Carbon emission prediction method and device based on COA-Elman model

Abstract

The invention relates to the field of intelligent building management, and particularly discloses a carbon emission prediction method and device based on a COA-Elman model. Comprising the steps of obtaining annual historical data of carbon emission related factors of a target building, constructing a carbon emission influence factor index system according to the annual historical data, performing influence factor correlation degree screening based on the carbon emission influence factor index system by using grey correlation analysis and a maximum information coefficient, determining key influence factors, and determining the carbon emission related factors of the target building according to the key influence factors. And based on the key influence factors and the annual historical data, performing optimization training on key parameters in an ELMAN neural network model by using a raccoon algorithm to obtain a carbon emission training model, and based on the key influence factors, the annual historical data and the carbon emission training model, performing carbon emission prediction to obtain a carbon emission prediction result. The objective of the invention is to improve global optimization capability and prediction precision of a network.

Description

Technical Field

[0001] The present invention relates to the field of intelligent building management, and in particular to a carbon emission prediction method and device based on a COA-Elman model. Background Art

[0002] As the global climate change problem becomes increasingly serious, the accurate prediction and control of carbon emissions plays a vital role in achieving the global carbon neutrality goal. Especially with the accelerated urbanization process today, carbon emissions in the construction sector have increased significantly. How to effectively predict and control carbon emissions from buildings has become a key issue in the construction of low-carbon cities. As an important energy consumption unit in the city, the carbon emissions of buildings account for a large part of the total emissions of the city. Therefore, conducting research on the prediction of building carbon emissions and then formulating effective carbon emission management strategies are of great practical significance for promoting the sustainable development of cities.

[0003] At present, the prediction methods of building carbon emissions mainly include traditional models based on statistics and machine learning models based on artificial intelligence. These models play an important role in the prediction of building carbon emissions, but there are also some significant deficiencies and shortcomings in practical applications, including insufficient prediction accuracy: due to its reliance on linear relationships, traditional statistical models have low prediction accuracy when facing complex building energy consumption data, and it is difficult to accurately capture the dynamic change characteristics of carbon emissions. Even advanced machine learning models face certain prediction errors when processing nonlinear and dynamic energy consumption data, and it is difficult to fully meet the needs of practical applications. Local optimal problem: traditional BP neural networks and some machine learning models are prone to fall into local optimal solutions during training and cannot effectively search for global optimal solutions. This leads to unstable prediction results of the model, inconsistent performance in different data sets or application scenarios, affecting the reliability and practicality of the prediction. High computational cost: with the continuous increase in the scale of building energy consumption data, the computational complexity of the model and the data processing cost have also increased significantly. When processing large-scale data, traditional machine learning models face the problems of high computing resource consumption and long training time, which limits their application in real-time prediction and dynamic management. Poor generalization ability: When machine learning models process diverse building data, the generalization ability of the model is poor. This means that when faced with new building types or data with different characteristics, the prediction effect of the model may drop significantly and cannot adapt to the diverse needs in actual applications. In addition, the model is highly dependent on parameters, requiring a lot of parameter tuning when applied, which further increases the complexity of model application. Summary of the invention

[0004] In view of the above problems, an object of the present invention is to provide a carbon emission prediction method based on the COA-Elman model.

[0005] The second object of the present invention is to provide a carbon emission prediction device based on the COA-Elman model.

[0006] The first technical solution adopted by the present invention is: S1, obtaining annual historical data of carbon emission related factors of the target building;

[0007] S2. Construct an indicator system of carbon emission influencing factors based on the annual historical data;

[0008] S3. Based on the carbon emission influencing factor index system, grey correlation analysis and maximum information coefficient are used to screen the correlation of influencing factors and determine the key influencing factors;

[0009] S4. Based on the key influencing factors and the annual historical data, the raccoon algorithm is used to optimize the key parameters in the ELMAN neural network model to obtain a carbon emission training model;

[0010] S5. Based on the key influencing factors, the annual historical data and the carbon emission training model, perform carbon emission forecasting to obtain a carbon emission forecasting result.

[0011] Optionally, the S3 includes:

[0012] S31, using grey correlation to perform correlation analysis on the carbon emission influencing factor index system to obtain the grey correlation degree of each influencing factor;

[0013] S32, based on the grey correlation degree, perform preliminary screening to determine actual relevant factors;

[0014] S33. Using the maximum information coefficient, perform a secondary screening on the actual relevant factors to obtain the key influencing factors.

[0015] Optionally, the S4 includes:

[0016] S41, taking the annual historical data as output variables and the carbon emission influencing factor index system as input variables, dividing them in chronological order to obtain a training set and a test set;

[0017] S42, normalizing the data in the training set and the test set to obtain a processed training set and a processed test set;

[0018] S43, embedding the raccoon algorithm into the ELMAN neural network model, and using the processed training set to train the embedded model to obtain a trained model;

[0019] S44, constructing an evaluation index system to evaluate the trained model, and determining the carbon emission training model according to the evaluation results.

[0020] Optionally, the evaluation index system of S44 includes:

[0021] The root mean square error is calculated as follows:

[0022]

[0023] The mean absolute error is calculated as follows:

[0024]

[0025] The mean absolute percentage error is calculated as,

[0026]

[0027] The absolute correlation coefficient is calculated as follows:

[0028]

[0029] In the formula, n is the number of samples, y is i , and are the average of the actual value, predicted value and actual value of the i-th sample respectively.

[0030] Optionally, the S43 includes:

[0031] S431, initializing the weights and thresholds of the ELMAN neural network model, and using them as the search space for raccoon individuals in the raccoon algorithm;

[0032] S432, corresponding to a set of Elman neural network parameters for each raccoon individual, and performing random initialization to determine the initial parameters;

[0033] S433, applying the initial parameters to the ELMAN neural network model to obtain input-output error values, and using the error values ​​to determine fitness;

[0034] S434, using the processed training set for training, after each round of iteration, the raccoon individual dynamically adjusts its position according to the current optimal position and the global optimal position, thereby updating the weights and thresholds of the Elman neural network model, and performing convergence condition determination at the same time, and outputting the optimal weights and thresholds when the convergence conditions are met.

[0035] The second technical solution adopted by the present invention is: a carbon emission prediction device based on the COA-Elman model, comprising: a data acquisition module for acquiring annual historical data of carbon emission related factors of a target building;

[0036] An indicator system construction module, used to construct an indicator system of carbon emission influencing factors based on the annual historical data;

[0037] A factor screening module is used to screen the correlation of influencing factors based on the carbon emission influencing factor index system and determine the key influencing factors by using grey correlation analysis and maximum information coefficient;

[0038] A model optimization module, used to optimize and train the key parameters in the ELMAN neural network model using the raccoon algorithm based on the key influencing factors and the annual historical data to obtain a carbon emission training model;

[0039] The prediction module is used to perform carbon emission prediction based on the key influencing factors, the annual historical data and the carbon emission training model to obtain a carbon emission prediction result.

[0040] Optionally, the factor screening module includes:

[0041] A grey correlation analysis module is used to perform a correlation analysis on the carbon emission influencing factor index system using grey correlation to obtain the grey correlation of each influencing factor;

[0042] A preliminary screening module, used for performing preliminary screening based on the grey correlation degree to determine actual relevant factors;

[0043] The secondary screening module is used to perform secondary screening on the actual relevant factors using the maximum information coefficient to obtain the key influencing factors.

[0044] Optionally, the model optimization module includes:

[0045] A data partitioning module is used to use the annual historical data as output variables and the carbon emission influencing factor indicator system as input variables, and to partition them in chronological order to obtain a training set and a test set;

[0046] A normalization processing module, used to perform normalization processing on the data in the training set and the test set to obtain a processed training set and a processed test set;

[0047] A training module, used for embedding the raccoon algorithm into the ELMAN neural network model, and training the embedded model using the processed training set to obtain a trained model;

[0048] The evaluation module is used to construct an evaluation index system to evaluate the trained model and determine the carbon emission training model according to the evaluation results.

[0049] Optionally, the evaluation index system in the evaluation module includes:

[0050] The root mean square error is calculated as follows:

[0051]

[0052] The mean absolute error is calculated as follows:

[0053]

[0054] The mean absolute percentage error is calculated as,

[0055]

[0056] The absolute correlation coefficient is calculated as follows:

[0057]

[0058] In the formula, n is the number of samples, y is i , and are the average of the actual value, predicted value and actual value of the i-th sample respectively.

[0059] Optionally, the training module includes:

[0060] An initialization module, used for initializing the weights and thresholds of the ELMAN neural network model and using it as a search space for raccoon individuals in the raccoon algorithm;

[0061] A parameter determination module, used to correspond a set of Elman neural network parameters to each raccoon individual, and to perform random initialization to determine the initial parameters;

[0062] An adaptation calculation module, used for applying the initial parameters to the ELMAN neural network model to obtain input-output error values, and using the error values ​​to perform fitness determination;

[0063] The dynamic adjustment module is used to perform training using the processed training set. After each round of iteration, the raccoon individual dynamically adjusts its position according to the current optimal position and the global optimal position, thereby updating the weights and thresholds of the Elman neural network model, and performing convergence condition judgment at the same time. When the convergence conditions are met, the optimal weights and thresholds are output.

[0064] Beneficial effects of the above technical solution:

[0065] This paper proposes a quantitative method for comprehensively measuring rural low-carbonization based on a relatively complete rural carbon emission inventory. A coupling coordination model based on the combined weighted-improved matter-element extension evaluation method is proposed, and a spatiotemporal analysis framework is established. Based on the GTWR model and the random effect panel Tobit model, the action mechanism of the internal subsystem and the external influencing factors of the coupling coordination are analyzed. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 A flow chart of a carbon emission prediction method based on the COA-Elman model provided by the present invention;

[0067] Figure 2 A schematic diagram of the structure of a carbon emission prediction device based on the COA-Elman model provided by the present invention;

[0068] Figure 3 This is a schematic diagram of the Elman neural network structure;

[0069] Figure 4 This is a schematic diagram of the classification structure of building carbon emission evaluation indicators;

[0070] Figure 5 This is a comparison chart of the predicted building carbon emissions from different models;

[0071] Figure 6 This is a comparison chart of relative errors in building carbon emission predictions using different models. DETAILED DESCRIPTION

[0072] The following detailed description of the embodiments of the present invention is provided in conjunction with the accompanying drawings and examples. The following detailed description of the embodiments and the accompanying drawings are used to illustrate the principles of the present invention, but cannot be used to limit the scope of the present invention, that is, the present invention is not limited to the preferred embodiments described, and the scope of the present invention is defined by the claims.

[0073] In the description of the present invention, it should be noted that, unless otherwise specified, “plurality” means two or more than two; the terms “first”, “second”, etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance; for ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances.

[0074] Embodiment 1

[0075] An embodiment of the present invention provides a carbon emission prediction method based on the COA-Elman model, such as Figure 1 As shown,

[0076] It should be noted that the Elman neural network is a typical recursive neural network. It has a special receiving layer that can delay the output of the hidden layer and feed it back to the hidden layer. This structure gives the Elman neural network the ability of short-term memory, which makes it have significant advantages in processing time series data.

[0077] The main components of the Elman neural network include the input layer, hidden layer, receiving layer and output layer, such as Figure 3The network obtains external data through the input layer, processes the information through the hidden layer, and the receiving layer stores the output of the hidden layer and adds calculations to the next input, and finally obtains the output result of the network through the output layer.

[0078] The operating mechanism of the Elman neural network is as follows:

[0079] Input layer: receives time series data. The input vector is usually represented by u(k).

[0080] Hidden layer: The hidden layer performs nonlinear mapping, and the output of the hidden layer is determined by the weight matrix, input data, and feedback from the receiving layer.

[0081] Continuation layer: Saves the output of the hidden layer at the previous moment and uses it as one of the inputs of the hidden layer at the current moment.

[0082] Output layer: According to the output of the hidden layer, the final network output result is obtained through weight mapping.

[0083] 1. Update equation of hidden layer output:

[0084] x(k)=f2(w 1 x c (k)+w 2 u(k-1))

[0085] Among them, w 1 and w 2 are the connection weights of the hidden layer and the input layer, respectively, c (k) is the output of the receiving layer.

[0086] 2. The output of the following layer is the output of the hidden layer at the previous moment.

[0087] 3. Update equation of the output layer:

[0088] y(k)=f1(w 3 x(k)

[0089] Among them, w 3 is the connection weight from the hidden layer to the output layer, and f1 is the activation function of the output layer.

[0090] 4. Activation Function

[0091] The activation functions of the Elman neural network often use the sigmoid function and the hyperbolic tangent function (tanh), which are:

[0092]

[0093] 5. Error Function

[0094] The error function of the network is used to measure the gap between the network output and the expected output. The commonly used error function is the mean square error:

[0095]

[0096] Among them, y i (k) is the actual output of the network, e i (k) is the expected output, and m is the total number of samples.

[0097] The Raccoon Algorithm (COA) is a swarm intelligence optimization algorithm based on the foraging behavior of raccoons. Raccoons search for food locations in the forest and use feedback from their peers to find the best food location. COA imitates this group foraging behavior, allowing individuals to jump out of the local optimum and gradually approach the global optimal solution.

[0098] Phase 1: Hunting and attack strategies against iguanas (exploration phase)

[0099] The population initialization method of COA is the same as that of most intelligent optimization algorithms, that is, randomly generating several candidate solutions in the search space as initial individuals. For details, please refer to the later push, and I will not go into details here.

[0100] In the search space, the first phase of updating the coati population is modeled based on a strategy that simulates their attack on an iguana. In this strategy, a group of coatis climb a tree to approach an iguana and expel it. Other coatis wait under the tree and attack the iguana directly on the ground. Once the iguana lands, the coatis attack it and eliminate it. This strategy allows COA to improve its global search capability in the search space, i.e., exploration.

[0101] In the COA design, the position x of the best individual in the current population is best Assume the position of the iguana. Assume that half of the coatis will climb the tree and the other half will wait for the iguanas to fall to the ground. Therefore, the following formula is used to mathematically simulate the position of the coatis in the tree:

[0102]

[0103] After the iguana lands, it is placed at a random position in the search space. Based on this random position, the coati on the ground moves in the search space and updates its position using the following formula:

[0104]

[0105] Where N is the population size, t is the current iteration number, lbj represents the lower bound of the □-th dimension variable, ubj represents the upper bound of the □-th dimension variable, fitness() is the calculated fitness, r represents a random number between [0, 1], and I represents a random integer from the set {1, 2}. Therefore, The above formula describes the behavior of the coatis when attacking the iguana, that is, half of the coatis climb the tree, and the other half wait for the iguana to land and use the formula to update their position.

[0106] If the updated position calculated by a coati improves the value of the objective function, the new position is acceptable, otherwise the coati remains at the previous position. In other words, a greedy selection is performed, as shown in the following formula:

[0107]

[0108] Phase 2: The process of escaping from predators (development phase)

[0109] The second phase of updating the coati's position in the search space is mathematically modeled based on the coati's natural behavior when fleeing from a predator. When a predator attacks a coati, the coati flees its position. The coati's actions in this strategy put it in a safe position close to its current position, which is the ability of COA to exploit in local search. To simulate this behavior, a random position is generated near each coati's position, as shown in the following formula:

[0110]

[0111] Among them, r represents a random number between [0, 1], and T is the maximum number of iterations. It can be seen that are the upper and lower bounds of the j-th dimension variable as the number of iterations is updated. If the newly calculated position improves the value of the objective function, the position is acceptable and is simulated using , that is, another greedy selection is performed.

[0112] S1. Obtain annual historical data on carbon emission-related factors of target buildings;

[0113] For example, collect carbon emission-related data for a building in a certain district of a province, covering carbon emission-related factor data for 12 months in 2022-2023.

[0114] S2. Construct an indicator system of carbon emission influencing factors based on the annual historical data;

[0115] According to the existing research on the design of carbon emission monitoring system in the building operation stage, multiple factors such as electricity consumption, water resource use, gas consumption, waste disposal and other environmental indicators are selected as building carbon emission evaluation indicators. The specific corresponding structure is as follows: Figure 4 As shown. Based on the various indicators of the carbon emission monitoring system in the building operation phase listed in the figure, an indicator system of factors that may affect emissions is constructed. These factors include electricity consumption (total electricity, lighting, air conditioning, power, etc.), water resource consumption (domestic water, public water, sewage discharge, etc.), gas usage, waste treatment volume and other indicators (such as dry bulb temperature, inlet and outlet cooling tower temperature, etc.).

[0116] S3. Based on the carbon emission influencing factor index system, grey correlation analysis and maximum information coefficient are used to screen the correlation of influencing factors and determine the key influencing factors;

[0117] The GRA (grey relational analysis) and MIC (maximum information coefficient) methods were used to screen out the key factors with the strongest correlation with building carbon emissions, and the factors with less correlation were eliminated. Secondly, the COA (raccoon algorithm) was used to optimize the key parameters in the ELMAN neural network model to improve the global search ability and prediction accuracy of the model.

[0118] For example, in order to deeply understand the key factors affecting carbon emissions of a building and accurately predict future carbon emissions, this paper uses GRA (grey relational analysis) and MIC (maximum information coefficient) methods to screen out factors with a high correlation with building carbon emissions.

[0119] First, GRA was used to conduct the initial screening of influencing factors. The analysis results and rankings are shown in Table 1. The cooling water pump of the air-conditioning system ranked first with a correlation of 0.92, and the lighting and hot water systems ranked second and third with 0.91 and 0.90 respectively. The correlation between waste treatment and building carbon emissions was the smallest, at 0.68. Since the grey correlation of the above influencing factors was above 0.5, it showed that there was a certain degree of correlation between them and the building carbon emissions.

[0120] In order to further analyze the strength of the nonlinear relationship between the above factors and building carbon emissions, the MIC method was selected for secondary screening. The MIC coefficients between each influencing factor and building carbon emissions are shown in Table 1. The results show that the MIC coefficients between 10 factors such as the cooling water pump, lighting, and hot water system of the air-conditioning system and the building carbon emissions are all 1. At the same time, combined with the results of gray correlation analysis, the inlet and outlet cooling tower temperatures, water pump inlet and outlet pressures, dry bulb temperature, etc. are relatively less correlated with building carbon emissions. In order to improve the prediction accuracy of the constructed model, these five influencing factors are eliminated, and the remaining 10 are the key influencing factors of building carbon emissions.

[0121] Table 1 Grey correlation analysis and MIC coefficient of building carbon emissions

[0122]

[0123] Specifically, S3 includes the following steps S31-S33:

[0124] S31, using grey correlation to perform correlation analysis on the carbon emission influencing factor index system to obtain the grey correlation degree of each influencing factor;

[0125] S32, based on the grey correlation degree, perform preliminary screening to determine actual relevant factors;

[0126] S33. Using the maximum information coefficient, perform a secondary screening on the actual relevant factors to obtain the key influencing factors.

[0127] S4. Based on the key influencing factors and the annual historical data, the raccoon algorithm is used to optimize the key parameters in the ELMAN neural network model to obtain a carbon emission training model;

[0128] The building carbon emission data and key influencing factors are input into COA's optimized ELMAN neural network model to train and predict carbon emissions. The predicted value of the building carbon emissions can be obtained by adding up the predicted results of each key influencing factor.

[0129] Specifically, S4 includes the following steps S41-S44:

[0130] S41, taking the annual historical data as output variables and the carbon emission influencing factor index system as input variables, dividing them in chronological order to obtain a training set and a test set;

[0131] S42, normalizing the data in the training set and the test set to obtain a processed training set and a processed test set;

[0132] S43, embedding the raccoon algorithm into the ELMAN neural network model, and using the processed training set to train the embedded model to obtain a trained model;

[0133] Specifically, S43 includes the following steps S431-S434:

[0134] S431, initializing the weights and thresholds of the ELMAN neural network model, and using them as the search space for raccoon individuals in the raccoon algorithm;

[0135] S432, corresponding to a set of Elman neural network parameters for each raccoon individual, and performing random initialization to determine the initial parameters;

[0136] S433, applying the initial parameters to the ELMAN neural network model to obtain input-output error values, and using the error values ​​to determine fitness;

[0137] S434, using the processed training set for training, after each round of iteration, the raccoon individual dynamically adjusts its position according to the current optimal position and the global optimal position, thereby updating the weights and thresholds of the Elman neural network model, and performing convergence condition determination at the same time, and outputting the optimal weights and thresholds when the convergence conditions are met.

[0138] S44, constructing an evaluation index system to evaluate the trained model, and determining the carbon emission training model according to the evaluation results.

[0139] For example, the carbon emissions of a residential building in a certain district of Hebei Province in 12 months from 2022 to 2023 are used as output variables, and the influencing factors of building carbon emissions are used as input variables. The training set and the test set are divided in chronological order according to a ratio of 4:1. In order to eliminate the dimensional influence between data features, normalization is performed to return the data to the interval [0, 1]. The basic parameters of the ELMAN neural network model are set as shown in Table 2.

[0140] Compared with manual parameter adjustment, COA (colony optimization algorithm) can explore the solution space more efficiently and comprehensively, and find the optimal parameter configuration that is more suitable for the current problem. Therefore, the COA algorithm is used to optimize the key parameters in the ELMAN neural network model to improve the prediction performance of the model. The relevant parameters are shown in Table 2.

[0141] Table 2 Model related parameters and corresponding values

[0142]

[0143] In order to verify the prediction effect of the model, this paper selects root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and absolute correlation coefficient (R2) as evaluation indicators. The smaller the values ​​of RMSE, MAE, and MAPE, the better the model prediction performance, and the closer R2 is to 1, the better the model fitting effect.

[0144] The evaluation index system includes:

[0145] The root mean square error is calculated as follows:

[0146]

[0147] The mean absolute error is calculated as follows:

[0148]

[0149] The mean absolute percentage error is calculated as,

[0150]

[0151] The absolute correlation coefficient is calculated as follows:

[0152]

[0153] In the formula, n is the number of samples, y is i , and are the average of the actual value, predicted value and actual value of the i-th sample respectively.

[0154] S5. Based on the key influencing factors, the annual historical data and the carbon emission training model, perform carbon emission forecasting to obtain a carbon emission forecasting result.

[0155] The present invention is a carbon emission prediction method based on the COA-Elman model, including S1, obtaining annual historical data of carbon emission related factors of target buildings, constructing a carbon emission influencing factor index system based on the annual historical data, and screening the correlation of influencing factors based on the carbon emission influencing factor index system by using grey correlation analysis and maximum information coefficient to determine the key influencing factors. Based on the key influencing factors and the annual historical data, the raccoon algorithm is used to optimize the key parameters in the ELMAN neural network model to obtain a carbon emission training model, and based on the key influencing factors, the annual historical data and the carbon emission training model, carbon emission prediction is performed to obtain carbon emission prediction results. Based on a relatively complete rural carbon emission inventory, a quantitative method for comprehensively measuring rural low-carbonization is proposed. A coupling coordination degree model based on a combined weighted-improved matter-element extension evaluation method is proposed, and a spatiotemporal analysis framework is established. Based on the GTWR model and the random effect panel Tobit model, the action mechanism of the internal subsystem and the external influencing factors of the coupling coordination degree are analyzed.

[0156] Embodiment 2

[0157] An embodiment of the present invention provides a carbon emission prediction device 200 based on the COA-Elman model, such as Figure 2 As shown, it includes: a data acquisition module 201, which is used to obtain annual historical data of carbon emission related factors of the target building;

[0158] An indicator system construction module 202 is used to construct an indicator system of carbon emission influencing factors based on the annual historical data;

[0159] The factor screening module 203 is used to screen the correlation of influencing factors based on the carbon emission influencing factor index system by using grey correlation analysis and maximum information coefficient to determine the key influencing factors;

[0160] A model optimization module 204 is used to optimize and train the key parameters in the ELMAN neural network model using the raccoon algorithm based on the key influencing factors and the annual historical data to obtain a carbon emission training model;

[0161] The prediction module 205 is used to perform carbon emission prediction based on the key influencing factors, the annual historical data and the carbon emission training model to obtain a carbon emission prediction result.

[0162] Furthermore, the factor screening module 203 includes:

[0163] A grey correlation analysis module 2031 is used to perform a correlation analysis on the carbon emission influencing factor index system using grey correlation to obtain the grey correlation of each influencing factor;

[0164] A preliminary screening module 2032 is used to perform preliminary screening based on the grey correlation degree to determine actual relevant factors;

[0165] The secondary screening module 2033 is used to perform secondary screening on the actual relevant factors using the maximum information coefficient to obtain the key influencing factors.

[0166] Furthermore, the model optimization module 204 includes:

[0167] The data partitioning module 2041 is used to use the annual historical data as output variables and the carbon emission influencing factor index system as input variables, and partition them in chronological order to obtain a training set and a test set;

[0168] A normalization processing module 2042 is used to perform normalization processing on the data in the training set and the test set to obtain a processed training set and a processed test set;

[0169] A training module 2043 is used to embed the raccoon algorithm into the ELMAN neural network model, and train the embedded model using the processed training set to obtain a trained model;

[0170] The evaluation module 2044 is used to construct an evaluation index system to evaluate the trained model and determine the carbon emission training model according to the evaluation results.

[0171] Furthermore, the evaluation index system in the evaluation module 2044 includes:

[0172] The root mean square error is calculated as follows:

[0173]

[0174] The mean absolute error is calculated as follows:

[0175]

[0176] The mean absolute percentage error is calculated as,

[0177]

[0178] The absolute correlation coefficient is calculated as follows:

[0179]

[0180] In the formula, n is the number of samples, y is i , and are the average of the actual value, predicted value and actual value of the i-th sample respectively.

[0181] Further, the training module 2043 includes:

[0182] Initialization module 20431, used to initialize the weights and thresholds of the ELMAN neural network model and use it as the search space for raccoon individuals in the raccoon algorithm;

[0183] The parameter determination module 20432 is used to make each raccoon individual correspond to a set of Elman neural network parameters, and perform random initialization to determine the initial parameters;

[0184] The adaptation calculation module 20433 is used to apply the initial parameters to the ELMAN neural network model to obtain input-output error values, and use the error values ​​to determine the fitness;

[0185] The dynamic adjustment module 20434 is used to perform training using the processed training set. After each round of iteration, the raccoon individual dynamically adjusts its position according to the current optimal position and the global optimal position, thereby updating the weights and thresholds of the Elman neural network model, and performing convergence condition determination at the same time. When the convergence conditions are met, the optimal weights and thresholds are output.

[0186] The present invention proposes a carbon emission prediction device based on the COA-Elman model, including a data acquisition module for acquiring the annual historical data of carbon emission related factors of the target building; an index system construction module for constructing a carbon emission influencing factor index system based on the annual historical data; a factor screening module for screening the correlation of influencing factors based on the carbon emission influencing factor index system using grey correlation analysis and maximum information coefficient to determine the key influencing factors; a model optimization module for optimizing and training the key parameters in the ELMAN neural network model using the raccoon algorithm based on the key influencing factors and the annual historical data to obtain a carbon emission training model; a prediction module for predicting carbon emissions based on the key influencing factors, the annual historical data and the carbon emission training model to obtain carbon emission prediction results. A coupling coordination degree model based on the combined weighted-improved matter-element extension evaluation method is established, and a spatiotemporal analysis framework is established. Based on the GTWR model and the random effect panel Tobit model, the action mechanism of the internal subsystem and the external influencing factors of the coupling coordination degree are analyzed.

[0187] Embodiment 3

[0188] In order to reflect the superior performance of the MIC-COA-ELMAN model constructed in this paper, the MIC-WOA-ELMAN (whale optimization algorithm), MIC-SSA-ELMAN (sparrow optimization algorithm), COA-ELMAN, MIC-ELMAN and MIC-KELM models are selected for comparative analysis.

[0189] According to the above steps, the model parameters are set, the normalized preprocessed data is input and the model is trained. After the output data is inversely normalized, the predicted carbon emissions of buildings in a certain district of Hebei Province for 12 months from 2022 to 2023 are obtained, and the prediction results of each model are analyzed. Figure 5 , relative error comparison see Figure 6 .

[0190] Depend on Figure 5 It can be seen that the prediction effect of the MIC-COA-ELMAN model is the best, followed by the MIC-COA-ELMAN model. The prediction effects of the MIC-SSA-ELMAN and MIC-WOA-ELMAN models are relatively close, ranking third and fourth. The prediction effects of the MIC-ELMAN and MIC-BP models are relatively poor. Figure 6It can be seen that the MIC-COA-ELMAN model has the smallest relative error in the prediction results of most years, and the average relative error of this model is 1.74%, which is smaller than other models, which once again proves that its prediction effect is the best. In order to further compare the prediction performance of each model, this paper selects MAE, RMSE, MAPE and R2 as performance evaluation indicators, and the results are shown in Table 3.

[0191] Table 3 Model performance evaluation indicators

[0192]

[0193] From the results in Table 3, it can be seen that the prediction performance of the MIC-COA-ELMAN model is significantly better than that of other models. Specifically, first, compared with the non-MIC model, the MIC-COA-ELMAN model performs better in all evaluation indicators. The MAE, RMSE and MAPE indicators of the MIC-COA-ELMAN model are lower than those of the COA-ELMAN model, which are reduced by 1.95 tons, 2.78 tons and 2.09% respectively, indicating that the MIC method has a significant effect on improving the performance of the model. Compared with the MIC-WOA-ELMAN, MIC-SSA-ELMAN and MIC-ELMAN models, the MAE, RMSE and MAPE indicators are reduced by 2.63 tons, 3.77 tons and 2.90%, 3.30 tons, 4.77 tons and 4.13%, and 2.25 tons, 3.98 tons and 2.23% respectively. These results show that the model optimized by the COA algorithm is significantly better than the WOA, SSA and single ELMAN neural network models in prediction performance. Secondly, by comparison, the ELMAN model has certain advantages over the BP model in predicting building carbon emissions. For example, the mean absolute error (MAE) of ELMAN is 6.14 tons, which is lower than the BP model's 6.86 tons, a decrease of 0.72 tons; the root mean square error (RMSE) is 6.92 tons, which is 0.15 tons less than the BP's 7.07 tons; the mean absolute percentage error (MAPE) is 5.96%, which is 0.33% higher than the BP's 5.63%, but performs better in absolute error. In addition, the coefficient of determination (R2) of the ELMAN model is 0.61, slightly higher than the BP's 0.57, indicating that ELMAN has more advantages in explaining data variability. Therefore, although BP performs better in MAPE, the ELMAN model appears to be superior in terms of absolute error and explanatory power. Finally, the R2 of the MIC-COA-ELMAN model is 0.9626, which is significantly higher than the other five comparative models, showing higher fit and prediction accuracy.

[0194] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A carbon emission prediction method based on the COA-Elman model, characterized in that: include: S1. Obtain annual historical data on carbon emission-related factors of target buildings; S2. Construct an indicator system of carbon emission influencing factors based on the annual historical data; S3. Based on the carbon emission influencing factor index system, grey correlation analysis and maximum information coefficient are used to screen the correlation of influencing factors and determine the key influencing factors; S4. Based on the key influencing factors and the annual historical data, the raccoon algorithm is used to optimize the key parameters in the ELMAN neural network model to obtain a carbon emission training model; S5. Based on the key influencing factors, the annual historical data and the carbon emission training model, perform carbon emission forecasting to obtain a carbon emission forecasting result.

2. The carbon emission prediction method based on the COA-Elman model according to claim 1 is characterized in that: The S3 includes: S31, using a grey correlation analysis method to perform a correlation analysis on the carbon emission influencing factor index system to obtain a grey correlation degree of each influencing factor; S32, based on the grey correlation degree, perform preliminary screening to determine actual relevant factors; S33. Using the maximum information coefficient, perform a secondary screening on the actual relevant factors to obtain the key influencing factors.

3. The carbon emission prediction method based on the COA-Elman model according to claim 1 is characterized in that: The S4 includes: S41, taking the annual historical data as output variables and the carbon emission influencing factor index system as input variables, dividing them in chronological order to obtain a training set and a test set; S42, normalizing the data in the training set and the test set to obtain a processed training set and a processed test set; S43, embedding the raccoon algorithm into the ELMAN neural network model, and using the processed training set to train the embedded model to obtain a trained model; S44, constructing an evaluation index system, evaluating the trained model, and determining the carbon emission training model according to the evaluation results.

4. The carbon emission prediction method based on the COA-Elman model according to claim 3 is characterized in that: The evaluation index system of S44 includes: The root mean square error is calculated as follows: The mean absolute error is calculated as follows: The mean absolute percentage error is calculated as, The absolute correlation coefficient is calculated as follows: In the formula, n is the number of samples, y is i , and are the average of the actual value, predicted value and actual value of the i-th sample respectively.

5. The carbon emission prediction method based on the COA-Elman model according to claim 3 is characterized in that: The S43 includes: S431, initializing the weights and thresholds of the ELMAN neural network model, and using them as the search space for raccoon individuals in the raccoon algorithm; S432, corresponding to a set of Elman neural network parameters for each raccoon individual, and performing random initialization to determine the initial parameters; S433, applying the initial parameters to the ELMAN neural network model to obtain input-output error values, and using the error values ​​to determine fitness; S434, using the processed training set for training, after each round of iteration, the raccoon individual dynamically adjusts its position according to the current optimal position and the global optimal position, thereby updating the weights and thresholds of the Elman neural network model, and performing convergence condition determination at the same time, and outputting the optimal weights and thresholds when the convergence conditions are met.

6. A carbon emission prediction device based on the COA-Elman model, characterized in that: include: A data acquisition module is used to obtain annual historical data of carbon emission-related factors of target buildings; An indicator system construction module, used to construct an indicator system of carbon emission influencing factors based on the annual historical data; A factor screening module is used to screen the correlation of influencing factors based on the carbon emission influencing factor index system and determine the key influencing factors by using grey correlation analysis and maximum information coefficient; A model optimization module, for optimizing and training key parameters in the ELMAN neural network model using a raccoon algorithm based on the key influencing factors and the annual historical data to obtain a carbon emission training model; The prediction module is used to perform carbon emission prediction based on the key influencing factors, the annual historical data and the carbon emission training model to obtain a carbon emission prediction result.

7. The carbon emission prediction device based on the COA-Elman model according to claim 6 is characterized in that: The factor screening module includes: A grey correlation analysis module is used to perform a correlation analysis on the carbon emission influencing factor index system using grey correlation to obtain the grey correlation of each influencing factor; A preliminary screening module, used for performing preliminary screening based on the grey correlation degree to determine actual relevant factors; The secondary screening module is used to perform secondary screening on the actual relevant factors using the maximum information coefficient to obtain the key influencing factors.

8. The carbon emission prediction device based on the COA-Elman model according to claim 7, characterized in that: The model optimization module includes: A data partitioning module is used to use the annual historical data as output variables and the carbon emission influencing factor indicator system as input variables, and to partition them in chronological order to obtain a training set and a test set; A normalization processing module, used to perform normalization processing on the data in the training set and the test set to obtain a processed training set and a processed test set; A training module, used for embedding the raccoon algorithm into the ELMAN neural network model, and training the embedded model using the processed training set to obtain a trained model; The evaluation module is used to construct an evaluation index system to evaluate the trained model and determine the carbon emission training model according to the evaluation results.

9. The carbon emission prediction device based on the COA-Elman model according to claim 8, characterized in that: The evaluation index system in the evaluation module includes: The root mean square error is calculated as follows: The mean absolute error is calculated as follows: The mean absolute percentage error is calculated as, The absolute correlation coefficient is calculated as follows: In the formula, n is the number of samples, y is i , and are the average of the actual value, predicted value and actual value of the i-th sample respectively.

10. The carbon emission prediction device based on the COA-Elman model according to claim 8, characterized in that: The training module includes: An initialization module, used for initializing the weights and thresholds of the ELMAN neural network model and using it as a search space for raccoon individuals in the raccoon algorithm; A parameter determination module, used to correspond a set of Elman neural network parameters to each raccoon individual, and to perform random initialization to determine the initial parameters; An adaptation calculation module, used for applying the initial parameters to the ELMAN neural network model to obtain input-output error values, and using the error values ​​to perform fitness determination; The dynamic adjustment module is used to perform training using the processed training set. After each round of iteration, the raccoon individual dynamically adjusts its position according to the current optimal position and the global optimal position, thereby updating the weights and thresholds of the Elman neural network model, and performing convergence condition judgment at the same time. When the convergence conditions are met, the optimal weights and thresholds are output.

Images