Carbon emission factor prediction method and system for multi-time granularity data coordination

Abstract

The present invention provides a carbon emission factor prediction method and system for multi-time granularity data coordination. The method comprises: acquiring first carbon emission data of an area to be predicted, and, according to time granularities, establishing a plurality of first prediction models for initially predicting a carbon emission factor for the first carbon emission data, each time granularity corresponding to one first prediction model; on the basis of the time granularities and a time period to be predicted, establishing a time hierarchical matrix, and, on the basis of the time hierarchical matrix, fusing the plurality of first prediction models to obtain an initial first coordination prediction model; and, on the basis of the first carbon emission data, updating parameters of the first coordination prediction model to obtain a trained second coordination prediction model, so that the carbon emission factor is predicted by means of the second coordination prediction model. The prediction accuracy for the carbon emission factor can be improved by using the present invention.

Description

Carbon emission factor prediction method and system based on coordinated multi-time granularity data Technical Field

[0001] The present invention relates to the field of carbon emission technology, and in particular to a carbon emission factor prediction method and system coordinated with multi-time granularity data. Background Art

[0002] Currently, as the development of the carbon market accelerates, economic incentives are being used to encourage businesses and organizations to reduce carbon emissions, thereby accelerating the achievement of the dual carbon goals. Within this market, the carbon emission factor, a numerical indicator, quantifies and assesses carbon emission levels, providing fundamental data and guidance for the design and operation of the carbon market. Carbon emission factor forecasts enable market managers to understand the changing trends of carbon emission factors across different activities and industries, setting appropriate carbon emission limits and pricing, and thus guiding and promoting emission reduction actions.

[0003] However, there is a temporal granularity issue with carbon emission factor data: published carbon emission factors are recorded on an "annual" basis, while local governments may hold carbon emission factor data at a finer granularity (such as "monthly"). Therefore, if existing carbon emission factor prediction methods are used to predict data at different time granularities, the resulting prediction results are often difficult to be self-consistent, that is, the aggregated fine-grained prediction results are inconsistent with the coarse-grained predictions. This makes the existing carbon emission factor prediction methods have low utilization rates for data at different time granularities, and often ignores the coupling relationship between time levels, resulting in low accuracy of the obtained carbon emission factors.

[0004] Summary of the Invention

[0005] The purpose of the present invention is to address the deficiencies of the above-mentioned existing related technologies and propose a carbon emission factor prediction method and system with coordinated multi-time granularity data, which can improve the accuracy of the carbon emission factor.

[0006] In a first aspect, the present invention provides a carbon emission factor prediction method based on coordinated data of multiple time granularities, comprising:

[0007] Collecting first carbon emission data of the area to be predicted, and establishing multiple first prediction models for initially predicting carbon emission factors based on the first carbon emission data at a time granularity; wherein each time granularity corresponds to a first prediction model;

[0008] Establishing a time hierarchy matrix according to the time granularity and the time period to be predicted, and fusing multiple first prediction models according to the time hierarchy matrix to obtain an initial first coordinated prediction model;

[0009] The parameters of the first coordinated prediction model are updated according to the first carbon emission data to obtain a trained second coordinated prediction model, so that the carbon emission factor is predicted by the second coordinated prediction model.

[0010] The present invention adopts the method of establishing a prediction model for different time granularities respectively, which can improve the utilization rate of multivariate time series data, and establishes a time hierarchy matrix according to the time granularity and the period to be predicted. The time hierarchy matrix is ​​used to fuse multiple prediction models, which can aggregate the prediction results of different fine-grainedness, strengthen the implicit influence of the coupling relationship of the time hierarchy on the prediction of carbon emission factors, and construct a self-consistent and coordinated carbon emission factor coordination prediction model, which can form consistent multi-time hierarchy prediction results, thereby improving the prediction accuracy of carbon emission factors, and providing a basis for insight analysis of the changing trend of carbon emission factors at different time scales.

[0011] Furthermore, the fusing of the multiple first prediction models according to the time hierarchy matrix to obtain an initial first coordinated prediction model includes:

[0012] According to the number of predicted values ​​of carbon emission factors at the finest time granularity, an initial coordination matrix is ​​constructed, and an output matrix composed of the outputs of multiple first prediction models is obtained. The initial first coordination prediction model is constructed in the order of point product of the time hierarchy matrix, the initial coordination matrix and the output matrix.

[0013] The present invention fuses the output matrix composed of multiple prediction model outputs using a time hierarchy matrix and a coordination matrix, which can aggregate the prediction results of different prediction models, strengthen the implicit influence of the coupling relationship of the time hierarchy on the prediction of carbon emission factors, and construct a self-consistent and coordinated carbon emission factor coordination prediction model, thereby improving the prediction accuracy of carbon emission factors.

[0014] Furthermore, the initial coordination matrix is ​​constructed based on the number of predicted values ​​of carbon emission factors at the finest time granularity, including:

[0015] A unit matrix is ​​constructed with the number of predicted values ​​of carbon emission factors at the finest time granularity, and a zero matrix is ​​constructed with the number of predicted values ​​of carbon emission factors as rows and the difference between the total predicted values ​​of carbon emission factors at all time granularities and the number of predicted values ​​of carbon emission factors as columns. The block matrix composed of the zero matrix and the unit matrix is ​​used as the initial coordination matrix.

[0016] Furthermore, the step of establishing a time hierarchy matrix according to the time granularity and the time period to be predicted includes:

[0017] If the time granularity includes: year, quarter and month, then the time granularity is 3. If the forecast period is the next 12 months, then the time hierarchy matrix is ​​expressed as:

[0018] in, 14=(1,…,1) 1×4 ;S M =I 12×12 , I is the identity matrix.

[0019] The present invention establishes three time granularities through year, season and month, and constructs a time hierarchy matrix composed of three time granularities through the period to be predicted, so as to facilitate the aggregation of prediction results of different fine-grainedness, improve the prediction accuracy of carbon emission factors, and improve the reliability and credibility of carbon emission factor prediction.

[0020] Furthermore, updating parameters of the first coordinated prediction model according to the first carbon emission data includes:

[0021] Sequentially dividing the second carbon emission data of the plurality of first prediction models into a training set and a validation set in proportion; the first carbon emission data includes: the second carbon emission data corresponding to each first prediction model;

[0022] The initial parameters and initial coordination matrix of the first coordination prediction model are updated according to the training set, the validation set and a pre-constructed two-layer optimization model; wherein the two-layer optimization model includes: a lower-layer optimization model and an upper-layer optimization model; the lower-layer optimization model optimizes the initial parameters according to the training set to obtain the optimal parameters; the upper-layer optimization model optimizes the initial coordination matrix according to the validation set and the optimal parameters.

[0023] Furthermore, the two-layer optimization model is expressed as:

[0024] in, is the lower-level optimization model, It is represented by the number of training set samples of the first prediction model i, i = 1, 2, ..., k; y j,1 is the true load of the jth training sample; G is the initial coordination matrix, is the initial parameter set of the k first prediction models; is the prediction result of the training set input to the first coordinated prediction model; is the prediction result of the test set to the first coordinated prediction model; It is represented by the number of validation set samples of the first prediction model i; j,2 is the true load of the j-th validation sample.

[0025] Furthermore, the second carbon emission data of the plurality of first prediction models are divided into a training set and a validation set in proportion in sequence, including:

[0026] The features of the second carbon emission data of multiple first prediction models are screened in turn to obtain feature screening results, the feature screening results are divided into a training set, a validation set and a test set in proportion, and the training set, the validation set and the test set are normalized by column to obtain a normalized training set, a normalized validation set and a normalized test set, so that training, validation and testing are finally performed according to the normalized training set, the normalized validation set and the normalized test set, respectively; wherein each first prediction model corresponds to a normalized training set, a normalized validation set and a normalized test set.

[0027] Furthermore, the test is performed based on the normalized test set, including:

[0028] Inputting multiple normalized test sets into the corresponding optimized second prediction model respectively, outputting corresponding prediction results according to the optimal parameters obtained after optimization, and inputting multiple prediction results into the second prediction model respectively, outputting the total prediction result according to the optimal coordination matrix obtained after optimization;

[0029] The prediction accuracy of each second prediction model is evaluated based on the actual load of each normalized test set and the total prediction result.

[0030] Furthermore, the prediction accuracy evaluation of each second prediction model includes:

[0031] The mean absolute percentage error is used as an evaluation index of the prediction accuracy of the second prediction model, and the evaluation index is expressed as:

[0032] Among them, j is the test sample number, t is the time point number of the period to be predicted, is the prediction result of the normalized test set, is the actual load corresponding to the test sample, is the number of test set samples of the i-th second prediction model, and T is the total number of time points in the period to be predicted.

[0033] In a second aspect, the present invention provides a carbon emission factor prediction system with coordinated data of multiple time granularities, comprising: a prediction model establishment module, a coordinated prediction model establishment module, and a training module; wherein,

[0034] The prediction model establishment module is used to collect first carbon emission data of the area to be predicted, and establish multiple first prediction models for initially predicting carbon emission factors based on the first carbon emission data according to time granularity; wherein each time granularity corresponds to a first prediction model;

[0035] The coordinated prediction model establishment module is used to establish a time hierarchy matrix according to the time granularity and the time period to be predicted, and to fuse multiple first prediction models according to the time hierarchy matrix to obtain an initial first coordinated prediction model;

[0036] The training module is used to update the parameters of the first coordinated prediction model according to the first carbon emission data to obtain a trained second coordinated prediction model, so as to predict the carbon emission factor through the second coordinated prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] FIG1 is a flow chart of a carbon emission factor prediction method using multi-time granularity data coordination provided by this embodiment;

[0038] FIG2 is a schematic diagram of a coordinated prediction model provided in this embodiment;

[0039] FIG3 is a schematic flow chart of a complete carbon emission factor prediction method using multi-time granularity data coordination provided by this embodiment;

[0040] FIG4 is a schematic diagram of the structure of a carbon emission factor prediction system with coordinated data of multiple time granularities provided in this embodiment. DETAILED DESCRIPTION

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0042] It is worth noting that the purpose of the present invention is to overcome the shortcomings of the existing technology and propose a carbon emission factor prediction method and system with coordinated data of multiple time granularities. First, the carbon emission factor of the area to be predicted is investigated, the different time granularities to be predicted are determined, and a time hierarchy matrix is ​​established; then, a prediction model of the carbon emission factor and a time hierarchy coordination model are established; then, the data set is preprocessed, the prediction model and the coordination matrix are updated, and a fitted prediction model of the carbon emission factor with multiple time granularities is obtained; finally, the prediction accuracy of the carbon emission factor prediction at different time granularities is calculated, and the prediction effect is evaluated; the use of the present invention helps to solve the problem of difficulty and incoordination in predicting carbon emission factors at different time granularities, thereby improving the prediction accuracy of the carbon emission factor. In order to better illustrate the technical solution of the present invention, it will be described in detail from the following examples.

[0043] Example 1

[0044] 1 is a flow chart of a carbon emission factor prediction method based on coordinated multi-time granularity data provided by this embodiment, which includes steps S11 to S13, specifically:

[0045] Step S11: collect first carbon emission data of the area to be predicted, and establish multiple first prediction models for initially predicting carbon emission factors based on the first carbon emission data according to time granularity; wherein each time granularity corresponds to a first prediction model.

[0046] In some embodiments, carbon emission data of the area to be predicted is collected, changes in carbon emission factors are analyzed, and combined with expert advice, k different time granularities and time periods to be predicted are determined; where k is a positive integer.

[0047] It is worth noting that the first carbon emission data is the historical carbon emission data of the area to be predicted, and the first prediction model is an initial prediction model corresponding to the time granularity.

[0048] In some embodiments, a time hierarchy matrix is ​​established based on the time granularity and the period to be predicted, including: if the time granularity includes: year, quarter and month, then the time granularity k=3; if the period to be predicted is the next 12 months, then the time hierarchy matrix is ​​expressed as:

[0049] in, 14=(1,…,1) 1×4 ;S M =I 12×12 , I is the identity matrix.

[0050] In some embodiments, a time hierarchy matrix is ​​established based on the selected time granularity and time period. Where n is the number of predicted values ​​under all time granularity periods; m is the number of predicted values ​​under the finest time granularity. The time hierarchy matrix S can be written as k block matrices S k , each block matrix represents the weight of the kth target time granularity prediction constructed with the predicted value at the finest time granularity. If the carbon emission factor prediction at the finest time granularity is known to be Where N is the total number of prediction points at the finest granularity, the total prediction at all time granularities can be expressed as T stands for transpose.

[0051] In some embodiments, the corresponding input feature space can be determined for the prediction model of the carbon emission factor at k time granularities according to the selected time granularity and the duration to be predicted. and output space

[0052] Therefore, the prediction model of carbon emission factors at different time granularities can be expressed as: [i] =f i (x [i] θ [i] )+∈ i ,i=1,2,…,k,

[0053] in, is the input and output data of the prediction model at the i-th time granularity; f i is the i-th parameterized neural network regression model, whose trainable parameter is θ [i] ∈ i is the error term.

[0054] In some embodiments, if f i Adjust its parameter θ [i] Perform fitting and finally obtain The predicted value of the carbon emission factor is:

[0055] Step S12: establishing a time hierarchy matrix according to the time granularity and the time period to be predicted, and fusing multiple first prediction models according to the time hierarchy matrix to obtain an initial first coordinated prediction model.

[0056] In some embodiments, the multiple first prediction models are fused according to the time hierarchy matrix to obtain an initial first coordinated prediction model, including: constructing an initial coordination matrix according to the number of carbon emission factor prediction values ​​at the finest time granularity, and obtaining an output matrix composed of the outputs of multiple first prediction models, and constructing the initial first coordinated prediction model in the order of point product of the time hierarchy matrix, the initial coordination matrix and the output matrix.

[0057] It is worth noting that the first coordinated prediction model is an initial coordinated prediction model established based on the time hierarchy matrix, multiple prediction models and the coordination matrix. The initial coordination matrix and the optimal coordination matrix are the initial coordination matrix and the optimal coordination matrix after optimization, respectively.

[0058] In some embodiments, see FIG2 , which is a schematic diagram of a coordinated prediction model provided by this embodiment. In FIG2 , a prediction model is established for each time granularity, and a time hierarchy matrix is ​​established based on the time granularity and the time period to be predicted. The coordinated prediction model is established based on the prediction model, the coordination matrix, and the time hierarchy matrix. When the parameters of the coordinated prediction model are updated, the parameters of the prediction model and the coordination matrix are essentially updated.

[0059] In some embodiments, the prediction results of k prediction models are aggregated to obtain an output matrix composed of carbon emission factors output by multiple prediction models: is the prediction result of the carbon emission factor of the kth prediction model, and T is the transpose. Then, combined with the output matrix The time hierarchy matrix S and the coordination matrix are used to construct the coordination prediction model, which can be expressed as:

[0060] Where G=(G ij ) m×n is a coordination matrix, if G=G0=(0 I m×m ), it means using only the most fine-grained prediction results Aggregating layer by layer until the most coarse-grained prediction is the simplest initial coordination matrix. G can be understood as a hyperparameter to adjust the output matrix of multiple prediction models. The final coordinated total prediction result is

[0061] In some embodiments, an initial coordination matrix is ​​constructed based on the number of predicted values ​​of carbon emission factors at the finest time granularity, including: constructing a unit matrix with the number of predicted values ​​of carbon emission factors at the finest time granularity, and constructing a zero matrix with the number of predicted values ​​of carbon emission factors as rows and the difference between the total number of predicted values ​​of carbon emission factors at all time granularities and the number of predicted values ​​of carbon emission factors as columns, and the block matrix composed of the zero matrix and the unit matrix is ​​the initial coordination matrix.

[0062] Step S13: Update the parameters of the first coordinated prediction model according to the first carbon emission data to obtain a trained second coordinated prediction model, so as to predict the carbon emission factor through the second coordinated prediction model.

[0063] In some embodiments, the parameters of the first coordination prediction model are updated according to the first carbon emission data, including: dividing the second carbon emission data of multiple first prediction models into training sets and validation sets in proportion in turn; the first carbon emission data include: the second carbon emission data corresponding to each first prediction model; according to the training set, the validation set and a pre-constructed two-layer optimization model, the initial parameters and the initial coordination matrix of the first coordination prediction model are updated; wherein, the two-layer optimization model includes: a lower-layer optimization model and an upper-layer optimization model; the lower-layer optimization model optimizes the initial parameters according to the training set to obtain the optimal parameters; the upper-layer optimization model optimizes the initial coordination matrix according to the validation set and the optimal parameters.

[0064] It is worth noting that the second carbon emission data is the carbon emission data used by each prediction model. According to different time granularities, the time scales between the second carbon emission data are different. The second coordinated prediction model is a coordinated prediction model after the initial first coordinated prediction model is trained.

[0065] In some embodiments, the second carbon emission data of multiple first prediction models are divided into training sets and validation sets in proportion in sequence, including: screening the features of the second carbon emission data of multiple first prediction models in sequence to obtain feature screening results, dividing the feature screening results into training sets, validation sets and test sets in proportion, and normalizing the training sets, validation sets and test sets by columns to obtain normalized training sets, normalized validation sets and normalized test sets, respectively, so that finally training, validation and testing are performed according to the normalized training sets, normalized validation sets and normalized test sets, respectively; wherein each first prediction model corresponds to a normalized training set, a normalized validation set and a normalized test set.

[0066] In some embodiments, according to the input space and the output space The total data set at the i-th time granularity can be constructed from the carbon emission factor data set Then D i According to the order of time, the training set is divided into 8:2:2 ratios. Validation set and test set Then, the input data needs to be normalized. The normalization method is:

[0067] Among them, x min is the minimum value on the column vector, x max is the maximum value on the column vector, and x are the elements in the column vector before and after normalization, respectively.

[0068] In some embodiments, the above normalization method is applied to each column of variables in the global training set, and the maximum and minimum values ​​of each column in the training set are memorized. The same normalization method is then applied to the validation set and the test set. Thus, a global dataset after data preprocessing is obtained.

[0069] In some embodiments, the training set loss function is defined separately and validation set loss function Establish a two-level optimization model:

[0070] in It is expressed as the number of training set samples of the i-th time granularity. When the coordination matrix G is given, the corresponding optimal prediction model parameters It can be expressed as:

[0071] In some embodiments, the coordination matrix G can be interpreted as a hyperparameter, so it can be calculated on the validation set according to the validation set loss function. Adjustment:

[0072] in, It is expressed as the number of validation set samples at the i-th time granularity.

[0073] It is worth noting that since the carbon emission data are divided into k sets according to the time granularity, a prediction model is established for each carbon emission data set. That is to say, one time granularity corresponds to one prediction model, and the i-th time granularity refers to the i-th prediction model.

[0074] In some embodiments, the two-level optimization model is represented as:

[0075] in, is the lower-level optimization model, It is represented by the number of training set samples of the first prediction model i, i = 1, 2, ..., k; y j,1 is the true load of the jth training sample; G is the initial coordination matrix, is the initial parameter set of the k first prediction models; is the prediction result of the training set input to the first coordinated prediction model; is the prediction result of the test set to the first coordinated prediction model; It is represented by the number of validation set samples of the first prediction model i; j,2 is the true load of the j-th validation sample.

[0076] It is worth noting that the lower-level optimization aims to give the optimal carbon emission factor prediction model parameter set for a given coordination matrix G based on the training set loss. The upper-level optimization aims to adjust the coordination matrix G according to the validation set loss, thereby jointly optimizing the prediction model parameters and the coordination matrix.

[0077] In some embodiments, testing is performed based on a normalized test set, including: inputting multiple normalized test sets into the corresponding optimized second prediction model, outputting corresponding prediction results based on the optimal parameters obtained after optimization, and inputting multiple prediction results into the second prediction model, outputting the total prediction result based on the optimal coordination matrix obtained after optimization; and evaluating the prediction accuracy of each second prediction model based on the actual load of each normalized test set and the total prediction result.

[0078] In some embodiments, after obtaining the optimal coordination matrix And the optimal carbon emission factor prediction model parameters at each time granularity After that, test the dataset at each time granularity Carry out carbon emission factor prediction and coordinate to obtain the coordinated total prediction result It can be decomposed into prediction results at various time granularities according to the time hierarchy structure

[0079] In some embodiments, the performing of prediction accuracy evaluation on each second prediction model includes: using the mean absolute percentage error as an evaluation index of the prediction accuracy of the second prediction model, and the evaluation index is expressed as:

[0080] Among them, j is the test sample number, t is the time point number of the period to be predicted, is the prediction result of the normalized test set, is the actual load corresponding to the test sample, is the number of test set samples of the i-th second prediction model, and T is the total number of time points in the period to be predicted.

[0081] It is worth noting that the smaller the MAPE value, the higher the prediction accuracy of the carbon emission factor. Therefore, the prediction results of the carbon emission factor at k time granularities can be evaluated.

[0082] The carbon emission factor prediction method provided in this real-time example, which coordinates data of multiple time granularities, can efficiently utilize data of different time granularities to build a coordinated and high-performance carbon emission factor prediction model, thereby providing a basis for insightful analysis of the changing trends of carbon emission factors at different time scales.

[0083] Example 2

[0084] 3 , which is a flow chart of a complete carbon emission factor prediction method based on coordinated multi-time granularity data provided by this embodiment, including steps S21 to S26 , specifically:

[0085] Step S21: Collect the carbon emission factors of the area to be predicted, determine k different time granularities to be predicted, and establish a time hierarchy matrix.

[0086] Step S22: Combine expert opinions and select input features for each time granularity to establish carbon emission factor prediction models f1,…,f k .

[0087] Step S23: Construct a coordinated prediction model G based on the prediction models and time hierarchy matrices at different time granularities.

[0088] Step S24: divide and normalize the data set to obtain a training set, a validation set, and a test set after data preprocessing.

[0089] Step S25: Input the training set and validation set, iteratively update the prediction model and coordination matrix, and obtain the fitted multi-time granularity carbon emission factor prediction model. This specifically includes two sub-steps: sub-step S251 and sub-step S252.

[0090] Sub-step S251, define the training set loss function respectively and validation set loss function Establish a two-level optimization model:

[0091] Sub-step S252: To train this model, the gradient descent method is used to iteratively update the model parameters θ and the coordination matrix G to obtain the optimal coordination matrix And the optimal carbon emission factor prediction model parameters at each time granularity

[0092] Step S26: Input the test set, calculate the prediction accuracy of the carbon emission factor prediction at different time granularities, and evaluate the prediction effect. Use the Mean Absolute Percentage Error (MAPE) as an indicator to evaluate the prediction accuracy of the i-th time granularity, and calculate the prediction results. With real load Substitute the following formula for calculation:

[0093] The smaller the MAPE value, the higher the prediction accuracy of the carbon emission factor. Therefore, the prediction results of the carbon emission factor at k time granularities can be evaluated.

[0094] This embodiment adopts the method of establishing a prediction model for different time granularities respectively, which can improve the utilization rate of multivariate time series data, and establish a time hierarchy matrix according to the time granularity and the period to be predicted. The time hierarchy matrix is ​​used to fuse multiple prediction models, which can aggregate the prediction results of different granularities, strengthen the implicit influence of the coupling relationship of the time hierarchy on the prediction of carbon emission factors, and construct a self-consistent and coordinated carbon emission factor coordination prediction model, which can form consistent multi-time hierarchy prediction results, thereby improving the prediction accuracy of carbon emission factors, and providing a basis for insight analysis of the changing trend of carbon emission factors at different time scales.

[0095] Example 3

[0096] 4 , which is a schematic structural diagram of a carbon emission factor prediction system with coordinated data of multiple time granularities provided in this embodiment, including: a prediction model establishment module 31 , a coordinated prediction model establishment module 32 , and a training module 33 .

[0097] It is worth noting that the prediction model establishment module 31 mainly divides the carbon emission data according to the time granularity, and establishes a prediction model for each time granularity, and transmits the number of time granularities and multiple prediction models to the coordinated prediction model establishment module 32; after the coordinated prediction model establishment module 32 receives the time granularity and multiple prediction models, it establishes a time hierarchy matrix in combination with the time period to be predicted, and fuses multiple prediction models according to the time hierarchy matrix to establish a coordinated prediction model, and transmits the coordinated prediction model to the training module 33; the training module 33 trains the received coordinated prediction model, and uses the trained coordinated prediction model to coordinate the prediction of the carbon emission factor.

[0098] A prediction model module 31 is established to collect first carbon emission data of the area to be predicted, and to establish multiple first prediction models for initial prediction of carbon emission factors according to the time granularity of the first carbon emission data; wherein each time granularity corresponds to a first prediction model.

[0099] In some embodiments, a time hierarchy matrix is ​​established based on the time granularity and the period to be predicted, including: if the time granularity includes: year, quarter and month, then the time granularity k=3; if the period to be predicted is the next 12 months, then the time hierarchy matrix is ​​expressed as:

[0100] in, 14=(1,…,1) 1×4 ;S M =I 12×12 , I is the identity matrix.

[0101] The coordinated prediction model establishment module 32 is used to establish a time hierarchy matrix according to the time granularity and the time period to be predicted, and to fuse multiple first prediction models according to the time hierarchy matrix to obtain an initial first coordinated prediction model.

[0102] In some embodiments, the multiple first prediction models are fused according to the time hierarchy matrix to obtain an initial first coordinated prediction model, including: constructing an initial coordination matrix according to the number of carbon emission factor prediction values ​​at the finest time granularity, and obtaining an output matrix composed of the outputs of multiple first prediction models, and constructing the initial first coordinated prediction model in the order of point product of the time hierarchy matrix, the initial coordination matrix and the output matrix.

[0103] In some embodiments, an initial coordination matrix is ​​constructed based on the number of predicted values ​​of carbon emission factors at the finest time granularity, including: constructing a unit matrix with the number of predicted values ​​of carbon emission factors at the finest time granularity, and constructing a zero matrix with the number of predicted values ​​of carbon emission factors as rows and the difference between the total number of predicted values ​​of carbon emission factors at all time granularities and the number of predicted values ​​of carbon emission factors as columns, and the block matrix composed of the zero matrix and the unit matrix is ​​the initial coordination matrix.

[0104] The training module 33 is configured to update the parameters of the first coordinated prediction model according to the first carbon emission data to obtain a trained second coordinated prediction model, so as to predict the carbon emission factor through the second coordinated prediction model.

[0105] In some embodiments, the parameters of the first coordination prediction model are updated according to the first carbon emission data, including: dividing the second carbon emission data of multiple first prediction models into training sets and validation sets in proportion in turn; the first carbon emission data include: the second carbon emission data corresponding to each first prediction model; according to the training set, the validation set and a pre-constructed two-layer optimization model, the initial parameters and the initial coordination matrix of the first coordination prediction model are updated; wherein, the two-layer optimization model includes: a lower-layer optimization model and an upper-layer optimization model; the lower-layer optimization model optimizes the initial parameters according to the training set to obtain the optimal parameters; the upper-layer optimization model optimizes the initial coordination matrix according to the validation set and the optimal parameters.

[0106] In some embodiments, the two-level optimization model is represented as:

[0107] in, is the lower-level optimization model, Represented as the number of training set samples of the i-th first prediction model, i∈[1,k]; y j,1 is the true load of the jth training sample; G is the initial coordination matrix, is the initial parameter set of the k first prediction models; is the prediction result of the training set input to the first coordinated prediction model; is the prediction result of the test set to the first coordinated prediction model; It is represented by the number of validation set samples of the first prediction model i; j,2 is the true load of the j-th validation sample.

[0108] In some embodiments, the second carbon emission data of multiple first prediction models are divided into training sets and validation sets in proportion in sequence, including: screening the features of the second carbon emission data of multiple first prediction models in sequence to obtain feature screening results, dividing the feature screening results into training sets, validation sets and test sets in proportion, and normalizing the training sets, validation sets and test sets by columns to obtain normalized training sets, normalized validation sets and normalized test sets, respectively, so that finally training, validation and testing are performed according to the normalized training sets, normalized validation sets and normalized test sets, respectively; wherein each first prediction model corresponds to a normalized training set, a normalized validation set and a normalized test set.

[0109] In some embodiments, testing is performed based on a normalized test set, including: inputting multiple normalized test sets into the corresponding optimized second prediction model, outputting corresponding prediction results based on the optimal parameters obtained after optimization, and inputting multiple prediction results into the second prediction model, outputting the total prediction result based on the optimal coordination matrix obtained after optimization; and evaluating the prediction accuracy of each second prediction model based on the actual load of each normalized test set and the total prediction result.

[0110] In some embodiments, the performing of prediction accuracy evaluation on each second prediction model includes: using the mean absolute percentage error as an evaluation index of the prediction accuracy of the second prediction model, and the evaluation index is expressed as:

[0111] Among them, j is the test sample number, t is the time point number of the period to be predicted, is the prediction result of the normalized test set, is the actual load corresponding to the test sample, is the number of test set samples of the i-th second prediction model, and T is the total number of time points in the period to be predicted.

[0112] This embodiment adopts the prediction model establishment module 31 to establish a prediction model for different time granularities respectively, which can improve the utilization rate of multivariate time series data, and adopts the coordination prediction model establishment module 32 to establish a time hierarchy matrix according to the time granularity and the time period to be predicted. The time hierarchy matrix is ​​used to fuse multiple prediction models, which can aggregate the prediction results of different fine-grained levels, strengthen the implicit influence of the coupling relationship of the time hierarchy on the prediction of the carbon emission factor, and construct a self-consistent and coordinated carbon emission factor coordination prediction model. Through training by the training module 33, it can form the optimal and consistent multi-time hierarchy prediction results, thereby improving the prediction accuracy of the carbon emission factor, and providing a basis for insight analysis of the changing trend of the carbon emission factor at different time scales.

[0113] Those skilled in the art will appreciate that the embodiments of the present application may also provide computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0114] The present application is described with reference to the flow chart and / or block diagram of the method, device (system), and computer program product according to the embodiment of the present application. It should be understood that each flow process and / or box in the flow chart and / or block diagram and the combination of the flow process and / or box in the flow chart and / or block diagram can be realized by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processing machine or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for realizing the function specified in one flow chart flow or multiple flows and / or one box or multiple boxes of the block diagram.

[0115] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction device that implements the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0116] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0117] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A carbon emission factor prediction method based on coordinated multi-time granularity data, characterized in that: include: Collecting first carbon emission data of the area to be predicted, and establishing multiple first prediction models for initially predicting carbon emission factors based on the first carbon emission data at a time granularity; wherein each time granularity corresponds to a first prediction model; Establishing a time hierarchy matrix according to the time granularity and the time period to be predicted, and fusing multiple first prediction models according to the time hierarchy matrix to obtain an initial first coordinated prediction model; The parameters of the first coordinated prediction model are updated according to the first carbon emission data to obtain a trained second coordinated prediction model, so that the carbon emission factor is predicted by the second coordinated prediction model.

2. The carbon emission factor prediction method based on multi-time granularity data coordination according to claim 1 is characterized in that: The fusing of the multiple first prediction models according to the time hierarchy matrix to obtain an initial first coordinated prediction model includes: According to the number of predicted values ​​of carbon emission factors at the finest time granularity, an initial coordination matrix is ​​constructed, and an output matrix composed of the outputs of multiple first prediction models is obtained. The initial first coordination prediction model is constructed in the order of point product of the time hierarchy matrix, the initial coordination matrix and the output matrix.

3. The carbon emission factor prediction method based on multi-time granularity data coordination as claimed in claim 2, characterized in that: The initial coordination matrix is ​​constructed based on the number of predicted values ​​of carbon emission factors at the finest time granularity, including: A unit matrix is ​​constructed with the number of predicted values ​​of carbon emission factors at the finest time granularity, and a zero matrix is ​​constructed with the number of predicted values ​​of carbon emission factors as rows and the difference between the total predicted values ​​of carbon emission factors at all time granularities and the number of predicted values ​​of carbon emission factors as columns. The block matrix composed of the zero matrix and the unit matrix is ​​used as the initial coordination matrix.

4. The carbon emission factor prediction method based on multi-time granularity data coordination according to claim 1, characterized in that: The step of establishing a time hierarchy matrix according to the time granularity and the time period to be predicted includes: If the time granularity includes: year, quarter and month, then the time granularity is 3. If the period to be predicted is the next 12 months, then the time hierarchy matrix is ​​expressed as: in, 14=(1,…,1) 1×4 ;S M =I 12×12 , I is the identity matrix.

5. The carbon emission factor prediction method based on multi-time granularity data coordination according to claim 1, characterized in that: The updating of parameters of the first coordinated prediction model according to the first carbon emission data includes: Sequentially dividing the second carbon emission data of the plurality of first prediction models into a training set and a validation set in proportion; the first carbon emission data includes: the second carbon emission data corresponding to each first prediction model; The initial parameters and initial coordination matrix of the first coordination prediction model are updated according to the training set, the validation set and a pre-constructed two-layer optimization model; wherein the two-layer optimization model includes: a lower-layer optimization model and an upper-layer optimization model; the lower-layer optimization model optimizes the initial parameters according to the training set to obtain the optimal parameters; the upper-layer optimization model optimizes the initial coordination matrix according to the validation set and the optimal parameters.

6. The carbon emission factor prediction method based on multi-time granularity data coordination according to claim 5 is characterized in that: The two-level optimization model is expressed as: in, is the lower-level optimization model, It is represented by the number of training set samples of the first prediction model i, i = 1, 2, ..., k; y j,1 is the jth The actual load of the training sample; G is the initial coordination matrix, is the initial parameter set of the k first prediction models; is the prediction result of the training set input to the first coordinated prediction model; is the prediction result of the test set to the first coordinated prediction model; It is represented by the number of validation set samples of the first prediction model i; j,2 is the true load of the j-th validation sample.

7. The carbon emission factor prediction method based on multi-time granularity data coordination according to claim 5, characterized in that: The second carbon emission data of the plurality of first prediction models are divided into a training set and a validation set in proportion in sequence, including: The features of the second carbon emission data of multiple first prediction models are screened in turn to obtain feature screening results, the feature screening results are divided into a training set, a validation set and a test set in proportion, and the training set, the validation set and the test set are normalized by column to obtain a normalized training set, a normalized validation set and a normalized test set, so that training, validation and testing are finally performed according to the normalized training set, the normalized validation set and the normalized test set, respectively; wherein each first prediction model corresponds to a normalized training set, a normalized validation set and a normalized test set.

8. The carbon emission factor prediction method based on multi-time granularity data coordination according to claim 7 is characterized in that: Tested on a normalized test set, including: Inputting multiple normalized test sets into the corresponding optimized second prediction model respectively, outputting corresponding prediction results according to the optimal parameters obtained after optimization, and inputting multiple prediction results into the second prediction model respectively, outputting the total prediction result according to the optimal coordination matrix obtained after optimization; The prediction accuracy of each second prediction model is evaluated based on the actual load of each normalized test set and the total prediction result.

9. The carbon emission factor prediction method based on multi-time granularity data coordination according to claim 8, characterized in that: The step of evaluating the prediction accuracy of each second prediction model includes: The mean absolute percentage error is used as an evaluation index of the prediction accuracy of the second prediction model, and the evaluation index is expressed as: Among them, j is the test sample number, t is the time point number of the period to be predicted, is the prediction result of the normalized test set, is the actual load corresponding to the test sample, is the number of test set samples of the i-th second prediction model, and T is the total number of time points in the period to be predicted.

10. A carbon emission factor prediction system with coordinated multi-time granularity data, characterized in that: include: Establish a prediction model module, establish a coordinated prediction model module and a training module; among them, The prediction model establishment module is used to collect first carbon emission data of the area to be predicted, and establish multiple first prediction models for initially predicting carbon emission factors based on the first carbon emission data according to time granularity; wherein each time granularity corresponds to a first prediction model; The coordinated prediction model establishment module is used to establish a time hierarchy matrix according to the time granularity and the time period to be predicted, and to fuse multiple first prediction models according to the time hierarchy matrix to obtain an initial first coordinated prediction model; The training module is used to update the parameters of the first coordinated prediction model according to the first carbon emission data to obtain a trained second coordinated prediction model, so as to predict the carbon emission factor through the second coordinated prediction model.

Images