structure is divided into different functional modules to complete all or part of the above described functions. The system is a method for predicting and analyzing energy-saving reconstruction of public buildings in different quarters according to the above embodiments.

[0112] Embodiment Four

[0113] As shown in Figure 5 In another embodiment of the present application, a computer readable storage medium and an electronic device are also provided, the computer readable storage medium stores a program, and the program is executed by a processor of the electronic device to implement the method for predicting and analyzing energy-saving reconstruction of public buildings in different quarters according to the above embodiments, specifically:

[0114] S1, obtaining energy consumption parameters of different types of public buildings to be predicted in four quarters, including date parameters, load parameters, and energy-saving measure parameters;

[0115] S2, constructing a multi-feature data set according to the obtained energy consumption parameters and performing data verification and cleaning, data feature screening, and data normalization processing;

[0116] S3, constructing an energy consumption prediction model, the energy consumption prediction model comprising a vector screening layer and a convolutional neural network connected in sequence, and a plurality of convolutional layers and pooling layers are added in the convolutional neural network according to different types of public institutions;

[0117] S4, using the preprocessed multi-feature data set to train and test the energy consumption prediction model;

[0118] S5, predicting the unit area energy consumption and the per capita energy consumption of multiple quarters of different types of public buildings by using the trained energy consumption prediction model;

[0119] S6, performing feature vector correlation analysis on the energy consumption of different seasons of each type of building based on the energy consumption parameters, the energy consumption prediction model, and the energy consumption prediction result, and a method of coupling analysis of Spearman and tree model.

[0120] It should be understood that parts of the present application can be realized by hardware, software, firmware or a combination thereof. In the above embodiments, a plurality of steps or methods can be realized by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if realized by hardware, and as in another embodiment, it can be realized by any one or a combination of the following technologies known in the art: discrete logic circuit with logic gate circuit for implementing logic function on data signal, special integrated circuit with suitable combination logic gate circuit, programmable gate array (PGA), field programmable gate array (FPGA) and the like.

[0121] The above embodiments are the preferred embodiments of the present application, but the embodiments of the present application are not limited by the above embodiments, and any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present application shall be equivalent replacement methods and shall be included in the protection scope of the present application.

Claims

1. A method for quarterly energy consumption forecasting and energy-saving renovation analysis of public buildings, characterized in that, Includes the following steps: Energy consumption parameters for different types of public buildings to be predicted are obtained for four quarters, including date parameters, load parameters, and energy-saving measure parameters. The public buildings to be predicted include educational buildings, administrative buildings, stadiums, and medical buildings. The date parameters include date, year, month, and quarter. The load parameters include per capita water consumption, building area, number of energy users, authorized staffing, number of gasoline vehicles, number of diesel vehicles, number of new energy vehicles, electricity consumption, coal consumption, natural gas consumption, oil consumption, other oil consumption, diesel consumption, liquefied petroleum gas consumption, other energy consumption, solar collector area, solar system installed capacity, and geothermal system installed capacity. The energy-saving measure parameters include whether energy-saving renovations have been carried out, the duration since the renovations, and the specific energy-saving renovation projects undertaken. Based on the obtained energy consumption parameters, a multi-feature dataset is constructed and data verification and cleaning, data feature filtering, and data normalization are performed. The specific data feature screening involves: using stepwise linear regression to analyze the correlation between per capita energy consumption and energy consumption per unit area of ​​public buildings and various influencing factors in the sequence S=[year, month, quarter, per capita water consumption PWC, building area BA, number of energy users NEU, number of authorized personnel H, number of gasoline vehicles NGV, number of diesel vehicles NDV, petroleum consumption GC, other oil consumption OC, diesel consumption DC, liquefied petroleum consumption CLP, other energy consumption OEC, solar collector area SCA, solar system installed capacity SSIC, geothermal system installed capacity GIC, energy-saving renovation ESR]. Factors with correlations higher than the preset value are selected, while factors with correlations lower than the preset value are deleted. Construct an energy consumption prediction model, which includes a vector filtering layer and a convolutional neural network connected in sequence, and the convolutional neural network includes several convolutional layers and pooling layers according to different types of public institutions. The energy consumption prediction model was trained and tested using the preprocessed multi-feature dataset; The trained energy consumption prediction model was used to predict the energy consumption per unit area and per capita energy consumption of different types of public buildings in multiple quarters. Based on energy consumption parameters, energy consumption prediction models, and energy consumption prediction results, and using a Spearman and tree model coupled analysis method, eigenvector correlation analysis is performed on the energy consumption of various types of buildings in different seasons. Specifically: Based on Spearman's formula, the correlation of energy consumption parameters of different types of public buildings is numerically calculated to obtain the linear eigenvector correlation order result. Based on the tree model formula, the correlation numerical calculation of energy consumption parameters of different types of public buildings is performed to obtain the nonlinear characteristic vector correlation order result. Important feature vectors are selected based on the correlation order of linear and nonlinear feature vectors.

2. The method for quarterly energy consumption forecasting and energy-saving renovation analysis of public buildings according to claim 1, characterized in that, Label classification of multi-feature datasets based on energy consumption parameters: 70% of the data in the multi-feature dataset is used to construct the training set, and 30% of the data in the multi-feature dataset is used to construct the test set. During the training phase, the attention weights and parameters of the convolutional neural network are adjusted by optimizing the algorithm; The core algorithm uses Adam, with a batch size of 100 / 800, 800 iterations, an initial learning rate of 0.01, and a learning rate decay factor and a period of 0.5 and 400, respectively. Energy consumption for each type of energy consumption is predicted for each quarter. Once the prediction results are evaluated by preset indicators, the energy consumption prediction model is considered to have been trained.

3. The method for quarterly energy consumption forecasting and energy-saving renovation analysis of public buildings according to claim 1, characterized in that, The data verification and cleaning specifically involve: The quartile method was used to verify and clean the acquired energy load parameters. Specifically, Q3 + 1.5IQR was taken as the upper threshold and Q1 - 1.5IQR as the lower threshold. Here, Q3 is the third quartile of the distribution interval, which is the number at the 75th percentile after arranging the data sequence from smallest to largest; Q1 is the first quartile of the distribution interval, which is the number at the 25th percentile after arranging the data sequence from smallest to largest; IQR is the interquartile range, IQR = Q3 - Q1, and values ​​outside Q1 to Q3 are discarded as outliers.

4. The method for quarterly energy consumption forecasting and energy-saving renovation analysis of public buildings according to claim 1, characterized in that, The data normalization process is specifically as follows: ； in, X For data to be processed, X max The maximum value in the dataset. X min The minimum value in the dataset. X n This is the normalized data.

5. The method for quarterly energy consumption forecasting and energy-saving renovation analysis of public buildings according to claim 1, characterized in that, The energy consumption prediction model includes a stepwise regression vector filtering layer, a stepwise linear regression layer, a normalization layer, several convolutional layers and pooling layers, a dropout layer and a fully connected layer, a regression layer, an inverse normalization layer, and an output layer, all connected in sequence. The stepwise regression vector filtering layer is used to assign different weights to features of different importance. The stepwise linear regression layer is used to perform a stepwise regression fit on each feature vector to obtain the weight of each feature vector, so as to determine the importance of the input elements; The normalization layer is used to normalize the order of magnitude of the variables in the input feature set; The pooling layer is used to reduce the dimensionality of the data; The dropout layer sets a dropout rate to enhance the stability of the model; The fully connected layer is used to perform matrix multiplication and bias addition operations on the input features and the connection weights between each neuron to obtain the output result; The regression layer is used to calculate the half mean square error loss of the regression task and returns the output layer of the energy consumption prediction model as a RegressionOutputLayer object. The denormalization layer is used to restore the normalized data to the original data; The output layer is used to output the energy consumption per unit area and per capita energy consumption of the public building to be predicted.

6. A system for quarterly energy consumption forecasting and energy-saving renovation analysis of public buildings, characterized in that, The method for quarterly energy consumption forecasting and energy-saving renovation analysis of public buildings applied to any one of claims 1-5 includes a data acquisition and processing module, a data preprocessing module, a model building and training module, a prediction module, and a correlation analysis module; The data acquisition and processing module is used to obtain energy consumption parameters for different types of public buildings to be predicted in four quarters, including date parameters, load parameters, and energy-saving measure parameters; and to construct a multi-feature dataset based on the obtained energy consumption parameters. The data preprocessing module is used to perform data verification and cleaning, data feature filtering, and data normalization on multi-feature datasets. The model building and training module is used to build an energy consumption prediction model, which includes a vector filtering layer and a convolutional neural network connected in sequence. The convolutional neural network has several convolutional layers and pooling layers added according to different types of public institutions. The energy consumption prediction model is trained and predicted using a preprocessed multi-feature dataset. The prediction module is used to predict the energy consumption per unit area and per capita energy consumption of different types of public buildings in multiple quarters using a trained energy consumption prediction model. The correlation analysis module is used to perform feature vector correlation analysis on the energy consumption of various types of buildings in different seasons based on energy consumption parameters, energy consumption prediction models and energy consumption prediction results, using the Spearman and tree model coupling analysis method.

7. A computer-readable storage medium and an electronic device, the computer-readable storage medium storing a program, characterized in that: When the program is executed by the processor of the electronic device, it implements the method for quarterly energy consumption forecasting and energy-saving renovation analysis of public buildings as described in any one of claims 1-5.

Images