A carbon emission prediction method and system based on improved long short-term memory network

Abstract

The present invention relates to the technical field of energy consumption prediction, and in particular to a carbon emission prediction method and system based on an improved long short-term memory network, the method comprising: obtaining training data of a node, the training data comprising a historical load sequence and a corresponding carbon emission factor; taking a hyperparameter of an LSTM network as a variable for Bayesian optimization, training the hyperparameter of the LSTM network according to the training data, and updating the hyperparameter of the LSTM network in combination with loss function evaluation and back propagation; determining a sampling point for the next iteration based on Gaussian process regression, and iteratively training the hyperparameters of the LSTM network to obtain a trained LSTM network; obtaining a real-time load sequence, and using the trained LSTM network to online predict a carbon emission factor corresponding to the real-time load sequence; the present invention can quickly predict the carbon emission factor.

Description

Technical Field

[0001] The present invention relates to the technical field of energy consumption prediction, and in particular to a carbon emission prediction method and system based on an improved long short-term memory network. Background Art

[0002] At present, there are many related studies devoted to the low-carbon operation of power systems, including environmental economic scheduling, optimal power flow considering carbon emissions of units, etc. However, the carbon emissions focused on by existing studies are mainly to reduce the total carbon emissions of power generation companies, that is, the power generation side, including the use of new energy such as wind power generation, solar power generation, and hydropower generation, and reducing traditional thermal power generation based on coal-fired technology.

[0003] However, the existing technology is mainly based on the carbon flow calculation model of power flow analysis, which requires the collection of information from each node for centralized power flow calculation. In addition, the calculation time of the system power flow and carbon flow is relatively long, and it needs to rely on the grid and topology information of the power system and high-precision parameters. With the complexity of the power system structure, when the load and power generation fluctuate violently, there are situations where information collection is unreliable and the calculation complexity is high, and it is impossible to use historical data to help the calculation and analysis of carbon flow and power flow. Summary of the invention

[0004] The present invention aims to provide a carbon emission prediction method and system based on an improved long short-term memory network, which can quickly predict carbon emission factors.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] In a first aspect, an embodiment of the present invention provides a carbon emission prediction method based on an improved long short-term memory network, the method comprising the following steps:

[0007] S100, obtaining training data of a node, wherein the training data includes a historical load sequence and a corresponding carbon emission factor;

[0008] S200, taking the hyperparameters of the LSTM network as variables for Bayesian optimization, training the hyperparameters of the LSTM network according to the training data, and updating the hyperparameters of the LSTM network in combination with loss function evaluation and back propagation;

[0009] S300, determining the sampling point of the next iteration based on Gaussian process regression, and iteratively training the hyperparameters of the LSTM network to obtain a trained LSTM network;

[0010] S400, obtaining a real-time load sequence, and using a trained LSTM network to online predict a carbon emission factor corresponding to the real-time load sequence.

[0011] Optionally, in S100, the acquiring node training data, the training data including historical load sequences and corresponding carbon emission factors, includes:

[0012] Get historical load series Among them, n s is the number of data samples;

[0013] The power flow equation is used to calculate and output the corresponding active energy, and then the carbon flow is calculated to obtain the corresponding carbon emission factor;

[0014] The historical load sequence is used as input data, the corresponding carbon emission factor is used as output data, and the input data and output data are used as training data.

[0015] Optionally, the power flow equation of a node in the power system is:

[0016]

[0017]

[0018] θ ij =θ i -θ j ;

[0019] Among them, N g is the total number of generators in the power system, N n is the total number of nodes in the power system, is the active power injected by the kth generator into the i-th node, is the reactive power injected by the kth generator to the ith node; LP i is the active load of the ith node, LQ i is the reactive load of the ith node; G ij is the conductance corresponding to the circuit ij, B ij is the susceptance corresponding to line ij; V i is the node voltage of the ith node, V j is the node voltage of the jth node; θ ij represents the voltage angle difference between node i and node j; θ i is the voltage angle at node i, θ j is the voltage angle at node j.

[0020] The constraint equation for the power flow is:

[0021]

[0022]

[0023]

[0024] in, are the minimum active power regulation and the maximum active power regulation of the j-th generator respectively; are the minimum reactive power regulation and the maximum reactive power regulation of the j-th generator respectively; are the minimum output voltage and maximum output voltage of the j-th generator respectively.

[0025] Optionally, the calculation formula of the carbon emission factor is:

[0026]

[0027] The carbon flow equation is:

[0028]

[0029] Among them, δ i (i=1,2,…N n ) is the solution set of the carbon flow equation, with a total of N n equations to be solved; is the diagonal component of the carbon flow equation, corresponding to the total power injected into the i-th node; is the active power injected from the jth node to the ith node along the ji line; g k is the carbon emission coefficient of the kth unit.

[0030] Optionally, the gating mechanism of the LSTM network includes a forget gate, an input gate, and an output gate. The expressions of the forget gate, the input gate, and the output gate in the LSTM network are:

[0031]

[0032]

[0033]

[0034]

[0035] Among them, sigmoid(*) represents the sigmoid operation, x t represents the feature input of the t-th time dimension, h t-1 represents the feature output of the (t-1)th time dimension, f t 、i t , g t and t Respectively represent the forgotten state, input state, new vector and output state of the t-th time dimension, w f 、w i 、w gand w o Represent the weight parameters of the LSTM network layer, b f 、b i 、b g and b o represents the bias parameter of the LSTM network layer, Represents the aggregation operation h of the last time series feature t-1 and the current time value x t .

[0036] The output data of the LSTM network is:

[0037] c t =f t ·c t-1 +i t ·g t ;

[0038] h t =o t ·σ(c t );

[0039] h t =LSTM t (h t-1 ,x t ,c t-1 ,W t ,B t );

[0040] Among them, σ represents the tanh operation, c t represents the memory state of the t-th time dimension, W t is the weight parameter set of the LSTM network, B t Represents the bias parameter set of the LSTM network, LSTM t Represents the t-th operation of the LSTM network.

[0041] Optionally, in S300, determining the sampling point of the next iteration based on Gaussian process regression, and iteratively training the hyperparameters of the LSTM network to obtain a trained LSTM network includes:

[0042] S310, selecting an initial learning rate and a learning rate reduction coefficient as search hyperparameters;

[0043] S320, selecting validation set data from the training data, selecting the mean absolute error function of the validation set data as the optimization objective function, and performing LSTM network training and loss function verification;

[0044] S330, constructing an acquisition function using Gaussian process regression to obtain sampling points for the next iteration;

[0045] S340, iterate and repeat S310 to S330 until the maximum iteration requirement is met.

[0046] Optionally, the hyperparameter combination of the LSTM network after Bayesian optimization is:

[0047]

[0048]

[0049] in, is the hyperparameter combination after Bayesian optimization; X represents the optimization space corresponding to the hyperparameter combination; f(x) represents the optimized objective function; y ij is the original data of the i-th dimension of the j-th sample, is the original data of the i-th dimension of the j-th sample.

[0050] Optionally, the output data is calculated as:

[0051]

[0052] in, is the output data predicted by the trained LSTM network; W is the total weight parameter setting; B is the total bias parameter setting; It is the input time series, including the load amount and carbon emission factor of the load node.

[0053] In a second aspect, an embodiment of the present invention provides a carbon emission prediction system based on an improved long short-term memory network, the system comprising:

[0054] at least one processor;

[0055] at least one memory for storing at least one program;

[0056] When the at least one program is executed by the at least one processor, the at least one processor implements the carbon emission prediction method based on the improved long short-term memory network as described in any one of the above items.

[0057] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium storing a program executable by a processor, wherein the program executable by the processor is used to execute a carbon emission prediction method based on an improved long short-term memory network as described in any one of the above when executed by the processor.

[0058] The beneficial effects of the present invention are as follows: the present invention discloses a carbon emission prediction method and system based on an improved long short-term memory network. The present invention uses an LSTM network to solve the carbon flow equation, which can improve the accuracy of carbon emission prediction and reduce the computational complexity. By using a Bayesian optimization algorithm to optimize the hyperparameters of the LSTM network, the prediction accuracy can be more effectively improved and the training time can be reduced. The method of the embodiment of the present invention can be applied to scenarios such as carbon emission prediction and formulation of carbon emission reduction measures to realize the function of automatic and rapid prediction of carbon emissions. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0060] Figure 1 It is a flow chart of a carbon emission prediction method based on an improved long short-term memory network in an embodiment of the present invention;

[0061] Figure 2 It is a structural schematic diagram of a carbon emission prediction system based on an improved long short-term memory network in an embodiment of the present invention. DETAILED DESCRIPTION

[0062] The following will be combined with the embodiments and drawings to clearly and completely describe the concept, specific structure and technical effects of the present invention, so as to fully understand the purpose, scheme and effect of the present invention. It should be noted that the embodiments and features in the embodiments of the present invention can be combined with each other without conflict.

[0063] See also Figure 1 , Figure 1 The present invention provides a carbon emission prediction method based on an improved long short-term memory network, the method comprising the following steps:

[0064] S100, obtaining training data of a node, wherein the training data includes a historical load sequence and a corresponding carbon emission factor;

[0065] S200, taking the hyperparameters of the LSTM network as variables for Bayesian optimization, training the hyperparameters of the LSTM network according to the training data, and updating the hyperparameters of the LSTM network in combination with loss function evaluation and back propagation;

[0066] S300, determining the sampling point of the next iteration based on Gaussian process regression, and iteratively training the hyperparameters of the LSTM network to obtain a trained LSTM network;

[0067] S400, obtaining a real-time load sequence, and using a trained LSTM network to online predict a carbon emission factor corresponding to the real-time load sequence.

[0068] It should be noted that the present invention is for the prediction of carbon flow based on time series prediction, and the prediction process is divided into four steps. Taking the PQ node as an example, the PQ node refers to the node with given active power P and reactive power Q, and the node voltage and phase are the quantities to be determined. Specifically, first collect the historical load sequence of the node as input data, use the Newton method to calculate the flow, then perform carbon flow analysis, and collect the carbon emission factors corresponding to the historical load sequence of the node in the power system as output data, then select the hyperparameters of the LSTM network as the variables for Bayesian optimization, and then train the parameters of the LSTM network according to the collected training data, and update the LSTM network parameters in combination with loss function evaluation and back propagation; then, use Gaussian process regression to construct the set function of the next sampling point, and repeat the hyperparameter optimization iteration and network parameter training; finally, according to the real-time load sequence, use the optimized LSTM network to predict the carbon emission coefficient online, so as to achieve rapid prediction of the carbon emission coefficient of each node.

[0069] The present invention aims at the current "dual carbon" problem, in order to quickly predict the real-time carbon emissions of nodes, reduce computing costs, improve computing stability and improve the generalization ability of the model, a long and short-term memory network based on Bayesian optimization (BayesianOptimization based-long and short term model, BO-LSTM network) is applied to predict the carbon emissions of the power system. The present invention provides a scheme for applying a long and short-term memory network based on Bayesian optimization to predict the carbon flow of power grids. For the power system network, when the Bo-LSTM network predicts the carbon emissions of user nodes in a large-scale power system, by analyzing the historical data of the source-grid-load parameters and carbon emission factors of the power system, the high-dimensional electrical quantity characteristics of the power system in time and space can be effectively extracted, and the nonlinear characterization of each quantity can be performed. Not only that, the BO-LSTM network can also meet a certain calculation accuracy while quickly predicting the carbon emissions of the power system. Therefore, the present invention applies a long and short-term memory network based on Bayesian optimization to the power system network, which can quickly predict the carbon emission factors of the load nodes of the power system.

[0070] In some preferred embodiments, in S100, the acquiring of node training data, the training data including historical load sequences and corresponding carbon emission factors, includes:

[0071] Get historical load series Among them, n s is the number of data samples;

[0072] The power flow equation is used to calculate and output the corresponding active energy, and then the carbon flow is calculated to obtain the corresponding carbon emission factor;

[0073] The historical load sequence is used as input data, the corresponding carbon emission factor is used as output data, and the input data and output data are used as training data.

[0074] It should be noted that, in some embodiments, obtaining the load sequence After that, the power flow equation is first used to calculate and output the corresponding active power, and then the carbon flow is calculated. Then, the input data and output data are selected for network training. s is the number of data samples.

[0075] In the present invention, the load sequence is used as input data, the carbon emission factor is used as output data, and the carbon emission factor includes the carbon emission factor of the load node.

[0076] Specifically, the collected load sequence is Among them, F d is the total sample set of the total load sequence, F i represents the load sequence sample at the i-th moment, f i is the load at the ith moment, n k is the length of a single load sequence of a node. For output data, the collected carbon emission factors include the carbon emission factors of the load nodes δ o ={δ p (1),δ p (2),…,δ p (n s )}, Among them, δo is the total output sample of carbon emission factor, δ p represents a sample carbon emission factor, Represents the carbon emission factor of the i-th load node.

[0077] It should be noted that for the calculation of power flow, the power flow of the power system is related to the voltage distribution. g generators and N n In a power system with multiple nodes, the power output of the nodes is divided into active output and reactive output.

[0078] In some preferred embodiments, the power flow equation of a node in the power system is:

[0079]

[0080]

[0081] θ ij =θi -θ j ;

[0082] Among them, N g is the total number of generators in the power system, N n is the total number of nodes in the power system, is the active power injected by the kth generator into the i-th node, is the reactive power injected by the kth generator to the ith node; LP i is the active load of the ith node, LQ i is the reactive load of the ith node; G ij is the conductance corresponding to the circuit ij, B ij is the susceptance corresponding to line ij; V i is the node voltage of the ith node, V j is the node voltage of the jth node; θ ij represents the voltage angle difference between node i and node j; θ i is the voltage angle at node i, θ j is the voltage angle at node j.

[0083] It should be noted that in order to ensure the stability of generator power supply, the calculation of power flow needs to take into account the power flow constraints of the generator, including active power constraints, reactive power constraints and voltage constraints.

[0084] In some preferred embodiments, the constraint equation for the power flow is:

[0085]

[0086]

[0087]

[0088] in, are the minimum active power regulation and the maximum active power regulation of the j-th generator respectively; are the minimum reactive power regulation and the maximum reactive power regulation of the j-th generator respectively; are the minimum output voltage and maximum output voltage of the j-th generator respectively.

[0089] It should be noted that for carbon flow analysis, the carbon emission factor of the node in the power system is related to the active power of the input source and the carbon emission factor of the corresponding input source node. The carbon emission factor can be calculated based on the above power flow results, and the reactive power flow data does not need to be considered in the calculation.

[0090] In some preferred embodiments, the calculation formula of the carbon emission factor is:

[0091]

[0092] The carbon flow equation is:

[0093]

[0094] Among them, δ i (i=1,2,…N n ) is the solution set of the carbon flow equation, with a total of N n equations to be solved; is the diagonal component of the carbon flow equation, corresponding to the total power injected into the i-th node; is the active power injected from the jth node to the ith node along the ji line; g k is the carbon emission coefficient of the kth unit.

[0095] In some preferred embodiments, the gating mechanism of the LSTM network includes a forget gate, an input gate, and an output gate. The expressions of the forget gate, the input gate, and the output gate in the LSTM network are:

[0096]

[0097]

[0098]

[0099]

[0100] Among them, sigmoid(*) represents the sigmoid operation, x t represents the feature input of the t-th time dimension, h t-1 represents the feature output of the (t-1)th time dimension, f t 、i t , g t and t Respectively represent the forgotten state, input state, new vector and output state of the t-th time dimension, w f 、w i 、w g and w o Represent the weight parameters of the LSTM network layer, b f 、b i 、b g and b o represents the bias parameter of the LSTM network layer, Represents the aggregation operation h of the last time series feature t-1 and the current time value x t .

[0101] The output data of the LSTM network is:

[0102] c t =f t ·c t-1 +i t ·g t ;

[0103] h t =o t ·σ(c t );

[0104] h t =LSTM t (h t-1 ,x t ,c t-1 ,W t ,B t );

[0105] Among them, σ represents the tanh operation, c t represents the memory state of the t-th time dimension, W t is the weight parameter set of the LSTM network, B t Represents the bias parameter set of the LSTM network, LSTM t Represents the t-th operation of the LSTM network.

[0106] It should be noted that compared with the traditional RNN network, the LSTM network uses three gates for information selection, which are the forget gate, input gate and output gate. The forget gate, input gate and output gate are three gating mechanisms in the LSTM network, which are used to control the input, output and forget the previous information. Among them, the forget gate is used to determine the degree to which the previously input information is forgotten, the input gate is used to control the degree to which the new input information is added, and the output gate is used to control the degree to which the current state is output. With the addition of these gates, the LSTM network is more capable of adjusting the state of the unit through historical memory to prevent the gradient from disappearing. Based on the calculation of the above gate values, the output data can be given by the state of the last unit and the above output state, according to the proposed memory state and the operation of the gate calculation.

[0107] In some preferred embodiments, in S300, determining the sampling point of the next iteration based on Gaussian process regression, and iteratively training the hyperparameters of the LSTM network to obtain a trained LSTM network includes:

[0108] S310, selecting an initial learning rate and a learning rate reduction coefficient as search hyperparameters;

[0109] S320, selecting validation set data from the training data, selecting the mean absolute error function of the validation set data as the optimization objective function, and performing LSTM network training and loss function verification;

[0110] S330, constructing an acquisition function using Gaussian process regression to obtain sampling points for the next iteration;

[0111] S340, iterate and repeat S310 to S330 until the maximum iteration requirement is met.

[0112] In some preferred embodiments, the hyperparameter combination of the LSTM network after Bayesian optimization is:

[0113]

[0114]

[0115] in, is the hyperparameter combination after Bayesian optimization; X represents the optimization space corresponding to the hyperparameter combination; f(x) represents the optimized objective function; y ij is the original data of the i-th dimension of the j-th sample, is the original data of the i-th dimension of the j-th sample.

[0116] The formula in this embodiment gives the optimization operation and search process of the hyperparameters in the LSTM network. The prediction of the system dynamic carbon emission factor can be quickly calculated by the optimized network according to the dynamic load sequence and the optimal network parameters.

[0117] In some preferred embodiments, the calculation formula for output data is:

[0118]

[0119] in, is the output data predicted by the trained LSTM network; W is the total weight parameter setting; B is the total bias parameter setting; It is the input time series, including the load amount and carbon emission factor of the load node.

[0120] Aiming at the problem of rapid prediction of carbon emission factors of user nodes in power systems, the present invention proposes a long short-term memory network prediction algorithm based on Bayesian optimization, uses Bayesian optimization to optimize network hyperparameters, and combines historical load and carbon emission factor data for network training, so that only real-time load information and optimal network parameters are needed in the carbon emission factor prediction stage. The algorithm has the advantages of low information collection cost, low calculation cost, high prediction accuracy, and strong model generalization ability. The four major characteristics of the BO-LSTM network algorithm are as follows:

[0121] (1) Through the two stages of offline network training and online network application, a faster carbon emission factor calculation process is achieved, which avoids the loss of some node information during the centralized calculation process and makes the carbon emission calculation results more stable.

[0122] (2) This method does not need to consider the parameters of the actual power system, but only needs to collect relevant historical data, which greatly reduces the cost of information collection.

[0123] (3) By applying Bayesian optimization to the network's hyperparameter optimization operation, the prediction accuracy and generalization ability of the LSTM network are improved.

[0124] (4) By mining the information of historical data of the power system, the real-time calculation of carbon emission factors is realized, which can provide a basis for energy conservation and emission reduction, and also meet the power system's monitoring needs for carbon emission factors of user nodes.

[0125] and Figure 1 Corresponding to the method, refer to Figure 2 The embodiment of the present invention provides a carbon emission prediction system based on an improved long short-term memory network, including:

[0126] at least one processor;

[0127] at least one memory for storing at least one program;

[0128] When the at least one program is executed by the at least one processor, the at least one processor implements the above method.

[0129] It can be seen that the contents of the above method embodiments are all applicable to the present system embodiments, the functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0130] In addition, an embodiment of the present invention further discloses a computer program product or a computer program, which is stored in a computer-readable storage medium. A processor of a computer device can read the computer program from a computer-readable storage medium, and the processor executes the computer program, so that the computer device executes the above method. Similarly, the contents of the above method embodiment are all applicable to the storage medium embodiment, and the functions specifically implemented by the storage medium embodiment are the same as those of the above method embodiment, and the beneficial effects achieved are also the same as those achieved by the above method embodiment.

[0131] It will be appreciated by those skilled in the art that all or some of the methods disclosed above, the system can be implemented as software, firmware, hardware and appropriate combinations thereof. Some physical components or all physical components can be implemented as software executed by a processor, such as a central processing unit, a digital signal processor or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, and the computer-readable medium can include a computer storage medium (or a non-transitory medium) and a communication medium (or a temporary medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0132] The above is a specific description of the preferred implementation of the present disclosure, but the present disclosure is not limited to the above-mentioned implementation mode. Technical personnel familiar with the field can also make various equivalent deformations or substitutions without violating the spirit of the present disclosure. These equivalent deformations or substitutions are all included in the scope defined by the claims of the present disclosure.

Claims

1. A carbon emission prediction method based on improved long short-term memory network, characterized in that: The method comprises the following steps: S100, obtaining training data of a node, wherein the training data includes a historical load sequence and a corresponding carbon emission factor; S200, taking the hyperparameters of the LSTM network as variables for Bayesian optimization, training the hyperparameters of the LSTM network according to the training data, and updating the hyperparameters of the LSTM network in combination with loss function evaluation and back propagation; S300, determining the sampling point of the next iteration based on Gaussian process regression, and iteratively training the hyperparameters of the LSTM network to obtain a trained LSTM network; S400, obtaining a real-time load sequence, and using a trained LSTM network to online predict a carbon emission factor corresponding to the real-time load sequence; In S100, the node training data is obtained, and the training data includes a historical load sequence and a corresponding carbon emission factor, including: Get historical load series Among them, n s is the number of data samples; The power flow equation is used to calculate and output the corresponding active energy, and then the carbon flow is calculated to obtain the corresponding carbon emission factor; Using the historical load sequence as input data, using the corresponding carbon emission factor as output data, and using the input data and output data as training data; The power flow equation of the node in the power system is: i ij =θ i -θ j ; Among them, N g is the total number of generators in the power system, N n is the total number of nodes in the power system, is the active power injected by the kth generator into the i-th node, is the reactive power injected by the kth generator to the ith node; LP i is the active load of the ith node, LQ i is the reactive load of the ith node; G ij is the conductance corresponding to the circuit ij, B ij is the susceptance corresponding to line ij; V i is the node voltage of the ith node, V j is the node voltage of the jth node; θ ij represents the voltage angle difference between node i and node j; θ i is the voltage angle at node i, θ j is the voltage angle of node j; The constraint equation for the power flow is: in, are the minimum active power regulation and the maximum active power regulation of the j-th generator respectively; are the minimum reactive power regulation and the maximum reactive power regulation of the j-th generator respectively; are the minimum output voltage and maximum output voltage of the j-th generator respectively; The calculation formula of the carbon emission factor is: The carbon flow equation is: Among them, δ i (i=1,2,…N n ) is the solution set of the carbon flow equation, with a total of N n equations to be solved; is the diagonal component of the carbon flow equation, corresponding to the total power injected into the i-th node; is the active power injected from the jth node to the ith node along the ji line; gk is the carbon emission coefficient of the kth unit; The gating mechanism of the LSTM network includes a forget gate, an input gate, and an output gate. The expressions of the forget gate, the input gate, and the output gate in the LSTM network are: Among them, sigmoid(*) represents the sigmoid operation, x t represents the feature input of the t-th time dimension, h t-1 represents the feature output of the (t-1)th time dimension, f t 、i t , g t and t Respectively represent the forgotten state, input state, new vector and output state of the t-th time dimension, w f 、w i 、w g and w o Represent the weight parameters of the LSTM network layer, b f 、b i 、b g and b o represents the bias parameter of the LSTM network layer, Represents the aggregation operation h of the last time series feature t-1 and the current time value x t ; The output data of the LSTM network is: c t =f t ·c t-1 +i t ·g t ; h t =o t ·σ(c t ); h t =LSTM t (h t-1 ,x t ,c t-1 ,W t ,B t ); Among them, σ represents the tanh operation, c t represents the memory state of the t-th time dimension, W t is the weight parameter set of the LSTM network, B t Represents the bias parameter set of the LSTM network, LSTM t Represents the t-th operation of the LSTM network.

2. According to claim 1, a carbon emission prediction method based on improved long short-term memory network is characterized in that: In S300, the step of determining the sampling point of the next iteration based on Gaussian process regression and iteratively training the hyperparameters of the LSTM network to obtain a trained LSTM network includes: S310, selecting an initial learning rate and a learning rate reduction coefficient as search hyperparameters; S320, selecting validation set data from the training data, selecting the mean absolute error function of the validation set data as the optimization objective function, and performing LSTM network training and loss function verification; S330, constructing an acquisition function using Gaussian process regression to obtain sampling points for the next iteration; S340, iterate and repeat S310 to S330 until the maximum iteration requirement is met.

3. The carbon emission prediction method based on improved long short-term memory network according to claim 2 is characterized in that: The hyperparameter combination of the LSTM network after Bayesian optimization is: in, is the hyperparameter combination after Bayesian optimization; X represents the optimization space corresponding to the hyperparameter combination; f(x) represents the optimized objective function; y ij is the original data of the i-th dimension of the j-th sample, is the original data of the i-th dimension of the j-th sample.

4. The carbon emission prediction method based on improved long short-term memory network according to claim 3 is characterized in that: The calculation formula for the output data is: in, is the output data predicted by the trained LSTM network; W is the total weight parameter setting; B is the total bias parameter setting; It is the input time series, including the load amount and carbon emission factor of the load node.

5. A carbon emission prediction system based on improved long short-term memory network, characterized in that: The system comprises: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the carbon emission prediction method based on the improved long short-term memory network as described in any one of claims 1 to 4.

6. A computer-readable storage medium storing a program executable by a processor, characterized in that: The processor-executable program is used to perform the method according to any one of claims 1 to 4 when executed by the processor.

Images