Electric carbon metering system and method based on neural network

Abstract

The invention discloses an electrical carbon metering system and method based on a neural network, and belongs to the technical field of electrical carbon metering. The system comprises a multi-modal data acquisition module, a data preprocessing module, a feature extraction and dimension reduction module, a modal potential energy calculation module, a modal tension matrix construction module, a modal neural state updating module, a fusion module and an electric carbon quantification module. According to the method, a multi-modal feature vector sequence, modal potential energy and inter-modal tension are obtained based on original multi-modal data, the neural state of the modal is updated, and the total carbon emission is obtained through a fusion state output mechanism. According to the method, the generalization ability and prediction stability of the model in a complex power scene can be enhanced, and the requirements of real-time control on robustness and reliability are met.

Description

Technical Field

[0001] This invention belongs to the field of electric carbon metering technology, and particularly relates to an electric carbon metering system and method based on neural networks. Background Technology

[0002] The power industry is a key sector for energy consumption and carbon emissions. In achieving the "dual-carbon" strategic goals, building a precise, efficient, and intelligent carbon emission measurement and management system is of significant practical importance. Currently, electricity carbon measurement not only needs to support carbon asset accounting and trading pricing, but also needs to provide real-time and reliable decision-making support for grid operation, energy efficiency optimization, and emission reduction control. Therefore, higher demands are placed on the accuracy, timeliness, and environmental adaptability of measurement models.

[0003] Traditional carbon emission measurement methods often rely on static parameter models or estimation methods based on fixed emission factors, such as calculating by simply multiplying electricity consumption by a unit emission factor. While these methods are easy to implement, their fundamental limitation lies in statically linearizing a dynamic nonlinear system, neglecting the complex impact of multimodal factors on carbon emissions during electricity consumption, such as differences in equipment type, load fluctuation characteristics, and power quality disturbances.

[0004] Furthermore, to improve time series prediction capabilities, existing technologies have introduced time series machine learning models such as LSTM and GRU. However, these models typically rely on fixed time evolution paths, and their core problem lies in the lack of a mechanism that can explicitly quantify the interactions between multimodal features and dynamically guide information propagation within the neural network. They struggle to effectively capture the interaction structure and dynamic coupling relationships between different modal features, resulting in the input features being largely treated as independent variables when dealing with complex systems, and the information propagation path within the network being predefined and rigid. In practical applications, especially when facing high-frequency interference and multimodal collaboration in complex power systems, this structural deficiency directly manifests as feature redundancy, insufficient generalization ability, and poor prediction stability. Simultaneously, the lack of an adaptive feedback mechanism based on modal relationships makes carbon emission prediction prone to distortion, failing to meet the robustness and reliability requirements of integrated real-time decision-making and intelligent control engineering applications. Summary of the Invention

[0005] This invention proposes a neural network-based carbon metering system and method. Its purpose is to enhance the generalization ability and predictive stability of the model in complex power scenarios, and to establish a mechanism that can perform adaptive feedback based on the dynamic relationship between modes, so as to meet the requirements of robustness and reliability of real-time control.

[0006] The technical solution of this invention is as follows:

[0007] A neural network-based electrocarbon metering system includes a multimodal data acquisition module, a data preprocessing module, a feature extraction and dimensionality reduction module, a modal potential energy calculation module, a modal tension matrix construction module, a modal neural state update module, a fusion module, and an electrocarbon quantization module.

[0008] The raw multimodal data acquired by the multimodal data acquisition module is sequentially processed by the data preprocessing module and the feature extraction and dimensionality reduction module to obtain the modal feature vector sequence. The modal feature vector sequence is processed by the modal potential energy calculation module, the modal tension matrix construction module and the modal neural state update module to obtain the modal potential energy, intermodal tension and neural state vector. The modal potential energy, intermodal tension and neural state vector are fused by the fusion module to obtain the total carbon emissions. The total carbon emissions are input to the electrocarbon quantification module to obtain the electrocarbon measurement result.

[0009] As a further improvement to the aforementioned neural network-based electric carbon metering system:

[0010] A multimodal data acquisition module is used to acquire raw multimodal data through power acquisition devices and status monitoring sensors;

[0011] The data preprocessing module is used to preprocess the raw multimodal data to obtain preprocessed multimodal data, which is then sent to the feature extraction and dimensionality reduction module.

[0012] The feature extraction and dimensionality reduction module is used to extract features from the preprocessed multimodal data to obtain preliminary feature data. Then, the extracted preliminary feature data is dimensionality reduced to obtain a multimodal feature vector sequence.

[0013] As a further improvement to the aforementioned neural network-based carbon metering system: a mode potential energy calculation module is used to calculate the mode potential energy value of each mode based on the multimodal feature vector sequence to quantify the degree of external modal interference currently experienced by each mode, thereby obtaining the mode potential energy.

[0014] As a further improvement to the aforementioned neural network-based electric carbon metering system: a modal tension matrix construction module is used to use modal potential energy as the original state representation of neural network nodes, and introduces a neural connection mechanism to use the intermodal tension calculated based on the modal potential energy as the connection weight.

[0015] As a further improvement to the aforementioned neural network-based electric carbon metering system, a modal neural state update module is used to update the neural state of the modes based on the multimodal feature vector sequence, combined with modal potential energy and intermodal tension, by introducing a modal neural state update formula to obtain the updated neural state vector.

[0016] As a further improvement to the neural network-based electric carbon metering system, a fusion module is used to construct a fusion state output mechanism based on the updated neural state vector, combining intermodal tension and modal potential energy, to obtain the total carbon emissions.

[0017] This invention also provides a method for measuring carbon electricity based on neural networks, comprising the following steps:

[0018] S1. Obtain the raw multimodal data. After preprocessing, perform feature extraction and dimensionality reduction on the preprocessed multimodal data to obtain a multimodal feature vector sequence. Calculate the modal potential energy based on the multimodal feature vector sequence and use the modal potential energy as the original state representation of the neural network nodes. At the same time, introduce a neural connection mechanism and use the intermodal tension calculated based on the modal potential energy as the connection weight.

[0019] S2. Based on the multimodal feature vector sequence, combined with modal potential energy and intermodal tension, a modal neural state update formula is introduced to update the neural state of the modality and obtain the updated neural state vector; further, a fusion state output mechanism is constructed to obtain the total carbon emissions, and then combined with the total electricity consumption to obtain the carbon emission measurement result.

[0020] As a further improvement to the neural network-based electrocarbon metering method, step S1 specifically includes:

[0021] Step S1-1. Acquire raw multimodal data through power acquisition devices and condition monitoring sensors, etc.

[0022] Step S1-2. Preprocess the original multimodal data to obtain preprocessed multimodal data;

[0023] Step S1-3. Extract features from the preprocessed multimodal data and perform dimensionality reduction on the extracted features to obtain a multimodal feature vector sequence;

[0024] Time The multimodal feature vector sequence is represented as ,in, Represents the total number of modes; elements Indicates the first Each mode in The features at each time step are represented and used as nodes in the neural network;

[0025] Step S1-4. Quantify the degree of external modal disturbance experienced by each mode by calculating the modal potential energy value of each mode, and use the modal potential energy as the original state representation of the neural network node; the modal potential energy is the weighted sum of the nonlinear interaction effects of all other modes on the current mode;

[0026] The modal potential energy Defined as all other modes in relation to the current mode The weighted sum of the nonlinear interactive effects is expressed mathematically as follows:

[0027] ;

[0028] in, It is the first The modality at time... The combined potential energy intensity formed by the combined perturbation of other modes, namely the modal potential energy, constitutes the original neural state representation of the neural network node; They represent the first , Each mode in Characteristic representation of time; It is the first The sliding mean of each mode within a pre-defined time window based on expert experience is used to measure the steady-state center. They represent the first , Each mode in The standard deviation is always within a pre-defined time window based on expert experience. It is a regularized small quantity, a constant that prevents numerical overflow caused by a denominator of zero; It is a regular insignificant, a constant that prevents the square root from being zero; It is the Euclidean squared distance, representing the distance between the current modal value and the center values ​​of other modalities, reflecting the degree of interference; It is a normalized modal difference sinusoidal mapping used to capture periodic deviations; It is a Gaussian decay term, which simulates the range of action of nonlinear mode pairs through logarithmic functions and Gaussian kernel mapping; By using the square root of the difference in modal standard deviations as a weighting adjustment factor, the hypothesis that the greater the difference in the fluctuation characteristics between modes, the stronger their interactive perturbation may be is emphasized, which reflects the dynamic energy coupling relationship.

[0029] Steps S1-5. Construct a neural connection mechanism by building a fully connected dynamic graph network structure with modalities as nodes and intermodal tension as edge weights, calculating the intermodal tension between different modalities, and constructing a tension matrix based on the intermodal tension.

[0030] The intermodal tension The first one is described The modality and the first The interaction strength between modes, modal tension, is defined as:

[0031]

[0032] in, It is the intermodal tension, representing the first... The modality and the first Between the modes The tension value at any given moment represents the strength of the interaction between modes. It can be understood as the "channel weight" when information propagates between two modes and is the main driver of neural state updates. They represent the first , The modality at time... The potential energy of the model; Represents the L1 norm; It is a hyperbolic cosine function; It is a distance smoothing constant to prevent singularities in the logarithmic and cosine functions when the L1 distance is 0; It is a hyperparameter that regulates the degree of nonlinear growth of tension.

[0033] The obtained intermodal tension Constructing the tension matrix .

[0034] As a further improvement to the neural network-based electrocarbon metering method, step S2 specifically includes:

[0035] S2-1. Based on modal potential energy and combined with intermodal tension, a modal neural state update formula is introduced to update the neural state of the neural network node, i.e., the modal neural state.

[0036] The modal neural state update formula introduces the relative distribution probability for simulating the mutual attraction between modes and the potential energy response function for enhancing the influence weight of important modes.

[0037] The formula for updating the modal neural state is as follows:

[0038] ;

[0039] in, Indicates the first The modality at time... The neural state vector represents the potential contribution of the current mode to the carbon emission prediction output; It is a non-linear compression function that ensures stable neural state output and prevents numerical overflow. The value range is 1 to ; It is the tension normalization lower bound term to avoid the denominator being zero; It is the weight matrix of the modal inputs to the neural network; This refers to the bias term input to the neural network for the modal input; It is the potential energy response function, which enhances the weighting of the influence of important modes, and is expressed as: ; It is the normalized distribution of intermodal tension, simulating the relative distribution probability of mutual attraction between modes; The nonlinear activation output is obtained after linear mapping of the current mode; Used to simulate potential high-frequency disturbances.

[0040] As a further improvement to the neural network-based electrocarbon metering method, step S2 further includes:

[0041] S2-2. Construct a fusion state output mechanism to calculate the total carbon emissions at the target time. The fusion state output mechanism is not only based on the neural state vector, but also introduces the tension derivative and the tension field stability integral to enhance the sensitivity and robustness of the inference.

[0042] The reasoning formula for the fusion-state output mechanism is as follows:

[0043] ;

[0044] in, At any moment Total carbon emissions; It is the partial derivative of the modal tension with respect to the input, representing the first... The input change of the first modality is related to its relationship with the first modality. The degree of influence of tension on each mode, the quantization sensitivity, is obtained using the first-order difference method; It is a stability adjustment factor that adjusts the strength of the integral term of the tension field disturbance and is sensitive to the robustness control of the electric carbon metering system. It is the rate of change of intermodal tension over time, representing the intensity of topological changes, and is calculated using the first-order finite difference method; It is the first The instantaneous rate of change of the potential energy of each mode represents the strength of the external disturbance input and is calculated by the first-order difference method. It represents the degree of accumulation of disturbances over time, reflecting the stability of the neural network modeling structure and its sensitivity to external disturbances.

[0045] S2-3. After obtaining the total carbon emissions at the target time, combine it with the total electricity consumption during the current period to obtain the carbon emission measurement result.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] 1. This invention introduces a modal potential energy construction mechanism to quantify the nonlinear interactions between multimodal features into potential energy values ​​with clear physical meaning. This ensures that each modal feature carries its relative position within the system's dynamic coupling structure when input into the neural network. This design fundamentally changes the traditional neural network's approach of treating input features as independent variables, effectively overcoming the "black box" problem caused by ignoring the intrinsic relationships between features, and significantly enhancing the model's ability to interpret the physical properties of input features. Based on this, the model can more accurately capture the impact mechanism of multiple factors synergistically affecting carbon emissions in complex power scenarios, thereby improving the model's generalization ability under different operating conditions.

[0048] 2. This invention further proposes a dynamic calculation and graph construction method for intermodal tension, transforming the modal potential energy differences into connection weights in a graph neural network, forming a tension-driven adaptive neural state update mechanism. This mechanism allows the state update of each modal node to no longer depend on a fixed, predefined network hierarchy, but rather dynamically adjusts the information propagation path based on the real-time calculated intermodal tension. When facing industrial energy systems with numerous modes and strong interference, this structure can autonomously suppress the interference of redundant or noisy modes, strengthen the contribution of key modes, thereby improving the convergence efficiency and stability of model training, avoiding overfitting caused by excessive participation of irrelevant modes, and enhancing the robustness of the model in complex dynamic environments.

[0049] 3. The fusion-state output mechanism constructed in this invention further introduces tension derivatives and tension field stability integral terms when integrating the neural states of various modalities for final carbon emission inference. This mechanism not only considers the instantaneous states of each modality but also enhances the response capability to sudden dynamics of the system and the smoothness of the prediction results through the sensitivity of tension change rate and the cumulative effect of historical disturbances. This design enables the model to effectively suppress abnormal fluctuations in predicted values ​​when facing transient processes such as power quality disturbances and load changes, thereby providing a more reliable and stable decision-making basis for online carbon emission measurement and real-time control. Attached Figure Description

[0050] Figure 1 This is a structural diagram of an electric carbon metering system based on a neural network according to the present invention;

[0051] Figure 2 This is a flowchart of a neural network-based carbon metering method according to the present invention. Detailed Implementation

[0052] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0053] Example 1

[0054] This embodiment provides an electric carbon metering system based on a neural network, such as... Figure 1 The system includes: a multimodal data acquisition module, a data preprocessing module, a feature extraction and dimensionality reduction module, a modal potential energy calculation module, a modal tension matrix construction module, a modal neural state update module, a fusion module, and an electrocarbon quantization module.

[0055] The raw multimodal data acquired by the multimodal data acquisition module is sequentially processed by the data preprocessing module and the feature extraction and dimensionality reduction module to obtain the modal feature vector sequence. The modal feature vector sequence is processed by the modal potential energy calculation module, the modal tension matrix construction module and the modal neural state update module to obtain the modal potential energy, intermodal tension and neural state vector. The modal potential energy, intermodal tension and neural state vector are fused by the fusion module to obtain the total carbon emissions. The total carbon emissions are input to the electrocarbon quantification module to obtain the electrocarbon measurement result.

[0056] The multimodal data acquisition module acquires raw multimodal data through equipment such as power acquisition devices and condition monitoring sensors deployed based on expert experience. This raw multimodal data includes electrical parameters such as voltage, current, power, active load, and harmonic spectrum, as well as equipment operating status, power quality disturbances, and room temperature.

[0057] The data preprocessing module preprocesses the original multimodal data to obtain preprocessed multimodal data, which is then sent to the feature extraction and dimensionality reduction module.

[0058] The feature extraction and dimensionality reduction module uses existing feature engineering techniques to extract features from the preprocessed multimodal data to obtain preliminary feature data. Then, the mutual information method and principal component analysis algorithm are used to reduce the dimensionality of the extracted preliminary feature data to obtain a multimodal feature vector sequence.

[0059] The mode potential energy calculation module calculates the mode potential energy value of each mode based on the multimodal feature vector sequence to quantify the degree of external modal disturbance that each mode is currently subjected to, and obtains the mode potential energy.

[0060] The modal tension matrix construction module uses modal potential energy as the original state representation of neural network nodes and introduces a neural connection mechanism, using the intermodal tension calculated based on modal potential energy as the connection weights.

[0061] The modal neural state update module, based on the multimodal feature vector sequence, combines modal potential energy and intermodal tension, introduces a modal neural state update formula, updates the neural state of the modality, and obtains the updated neural state vector.

[0062] The fusion module, based on the updated neural state vector, combines intermodal tension and modal potential energy to construct a fusion state output mechanism and obtain the total carbon emissions.

[0063] The electricity carbon quantification module processes the total carbon emissions and the total electricity consumption in the current period using existing electricity carbon metering models to obtain electricity carbon metering results, thus achieving electricity carbon metering.

[0064] Example 2

[0065] This embodiment provides a neural network-based method for measuring carbon electricity, such as... Figure 2 As shown, the method includes the following steps:

[0066] S1. Obtain the original multimodal data. After preprocessing, perform feature extraction and dimensionality reduction on the preprocessed multimodal data to obtain a multimodal feature vector sequence. Calculate the modal potential energy based on the multimodal feature vector sequence and use the modal potential energy as the original state representation of the neural network nodes. At the same time, introduce a neural connection mechanism and use the intermodal tension calculated based on the modal potential energy as the connection weight.

[0067] The specific process is as follows:

[0068] Step S1-1. Obtain raw multimodal data by deploying devices such as power acquisition devices and status monitoring sensors based on expert experience. The raw data includes electrical parameters such as voltage, current, power, active load, and harmonic spectrum, as well as equipment operating status, power quality disturbances, and room temperature.

[0069] Step S1-2. Perform preprocessing on the original multimodal data, such as denoising, cleaning, time synchronization, standardization, and normalization, to obtain preprocessed multimodal data. The preprocessing process is a technique well known to those skilled in the art and will not be described in detail here.

[0070] Step S1-3. Use existing feature engineering techniques, such as statistical analysis, to extract features from the preprocessed multimodal data, and use existing mutual information methods and principal component analysis algorithms to reduce the dimensionality of the extracted features to obtain a multimodal feature vector sequence.

[0071] Time The multimodal feature vector sequence is represented as ,in, Represents the total number of modes; elements Indicates the first Each mode in The features at each time point are represented and used as nodes in the neural network.

[0072] Step S1-4. In order to describe the physical interpretation capability of the above multimodal feature vector sequence and make the influence between nonlinear modes explicit, the degree of external modal interference currently experienced by each mode is quantified by calculating the modal potential energy value of each mode.

[0073] The modal potential energy Defined as all other modes in relation to the current mode The weighted sum of the nonlinear interactive effects is expressed mathematically as follows:

[0074] ;

[0075] in, It is the first The modality at time... The combined potential energy intensity formed by the combined perturbation of other modes, namely the modal potential energy, constitutes the original neural state representation of the neural network node; They represent the first , Each mode in Characteristic representation of time; It is the first The sliding mean of each mode within a pre-defined time window based on expert experience is used to measure the steady-state center. They represent the first , Each mode in The standard deviation (fluctuation range) is always within a time window pre-set according to the expert experience method. It is a regularized small quantity, a constant that prevents numerical overflow caused by a denominator of zero; It is a regular insignificant, a constant that prevents the square root from being zero; It is the Euclidean squared distance, representing the distance between the current modal value and the center values ​​of other modalities, reflecting the degree of interference; It is a normalized modal difference sinusoidal mapping used to capture periodic deviations; It is a Gaussian decay term, which simulates the range of action of nonlinear mode pairs through logarithmic functions and Gaussian kernel mapping; By using the square root of the difference in modal standard deviations as a weighting adjustment factor, the hypothesis that the greater the difference in fluctuation characteristics between modes, the stronger their interactive perturbation may be is emphasized, reflecting the dynamic energy coupling relationship.

[0076] Steps S1-5. After calculating the potential energy of all modes, i.e. the original neural state representation of the nodes, a neural connection mechanism is constructed based on the above potential energy of modes. This is achieved through a graph structure, i.e., a fully connected dynamic graph network structure with modes as nodes and inter-modal tension as edge weights is constructed.

[0077] The intermodal tension The first one is described The modality and the first The interaction strength between modes, modal tension, is defined as:

[0078]

[0079] in, It is the intermodal tension, representing the first... The modality and the first Between the modes The tension value at any given moment represents the strength of the interaction between modes. It can be understood as the "channel weight" when information propagates between two modes and is the main driver of neural state updates. They represent the first , The modality at time... The potential energy of the model; Represents the L1 norm; It is a hyperbolic cosine function; It is a distance smoothing constant to prevent singularities in the logarithmic and cosine functions when the L1 distance is 0; It is a hyperparameter that adjusts the degree of nonlinear tension growth, determined based on expert experience, with a reference range of values. .

[0080] The obtained intermodal tension Constructing the tension matrix .

[0081] S2. Based on the multimodal feature vector sequence, combined with modal potential energy and intermodal tension, a modal neural state update formula is introduced to update the neural state of the modality and obtain the updated neural state vector; further, a fusion state output mechanism is constructed to obtain the total carbon emissions, and then combined with the total electricity consumption to obtain the carbon emission measurement result.

[0082] The specific process is as follows:

[0083] S2-1. Based on modal potential energy and combined with intermodal tension, a modal neural state update formula is introduced to update the neural state of the neural network node, i.e., the modal neural state.

[0084] The formula for updating the modal neural state is as follows:

[0085] ;

[0086] in, Indicates the first The modality at time... The neural state vector represents the potential contribution of the current mode to the carbon emission prediction output; It is a non-linear compression function that ensures stable neural state output and prevents numerical overflow. The value range is 1 to ; It is the tension normalization lower bound term to avoid the denominator being zero; It is the weight matrix of the modal inputs to the neural network; This refers to the bias term input to the neural network for the modal input; It is the potential energy response function, which enhances the weighting of the influence of important modes, and is expressed as: ; It is the normalized distribution of intermodal tension, simulating the relative distribution probability of mutual attraction between modes; The nonlinear activation output is obtained after linear mapping of the current mode; Used to simulate potential high-frequency disturbances.

[0087] S2-2. After all modal states have been updated, a fusion state output mechanism is constructed to further achieve unified reasoning on carbon emissions. This fusion state output mechanism is not only based on each neural state vector, but also introduces tension derivatives and tension field stability integrals to enhance the sensitivity and robustness of the reasoning.

[0088] The reasoning formula for the fusion-state output mechanism is as follows:

[0089] ;

[0090] in, At any moment Total carbon emissions; It is the partial derivative of the modal tension with respect to the input, representing the first... The input change of the first modality is related to its relationship with the first modality. The degree of influence of tension on each mode, the quantization sensitivity, is obtained using the first-order difference method; It is a stability adjustment factor that adjusts the strength of the integral term of the tension field disturbance and is sensitive to the robustness control of the electric carbon metering system. It is the rate of change of intermodal tension over time, representing the intensity of topological changes, and is calculated using the first-order finite difference method; It is the first The instantaneous rate of change of the potential energy of each mode represents the strength of the external disturbance input and is calculated by the first-order difference method. It represents the degree of accumulation of disturbances over time, reflecting the stability of the neural network modeling structure and its sensitivity to external disturbances.

[0091] In the above formula, firstly, the neural state vector... Multiplied by an adjustment factor based on its tension partial derivative with respect to the input. This is used to simulate the sensitivity of modal responses to changes in overall tension; secondly, it introduces... The redundant effects of low-potential-energy modes are suppressed. Finally, the integral norm of the overall system-risk (i.e., the product of structural disturbance and potential energy disturbance) in the time dimension is introduced as a "system stability term". If the system tension disturbance is large, this term is increased, thereby penalizing the final prediction value and preventing prediction anomalies.

[0092] S2-3. After obtaining the total carbon emissions at the target time, combine it with the total electricity consumption in the current period, process it using the existing carbon emission metering model, obtain the carbon emission metering result, and realize carbon emission metering.

[0093] It should be noted that Embodiments 1 and 2 are not independent of each other. In implementation, the corresponding modules in Embodiment 1 can be implemented by referring to the method steps in Embodiment 2, or the module division method in Embodiment 1 can be used to plan and design the program product for implementing Embodiment 2.

[0094] The order of the embodiments is for illustrative purposes only and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0095] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0096] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A neural network-based electric carbon metering system, characterized in that: It includes a multimodal data acquisition module, a data preprocessing module, a feature extraction and dimensionality reduction module, a modal potential energy calculation module, a modal tension matrix construction module, a modal neural state update module, a fusion module, and an electrocarbon quantization module. The raw multimodal data acquired by the multimodal data acquisition module is sequentially processed by the data preprocessing module and the feature extraction and dimensionality reduction module to obtain the modal feature vector sequence. The modal feature vector sequence is processed by the modal potential energy calculation module, the modal tension matrix construction module and the modal neural state update module to obtain the modal potential energy, intermodal tension and neural state vector. The modal potential energy, intermodal tension and neural state vector are fused by the fusion module to obtain the total carbon emissions. The total carbon emissions are input to the electrocarbon quantification module to obtain the electrocarbon measurement result.

2. The neural network-based electric carbon metering system as described in claim 1, characterized in that: A multimodal data acquisition module is used to acquire raw multimodal data through power acquisition devices and status monitoring sensors; The data preprocessing module is used to preprocess the raw multimodal data to obtain preprocessed multimodal data, which is then sent to the feature extraction and dimensionality reduction module. The feature extraction and dimensionality reduction module is used to extract features from the preprocessed multimodal data to obtain preliminary feature data. Then, the extracted preliminary feature data is dimensionality reduced to obtain a multimodal feature vector sequence.

3. The neural network-based electric carbon metering system as described in claim 1 or 2, characterized in that: The mode potential energy calculation module is used to calculate the mode potential energy value of each mode based on the multimodal feature vector sequence to quantify the degree of external modal disturbance that each mode is currently subjected to, and obtain the mode potential energy.

4. The neural network-based electric carbon metering system as described in claim 3, characterized in that: The modal tension matrix construction module is used to use modal potential energy as the original state representation of neural network nodes and introduces a neural connection mechanism to use the intermodal tension calculated based on the modal potential energy as the connection weight.

5. The neural network-based electric carbon metering system as described in claim 4, characterized in that: The modal neural state update module is used to update the neural state of a modality based on a multimodal feature vector sequence, combined with modal potential energy and intermodal tension, by introducing a modal neural state update formula, and obtain the updated neural state vector.

6. The neural network-based electric carbon metering system as described in claim 5, characterized in that: The fusion module is used to construct a fusion state output mechanism based on the updated neural state vector, combined with intermodal tension and modal potential energy, to obtain the total carbon emissions.

7. A method for metering electric carbon based on neural networks, characterized in that the steps include... include: S1. Obtain the raw multimodal data. After preprocessing, perform feature extraction and dimensionality reduction on the preprocessed multimodal data to obtain a multimodal feature vector sequence. Calculate the modal potential energy based on the multimodal feature vector sequence and use the modal potential energy as the original state representation of the neural network nodes. At the same time, introduce a neural connection mechanism and use the intermodal tension calculated based on the modal potential energy as the connection weight. S2. Based on the multimodal feature vector sequence, combined with modal potential energy and intermodal tension, a modal neural state update formula is introduced to update the neural state of the modality and obtain the updated neural state vector; further, a fusion state output mechanism is constructed to obtain the total carbon emissions, and then combined with the total electricity consumption to obtain the carbon emission measurement result.

8. The neural network-based carbon metering method as described in claim 7, characterized in that: Step S1 specifically includes: Step S1-1. Acquire raw multimodal data through power acquisition devices and condition monitoring sensors, etc. Step S1-2. Preprocess the original multimodal data to obtain preprocessed multimodal data; Step S1-3. Extract features from the preprocessed multimodal data and perform dimensionality reduction on the extracted features to obtain a multimodal feature vector sequence; Step S1-4. Quantify the degree of external modal disturbance experienced by each mode by calculating the modal potential energy value of each mode, and use the modal potential energy as the original state representation of the neural network node; the modal potential energy is the weighted sum of the nonlinear interaction effects of all other modes on the current mode; Steps S1-5. Construct a neural connection mechanism by building a fully connected dynamic graph network structure with modalities as nodes and intermodal tension as edge weights, calculating the intermodal tension between different modalities, and constructing a tension matrix based on the intermodal tension.

9. The neural network-based carbon metering method as described in claim 7 or 8, characterized in that, Step S2 specifically includes: S2-1. Based on modal potential energy and combined with intermodal tension, a modal neural state update formula is introduced to update the neural state of the neural network node, i.e., the modal neural state. The modal neural state update formula introduces the relative distribution probability for simulating the mutual attraction between modes and the potential energy response function for enhancing the influence weight of important modes.

10. The neural network-based carbon metering method as described in claim 9, characterized in that, Step S2 also includes: S2-2. Construct a fusion state output mechanism to calculate the total carbon emissions at the target time. The fusion state output mechanism is not only based on the neural state vector, but also introduces the tension derivative and the tension field stability integral to enhance the sensitivity and robustness of the inference. S2-3. After obtaining the total carbon emissions at the target time, combine it with the total electricity consumption during the current period to obtain the carbon emission measurement result.

Images