EMD decomposition-based carbon emission prediction method

Abstract

The invention discloses a carbon emission prediction method based on EMD decomposition, and the method comprises the following steps: S1, collecting historical carbon emission data of a prediction target in a preset time period, and constructing a historical carbon emission sequence of the prediction target; s2, obtaining an initial training sample, and constructing an initial training sample set; s3, for training samples in the initial training sample set, processing sample features based on EMD decomposition and multi-scale permutation entropy, and constructing an optimized sample set reflecting different time scale fluctuation characteristics and time sequence complexity; s4, constructing a carbon emission prediction model based on an LSTM network, and training the carbon emission prediction model by using the optimized sample set to obtain a trained carbon emission prediction model; and S5, predicting the carbon emission of the prediction target by using the trained carbon emission prediction model. According to the method, the samples reflecting different time scale fluctuation characteristics and time sequence complexity are obtained, and the carbon emission prediction model is trained on the basis, so that the prediction result is more accurate.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of carbon emission prediction, and particularly to a carbon emission prediction method based on EMD decomposition. BACKGROUND

[0002] In order to cope with the problems of greenhouse effect and global warming, it is necessary to accurately understand the situation of carbon emission and predict its future trend, so as to take effective emission reduction measures. With the accumulation of massive historical data of carbon emission in various regions and the development of artificial intelligence technology, the prediction of carbon emission is provided with favorable conditions. However, in the existing prediction methods, the time series of carbon emission is directly trained and predicted based on neural network and carbon emission composition, without considering the fluctuation characteristics of different time scales and the complexity of time series, so that the prediction result of carbon emission has the problem of insufficient accuracy. SUMMARY

[0003] The present application aims to overcome the shortcomings of the prior art and provide a carbon emission prediction method based on EMD decomposition. The time series of carbon emission is processed by EMD decomposition and multi-scale permutation entropy to obtain samples reflecting the fluctuation characteristics of different time scales and the complexity of time series, and the carbon emission prediction model is trained based on the samples, so that the prediction result is more accurate.

[0004] The present application is achieved by the following technical scheme: a carbon emission prediction method based on EMD decomposition, comprising the following steps:

[0005] S1. Collecting historical carbon emission data of the prediction target in a preset time period to construct a historical carbon emission sequence of the prediction target;

[0006] S2. Obtaining initial training samples from the historical carbon emission sequence to construct an initial training sample set;

[0007] S3. For the training samples in the initial training sample set, processing the sample features based on EMD decomposition and multi-scale permutation entropy, and constructing an optimized sample set reflecting the fluctuation characteristics of different time scales and the complexity of time series;

[0008] S4. Constructing a carbon emission prediction model based on LSTM network, and training the carbon emission prediction model using the optimized sample set to obtain a trained carbon emission prediction model;

[0009] S5. Using the trained carbon emission prediction model to predict the carbon emission of the prediction target.

[0010] The beneficial effects of the present application are: firstly, the present application decomposes the time series of carbon emissions into a series of IMF components and residual terms through EMD decomposition, then calculates the multi-scale permutation entropy of each IMF component and residual term, and optimizes the training samples based on the IMF components, residual terms and corresponding multi-scale permutation entropy, so that the training samples can reflect the fluctuation characteristics of different time scales and the complexity of time series (irregularity and randomness in different time scales), and the training of the carbon emission prediction model based on the optimized samples can make the prediction result more accurate. BRIEF DESCRIPTION OF DRAWINGS

[0011] Figure 1 The method flowchart of the present application. DETAILED DESCRIPTION

[0012] The technical solutions of the present application will be described in further detail below in conjunction with the drawings, but the protection scope of the present application is not limited to the following description.

[0013] As shown in the figure, a carbon emission prediction method based on EMD decomposition includes the following steps: Figure 1

[0014] S1. Collect the historical carbon emission data of the prediction target in the preset time period to construct the historical carbon emission sequence of the prediction target;

[0015] S2. Obtain the initial training samples from the historical carbon emission sequence to construct the initial training sample set;

[0016] S3. For the training samples in the initial training sample set, process the sample features based on EMD decomposition and multi-scale permutation entropy, and construct an optimized sample set reflecting the fluctuation characteristics of different time scales and the complexity of time series;

[0017] S4. Construct a carbon emission prediction model based on the LSTM network, and train the carbon emission prediction model using the optimized sample set to obtain a trained carbon emission prediction model;

[0018] S5. Use the trained carbon emission prediction model to predict the carbon emissions of the prediction target.

[0019] In the embodiments of the present application, the preset time period is composed of a plurality of sub-time periods, each sub-time period has the same time length, and the historical carbon emissions in each sub-time period form the historical carbon emission sequence in chronological order , denoted as:

[0020] ;

[0021] wherein, Em(k) represents the carbon emissions in the kth sub-time period, ​K represents the number of sub-periods.

[0022] In embodiments of the present application, the prediction target is an industrial park, a school or a city.

[0023] In embodiments of the present application, the step S2 comprises:

[0024] S201. From the historical carbon emission sequence, the carbon emissions of consecutive M sub-periods are selected to form a sequence :

[0025] ;

[0026] wherein, represents the carbon emissions of the selected mthsub-period, ; wherein M < K;

[0027] Taking as the sample feature, and taking as the sample label, an initial training sample is constructed.

[0028] S202. The step S201 is repeatedly executed to construct multiple training samples to form an initial training sample set.

[0029] In embodiments of the present application, the step S3 comprises:

[0030] S301. For any one training sample in the initial training sample set :

[0031] (1) The sample feature is decomposed to obtain n IMF components and a residual term ;

[0032] (2) For any one IMF component , , the multi-scale permutation entropy calculation is performed:

[0033] At the scale factor , the is respectively coarsely granulated to obtain time series at different scales, and then according to the time series at different scales, the permutation entropy at each scale is calculated to form a vector , which is the multi-scale permutation entropy of ; wherein, represents the permutation entropy calculated at the scale factor s;

[0034] (3) In ​, repeat (2) to obtain the multi-scale permutation entropy of each IMF component;

[0035] (4) For the residual term , calculate the multi-scale permutation entropy according to (2):

[0036] At the scale factor , the residual term is coarsely granulated to obtain a time series at different scales, and then the permutation entropy at each scale is calculated according to the time series at different scales, and a vector is formed, which is the multi-scale permutation entropy of the residual term ;

[0037] (5) Replace the sample feature with the optimized sample feature composed of each IMF component, residual term and corresponding multi-scale permutation entropy, denoted as:

[0038] ;

[0039] In the embodiments of the present application, a threshold of entropy can be set. For any IMF component, if its multi-scale permutation entropy has an element greater than the threshold, the IMF component and the corresponding multi-scale permutation entropy can be removed from the sample feature; for example, if there is an element greater than the threshold in , then is removed from the sample feature, which can reduce irregularity and randomness at different time scales.

[0040] Thus, a new training sample is obtained, denoted as:

[0041] ;

[0042] S302. For each training sample in the initial training sample set, repeat step S301 to form an optimized sample set reflecting the fluctuation characteristics of different time scales and the complexity of time series.

[0043] In the embodiments of the present application, in step S5, the sample feature of each training sample in the optimized sample set is taken as the input of the carbon emission prediction model, the label of each training sample is taken as the expected output, the mean square error MSE is taken as the loss function, and the MSprop or Adam optimizer is used to update the carbon emission prediction model. When the training of all samples in the optimized sample set is completed, or the loss function is less than a preset threshold, it is considered that the trained carbon emission prediction model is obtained.

[0044] In the embodiments of the present application, the step S5 comprises: obtaining carbon emissions in the last M-1 sub-time periods, constructing sample features in chronological order, then processing the sample features according to the step S301 to obtain optimized sample features, inputting the optimized sample features into the trained carbon emission prediction model to obtain the carbon emission prediction result of a future sub-time period.

[0045] In the embodiments of the present application, the length of each sub-time period is one day, one week or one month.

[0046] The above is the preferred embodiment of the present application, it should be understood that the present application is not limited to the form disclosed herein, should not be seen as excluding other embodiments, but can be used in other combinations, modifications and environment, and can be within the scope of the concept described herein, by the above teaching or related art or knowledge to make changes. The changes and variations made by the person skilled in the art without departing from the spirit and scope of the present application, shall be within the scope of the appended claims of the present application.

Claims

1. A carbon emission prediction method based on EMD decomposition, characterized in that: Includes the following steps: S1. Collect historical carbon emission data of the predicted target within a preset time period and construct the historical carbon emission sequence of the predicted target; S2. Obtain initial training samples from historical carbon emission sequences and construct an initial training sample set; S3. For the training samples in the initial training sample set, the sample features are processed based on EMD decomposition and multi-scale permutation entropy, and an optimized sample set reflecting the fluctuation characteristics and time series complexity at different time scales is constructed. S4. Construct a carbon emission prediction model based on an LSTM network, and train the carbon emission prediction model using an optimized sample set to obtain a trained carbon emission prediction model. S5. Use the trained carbon emission prediction model to predict the carbon emissions of the target.

2. The carbon emission prediction method based on EMD decomposition according to claim 1, characterized in that: The preset time period consists of multiple sub-time periods, each of equal length. The historical carbon emissions within each sub-time period are arranged in chronological order to form the historical carbon emission sequence. , denoted as: ； in, This represents the carbon emissions in the k-th sub-time period. K represents the number of sub-time periods.

3. The carbon emission prediction method based on EMD decomposition according to claim 1, characterized in that: The predicted targets are industrial parks, schools, or cities.

4. The carbon emission prediction method based on EMD decomposition according to claim 2, characterized in that: Step S2 includes: S201. From the historical carbon emission series, select M consecutive sub-time periods of carbon emissions to form a sequence. : ； in, This represents the carbon emissions for the selected m-th sub-time period. ; where M <K; Will As a sample feature, As sample labels, construct the initial training samples. ; S202. Repeat step S201 to construct multiple training samples and form an initial training sample set.

5. The carbon emission prediction method based on EMD decomposition according to claim 4, characterized in that: Step S3 includes: S301. For any training sample in the initial training sample set... : (1) Sample characteristics Decompose to obtain n IMF components. and a residual term ; (2) For any IMF component , Perform multi-scale permutation entropy calculation: In scale factor At that time, respectively for Coarse-graining is performed to obtain time series at different scales. Then, based on the time series at different scales, the permutation entropy at each scale is calculated and a vector is constructed. This vector is The multi-scale permutation entropy; among which... express The permutation entropy is calculated when the scale factor is s; (3) In Repeat step (2) to obtain the multiscale permutation entropy of each IMF component; (4) For the residual term Calculate the multi-scale permutation entropy according to (2): In scale factor At that time, for the residual terms respectively Coarse-graining is performed to obtain time series at different scales. Then, based on the time series at different scales, the permutation entropy at each scale is calculated and a vector is constructed. This vector is the residual term. Multiscale arrangement entropy; (5) Sample features Replace it with optimized sample features consisting of each IMF component, residual term, and corresponding multi-scale permutation entropy, denoted as: ； Thus, a new training sample is obtained, denoted as: ； S302. For each training sample in the initial training sample set, repeat step S301 to form an optimized sample set that reflects the fluctuation characteristics and time series complexity at different time scales.

6. The carbon emission prediction method based on EMD decomposition according to claim 5, characterized in that: In step S5, the sample features of each training sample in the optimization sample set are used as the input of the carbon emission prediction model, the label of each training sample is used as the expected output, the mean squared error (MSE) is used as the loss function, and the MSprop or Adam optimizer is used to update the carbon emission prediction model. When all samples in the optimization sample set have been trained, or when the loss function is less than a preset threshold, it is considered that a well-trained carbon emission prediction model has been obtained.

7. The carbon emission prediction method based on EMD decomposition according to claim 6, characterized in that: Step S5 includes: obtaining the carbon emissions in the most recent M-1 sub-time periods, constructing sample features in chronological order, then processing the sample features according to step S301 to obtain optimized sample features, inputting them into the trained carbon emission prediction model to obtain the carbon emission prediction result for the next sub-time period.

8. The carbon emission prediction method based on EMD decomposition according to claim 2, characterized in that: The length of each sub-time period is one day, one week, or one month.

Images