CN116050579A - Building energy consumption prediction method and system based on depth feature fusion network - Google Patents

Abstract

The invention relates to the technical field of energy consumption prediction, and discloses a building energy consumption prediction method and system based on a depth feature fusion network. According to the building energy consumption prediction method based on the depth feature fusion network, the one-dimensional sequence is converted into the two-dimensional space diagram according to the set time span, the periodic information of the building energy consumption sequence is enhanced, and the modified multi-scale convolution is embedded in the module, so that the space information and the time features of the input sequence can be further mined. Meanwhile, the embedded features of the time factors, the weather factors and the historical energy consumption sequences are fused, so that the correlation among various heterogeneous factors and the influence of the correlation on future energy consumption are captured, useful information in the influence factors is extracted, and the interference of a large amount of redundant information on a prediction task is avoided.

Description

Building energy consumption prediction method and system based on depth feature fusion network

Technical Field

The invention relates to the technical field of energy consumption prediction, in particular to a building energy consumption prediction method and system based on a depth feature fusion network.

Background

The accurate energy consumption prediction has important significance for the design planning of the power system, the establishment of policies such as power distribution and the like, and can also provide assistance for user portraits, anomaly detection and the like. With the development of smart cities, big data becomes an important feature of the era, and research on building energy consumption prediction methods is no longer constrained by data volume, which promotes the development and application of data-driven models in the field. In recent years, neural network-based approaches have received a great deal of attention from researchers, which benefit from their modeling capabilities for complex scenes and adaptability to various heterogeneous data. The artificial neural network ANN is used as a classical model, has the performance in the field of building energy consumption prediction obviously superior to that of the traditional statistical regression-based method, and has been widely applied. In recent years, researchers in the field explore deep network model application, and representative convolutional neural networks CNN, long-short-term memory cyclic neural networks LSTM, variants Bi-LSTM thereof and the like, and research and construct CNN-LSTM, and capture more information in an input sequence by utilizing a deep network structure by means of the local perceptibility of CNN and the time sequence modeling capability of LSTM.

In addition, given that trends in building energy consumption are significantly affected by occupant behavior, presenting regular changes, some studies have introduced relevant factors (such as time factors and weather factors) that affect occupant behavior as part of the model input. However, the existing method ignores the characteristics of potential correlation between the influencing factors and the energy consumption, introduces a large amount of interference information while enriching input features, and limits the improvement of prediction accuracy.

Disclosure of Invention

Therefore, the invention aims to solve the technical problems of insufficient feature extraction and easy interference of redundant information in the energy consumption prediction in the prior art.

In order to solve the technical problems, the invention provides a building energy consumption prediction method based on a depth feature fusion network, which comprises the following steps:

s1, respectively performing one-hot coding on historical time factor data and future time factor data, and respectively generating

And->

Wherein c _t Representing the number of encoded temporal features, l _i Representing the length of the input sequence, l _o Representing the predicted sequence length;

s2, carrying out normalization processing on continuous variable in meteorological factor data to generate

Carrying out one-hot coding on discrete variables in the meteorological factor data to generate +.>

Wherein c _s And c _d Respectively representing the number of continuous weather features and the number of discrete weather features;

s3, T is taken as ^h 、T ^f 、W ^s 、W ^d Respectively inputting the characteristic embedding networks and respectively generating time embedding characteristics f ^th Time embedded feature f ^tf Weather embedded feature f ^ws And weather embedded feature f ^wd ；

S4, carrying out normalization processing on the historical energy consumption sequence data to generate a historical energy consumption sequence E;

s5, according to different time spans k, converting the historical energy consumption sequence E into graphs with different sizes in a two-dimensional space, then sending the converted graphs into a multi-scale convolution network to perform feature extraction, and finally calculating and generating energy consumption embedded features f through weighting and combining operation ^e ；

S6, connection feature f ^th 、f ^tf 、f ^ws 、f ^wd 、f ^e Performing feature fusion by using a linear layer of the neural network to obtain a feature F;

s7, setting a trainable weight parameter, carrying out weighted addition on the historical energy consumption sequence E and the characteristic F, and then inputting the weighted addition into a multi-layer perceptron to obtain a final prediction result.

In one embodiment of the invention, the feature embedding network consists of one-dimensional convolution layers with convolution kernel size of a×a and full connection layers FC, and each convolution layer is followed by a ReLU activation function; wherein a is more than or equal to 3.

In one embodiment of the present invention, step S5 includes:

according to different time spans k and preset graph height H, performing two-dimensional conversion on the historical energy consumption sequence E to generate a graph G; g generated based on different time spans is then fed into a multi-scale convolutional network consisting of a 1 x 1 convolutional layer, a 3 x 3 convolutional layer, and a pooling layer, respectively, wherein the pooling layer consists of an average pooling layer and a maximum pooling layer.

In one embodiment of the present invention, the two-dimensional conversion of the historical energy consumption sequence E generates a graph G as follows:

G[i，：]＝E[(i×k)：(i×k+W)]

wherein w=l _i -H+1 is the width of the two-dimensional plot, and i×k+W.ltoreq.l _i The method comprises the steps of carrying out a first treatment on the surface of the The sequence is traversed using sliding windows during the conversion, and the time span determines the interval between each row of data in the two-dimensional map, i.e., the time difference of separation.

In one embodiment of the invention, the normalization is formulated as follows:

wherein x is _t The input variables at the time t are represented, and max and min represent preset maximum and minimum values respectively.

In one embodiment of the invention, the time factor includes month, hour, week, day type; discrete variables in the meteorological factor data include weather, outdoor temperature, relative humidity, atmospheric pressure.

The invention also provides a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps of any of the methods described above when the program is executed.

The invention also provides a computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of any of the methods described above.

The invention also provides a processor for running a program, wherein the program runs to execute the method of any one of the above.

The invention also provides a building energy consumption prediction system based on the depth feature fusion network, which comprises the following modules:

the time factor module is used for respectively carrying out one-hot coding on the historical time factor data and the future time factor data and respectively generating

And->

Wherein c _t Representing the number of encoded temporal features, l _i Representing the length of the input sequence, l _o Representing the predicted sequence length;

the weather factor module is used for carrying out normalization processing on continuous variables in the weather factor data to generate

Performing one-hot coding on discrete variables in the meteorological factor data to generate

Wherein c _s And cx represents the continuous and discrete weather feature quantity, respectively;

time and weather embedding feature generation module for embedding T ^h 、T ^f 、W ^s 、W ^d Respectively inputting the characteristic embedding network and respectively generating time embedding characteristicsf ^th Time embedded feature f ^tf Weather embedded feature f ^ws And weather embedded feature f ^wd ；

The historical energy consumption module is used for carrying out normalization processing on the historical energy consumption sequence data to generate a historical energy consumption sequence E;

the multi-span time fusion module is used for converting the historical energy consumption sequence E into graphs with different sizes in a two-dimensional space according to different time spans k, then sending the converted graphs into the multi-scale convolution network for feature extraction, and finally calculating and generating energy consumption embedded features f through weighted combination operation ^e ；

A feature fusion module for connecting the features f ^th 、t ^ft 、f ^ws 、f ^wd 、f ^e Performing feature fusion by using a linear layer of the neural network to obtain a feature F;

and the prediction module is used for setting a trainable weight parameter, carrying out weighted addition on the historical energy consumption sequence E and the characteristic F, and then inputting the weighted addition into the multi-layer perceptron to obtain a final prediction result.

Compared with the prior art, the technical scheme of the invention has the following advantages:

according to the building energy consumption prediction method based on the depth feature fusion network, the one-dimensional sequence is converted into the two-dimensional space diagram according to the set time span, the periodic information of the building energy consumption sequence is enhanced, and the modified multi-scale convolution is embedded in the module, so that the space information and the time features of the input sequence can be further mined.

Meanwhile, the embedded features of the time factors, the weather factors and the historical energy consumption sequences are fused, so that the correlation among various heterogeneous factors and the influence of the correlation on future energy consumption are captured, useful information in the influence factors is extracted, and the interference of a large amount of redundant information on a prediction task is avoided.

The foregoing description is only an overview of the present invention, and is intended to be implemented in accordance with the teachings of the present invention, as well as the preferred embodiments thereof, together with the following detailed description of the invention, given by way of illustration only, together with the accompanying drawings.

Drawings

In order that the invention may be more readily understood, a more particular description of the invention will be rendered by reference to specific embodiments thereof that are illustrated in the appended drawings, in which

FIG. 1 is a flow chart of a method for predicting building energy consumption based on a depth feature fusion network in an embodiment of the invention;

FIG. 2 is a schematic diagram of a building energy consumption prediction method based on a depth feature fusion network in an embodiment of the invention;

FIG. 3 is a block diagram of a feature embedding network in an embodiment of the present invention;

FIG. 4 is a block diagram of a multi-span time fusion module in an embodiment of the invention;

FIG. 5 is a schematic diagram of a pooling operation in an embodiment of the invention;

FIG. 6 is a schematic diagram of a sequence conversion process in an embodiment of the invention;

FIG. 7 is historical power consumption data of experimental samples in an embodiment of the invention;

fig. 8 is a visual diagram of a test result of a building energy consumption prediction method based on a depth feature fusion network on a sample in an embodiment of the invention.

Detailed Description

The present invention will be further described with reference to the accompanying drawings and specific examples, which are not intended to be limiting, so that those skilled in the art will better understand the invention and practice it.

Example 1

Referring to fig. 1-2, the embodiment discloses a building energy consumption prediction method based on a depth feature fusion network, which comprises the following steps:

step S1, respectively performing one-hot coding on the historical time factor data and the future time factor data, and respectively generating

And->

Wherein c _t Representing the number of encoded temporal features, l _i Representing the length of the input sequence, l _o Representing the predicted sequence length;

the specific method for the one-hot coding is as follows:

in order to convert the class variable into a form which is easy to use by an algorithm model, the discrete variable is binarized by using one-hot coding, specifically, for a feature with n states, the discrete variable is converted into a one-dimensional vector with the length of n, and only one bit of the vector is 1 in each sample, and the other bits are 0 to represent one state of the feature. For example, day type may be defined as two states, workday and non-workday, encoded as (1, 0) and (0, 1), respectively.

S2, carrying out normalization processing on continuous variables in the meteorological factor data to generate

Carrying out one-hot coding on discrete variables in the meteorological factor data to generate +.>

Wherein c _s And c _d Respectively representing the number of continuous weather features and the number of discrete weather features;

optionally, the normalization is formulated as follows:

wherein x is _t The input variables at the time t are represented, and max and min represent preset maximum and minimum values respectively.

Alternatively, referring to Table 1, the time factors include month, hour, week, day type, etc.; discrete variables in the meteorological factor data include weather, outdoor temperature, relative humidity, barometric pressure, and the like.

Table 1 summary of input variables

Step S3, T is carried out ^h 、T ^f 、W ^s 、W ^d Respectively inputting the feature embedding networks EN and respectively generating time embedding features f ^th Time embedded feature f ^tf Weather embedded feature f ^ws And weather embedded feature f ^wd ；

Referring to fig. 3, optionally, the feature embedding network is composed of one-dimensional convolution layers with convolution kernel size a×a and full connection layers FC, with each convolution layer followed by a ReLU activation function; wherein a is more than or equal to 3.

S4, carrying out normalization processing on the historical energy consumption sequence data to generate a historical energy consumption sequence

The formula of the normalization process may be as shown in step S2 described above.

S5, converting the historical energy consumption sequence E into graphs with different sizes in a two-dimensional space according to different time spans k, then sending the converted graphs into a multi-scale convolution network for feature extraction, and finally calculating and generating energy consumption embedded features f through weighted combination operation ^e ；

In one embodiment, the structure of the multi-span time fusion module performing step S5 refers to fig. 4. The step S5 specifically comprises the following steps: according to different time spans k and preset graph height H, performing two-dimensional conversion on the historical energy consumption sequence E to generate a graph G; then, G generated based on different time spans is fed into a multi-scale convolutional network composed of a 1×1 convolutional layer, a 3×3 convolutional layer, and a pooling layer, respectively, wherein the pooling layer is composed of an average pooling layer (average pooling) and a maximum pooling layer (max pooling) for performing a pooling operation, referring to fig. 5.

Step S6, connecting feature f ^th 、f ^tf 、f ^ws 、f ^wd 、f ^e Generating (f) ^th ,f ^tf ,f ^ws ,f ^wd ,f ^e ) And performing feature fusion by using a linear layer of the neural network to obtain features

In one embodiment of the present invention, the two-dimensional conversion of the historical energy consumption sequence E generates a graph G as follows:

G[i，：]＝E[(i×k)：(i×k+W)]

wherein w=l _i -H+1 is the width of the two-dimensional plot, and i×k+W.ltoreq.l _i The method comprises the steps of carrying out a first treatment on the surface of the The conversion process is shown in FIG. 6, with a time span k ₁ =1 and k ₂ For example, =2, the sequence is traversed using a sliding window during the conversion, and the time span determines the interval between each row of data in the two-dimensional map, i.e. the time difference of separation.

S7, setting a trainable weight parameter alpha, carrying out weighted addition on the historical energy consumption sequence E and the characteristic F, and then inputting the weighted addition into a multi-layer perceptron to obtain a final prediction result

According to the building energy consumption prediction method based on the depth feature fusion network, the one-dimensional sequence is converted into the two-dimensional space diagram according to the set time span, the periodic information of the building energy consumption sequence is enhanced, and the modified multi-scale convolution is embedded in the module, so that the space information and the time features of the input sequence can be further mined.

Meanwhile, the embedded features of the time factors, the weather factors and the historical energy consumption sequences are fused, so that the correlation among various heterogeneous factors and the influence of the correlation on future energy consumption are captured, useful information in the influence factors is extracted, and the interference of a large amount of redundant information on a prediction task is avoided.

To verify the accuracy and robustness of the present invention, experiments were performed based on the public dataset HUE (https:// dataset harvard. Edu/dataset. Xhtmlppersstentid = doi:10.7910/DVN/N3 HGRN).

HUE dataset contains 22 buildings in Vancouver areaPower data and weather data for the nearest weather station we selected the power data for 2018-11-01 to 2019-11-20 in the composition_2 and weather data for the Vancouver corresponding weather station as experimental samples to verify the model. The power load curve during this period is shown in fig. 7. Wherein the maximum value max of the outdoor temperature _T = 29.20 ℃, minimum min _T -8.60 ℃, maximum value max of relative humidity _RH =100%, minimum min _RH =18% maximum atmospheric pressure max _P = 103.94kPa, minimum min _P =98.15 kPa; maximum value max of historical energy consumption _E =1.09 kWh, minimum min _T ＝0.09kWh；a＝3；l _i ＝168，k＝(k ₁ ，k ₂ ，k ₃ )＝(1，12，24)，H＝72。

In training, a sliding window with a sliding step length of one hour is used for reconstructing a data set, the sliding window is 168+24 in size, the time span of the input historical sequence is 168 hours, the time span to be predicted is 24 hours, and experimental parameter settings are shown in table 2. The initial learning rate during training was 0.001 and the last 10 rounds were reduced to 0.0001.

Table 2 experimental parameter settings

Training sample number	Number of test samples	Learning rate	Number of iterations
				8124	732	0.001	40

Table 3 shows comparative experiments conducted by the present invention to demonstrate the advantages of the method of the present invention, and the comparative subjects are popular methods of predicting energy consumption based on ANN, LSTM, CNN-LSTM and CNN-BiLSTM. The mean absolute error MAE and the root mean square error RMSE are selected as evaluation indices. The method provided by the invention achieves the optimal performance on both indexes.

Table 3 comparison of the present invention with other popular methods

Method	MAE	RMSE
			The invention is that	0.076	0.091
ANN	0.095	0.137
			LSTM	0.099	0.147
CNN-LSTM	0.099	0.140
			CNN-BiLSTM	0.103	0.142

Referring to fig. 8, a visual diagram of a test result of a building energy consumption prediction method based on a depth feature fusion network in an embodiment of the present invention shows that the energy consumption value predicted by the method of the present invention has very small error from the true value, and the prediction accuracy is very high.

Example two

The invention also provides a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps of the method described in embodiment one when executing the program.

Example III

The invention also provides a computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of the method described in embodiment one.

Example IV

The invention also provides a processor for running a program, wherein the program runs to execute the method in the first embodiment.

Example five

The embodiment provides a building energy consumption prediction system based on a depth feature fusion network, which comprises the following modules:

the time factor module is used for respectively carrying out one-hot coding on the historical time factor data and the future time factor data and respectively generating

And->

Wherein c _t Representing the number of encoded temporal features, l _i Representing the length of the input sequence, l _o Representing the predicted sequence length;

a weather factor module for determining weather factor dataNormalizing the continuous variable to generate

Performing one-hot coding on discrete variables in the meteorological factor data to generate

Wherein c _s And c _d Respectively representing the number of continuous weather features and the number of discrete weather features;

time and weather embedding feature generation module for embedding T ^h 、T ^f 、W ^s 、W ^d Respectively inputting the characteristic embedding networks and respectively generating time embedding characteristics f ^th Time embedded feature f ^tf Weather embedded feature f ^ws And weather embedded feature f ^wd ；

The historical energy consumption module is used for carrying out normalization processing on the historical energy consumption sequence data to generate a historical energy consumption sequence E;

the multi-span time fusion module is used for converting the historical energy consumption sequence E into graphs with different sizes in a two-dimensional space according to different time spans k, then sending the converted graphs into the multi-scale convolution network for feature extraction, and finally calculating and generating energy consumption embedded features f through weighted combination operation ^e The method comprises the steps of carrying out a first treatment on the surface of the Refer to fig. 4.

A feature fusion module for connecting the features f ^th 、f ^tf 、f ^ws 、f ^wd 、f ^e Performing feature fusion by using a linear layer of the neural network to obtain a feature F;

and the prediction module is used for setting a trainable weight parameter, carrying out weighted addition on the historical energy consumption sequence E and the characteristic F, and then inputting the weighted addition into the multi-layer perceptron to obtain a final prediction result.

The building energy consumption prediction system based on the depth feature fusion network in the embodiment of the invention is used for realizing the building energy consumption prediction method based on the depth feature fusion network, so that the specific implementation mode of the system can be seen from the embodiment part of the building energy consumption prediction method based on the depth feature fusion network in the foregoing, and therefore, the specific implementation mode can be referred to the description of the corresponding embodiment of each part and is not further described herein.

In addition, since the building energy consumption prediction system based on the depth feature fusion network of the present embodiment is used to implement the foregoing building energy consumption prediction method based on the depth feature fusion network, the function of the system corresponds to that of the foregoing method, and the description thereof is omitted herein.

It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

It is apparent that the above examples are given by way of illustration only and are not limiting of the embodiments. Other variations and modifications of the present invention will be apparent to those of ordinary skill in the art in light of the foregoing description. It is not necessary here nor is it exhaustive of all embodiments. And obvious variations or modifications thereof are contemplated as falling within the scope of the present invention.

Claims (10)

1. The building energy consumption prediction method based on the depth feature fusion network is characterized by comprising the following steps of:

s1, respectively performing one-hot coding on historical time factor data and future time factor data, and respectively generating

And->

Wherein c _t Representing the number of encoded temporal features, l _i Representing the length of the input sequence, l _o Representing the predicted sequence length;

s2, carrying out normalization processing on continuous variable in meteorological factor data to generate

Carrying out one-hot coding on discrete variables in the meteorological factor data to generate +.>

Wherein c _s And c _d Respectively representing the number of continuous weather features and the number of discrete weather features;

s3, T is taken as ^h 、T ^f 、W ^s 、W ^d Respectively inputting the characteristic embedding networks and respectively generating time embedding characteristics f ^th Time embedded feature f ^tf Weather embedded feature f ^ws And weather embedded feature f ^wd ；

S4, carrying out normalization processing on the historical energy consumption sequence data to generate a historical energy consumption sequence E;

s5, according to different time spans k, converting the historical energy consumption sequence E into graphs with different sizes in a two-dimensional space, then sending the converted graphs into a multi-scale convolution network to perform feature extraction, and finally calculating and generating energy consumption embedded features f through weighting and combining operation ^e ；

S6, connection feature f ^th 、f ^tf 、f ^ws 、f ^wd 、f ^e Performing feature fusion by using a linear layer of the neural network to obtain a feature F;

s7, setting a trainable weight parameter, carrying out weighted addition on the historical energy consumption sequence E and the characteristic F, and then inputting the weighted addition into a multi-layer perceptron to obtain a final prediction result.

2. The method for predicting building energy consumption based on depth feature fusion network according to claim 1, wherein the feature embedding network is composed of one-dimensional convolution layers with convolution kernel size of a×a and full connection layers FC, and each convolution layer is followed by a ReLU activation function; wherein a is more than or equal to 3.

3. The method for predicting building energy consumption based on depth feature fusion network of claim 1, wherein step S5 comprises:

according to different time spans k and preset graph height H, performing two-dimensional conversion on the historical energy consumption sequence E to generate a graph G; g generated based on different time spans is then fed into a multi-scale convolutional network consisting of a 1 x 1 convolutional layer, a 3 x 3 convolutional layer, and a pooling layer, respectively, wherein the pooling layer consists of an average pooling layer and a maximum pooling layer.

4. A method for predicting building energy consumption based on depth feature fusion network according to claim 3, wherein the two-dimensional conversion of the historical energy consumption sequence E generates a graph G as follows:

G[i，：]＝E[(i×k)：(i×k+W)]

wherein w=l _i -H+1 is the width of the two-dimensional plot, and i×k+W.ltoreq.l _i The method comprises the steps of carrying out a first treatment on the surface of the The sequence is traversed using sliding windows during the conversion, and the time span determines the interval between each row of data in the two-dimensional map, i.e., the time difference of separation.

5. The depth feature fusion network-based building energy consumption prediction method according to claim 1, wherein the normalization process is formulated as follows:

wherein x is _t The input variables at the time t are represented, and max and min represent preset maximum and minimum values respectively.

6. The method for predicting building energy consumption based on depth feature fusion network of claim 1, wherein the time factors comprise month, hour, week, day types; discrete variables in the meteorological factor data include weather, outdoor temperature, relative humidity, atmospheric pressure.

7. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of the method according to any of claims 1 to 6 when the program is executed.

8. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the steps of the method according to any one of claims 1 to 6.

9. A processor for running a program, wherein the program when run performs the method of any one of claims 1 to 6.

10. The building energy consumption prediction system based on the depth feature fusion network is characterized by comprising the following modules:

the time factor module is used for respectively carrying out one-hot coding on the historical time factor data and the future time factor data and respectively generating

And->

Wherein c _t Representing the number of encoded temporal features, l _i Representing the length of the input sequence, l _o Representing the predicted sequence length;

the weather factor module is used for carrying out normalization processing on continuous variables in the weather factor data to generate

Performing one-hot coding on discrete variables in the meteorological factor data to generate

Wherein c _s And c _d Respectively representing the number of continuous weather features and the number of discrete weather features;

time and weather embedding feature generation module for embedding T ^h 、T ^f 、W ^s 、W ^d Respectively inputting the characteristic embedding networks and respectively generating time embedding characteristics f ^th Time embedded feature f ^tf Weather embedded featuresf ^ws And weather embedded feature f ^wd ；

The historical energy consumption module is used for carrying out normalization processing on the historical energy consumption sequence data to generate a historical energy consumption sequence E;

the multi-span time fusion module is used for converting the historical energy consumption sequence E into graphs with different sizes in a two-dimensional space according to different time spans k, then sending the converted graphs into the multi-scale convolution network for feature extraction, and finally calculating and generating energy consumption embedded features f through weighted combination operation ^e ；

A feature fusion module for connecting the features f ^th 、f ^tf 、f ^ws 、f ^wd 、f ^e Performing feature fusion by using a linear layer of the neural network to obtain a feature F;

and the prediction module is used for setting a trainable weight parameter, carrying out weighted addition on the historical energy consumption sequence E and the characteristic F, and then inputting the weighted addition into the multi-layer perceptron to obtain a final prediction result.

Images