Carbon emissions monitoring method and system based on artificial intelligence

Abstract

The present invention discloses a carbon emission monitoring method and system based on artificial intelligence, and the system includes a data acquisition module, a data preprocessing module, a sequence decomposition module, a complexity quantification module, a parameter optimization module, an optimal parameter calculation module and a carbon emission prediction module. The present invention belongs to the technical field of carbon emission monitoring, and specifically refers to a carbon emission monitoring method and system based on artificial intelligence. The present scheme optimizes the parameters K3 and α of the empirical mode decomposition that usually need to be manually set by the gray wolf optimization algorithm, so as to quickly determine the best parameters K3 and α without the need for manual setting, thereby solving the technical problem that it is difficult to quickly determine the best parameters, and usually a large number of experiments and adjustments are required to obtain the best decomposition effect. At the same time, the present scheme can effectively remove residual noise such as residual signals, and accurately analyze carbon emission information.

Description

Technical Field

[0001] The present invention belongs to the technical field of carbon emission monitoring, and specifically refers to a carbon emission monitoring method and system based on artificial intelligence. Background Art

[0002] Artificial intelligence-based carbon emissions monitoring methods and systems are mainly used to predict and monitor carbon emissions in order to conduct accurate carbon emissions assessment and management.

[0003] This method and system usually use artificial intelligence technology to collect raw data sequences related to carbon emissions, decompose the data sequences related to carbon emissions through empirical mode decomposition, and predict carbon emissions through extreme learning machines. However, the existing prediction methods still have many defects: some parameters of empirical mode decomposition usually need to be set manually, resulting in the decomposition results being greatly affected by personal subjectivity, and it is difficult to quickly determine the optimal parameters in this method, and usually a large number of experiments and adjustments are required to obtain the best decomposition effect; since the raw data sequence usually has the characteristics of nonlinearity and non-stationarity, the existing time-frequency domain analysis methods such as empirical mode decomposition cannot accurately analyze carbon emissions. The information released is difficult to eliminate residual noise such as residual signals; when predicting carbon emissions, the extreme learning machine based on particle swarm optimization is suitable for predicting high-complexity intrinsic mode functions, but not for predicting low-complexity intrinsic mode functions. The original extreme learning machine has a high prediction accuracy for low-complexity intrinsic mode functions, but is not suitable for predicting high-complexity intrinsic mode functions. The distributed adaptive extreme learning machine can predict high-complexity and low-complexity intrinsic mode functions, but has the defects of over-training and low convergence, which affects the accuracy of the prediction results. Therefore, it is difficult to accurately predict carbon emissions using the above extreme learning machines. Summary of the invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a carbon emission monitoring method and system based on artificial intelligence, which aims at the technical problem that the decomposition results caused by the need to manually set some parameters of the empirical mode decomposition are greatly affected by personal subjectivity, and it is difficult to quickly determine the optimal parameters, and usually a large number of experiments and adjustments are required to obtain the best decomposition effect. This scheme optimizes the parameters K3 and parameter α of the empirical mode decomposition that usually need to be manually set by the gray wolf optimization algorithm, thereby quickly determining the optimal parameters K3 and parameter α, and obtaining the optimized empirical mode decomposition without the need for manual setting, thereby solving the problem of decomposition. The decomposition results are greatly affected by personal subjectivity, and it is difficult to quickly determine the optimal parameters. Usually, a large number of experiments and adjustments are required to obtain the best decomposition effect. In view of the technical problems that the existing time-frequency domain analysis methods such as empirical mode decomposition cannot accurately analyze the information of carbon emissions due to the nonlinear and non-stationary characteristics of the original data sequence, and it is difficult to eliminate residual noise such as residual signals, this solution optimizes the empirical mode decomposition of the data sequence after preprocessing and adding a Gaussian white noise sequence, obtains the intrinsic mode function set and the residual signal sequence, and updates the intrinsic mode function set to obtain To the updated intrinsic mode function set, through the above operation, residual noise such as residual signal can be effectively removed. At the same time, this scheme quantifies the complexity of the intrinsic mode function in the updated intrinsic mode function set, and classifies the intrinsic mode functions in the updated intrinsic mode function set into high-complexity intrinsic mode functions and low-complexity intrinsic mode functions according to the complexity, so as to accurately analyze the information of carbon emissions; the existing extreme learning machine based on particle swarm optimization is not suitable for predicting low-complexity intrinsic mode functions, and the original extreme learning machine is not suitable for predicting high-complexity intrinsic mode functions, and the distribution The technical problem that the adaptive extreme learning machine has the defects of over-training and low convergence is solved. This scheme calculates the optimal weight and threshold parameters of the original extreme learning machine through the particle swarm optimization algorithm, calculates the optimal matrix weight parameters of the original extreme learning machine through the slime mold algorithm and the low-complexity intrinsic mode function, and optimizes the parameters of the original extreme learning machine according to the optimal weight and threshold parameters and the optimal matrix weight parameters to obtain the optimized original extreme learning machine. Through the above operations, the optimized original extreme learning machine can simultaneously predict the intrinsic mode functions of high complexity and low complexity, and does not have the defects of over-training and low convergence.

[0005] The technical solution adopted by the present invention is as follows: The carbon emission monitoring method based on artificial intelligence provided by the present invention comprises:

[0006] Step S1: data collection, specifically, collecting original data sequences;

[0007] Step S2: parameter optimization, specifically, optimizing the parameter K3 and parameter α of the empirical mode decomposition by using the Grey Wolf Optimization Algorithm to obtain an optimized empirical mode decomposition;

[0008] Step S3: Data preprocessing, specifically, preprocessing the original data sequence to remove trend and periodic components to obtain a preprocessed original data sequence, and adding a Gaussian white noise sequence to the preprocessed original data sequence to obtain a data sequence after preprocessing and adding the Gaussian white noise sequence. The formula used is as follows:

[0009] X i (t) = X(t) + βE(δ i (t)), i = 1, 2, ..., K;

[0010] In the formula, X(t) represents the original data sequence at time t, t represents the time variable, and X i (t) represents the data sequence after preprocessing and adding Gaussian white noise sequence at time t, β represents the Gaussian white noise weight coefficient, δ i (t) represents the i-th Gaussian white noise sequence at time t, E(δ i (t)) represents the i (t) the intrinsic mode function obtained by optimizing the empirical mode decomposition, K represents the number of Gaussian white noise sequences;

[0011] Step S4: decomposing the data sequence, specifically, performing optimized empirical mode decomposition on the data sequence after preprocessing and adding the Gaussian white noise sequence, obtaining an intrinsic mode function set and a residual signal sequence, and updating the intrinsic mode function set to obtain an updated intrinsic mode function set;

[0012] Step S5: quantifying the complexity, specifically, quantifying the complexity of the eigenmode functions in the updated eigenmode function set, and classifying the eigenmode functions in the updated eigenmode function set into eigenmode functions with high complexity and eigenmode functions with low complexity according to the complexity;

[0013] Step S6: Calculating optimal parameters, specifically, calculating the optimal weight and threshold parameters of the original extreme learning machine by a particle swarm optimization algorithm, calculating the optimal matrix weight parameters of the original extreme learning machine by a slime mold algorithm and a low-complexity intrinsic mode function, and optimizing the parameters of the original extreme learning machine by the optimal weight and threshold parameters and the optimal matrix weight parameters to obtain an optimized original extreme learning machine;

[0014] Step S7: predicting carbon emissions, specifically, inputting the high-complexity intrinsic mode function and the low-complexity intrinsic mode function into the optimized original extreme learning machine for training to obtain a carbon emission prediction result.

[0015] As a further improvement of this solution, in step S2, the parameter optimization step includes:

[0016] Step S21: Initialize the maximum number of optimization iterations, parameter α and parameter K3;

[0017] Step S22: Calculate the decomposition fitness, the calculation formula of the decomposition fitness is:

[0018] T n =-K t +H pe ;

[0019] Where, T n represents the decomposition fitness, K t Denotes kurtosis, H pe represents the permutation entropy;

[0020] Step S23: Update parameter α and parameter K3 according to the decomposition fitness;

[0021] Step S24: iteratively execute steps S52 to S53. When the maximum number of optimization iterations is reached, the iteration is stopped, and the updated parameters K3 and α generated in the last iteration are used to modify the parameters K3 and α of the empirical mode decomposition to obtain an optimized empirical mode decomposition.

[0022] As a further improvement of this solution, in step S4, the step of decomposing the data sequence includes:

[0023] Step S41: Calculate a first group of residual signals. The calculation formula for the first group of residual signals is:

[0024] R1=N(X i (t));

[0025] Where R1 represents the first group of residual signals, X i (t) represents the data sequence after preprocessing and adding Gaussian white noise sequence at time t, N(X i (t)) represents X i The local mean of (t);

[0026] Step S42: Calculate a first group of modal components. The calculation formula of the first group of modal components is:

[0027] d1=X i (t)-R1;

[0028] Where d1 represents the first group of modal components, R1 represents the first group of residual signals, and X i(t) represents the data sequence after preprocessing and adding Gaussian white noise sequence at time t;

[0029] Step S43: Add the Gaussian white noise sequence again, and calculate the second group of residual signals by local mean, the calculation formula of the second group of residual signals is:

[0030] R2=R1+R1E(δ i (t));

[0031] In the formula, R2 represents the second group of residual signals, R1 represents the first group of residual signals, and δ i (t) represents the i-th Gaussian white noise sequence at time t, E(δ i (t)) represents the i (t) the intrinsic mode function obtained by optimizing the empirical mode decomposition;

[0032] Step S44: Calculate the second group of modal components. The calculation formula of the second group of modal components is:

[0033] d2=R1-R2=R1-(N(R1+βE(δ i (t))));

[0034] In the formula, d2 represents the second group of modal components, R1 represents the first group of residual signals, R2 represents the second group of residual signals, β represents the Gaussian white noise weight coefficient, δ i (t) represents the i-th Gaussian white noise sequence at time t, E(δ i (t)) represents the i (t) The intrinsic mode function obtained by optimizing the empirical mode decomposition, N(R1+β1E(δ i (t))) represents R1+β1E(δ i (t)) local mean;

[0035] Step S45: presetting the initial value of k to 3, and setting a deviation threshold;

[0036] Step S46: Calculate the kth residual signal and the kth modal component, and go to step S47. The calculation formula of the kth residual signal and the kth modal component is:

[0037] R k =N(R k-1 +βE(δ k (t)), k = 3, 4, ..., K;

[0038] d k =R k-1 -R k , k=3,4,…,K;

[0039] In the formula, R k represents the kth residual signal, d k represents the kth modal component, K represents the number of Gaussian white noise sequences, δ k (t) represents the i-th Gaussian white noise sequence at time t, E(δ k (t)) represents the k (t) is the intrinsic mode function obtained by optimizing the empirical mode decomposition, that is, the kth intrinsic mode function, β represents the Gaussian white noise weight coefficient, R k-1 represents the k-1th residual signal, N(R k-1 +β k-1 E(δ k (t))) represents R k-1 +β k-1 E(δ k (t)) local mean;

[0040] Step S47: forming a residual signal sequence according to the residual signal, setting k=k+1, and judging whether the termination condition is satisfied. If the termination condition is satisfied, go to step S48; otherwise, go to step S46;

[0041] The termination condition is that the residual signal sequence decreases monotonically and the standard deviation is less than the deviation threshold. The calculation formula of the standard deviation is:

[0042]

[0043] In the formula, σ represents the standard deviation, I k represents the kth eigenmode function, I k+1 represents the k+1th intrinsic mode function, ||I k ||2 represents the L2 norm of k intrinsic mode functions, ||I k+1 -I k ||2 represents the L2 norm of the difference between the k+1th eigenmode function and the kth eigenmode function;

[0044] Step S48: presetting a mean square threshold, constructing an eigenmode function set according to the eigenmode function, and calculating the root mean square value of each eigenmode function, traversing all eigenmode functions, if the root mean square value of the eigenmode function is less than or equal to the mean square threshold, then removing the eigenmode function from the eigenmode function set, and adding the eigenmode function to the remaining items;

[0045] Step S49: adding the remaining items as eigenmode functions to the eigenmode function set to obtain an updated eigenmode function set.

[0046] As a further improvement of this solution, in step S5, the step of quantifying the complexity includes:

[0047] Step S51: presetting the embedding parameter t1 and the spatial offset w1 of the original data sequence;

[0048] Step S52: traversing the eigenmode functions in the updated eigenmode function set, and constructing a recursive graph by calculating the Chebyshev distance of the eigenmode functions;

[0049] Step S53: Analyze the recursive graph using the binary co-occurrence matrix to obtain the recursive mapping result

[0050] Step S54: Increase the value of the embedding parameter t1, that is, t1=t1+1, and repeat steps S52 and S53 to obtain a new recursive mapping result.

[0051] Step S55: Calculate the spatial correlation recursive sample entropy SdrSampEn(t1, w1) of each intrinsic mode function, and evaluate the complexity of each intrinsic mode function according to the spatial correlation recursive sample entropy SdrSampEn(t1, w1) of each intrinsic mode function. The calculation formula of the spatial correlation recursive sample entropy SdrSampEn(t1, w1) of each intrinsic mode function is:

[0052]

[0053] Where SdrSampEn(t1,w1) represents the spatial correlation recursive sample entropy of the intrinsic mode function, Represents the recursive mapping result, represents the new recursive mapping result, t1 represents the embedding parameter, and w1 represents the spatial offset of the original data sequence;

[0054] Step S56: classifying each eigenmode function into an eigenmode function with high complexity and an eigenmode function with low complexity according to the complexity of each eigenmode function.

[0055] As a further improvement of this solution, in step S6, the step of calculating the optimal parameters includes:

[0056] Step S61: Calculate the optimal weight and threshold parameters, specifically, calculate the optimal weight and threshold parameters of the original extreme learning machine by using a particle swarm optimization algorithm;

[0057] Step S62: Calculate the optimal matrix weight parameters of the original extreme learning machine, specifically, calculate the optimal matrix weight parameters of the original extreme learning machine by using a slime mold algorithm and a low-complexity intrinsic mode function;

[0058] Step S63: Optimizing the parameters of the original extreme learning machine according to the optimal weight and threshold parameters and the optimal matrix weight parameters to obtain an optimized original extreme learning machine.

[0059] As a further improvement of this solution, in step S61, the step of calculating the optimal weight and threshold parameters includes:

[0060] Step S611: Parameter initialization, specifically, initializing the parameters of the particle swarm optimization algorithm and the original extreme learning machine, the parameters of the particle swarm optimization algorithm include the maximum number of iterations, the number of particles N', the learning factors c1 and c2, and the inertia weight factor w2, and the parameters of the original extreme learning machine include the weight matrix from the input layer to the hidden layer and the weight matrix from the hidden layer to the output layer;

[0061] Step S612: Generate an initial particle swarm, specifically, randomly generate the speed V' and position X' of N' particles, and initialize the optimal position P of each particle best and the global optimal position G best ;

[0062] Step S613: Update the speed and position of each particle. The formula used is as follows:

[0063]

[0064] X′ i (t+1)=X′ i (t)+V′ i (t+1);

[0065] In the formula, w2 represents the inertia weight factor, c1 and c2 represent the learning factor, P best represents the optimal position of each particle, G best represents the global optimal position, V' i (t) represents the velocity of the ith particle at time t, V' i (t+1) represents the speed of the ith particle at time t+1, that is, the updated speed of the ith particle, X' i (t) represents the position of the i-th particle at time t, X' i (t+1) represents the position of the i-th particle at time t+1, that is, the position of the i+1-th particle after update, rand() represents a random number between 0 and 1, and * represents a multiplication operation;

[0066] Step S614: Update the parameters of the original extreme learning machine. Specifically, set the updated position of each particle to the value of the weight matrix from the input layer to the hidden layer and the weight matrix from the hidden layer to the output layer, train the original extreme learning machine using the training data set, and calculate the performance index of each particle on the validation set. If the performance index of the particle on the validation set is better than the optimal position P of the particle, best , then the optimal position of the particle P best Update to the updated position of the particle, if the performance index of the particle on the validation set is better than the global best position G best , then the global optimal position G best Update to the updated position of the particle;

[0067] Step S615: Iterate steps S613 to S614. When the maximum number of iterations is reached, the iteration process ends, and the values ​​of the weight matrix from the input layer to the hidden layer and the weight matrix from the hidden layer to the output layer at the last iteration are used as the optimal weight and threshold parameters.

[0068] As a further improvement of this solution, in step S62, the step of calculating the optimal matrix weight parameters of the original extreme learning machine includes:

[0069] Step S621: Initializing the parameters of the slime mold algorithm, wherein the parameters of the slime mold algorithm include the number of low-complexity iterations and the weight coefficient w3 of the slime mold algorithm;

[0070] Step S622: Initialize individual position X'' and individual velocity V'', wherein the individual position X'' is a matrix with M rows and N columns, and the element x'' in the individual position X'' is ij represents the position of the i-th individual in the j-th dimension. The individual velocity V″ is a matrix with M rows and N columns. The element v″ in the individual velocity V″ ij represents the velocity of the i-th individual in the j-th dimension;

[0071] Step S623: Calculate the fitness value of individual position X”;

[0072] Step S624: Update the individual position X'' and the individual speed V'' to obtain the updated individual position X'' and the updated individual speed V''. The calculation formula for updating the individual position X'' and the individual speed V'' is:

[0073] x″ ij (t+1)=x″ ij (t)+v″ ij (t+1);

[0074] v″ ij (t+1)=w3*v″ ij (t)+c3*r3*(pbestij (t)-x″ ij (t))+c4*r4*(gbest ij (t)-x″ ij (t));

[0075] In the formula, x″ ij (t) represents the position of the i-th individual in the j-th dimension at time t, x″ ij (t+1) represents the position of the i-th individual in the j-th dimension at time t+1, that is, the position of the i-th individual in the j-th dimension after update, v″ ij (t) represents the velocity of the i-th individual in the j-th dimension at time t, v″ ij (t+1) represents the speed of the i-th individual in the j-th dimension at time t+1, w3 represents the weight coefficient of the slime mold algorithm, c3 and c4 represent learning factors, r3 and r4 represent random numbers between 0 and 1, and pbest ij (t) represents the optimal position of the i-th individual in the j-th dimension at time t, gbest ij (t) represents the global optimal position of the i-th individual in the j-th dimension at time t;

[0076] Step S625: Calculate the fitness value of the updated individual position X' according to the updated individual position X';

[0077] Step S626: iteratively execute steps S624 to S625. After reaching the low-complexity iteration number, the iteration process ends, and the optimal individual position X" is selected according to the fitness value as the optimal matrix weight parameter of the original extreme learning machine.

[0078] The carbon emission monitoring system based on artificial intelligence provided by the present invention includes a data acquisition module, a data preprocessing module, a sequence decomposition module, a complexity quantification module, a parameter optimization module, an optimal parameter calculation module and a carbon emission prediction module;

[0079] The data acquisition module is used for data acquisition, specifically, to acquire the original data sequence and send the original data sequence to the data preprocessing module;

[0080] The parameter optimization module is used for parameter optimization, specifically, optimizing the parameter K3 and parameter α of the empirical mode decomposition by using the Grey Wolf Optimization Algorithm to obtain an optimized empirical mode decomposition;

[0081] The data preprocessing module is used for data preprocessing, specifically, preprocessing the original data sequence to remove trend and periodic components to obtain a preprocessed original data sequence, and adding a Gaussian white noise sequence to the preprocessed original data sequence to obtain a data sequence after preprocessing and adding the Gaussian white noise sequence, and sending the data sequence after preprocessing and adding the Gaussian white noise sequence to the sequence decomposition module;

[0082] The sequence decomposition module is used to decompose the data sequence, specifically, to perform an optimized empirical mode decomposition on the data sequence after preprocessing and adding a Gaussian white noise sequence, to obtain an intrinsic mode function set and a residual signal sequence, and to update the intrinsic mode function set to obtain an updated intrinsic mode function set, and to send the updated intrinsic mode function set to the complexity quantification module;

[0083] The complexity quantification module is used to quantify the complexity, specifically, to quantify the complexity of the intrinsic mode functions in the updated intrinsic mode function set, and classify the intrinsic mode functions in the updated intrinsic mode function set into high-complexity intrinsic mode functions and low-complexity intrinsic mode functions according to the complexity, send the high-complexity intrinsic mode functions to the carbon emission prediction module, and send the low-complexity intrinsic mode functions to the optimal parameter calculation module and the carbon emission prediction module;

[0084] The optimal parameter calculation module is used to calculate the optimal parameters, specifically, to calculate the optimal weight and threshold parameters of the original extreme learning machine through the particle swarm optimization algorithm, to calculate the optimal matrix weight parameters of the original extreme learning machine through the slime mold algorithm and the low-complexity intrinsic mode function, and to optimize the parameters of the original extreme learning machine through the optimal weight and threshold parameters and the optimal matrix weight parameters to obtain the optimized original extreme learning machine, and to send the optimized original extreme learning machine to the carbon emission prediction module;

[0085] The carbon emission prediction module is used to predict carbon emissions. Specifically, the high-complexity intrinsic mode function and the low-complexity intrinsic mode function are input into the optimized original extreme learning machine for training to obtain the carbon emission prediction result.

[0086] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0087] (1) In view of the technical problem that the decomposition results are greatly affected by personal subjectivity due to the need to manually set some parameters of the empirical mode decomposition, and it is difficult to quickly determine the optimal parameters, and a large number of experiments and adjustments are usually required to obtain the best decomposition effect, this scheme optimizes the parameters K3 and parameter α of the empirical mode decomposition that usually need to be manually set through the Gray Wolf Optimization Algorithm, so as to quickly determine the optimal parameters K3 and parameter α, and obtain the optimized empirical mode decomposition without the need for manual setting, thereby solving the technical problem that the decomposition results are greatly affected by personal subjectivity, and it is difficult to quickly determine the optimal parameters, and a large number of experiments and adjustments are usually required to obtain the best decomposition effect.

[0088] (2) In order to solve the technical problem that the existing time-frequency domain analysis methods such as empirical mode decomposition cannot accurately analyze the information of carbon emissions due to the nonlinear and non-stationary characteristics of the original data sequence, and it is difficult to eliminate residual noise such as residual signals, this scheme optimizes the empirical mode decomposition of the data sequence after preprocessing and adding a Gaussian white noise sequence to obtain an intrinsic mode function set and a residual signal sequence, and updates the intrinsic mode function set to obtain an updated intrinsic mode function set. Through the above operations, residual noise such as residual signals can be effectively removed. At the same time, this scheme quantifies the complexity of the intrinsic mode functions in the updated intrinsic mode function set, and classifies the intrinsic mode functions in the updated intrinsic mode function set into high-complexity intrinsic mode functions and low-complexity intrinsic mode functions according to the complexity, so as to accurately analyze the information of carbon emissions.

[0089] (3) Aiming at the technical problems that the existing extreme learning machine based on particle swarm optimization is not suitable for predicting low-complexity intrinsic mode functions, the original extreme learning machine is not suitable for predicting high-complexity intrinsic mode functions, and the distributed adaptive extreme learning machine has the defects of over-training and low convergence, this scheme calculates the optimal weight and threshold parameters of the original extreme learning machine through the particle swarm optimization algorithm, calculates the optimal matrix weight parameters of the original extreme learning machine through the slime mold algorithm and the low-complexity intrinsic mode function, and optimizes the parameters of the original extreme learning machine according to the optimal weight and threshold parameters and the optimal matrix weight parameters to obtain the optimized original extreme learning machine. Through the above operations, the optimized original extreme learning machine can simultaneously predict high-complexity and low-complexity intrinsic mode functions without the defects of over-training and low convergence. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] Figure 1 A schematic diagram of the process of the carbon emission monitoring method based on artificial intelligence provided by the present invention;

[0091] Figure 2 is a schematic flow chart of step S2;

[0092] Figure 3 is a schematic flow chart of step S4;

[0093] Figure 4 is a schematic flow chart of step S5;

[0094] Figure 5 is a schematic diagram of the process of step S6;

[0095] Figure 6 is a flow chart of step S61;

[0096] Figure 7 is a flow chart of step S62;

[0097] Figure 8 This is a schematic diagram of the structure of the artificial intelligence-based carbon emissions monitoring system provided by the present invention.

[0098] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0099] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0100] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0101] Example 1, see Figure 1 The carbon emission monitoring method based on artificial intelligence provided by the present invention includes:

[0102] Step S1: data collection, specifically, collecting original data sequences;

[0103] Step S2: parameter optimization, specifically, optimizing the parameter K3 and parameter α of the empirical mode decomposition by using the Grey Wolf Optimization Algorithm to obtain an optimized empirical mode decomposition;

[0104] Step S3: Data preprocessing, specifically, preprocessing the original data sequence to remove trend and periodic components to obtain a preprocessed original data sequence, and adding a Gaussian white noise sequence to the preprocessed original data sequence to obtain a data sequence after preprocessing and adding the Gaussian white noise sequence. The formula used is as follows:

[0105] X i (t) = X(t) + βE(δ i (t)), i = 1, 2, ..., K;

[0106] In the formula, X(t) represents the original data sequence at time t, t represents the time variable, and X i (t) represents the data sequence after preprocessing and adding Gaussian white noise sequence at time t, β represents the Gaussian white noise weight coefficient, δ i (t) represents the i-th Gaussian white noise sequence at time t, E(δ i (t)) represents the i (t) the intrinsic mode function obtained by optimizing the empirical mode decomposition, K represents the number of Gaussian white noise sequences;

[0107] Step S4: decomposing the data sequence, specifically, performing optimized empirical mode decomposition on the data sequence after preprocessing and adding the Gaussian white noise sequence, obtaining an intrinsic mode function set and a residual signal sequence, and updating the intrinsic mode function set to obtain an updated intrinsic mode function set;

[0108] Step S5: quantifying the complexity, specifically, quantifying the complexity of the eigenmode functions in the updated eigenmode function set, and classifying the eigenmode functions in the updated eigenmode function set into eigenmode functions with high complexity and eigenmode functions with low complexity according to the complexity;

[0109] Step S6: Calculating optimal parameters, specifically, calculating the optimal weight and threshold parameters of the original extreme learning machine by a particle swarm optimization algorithm, calculating the optimal matrix weight parameters of the original extreme learning machine by a slime mold algorithm and a low-complexity intrinsic mode function, and optimizing the parameters of the original extreme learning machine by the optimal weight and threshold parameters and the optimal matrix weight parameters to obtain an optimized original extreme learning machine;

[0110] Step S7: predicting carbon emissions, specifically, inputting the high-complexity intrinsic mode function and the low-complexity intrinsic mode function into the optimized original extreme learning machine for training to obtain a carbon emissions prediction result;

[0111] In the above operation, the decomposition results are greatly affected by personal subjectivity due to the need to manually set some parameters of the empirical mode decomposition, and it is difficult to quickly determine the optimal parameters, and usually a large number of experiments and adjustments are required to obtain the best decomposition effect. The technical problem is that the decomposition results are greatly affected by personal subjectivity, and it is difficult to quickly determine the optimal parameters, and usually a large number of experiments and adjustments are required to obtain the best decomposition effect. This solution uses the Gray Wolf Optimization Algorithm to optimize the parameters K3 and parameter α of the empirical mode decomposition that usually need to be set manually, so as to quickly determine the optimal parameters K3 and parameter α, and obtain the optimized empirical mode decomposition without the need for manual setting, thereby solving the technical problems that the decomposition results are greatly affected by personal subjectivity, and it is difficult to quickly determine the optimal parameters, and usually a large number of experiments and adjustments are required to obtain the best decomposition effect; at the same time, the original data sequence has the characteristics of nonlinearity and non-stationarity. The existing time-frequency domain analysis methods such as empirical mode decomposition cannot accurately analyze the information of carbon emissions, and it is difficult to eliminate the technical problem that residual noise such as residual signals. This scheme obtains an intrinsic mode function set and a residual signal sequence by performing optimized empirical mode decomposition on the data sequence after preprocessing and adding a Gaussian white noise sequence, and updates the intrinsic mode function set to obtain an updated intrinsic mode function set. Through the above operations, residual noise such as residual signals can be effectively removed. At the same time, this scheme quantifies the complexity of the intrinsic mode functions in the updated intrinsic mode function set, and classifies the intrinsic mode functions in the updated intrinsic mode function set into high-complexity intrinsic mode functions and low-complexity intrinsic mode functions according to the complexity, so as to accurately analyze the information of carbon emissions.

[0112] Embodiment 2, this embodiment is based on the above embodiment, see Figure 2 In step S2, the parameter optimization step includes:

[0113] Step S21: Initialize the maximum number of optimization iterations, parameter α and parameter K3;

[0114] Step S22: Calculate the decomposition fitness, the calculation formula of the decomposition fitness is:

[0115] T n =-K t +H pe ;

[0116] Where, T n represents the decomposition fitness, K t Denotes kurtosis, H pe represents the permutation entropy;

[0117] Step S23: Update parameter α and parameter K3 according to the decomposition fitness;

[0118] Step S24: iteratively execute steps S52 to S53. When the maximum number of optimization iterations is reached, the iteration is stopped, and the updated parameters K3 and α generated in the last iteration are used to modify the parameters K3 and α of the empirical mode decomposition to obtain an optimized empirical mode decomposition.

[0119] Embodiment 3, this embodiment is based on the above embodiment, see Figure 3 In step S4, the step of decomposing the data sequence comprises:

[0120] Step S41: Calculate a first group of residual signals. The calculation formula for the first group of residual signals is:

[0121] R1=N(X i (t));

[0122] Where R1 represents the first group of residual signals, X i (t) represents the data sequence after preprocessing and adding Gaussian white noise sequence at time t, N(X i (t)) represents X i The local mean of (t);

[0123] Step S42: Calculate a first group of modal components. The calculation formula of the first group of modal components is:

[0124] d1=X i (t)-R1;

[0125] Where d1 represents the first group of modal components, R1 represents the first group of residual signals, and X i (t) represents the data sequence after preprocessing and adding Gaussian white noise sequence at time t;

[0126] Step S43: Add the Gaussian white noise sequence again, and calculate the second group of residual signals by local mean, the calculation formula of the second group of residual signals is:

[0127] R2=R1+R1E(δ i (t));

[0128] In the formula, R2 represents the second group of residual signals, R1 represents the first group of residual signals, and δ i (t) represents the i-th Gaussian white noise sequence at time t, E(δ i (t)) represents the i (t) the intrinsic mode function obtained by optimizing the empirical mode decomposition;

[0129] Step S44: Calculate the second group of modal components. The calculation formula of the second group of modal components is:

[0130] d2=R1-R2=R1-(N(R1+βE(δ i (t))));

[0131] In the formula, d2 represents the second group of modal components, R1 represents the first group of residual signals, R2 represents the second group of residual signals, β represents the Gaussian white noise weight coefficient, δ i (t) represents the i-th Gaussian white noise sequence at time t, E(δ i (t)) represents the i (t) The intrinsic mode function obtained by optimizing the empirical mode decomposition, N(R1+β1E(δ i (t))) represents R1+β1E(δ i (t)) local mean;

[0132] Step S45: presetting the initial value of k to 3, and setting a deviation threshold;

[0133] Step S46: Calculate the kth residual signal and the kth modal component, and go to step S47. The calculation formula of the kth residual signal and the kth modal component is:

[0134] R k =N(R k-1 +βE(δ k (t)), k = 3, 4, ..., K;

[0135] d k =R k-1 -R k , k=3,4,…,K;

[0136] In the formula, R k represents the kth residual signal, d k represents the kth modal component, K represents the number of Gaussian white noise sequences, δ k (t) represents the i-th Gaussian white noise sequence at time t, E(δ k (t)) represents the k (t) is the intrinsic mode function obtained by optimizing the empirical mode decomposition, that is, the kth intrinsic mode function, β represents the Gaussian white noise weight coefficient, R k-1 represents the k-1th residual signal, N(R k-1 +β k-1 E(δ k (t))) represents R k-1 +β k-1 E(δ k (t)) local mean;

[0137] Step S47: forming a residual signal sequence according to the residual signal, setting k=k+1, and judging whether the termination condition is satisfied. If the termination condition is satisfied, go to step S48; otherwise, go to step S46;

[0138] The termination condition is that the residual signal sequence decreases monotonically and the standard deviation is less than the deviation threshold. The calculation formula of the standard deviation is:

[0139]

[0140] In the formula, σ represents the standard deviation, I k represents the kth eigenmode function, I k+1 represents the k+1th intrinsic mode function, ||I k ||2 represents the L2 norm of k intrinsic mode functions, ||I k+1 -I k ||2 represents the L2 norm of the difference between the k+1th eigenmode function and the kth eigenmode function;

[0141] Step S48: presetting a mean square threshold, constructing an eigenmode function set according to the eigenmode function, and calculating the root mean square value of each eigenmode function, traversing all eigenmode functions, if the root mean square value of the eigenmode function is less than or equal to the mean square threshold, then removing the eigenmode function from the eigenmode function set, and adding the eigenmode function to the remaining items;

[0142] Step S49: adding the remaining items as eigenmode functions to the eigenmode function set to obtain an updated eigenmode function set.

[0143] Embodiment 4, this embodiment is based on the above embodiment, see Figure 4 In step S5, the step of quantizing the complexity includes:

[0144] Step S51: presetting the embedding parameter t1 and the spatial offset w1 of the original data sequence;

[0145] Step S52: traversing the eigenmode functions in the updated eigenmode function set, and constructing a recursive graph by calculating the Chebyshev distance of the eigenmode functions;

[0146] Step S53: Analyze the recursive graph using the binary co-occurrence matrix to obtain the recursive mapping result

[0147] Step S54: Increase the value of the embedding parameter t1, that is, t1=t1+1, and repeat steps S52 and S53 to obtain a new recursive mapping result.

[0148] Step S55: Calculate the spatial correlation recursive sample entropy SdrSampEn(t1, w1) of each intrinsic mode function, and evaluate the complexity of each intrinsic mode function according to the spatial correlation recursive sample entropy SdrSampEn(t1, w1) of each intrinsic mode function. The calculation formula of the spatial correlation recursive sample entropy SdrSampEn(t1, w1) of each intrinsic mode function is:

[0149]

[0150] Where SdrSampEn(t1,w1) represents the spatial correlation recursive sample entropy of the intrinsic mode function, Represents the recursive mapping result, represents the new recursive mapping result, t1 represents the embedding parameter, and w1 represents the spatial offset of the original data sequence;

[0151] Step S56: classifying each eigenmode function into an eigenmode function with high complexity and an eigenmode function with low complexity according to the complexity of each eigenmode function.

[0152] Embodiment 5, this embodiment is based on the above embodiment, see Figure 5 , Figure 6 and Figure 7 In step S6, the step of calculating the optimal parameters includes:

[0153] Step S61: Calculate the optimal weight and threshold parameters, specifically, calculate the optimal weight and threshold parameters of the original extreme learning machine by using a particle swarm optimization algorithm;

[0154] Step S62: Calculate the optimal matrix weight parameters of the original extreme learning machine, specifically, calculate the optimal matrix weight parameters of the original extreme learning machine by using a slime mold algorithm and a low-complexity intrinsic mode function;

[0155] Step S63: Optimizing the parameters of the original extreme learning machine according to the optimal weight and threshold parameters and the optimal matrix weight parameters to obtain an optimized original extreme learning machine;

[0156] In step S61, the step of calculating the optimal weight and threshold parameters includes:

[0157] Step S611: Parameter initialization, specifically, initializing the parameters of the particle swarm optimization algorithm and the original extreme learning machine, the parameters of the particle swarm optimization algorithm include the maximum number of iterations, the number of particles N', the learning factors c1 and c2, and the inertia weight factor w2, and the parameters of the original extreme learning machine include the weight matrix from the input layer to the hidden layer and the weight matrix from the hidden layer to the output layer;

[0158] Step S612: Generate an initial particle swarm, specifically, randomly generate the speed V' and position X' of N' particles, and initialize the optimal position P of each particle best and the global optimal position G best ;

[0159] Step S613: Update the speed and position of each particle. The formula used is as follows:

[0160]

[0161] X′ i (t+1)=X′ i (t)+V′ i (t+1);

[0162] In the formula, w2 represents the inertia weight factor, c1 and c2 represent the learning factor, P best represents the optimal position of each particle, G best represents the global optimal position, V' i (t) represents the velocity of the ith particle at time t, V' i (t+1) represents the speed of the ith particle at time t+1, that is, the updated speed of the ith particle, X' i (t) represents the position of the i-th particle at time t, X' i (t+1) represents the position of the i-th particle at time t+1, that is, the position of the i+1-th particle after update, rand() represents a random number between 0 and 1, and * represents a multiplication operation;

[0163] Step S614: Update the parameters of the original extreme learning machine. Specifically, set the updated position of each particle to the value of the weight matrix from the input layer to the hidden layer and the weight matrix from the hidden layer to the output layer, train the original extreme learning machine using the training data set, and calculate the performance index of each particle on the validation set. If the performance index of the particle on the validation set is better than the optimal position P of the particle, best , then the optimal position of the particle P best Update to the updated position of the particle, if the performance index of the particle on the validation set is better than the global best position G best , then the global optimal position G best Update to the updated position of the particle;

[0164] Step S615: iteratively execute steps S613 to S614. When the maximum number of iterations is reached, the iteration process ends, and the values ​​of the weight matrix from the input layer to the hidden layer and the weight matrix from the hidden layer to the output layer at the last iteration are used as the optimal weight and threshold parameters;

[0165] In step S62, the step of calculating the optimal matrix weight parameters of the original extreme learning machine includes:

[0166] Step S621: Initializing the parameters of the slime mold algorithm, wherein the parameters of the slime mold algorithm include the number of low-complexity iterations and the weight coefficient w3 of the slime mold algorithm;

[0167] Step S622: Initialize individual position X'' and individual velocity V'', wherein the individual position X'' is a matrix with M rows and N columns, and the element x'' in the individual position X'' is ij represents the position of the i-th individual in the j-th dimension. The individual velocity V” is a matrix with M rows and N columns. The element v” in the individual velocity V” ij represents the velocity of the i-th individual in the j-th dimension;

[0168] Step S623: Calculate the fitness value of individual position X”;

[0169] Step S624: Update the individual position X'' and the individual speed V'' to obtain the updated individual position X'' and the updated individual speed V''. The calculation formula for updating the individual position X'' and the individual speed V'' is:

[0170] x″ ij (t+1)=x″ ij (t)+v″ ij (t+1);

[0171] v″ ij (t+1)=w3*v″ ij (t)+c3*r3*(pbest ij (t)-x″ ij (t))+c4*r4*(gbest ij (t)-x″ ij (t));

[0172] In the formula, x″ ij (t) represents the position of the i-th individual in the j-th dimension at time t, x″ ij (t+1) represents the position of the i-th individual in the j-th dimension at time t+1, that is, the position of the i-th individual in the j-th dimension after update, v″ ij (t) represents the velocity of the i-th individual in the j-th dimension at time t, v″ ij (t+1) represents the speed of the i-th individual in the j-th dimension at time t+1, w3 represents the weight coefficient of the slime mold algorithm, c3 and c4 represent learning factors, r3 and r4 represent random numbers between 0 and 1, and pbest ij (t) represents the optimal position of the i-th individual in the j-th dimension at time t, gbest ij(t) represents the global optimal position of the i-th individual in the j-th dimension at time t;

[0173] Step S625: Calculate the fitness value of the updated individual position X' according to the updated individual position X';

[0174] Step S626: iteratively execute steps S624 to S625, and after reaching the low-complexity iteration number, terminate the iteration process, and select the optimal individual position X" as the optimal matrix weight parameter of the original extreme learning machine according to the fitness value;

[0175] In the above operation, in view of the technical problems that the existing extreme learning machine based on particle swarm optimization is not suitable for predicting low-complexity intrinsic mode functions, the original extreme learning machine is not suitable for predicting high-complexity intrinsic mode functions, and the distributed adaptive extreme learning machine has the defects of over-training and low convergence, this scheme calculates the optimal weight and threshold parameters of the original extreme learning machine through the particle swarm optimization algorithm, calculates the optimal matrix weight parameters of the original extreme learning machine through the slime mold algorithm and the low-complexity intrinsic mode function, and optimizes the parameters of the original extreme learning machine according to the optimal weight and threshold parameters and the optimal matrix weight parameters to obtain the optimized original extreme learning machine. Through the above operation, the optimized original extreme learning machine can simultaneously predict high-complexity and low-complexity intrinsic mode functions without the defects of over-training and low convergence.

[0176] Embodiment 6, this embodiment is based on the above embodiment, see Figure 8 , the carbon emission monitoring system based on artificial intelligence provided by the present invention includes a data acquisition module, a data preprocessing module, a sequence decomposition module, a complexity quantification module, a parameter optimization module, an optimal parameter calculation module and a carbon emission prediction module;

[0177] The data acquisition module is used for data acquisition, specifically, to acquire the original data sequence and send the original data sequence to the data preprocessing module;

[0178] The parameter optimization module is used for parameter optimization, specifically, optimizing the parameter K3 and parameter α of the empirical mode decomposition by using the Grey Wolf Optimization Algorithm to obtain an optimized empirical mode decomposition;

[0179] The data preprocessing module is used for data preprocessing, specifically, preprocessing the original data sequence to remove trend and periodic components to obtain a preprocessed original data sequence, and adding a Gaussian white noise sequence to the preprocessed original data sequence to obtain a data sequence after preprocessing and adding the Gaussian white noise sequence, and sending the data sequence after preprocessing and adding the Gaussian white noise sequence to the sequence decomposition module;

[0180] The sequence decomposition module is used to decompose the data sequence, specifically, to perform an optimized empirical mode decomposition on the data sequence after preprocessing and adding a Gaussian white noise sequence, to obtain an intrinsic mode function set and a residual signal sequence, and to update the intrinsic mode function set to obtain an updated intrinsic mode function set, and to send the updated intrinsic mode function set to the complexity quantification module;

[0181] The complexity quantification module is used to quantify the complexity, specifically, to quantify the complexity of the intrinsic mode functions in the updated intrinsic mode function set, and classify the intrinsic mode functions in the updated intrinsic mode function set into high-complexity intrinsic mode functions and low-complexity intrinsic mode functions according to the complexity, send the high-complexity intrinsic mode functions to the carbon emission prediction module, and send the low-complexity intrinsic mode functions to the optimal parameter calculation module and the carbon emission prediction module;

[0182] The optimal parameter calculation module is used to calculate the optimal parameters, specifically, to calculate the optimal weight and threshold parameters of the original extreme learning machine through the particle swarm optimization algorithm, to calculate the optimal matrix weight parameters of the original extreme learning machine through the slime mold algorithm and the low-complexity intrinsic mode function, and to optimize the parameters of the original extreme learning machine through the optimal weight and threshold parameters and the optimal matrix weight parameters to obtain the optimized original extreme learning machine, and to send the optimized original extreme learning machine to the carbon emission prediction module;

[0183] The carbon emission prediction module is used to predict carbon emissions. Specifically, the high-complexity intrinsic mode function and the low-complexity intrinsic mode function are input into the optimized original extreme learning machine for training to obtain the carbon emission prediction result.

[0184] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0185] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0186] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A carbon emissions monitoring method based on artificial intelligence, characterized in that: include: Step S1: data collection, specifically, collecting raw data sequences related to carbon emissions; Step S2: parameter optimization, specifically, optimizing the parameter K3 and parameter α of the empirical mode decomposition by using the Grey Wolf Optimization Algorithm to obtain an optimized empirical mode decomposition; Step S3: Data preprocessing, specifically, preprocessing the original data sequence to remove trend and periodic components to obtain a preprocessed original data sequence, and adding a Gaussian white noise sequence to the preprocessed original data sequence to obtain a data sequence after preprocessing and adding the Gaussian white noise sequence. The formula used is as follows: X i (t)=X(t)+βE(δ i (t)),i=1,2,...,K; In the formula, X(t) represents the original data sequence at time t, t represents the time variable, and X i (t) represents the data sequence after preprocessing and adding Gaussian white noise sequence at time t, β represents the Gaussian white noise weight coefficient, δ i (t) represents the i-th Gaussian white noise sequence at time t, E(δ i (t)) represents the i (t) the intrinsic mode function obtained by optimizing the empirical mode decomposition, K represents the number of Gaussian white noise sequences; Step S4: decomposing the data sequence, specifically, performing optimized empirical mode decomposition on the data sequence after preprocessing and adding the Gaussian white noise sequence, obtaining an intrinsic mode function set and a residual signal sequence, and updating the intrinsic mode function set to obtain an updated intrinsic mode function set; Step S5: quantifying the complexity, specifically, quantifying the complexity of the eigenmode functions in the updated eigenmode function set, and classifying the eigenmode functions in the updated eigenmode function set into eigenmode functions with high complexity and eigenmode functions with low complexity according to the complexity; Step S6: Calculating optimal parameters, specifically, calculating the optimal weight and threshold parameters of the original extreme learning machine by a particle swarm optimization algorithm, calculating the optimal matrix weight parameters of the original extreme learning machine by a slime mold algorithm and a low-complexity intrinsic mode function, and optimizing the parameters of the original extreme learning machine by the optimal weight and threshold parameters and the optimal matrix weight parameters to obtain an optimized original extreme learning machine; Step S7: predicting carbon emissions, specifically, inputting the high-complexity intrinsic mode function and the low-complexity intrinsic mode function into the optimized original extreme learning machine for training to obtain a carbon emissions prediction result; In step S2, the parameter optimization step includes: Step S21: Initialize the maximum number of optimization iterations, parameter α and parameter K3; Step S22: Calculate the decomposition fitness, the calculation formula of the decomposition fitness is: T n =-K t +H pe ; Where, T n represents the decomposition fitness, K t Denotes kurtosis, H pe represents the permutation entropy; Step S23: Update parameter α and parameter K3 according to the decomposition fitness; Step S24: iteratively execute steps S52 to S53. When the maximum number of optimization iterations is reached, the iteration is stopped, and the updated parameters K3 and α generated in the last iteration are added to the parameters K3 and α of the empirical mode decomposition to obtain an optimized empirical mode decomposition. In step S4, the step of decomposing the data sequence includes: Step S41: Calculate a first group of residual signals. The calculation formula for the first group of residual signals is: R1=N(X i (t)); Where R1 represents the first group of residual signals, X i (t) represents the data sequence after preprocessing and adding Gaussian white noise sequence at time t, N(X i (t)) represents X i The local mean of (t); Step S42: Calculate a first group of modal components. The calculation formula of the first group of modal components is: d1=X i (t)-R1; Where d1 represents the first group of modal components, R1 represents the first group of residual signals, and X i (t) represents the data sequence after preprocessing and adding Gaussian white noise sequence at time t; Step S43: Add the Gaussian white noise sequence again, and calculate the second group of residual signals by local mean, the calculation formula of the second group of residual signals is: R2=R1+R1E(δ i (t)); In the formula, R2 represents the second group of residual signals, R1 represents the first group of residual signals, and δ i (t) represents the i-th Gaussian white noise sequence at time t, E(δ i (t)) represents the i (t) the intrinsic mode function obtained by optimizing the empirical mode decomposition; Step S44: Calculate the second group of modal components. The calculation formula of the second group of modal components is: d2=R1-R2=R1-(N(R1+βE(δ i (t)))); In the formula, d2 represents the second group of modal components, R1 represents the first group of residual signals, R2 represents the second group of residual signals, β represents the Gaussian white noise weight coefficient, δ i (t) represents the i-th Gaussian white noise sequence at time t, E(δ i (t)) represents the i (t) The intrinsic mode function obtained by optimizing the empirical mode decomposition, N(R1+β1E(δ i (t))) represents R1+β1E(δ i (t)) local mean; Step S45: presetting the initial value of k to 3, and setting a deviation threshold; Step S46: Calculate the kth residual signal and the kth modal component, and go to step S47. The calculation formula of the kth residual signal and the kth modal component is: R k =N(R k-1 +βE(δ k (t))),k=3,4,...,K; d k =R k-1 -R k ,k=3,4,...,K; In the formula, R k represents the kth residual signal, dk represents the kth modal component, K represents the number of Gaussian white noise sequences, δ k (t) represents the i-th Gaussian white noise sequence at time t, E(δ k (t)) represents the k (t) is the intrinsic mode function obtained by optimizing the empirical mode decomposition, that is, the kth intrinsic mode function, β represents the Gaussian white noise weight coefficient, R k-1 represents the k-1th residual signal, N(R k-1 +β k-1 E(δ k (t))) represents R k-1 +β k-1 E(δ k (t)) local mean; Step S47: forming a residual signal sequence according to the residual signal, setting k=k+1, and judging whether the termination condition is satisfied. If the termination condition is satisfied, go to step S48; otherwise, go to step S46; The termination condition is that the residual signal sequence decreases monotonically and the standard deviation is less than the deviation threshold. The calculation formula of the standard deviation is: In the formula, σ represents the standard deviation, I k represents the kth eigenmode function, I k+1 represents the k+1th intrinsic mode function, ||I k ||2 represents the L2 norm of k intrinsic mode functions, ||I k+1 -I k ||2 represents the L2 norm of the difference between the k+1th eigenmode function and the kth eigenmode function; Step S48: presetting a mean square threshold, constructing an eigenmode function set according to the eigenmode function, and calculating the root mean square value of each eigenmode function, traversing all eigenmode functions, if the root mean square value of the eigenmode function is less than or equal to the mean square threshold, then removing the eigenmode function from the eigenmode function set, and adding the eigenmode function to the remaining items; Step S49: adding the remaining items as eigenmode functions to the eigenmode function set to obtain an updated eigenmode function set; In step S5, the step of quantifying the complexity includes: Step S51: presetting the embedding parameter t1 and the spatial offset w1 of the original data sequence; Step S52: traversing the eigenmode functions in the updated eigenmode function set, and constructing a recursive graph by calculating the Chebyshev distance of the eigenmode functions; Step S53: Analyze the recursive graph using the binary co-occurrence matrix to obtain the recursive mapping result Step S54: Increase the value of the embedding parameter t1, that is, t1=t1+1, and repeat steps S52 and S53 to obtain a new recursive mapping result. Step S55: Calculate the spatial correlation recursive sample entropy SdrSampEn(t1, W1) of each intrinsic mode function, and evaluate the complexity of each intrinsic mode function according to the spatial correlation recursive sample entropy SdrSampEn(t1, W1) of each intrinsic mode function. The calculation formula of the spatial correlation recursive sample entropy SdrSampEn(t1, W1) of each intrinsic mode function is: Where SdrSampEn(t1, W1) represents the spatial correlation recursive sample entropy of the intrinsic mode function, Represents the recursive mapping result, represents the new recursive mapping result, t1 represents the embedding parameter, and w1 represents the spatial offset of the original data sequence; Step S56: classifying each eigenmode function into an eigenmode function of high complexity and an eigenmode function of low complexity according to the complexity of each eigenmode function; In step S6, the step of calculating the optimal parameters includes: Step S6 1: Calculate the optimal weight and threshold parameters, specifically, calculate the optimal weight and threshold parameters of the original extreme learning machine by using a particle swarm optimization algorithm; Step S62: Calculate the optimal matrix weight parameters of the original extreme learning machine, specifically, calculate the optimal matrix weight parameters of the original extreme learning machine by using a slime mold algorithm and a low-complexity intrinsic mode function; Step S63: Optimizing the parameters of the original extreme learning machine according to the optimal weight and threshold parameters and the optimal matrix weight parameters to obtain an optimized original extreme learning machine; In step S61, the step of calculating the optimal weight and threshold parameters includes: Step S611: Parameter initialization, specifically, initializing the parameters of the particle swarm optimization algorithm and the original extreme learning machine, the parameters of the particle swarm optimization algorithm include the maximum number of iterations, the number of particles N′, the learning factors c1 and c2, and the inertia weight factor w2, and the parameters of the original extreme learning machine include the weight matrix from the input layer to the hidden layer and the weight matrix from the hidden layer to the output layer; Step S612: Generate an initial particle swarm, specifically, randomly generate the speed V' and position X' of N' particles, and initialize the optimal position P of each particle best and the global optimal position G best ; Step S613: Update the speed and position of each particle. The formula used is as follows: X′ i (t+1)=X′ i (t)+V′ i (t+1); In the formula, w2 represents the inertia weight factor, c1 and c2 represent the learning factor, P best represents the optimal position of each particle, G best represents the global optimal position, V′ i (t) represents the velocity of the ith particle at time t, V′ i (t+1) represents the speed of the ith particle at time t+1, that is, the updated speed of the ith particle, X′ i (t) represents the position of the ith particle at time t, X′ i (t+1) represents the position of the i-th particle at time t+1, that is, the position of the i+1-th particle after update, rand() represents a random number between 0 and 1, and * represents a multiplication operation; Step S614: Update the parameters of the original extreme learning machine. Specifically, set the updated position of each particle to the value of the weight matrix from the input layer to the hidden layer and the weight matrix from the hidden layer to the output layer, train the original extreme learning machine using the training data set, and calculate the performance index of each particle on the validation set. If the performance index of the particle on the validation set is better than the optimal position P of the particle, best , then the optimal position of the particle P best Update to the updated position of the particle, if the performance index of the particle on the validation set is better than the global best position G best , then the global optimal position G best Update to the updated position of the particle; Step S615: iteratively execute steps S613 to S614. When the maximum number of iterations is reached, the iteration process ends, and the values ​​of the weight matrix from the input layer to the hidden layer and the weight matrix from the hidden layer to the output layer at the last iteration are used as the optimal weight and threshold parameters; In step S62, the step of calculating the optimal matrix weight parameters of the original extreme learning machine includes: Step S621: Initializing the parameters of the slime mold algorithm, wherein the parameters of the slime mold algorithm include the number of low-complexity iterations and the weight coefficient W3 of the slime mold algorithm; Step S622: Initialize individual position X″ and individual velocity V″, wherein the individual position X″ is a matrix with M rows and N columns, and the element x″ in the individual position X″ is ij represents the position of the i-th individual in the j-th dimension. The individual velocity V″ is a matrix with M rows and N columns. The element v″ in the individual velocity V″ ij represents the velocity of the i-th individual in the j-th dimension; Step S623: Calculate the fitness value of the individual position X″; Step S624: Update the individual position X″ and the individual speed V″ to obtain an updated individual position X″ and an updated individual speed V″. The calculation formula for updating the individual position X″ and the individual speed V″ is: x″ ij (t+1)=x″ ij (t)+v″ ij (t+1); v″ ij (t+1)=w3*v″ ij (t)+c3*r3*(pbest ij (t)-x″ ij (t))+c4*r4*(gbest ij (t)-xii″t; In the formula, x″ ij (t) represents the position of the i-th individual in the j-th dimension at time t, x″ ij (t+1) represents the position of the i-th individual in the j-th dimension at time t+1, that is, the position of the i-th individual in the j-th dimension after update, v″ ij (t) represents the velocity of the i-th individual in the j-th dimension at time t, v″ ij (t+1) represents the speed of the i-th individual in the j-th dimension at time t+1, w3 represents the weight coefficient of the slime mold algorithm, c3 and c4 represent learning factors, r3 and r4 represent random numbers between 0 and 1, and pbest ij (t) represents the optimal position of the i-th individual in the j-th dimension at time t, gbest ij (t) represents the global optimal position of the i-th individual in the j-th dimension at time t; Step S625: Calculate the fitness value of the updated individual position X″ according to the updated individual position X″; Step S626: iteratively execute steps S624 to S625. After reaching the low-complexity iteration number, the iteration process ends, and the optimal individual position X″ is selected as the optimal matrix weight parameter of the original extreme learning machine according to the fitness value.

2. An artificial intelligence-based carbon emissions monitoring system, used to implement the artificial intelligence-based carbon emissions monitoring method as claimed in claim 1, characterized in that: It includes data acquisition module, data preprocessing module, sequence decomposition module, complexity quantification module, parameter optimization module, optimal parameter calculation module and carbon emission prediction module; The data acquisition module is used for data acquisition, specifically, to acquire the original data sequence and send the original data sequence to the data preprocessing module; The parameter optimization module is used for parameter optimization, specifically, optimizing the parameter K3 and parameter α of the empirical mode decomposition by using the Grey Wolf Optimization Algorithm to obtain an optimized empirical mode decomposition; The data preprocessing module is used for data preprocessing, specifically, preprocessing the original data sequence to remove trend and periodic components to obtain a preprocessed original data sequence, and adding a Gaussian white noise sequence to the preprocessed original data sequence to obtain a data sequence after preprocessing and adding the Gaussian white noise sequence, and sending the data sequence after preprocessing and adding the Gaussian white noise sequence to the sequence decomposition module; The sequence decomposition module is used to decompose the data sequence, specifically, to perform an optimized empirical mode decomposition on the data sequence after preprocessing and adding a Gaussian white noise sequence, to obtain an intrinsic mode function set and a residual signal sequence, and to update the intrinsic mode function set to obtain an updated intrinsic mode function set, and to send the updated intrinsic mode function set to the complexity quantification module; The complexity quantification module is used to quantify the complexity, specifically, to quantify the complexity of the intrinsic mode functions in the updated intrinsic mode function set, and classify the intrinsic mode functions in the updated intrinsic mode function set into high-complexity intrinsic mode functions and low-complexity intrinsic mode functions according to the complexity, send the high-complexity intrinsic mode functions to the carbon emission prediction module, and send the low-complexity intrinsic mode functions to the optimal parameter calculation module and the carbon emission prediction module; The optimal parameter calculation module is used to calculate the optimal parameters, specifically, to calculate the optimal weight and threshold parameters of the original extreme learning machine through the particle swarm optimization algorithm, to calculate the optimal matrix weight parameters of the original extreme learning machine through the slime mold algorithm and the low-complexity intrinsic mode function, and to optimize the parameters of the original extreme learning machine through the optimal weight and threshold parameters and the optimal matrix weight parameters to obtain the optimized original extreme learning machine, and to send the optimized original extreme learning machine to the carbon emission prediction module; The carbon emission prediction module is used to predict carbon emissions. Specifically, the high-complexity intrinsic mode function and the low-complexity intrinsic mode function are input into the optimized original extreme learning machine for training to obtain the carbon emission prediction result.

Images