An electric carbon metering heterogeneous data fusion and deviation compensation method

Abstract

The application discloses an electric carbon metering heterogeneous data fusion and deviation compensation method and relates to the field of electric carbon metering heterogeneous data processing. The application carries out point-by-point correction on the historical input sequence of each data source by constructing a nonlinear compensation function of a fusion hyperbolic tangent variance response suppression term and a sinusoidal period modulation term, constructs a local residual trajectory tensor based on the corrected sequence, calculates residual resonance mapping values and total deviation potential energy between any two sources, converts the potential energy into a fusion weight by using an exponential normalization model, simultaneously constructs a residual cooperative covariance matrix and extracts a first principal feature vector to obtain a principal direction projection result, obtains a fusion estimated value by weighted summation, and finally constructs a weighted deviation estimation and introduces a compensation coefficient to compensate the estimated value. The application can process inconsistent measurement bases and dynamic deviations between heterogeneous devices, suppress periodic disturbances and noise, and improve the accuracy of the electric carbon metering fusion result.

Description

Technical Field

[0001] This invention relates to the field of heterogeneous data processing for carbon metering, and in particular to a method for fusing and compensating for heterogeneous carbon metering data. Background Technology

[0002] With the continued advancement of the "dual-carbon" strategy, electricity carbon metering, as a crucial infrastructure supporting the coordinated management of carbon emission accounting and electricity consumption, has been widely applied in industries such as manufacturing, construction, transportation, and energy. In actual deployments, electricity carbon metering systems typically consist of various devices, including electricity carbon meters, carbon emission estimation models, edge sensors, power analyzers, and energy gateways. These devices differ in their technical standards, measurement mechanisms, data interfaces, and installation locations, forming a typical heterogeneous metering system architecture. For the same electricity carbon index (such as active power, cumulative electricity consumption, or carbon emission intensity) for the same energy user, measurement results from multiple heterogeneous devices often exist simultaneously.

[0003] Due to differences in the metering accuracy, drift characteristics, anti-interference capabilities, and output benchmarks of heterogeneous equipment, problems arise such as inconsistent measurement benchmarks, difficulty in modeling dynamic biases, and data distortion or missing data from some equipment among multiple sets of observations of the same electric carbon index. Furthermore, electric carbon metering data often exhibits significant periodic variations (such as diurnal load fluctuations) and sudden disturbances (such as spikes caused by equipment start-up and shutdown) during actual operation. Existing methods lack dynamic compensation mechanisms for these time-series characteristics, making it difficult to establish effective bias relationship models among multi-source observations. Directly using simple weighted averaging or filtering can easily cause the fusion results to deviate from physical reality, or fail to effectively suppress the influence of high-biased data sources on the results. Therefore, how to uniformly model, dynamically identify and compensate for, and ensure the accuracy and physical rationality of the fusion results for multiple sets of observations of the same electric carbon index under multi-source heterogeneous conditions has become a key technical problem urgently needing to be solved in this field. Summary of the Invention

[0004] This invention proposes a method for fusing and compensating for heterogeneous data in carbon metering. Its purpose is to: (1) solve the problem in the prior art that the measurement benchmarks of the same carbon index are inconsistent and dynamic deviations are difficult to model due to differences in the metering accuracy, drift characteristics and output benchmarks of heterogeneous equipment; (2) overcome the shortcomings of the existing methods that lack dynamic compensation mechanisms for periodic changes and sudden disturbances, which make the fusion results deviate from the physical reality or make it difficult to effectively suppress the influence of high-bias data sources on the results.

[0005] The technical solution of this invention is as follows:

[0006] A method for fusing and compensating for heterogeneous data in carbon metering includes step S1: acquiring data from... Carbon metering data of the same object collected by two heterogeneous carbon metering sources;

[0007] In step S1, the carbon metering data of each data source are preprocessed according to time sequence, and a multi-source historical input sequence under a unified sliding window is constructed.

[0008] The method also includes:

[0009] Step S2: Construct a nonlinear compensation function that integrates the hyperbolic tangent variance response suppression term and the sinusoidal periodic modulation term, correct each sample value in the historical input sequence of each data source point by point to obtain the corrected historical sequence, and construct the local residual trajectory tensor of each point of each data source based on the corrected historical sequence.

[0010] Step S3: Calculate the residual resonance mapping value between any two data sources based on the local residual trajectory tensor; then calculate the total deviation potential energy based on the residual resonance mapping value to quantify the position of each data source in the overall deviation field;

[0011] Step S4: Use the exponential normalization suppression model to convert the total bias potential of each data source into fusion weights; construct the residual covariance matrix based on all corrected historical sequences, and perform singular value decomposition to extract the first principal eigenvector as the principal fluctuation direction of the residual trajectory to obtain the principal projection results of each data source; then perform weighted summation based on the fusion weights and principal projection results of each data source to obtain the fusion estimate.

[0012] Step S5: Construct a weighted deviation estimate between the fused estimate and the corrected values ​​from each data source, and introduce a deviation compensation coefficient to perform deviation compensation on the fused estimate, and output the final carbon metering result after compensation.

[0013] As a further improvement to the aforementioned method for fusing and compensating for heterogeneous data in carbon metering: In step S1, for the first... A data source, at the current moment The length of a sliding window is defined as follows: The historical input sequence is represented as:

[0014]

[0015] In the above formula, Indicates the first Data sources at time The historical input sequence; Indicates the first Data sources at time Preprocessed carbon metering data obtained from sampling ; It is the length of the sliding window; Indicates transpose;

[0016] Step S2 specifically includes step S2-1: introducing a stability penalty coefficient. Variance response sensitivity coefficient and periodic disturbance compensation amplitude control factor A nonlinear compensation function is constructed to correct the historical input sequence, resulting in the corrected historical sequence.

[0017] For the Historical input sequences of data sources At any time in raw value The correction calculation method is as follows:

[0018]

[0019] In the above formula, It is the revised version of the first Data source at time Carbon metering data; For the first Stability penalty coefficient of the data source; It is the variance response sensitivity coefficient; Represents the historical input sequence The variance; It is the hyperbolic tangent function; It is the periodic disturbance compensation amplitude control factor; The period of environmental disturbance;

[0020] The corrected carbon metering data corresponding to each element in the historical input sequence constitute the corrected historical sequence. .

[0021] As a further improvement to the aforementioned method for fusing and compensating for heterogeneous data in carbon metering, step S2 further includes step S2-2: calculating the... The corrected historical series mean of each data source And calculate the local residual trajectory tensor of each point in the corrected historical sequence one by one. :

[0022]

[0023] In the above formula, Indicates the first In the corrected historical sequence of the data source, the first... The local residual trajectory tensor of a point.

[0024] As a further improvement to the method for fusing and compensating for heterogeneous data in carbon metering, in step S3, for any two data sources... and Calculate its residual resonance mapping value :

[0025]

[0026] In the above formula, It is the first The data source and the first Data sources at time The residual resonance mapping value; and They are the first The and the first The first data source The local residual trajectory tensor of a point; It is a preset stability constant; and They are the first The and the first The mean of the corrected historical series from each data source; It is the cosine frequency modulation coefficient; This is the size of the sliding window.

[0027] As a further improvement to the aforementioned method for fusing and compensating for heterogeneous data in carbon metering, in step S3, the first... Total deviation potential of each data source The calculation method is as follows:

[0028]

[0029] In the above formula, It is the first Total deviation potential of each data source; It is the first The data source and the first Data sources at the current moment The residual resonance mapping value; It is the natural logarithm function; and They are the first The and the first The mean of the corrected historical series from each data source.

[0030] As a further improvement to the method for fusing and compensating for heterogeneous data in carbon metering, the method for calculating the fusion weight in step S4 is as follows:

[0031]

[0032] In the above formula, For the first Data sources at time The fusion weight; It is a natural constant; It is the first The total deviation potential of each data source.

[0033] As a further improvement to the method for fusing and compensating for heterogeneous data in carbon metering, step S4 involves constructing a residual covariance matrix. The method is as follows:

[0034]

[0035] In the above formula, It is the first The corrected historical sequence from a data source; It is a sequence of the mean values ​​from all data sources; This indicates transpose.

[0036] As a further improvement to the method for fusing and compensating for heterogeneous data in carbon metering, step S4 involves adjusting the covariance matrix. Perform singular value decomposition:

[0037]

[0038] In the above formula, It is a singular vector matrix. Given a singular value diagonal matrix, extract the first principal eigenvector corresponding to the largest singular value from it. The corrected historical sequence of each data source is projected onto the main fluctuation direction to obtain its main projection result. :

[0039]

[0040] In the above formula, Indicates the first The principal projection results of each data source; It is the first The corrected historical sequence from a data source; It is the mean sequence of all data sources.

[0041] As a further improvement to the method for fusing and compensating for heterogeneous data in carbon metering, step S4 involves calculating the fusion estimate. The method is as follows:

[0042]

[0043] In the above formula, For a moment The integrated valuation, For the first Data sources at time The fusion weight, Indicates the first The main projection results of the data source.

[0044] As a further improvement to the aforementioned method for fusing and compensating for heterogeneous data in carbon metering, step S5 specifically involves the following process:

[0045] Step S5-1: Based on the fusion valuation obtained in step S4 In addition to the corrected historical sequences of each data source obtained in step S2, a deviation estimate is constructed to characterize the overall deviation between the fusion result and the multi-source observation data. :

[0046]

[0047] In the above formula, Indicates the current moment The overall deviation; Indicates the first In the corrected historical sequence of data sources, at the current moment The carbon metering value is taken from the corrected historical series. The latest sampling point; The result calculated in step S4 is the same as the first... Current time of each data source The corresponding fusion weights;

[0048] Step S5-2: Introduce deviation compensation coefficient The fusion valuation is then corrected to obtain the final output. :

[0049]

[0050] In the above formula, The current moment after compensation The results of the carbon metering; This is the deviation compensation coefficient.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] 1. Based on acquiring multi-source heterogeneous electrocarbon metering data and constructing a unified sliding window historical sequence, this invention introduces a nonlinear compensation function that integrates a hyperbolic tangent variance response suppression term and a sinusoidal periodic modulation term. This function corrects the original sampled values ​​from each data source point by point before participating in subsequent analysis. Unlike the conventional approach of passively compensating after deviations occur, this mechanism proactively weakens nonlinear distortion caused by local fluctuations, high variance, and periodic disturbances in the initial stage of the data fusion process. It reduces the traction effect of abnormal sampled values ​​on downstream processing from the source, ensuring that subsequent operations such as residual analysis and weight calculation within the unified multi-source framework are based on more stable and aligned data.

[0053] 2. To address the difficulty in effectively measuring and modeling the dynamic deviation between heterogeneous electrocarbon metering sources, this invention constructs a local residual trajectory tensor based on the corrected sequence and designs a residual resonance mapping function accordingly. This function integrates the normalized residual difference, mean offset suppression term, and cosine frequency modulation factor, enabling it to capture the dynamic coupling strength of the residual behavior of any two data sources within a local time window, rather than relying solely on static correlation coefficients. This measurement method is more sensitive to changes in the behavioral trajectories between data sources. When the system faces periodic load fluctuations or sudden disturbances, it can identify data sources with deep deviations or abnormal coupling from the overall trend in real time, providing a structural basis for subsequent potential energy assessment and weight suppression.

[0054] 3. Building upon the aforementioned residual resonance mapping, this invention further defines the total deviation potential energy of each data source. By measuring the combined cumulative residual tension intensity between a given source and the others, its position in the overall deviation field is quantified. An exponential normalization suppression model is then used to transform this potential energy into fusion weights. This weighting mechanism possesses nonlinear mapping characteristics, amplifying the suppression effect of high-deviation potential energy data sources during fusion, ensuring that the fusion estimate is not easily dominated by individual data sources with high noise levels and severe drift. Simultaneously, this invention constructs a residual covariance matrix and extracts its principal fluctuation direction. The corrected sequences of each data source are projected onto this principal direction before being weighted and fused. This approach uses both the internal structural consistency of the data sources and the external potential energy situation as a fusion benchmark, enabling the fusion estimate to resist high-deviation interference while maintaining consistency with the overall fluctuation trend of the multi-source data.

[0055] 4. After obtaining the fused estimate, this invention constructs a weighted bias estimate using the fusion weights and correction values ​​of each data source, and introduces a bias compensation coefficient to perform targeted compensation on the fused estimate, ultimately outputting the carbon metering result. This step is not a simple scaling or fixed bias adjustment of the fused estimate, but rather a dynamic compensation amount formed based on the overall deviation between the multi-source data and the fused result. This allows the output result to inherit the stability of the fused estimate while making targeted corrections to systematic residual biases, ultimately improving the consistency between the metering result and the actual physical condition. Attached Figure Description

[0056] Figure 1 This is a flowchart of a method for fusing heterogeneous data and compensating for bias in carbon metering. Detailed Implementation

[0057] 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.

[0058] like Figure 1 As shown, a method for fusing and compensating for heterogeneous data in carbon metering specifically includes the following steps:

[0059] Step S1: Obtain from The carbon metering data of the same object collected by the heterogeneous carbon metering sources are preprocessed according to time sequence for each data source, and then a multi-source historical input sequence under a unified sliding window is constructed.

[0060] In this step, the heterogeneous electric carbon metering devices involved in the integration include, but are not limited to, electric carbon meters, carbon emission estimation models, edge sensors, power analyzers, and energy gateways. The data collected by these devices all characterize the energy consumption or carbon emission results of the same energy user per unit time, forming a unified electric carbon metering semantic space.

[0061] The preprocessing includes denoising, cleaning, missing value imputation, outlier pruning, time alignment, standardization, and normalization of the data to eliminate inconsistencies between data sources caused by differences in units, sampling frequencies, and transmission protocols.

[0062] For the A data source, at the current moment The length of a sliding window is defined as follows: The historical input sequence is represented as:

[0063]

[0064] In the above formula, Indicates the first Data sources at time The historical input sequence; Indicates the first Data sources at time Preprocessed carbon metering data obtained from sampling ; It is the length of the sliding window; This indicates transpose.

[0065] Step S2: Construct a nonlinear compensation function that integrates the hyperbolic tangent variance response suppression term and the sinusoidal periodic modulation term, correct each sample value in the historical input sequence of each data source point by point to obtain the corrected historical sequence, and construct the local residual trajectory tensor of each point of each data source based on the corrected historical sequence.

[0066] The purpose of this step is to suppress and correct the source data in the early stages of bias propagation, reducing nonlinear distortion caused by local fluctuations, periodic disturbances, and high variance, thereby improving the comparability and alignment of multi-source data within a unified fusion framework. The specific process includes:

[0067] Step S2-1: Introduce a stability penalty coefficient Variance response sensitivity coefficient and periodic disturbance compensation amplitude control factor A nonlinear compensation function is constructed to correct the historical input sequence, resulting in the corrected historical sequence.

[0068] Specifically, for the first Historical input sequences of data sources At any time in raw value The correction calculation method is as follows:

[0069]

[0070] In the above formula, It is the revised version of the first Data source at time Carbon metering data; For the first The stability penalty coefficient for the data source is determined based on expert experience, with a reference value of [value missing]. ; This is the variance response sensitivity coefficient, which controls the degree of nonlinearity of the variance response. It is determined through experimental fitting, and a reference value is [value missing]. ; Represents the historical input sequence The variance; It is the hyperbolic tangent function, used to achieve nonlinear variance suppression; This is the periodic disturbance compensation amplitude control factor, reflecting the amplitude of compensation for periodic fluctuations. Its value is set empirically, with a reference range of [missing information]. ,in , representing the historical input sequence The mean; The period of environmental disturbances, such as the diurnal cycle of a power system; It is a sine function used to simulate the modulation term of periodic fluctuations.

[0071] The corrected carbon metering data corresponding to each element in the historical input sequence constitute the corrected historical sequence. .

[0072] Step S2-2: Calculate the first... The corrected historical series mean of each data source And calculate the local residual trajectory tensor of each point in the corrected historical sequence one by one. :

[0073]

[0074] In the above formula, Indicates the first In the corrected historical sequence of the data source, the first... The local residual trajectory tensor of a point reflects the fluctuation of the data source relative to its own trend within a local time period, and is the basis for subsequent calculation of the deviation coupling metric.

[0075] Step S3: Introduce the residual resonance mapping function to measure the synergy and coupling magnitude of the residual behavior represented by the local residual trajectory tensor between any two data sources within the sliding window, and obtain the residual resonance mapping value; then calculate the total deviation potential energy based on the residual resonance mapping value to quantify the position of each data source in the overall deviation field.

[0076] This step replaces traditional static correlation metrics with a dynamic, asymmetric, and structurally adjustable behavioral coupling model, enabling real-time identification of highly coupled, low-quality data sources. The specific process includes:

[0077] Step S3-1: Using the square of the difference between the local residual trajectory tensors at each point from the two data sources as the numerator, and the difference in the corrected historical sequence mean as the denominator suppression term, a cosine modulation function is superimposed to adjust the importance distribution at different time points within the window, thus constructing the residual resonance mapping function. For any two data sources... and Calculate its residual resonance mapping value :

[0078]

[0079] In the above formula, It is the first The data source and the first Data sources at time The larger the residual resonance mapping value, the stronger the coupling between the behavioral deviations of the two. and They are the first The and the first The first data source The local residual trajectory tensor of a point; It is a preset minimum stability constant used to prevent the denominator from being zero; for example, it takes the value of... ; and They are the first The and the first The mean of the corrected historical series from each data source; It is the cosine frequency modulation coefficient, which controls the importance distribution of time points within the control window. For example, it can be set to... ; It is a cosine function.

[0080] Step S3-2: For each data source, logarithmically weighted sum of the mean differences between its residual resonance mapping values ​​and those of all other data sources is performed to construct the total deviation potential energy function, and the total deviation potential energy of that data source is output. This potential quantifies the overall tension intensity of the residual behavior between the current data source and all other sources:

[0081]

[0082] In the above formula, It is the first Total deviation potential of each data source; This represents the total number of data sources; It is the natural logarithm function; and They are the first The and the first The mean of the corrected historical series from each data source. In a physical sense, The larger the value, the greater the deviation of the data source's behavior from other data sources, the lower its reliability, and the less weight it should be given in subsequent fusion processes.

[0083] Optionally, it also includes:

[0084] Step S3-3: Superimpose the total deviation potential energies of all data sources obtained in all steps S3-2 to calculate the total intensity of the global deviation potential energy. :

[0085]

[0086] In the above formula, It reflects the overall deviation level of the entire fusion system within the current sliding window, and can be used in the subsequent parameter optimization process to control the deviation constraints of the overall fusion behavior.

[0087] Step S4: Use the exponential normalization suppression model to convert the total bias potential of each data source into fusion weights; construct the residual covariance matrix based on all corrected historical sequences, and perform singular value decomposition to extract the first principal eigenvector as the principal fluctuation direction of the residual trajectory to obtain the principal projection results of each data source; then perform weighted summation based on the fusion weights and principal projection results of each data source to obtain the fusion estimate.

[0088] This step establishes a non-linear, dynamic weight scheduling mechanism that assigns lower weights to data sources with high potential energy, thus preventing high-noise data from dominating the fusion valuation.

[0089] The method for calculating the fusion weights is as follows:

[0090]

[0091] In the above formula, For the first Data sources at time The fusion weights; It is a natural constant; It is the first The total deviation potential energy of all data sources; the denominator is the sum of the negative exponents of the potential energy of all data sources, used to achieve normalization.

[0092] This weighting implements a non-linear penalty based on deviation potential energy: the greater the deviation potential energy of the source data, the smaller the weight, thereby automatically suppressing the voice of low-quality data sources in the fusion valuation.

[0093] To characterize the joint fluctuation behavior of the deviation trajectories of all data sources over time, a residual covariance matrix is ​​constructed. The method is as follows:

[0094]

[0095] In the above formula, It is the covariance matrix; It is the first The corrected historical sequence of data sources; It is the mean sequence of all data sources, representing the average trajectory of all data sources within the sliding window; This indicates transpose.

[0096] For covariance matrix Perform singular value decomposition:

[0097]

[0098] In the above formula, It is a singular vector matrix. Let be a diagonal matrix of singular values. Extract the first principal eigenvector corresponding to the largest singular value from it. This vector represents the principal fluctuation direction of the residual trajectory in the time dimension. Then, the corrected historical sequence of each data source is projected onto this principal fluctuation direction to obtain its principal projection result. :

[0099]

[0100] In the above formula, Indicates the first The principal projection results of each data source reflect the consistency between the fluctuation pattern and the principal fluctuation direction after removing the average trend.

[0101] Computational fusion valuation The method is as follows:

[0102]

[0103] In the above formula, For a moment The fusion valuation relies not only on the original data itself, but also on the projection performance of each data source in the main direction of the collaborative residual space, achieving a globally optimal estimate that conforms to the physical coupling relationship.

[0104] Step S5: Construct a weighted deviation estimate between the fused estimate and the corrected values ​​from each data source, and introduce a deviation compensation coefficient to perform deviation compensation on the fused estimate, and output the final carbon metering result after compensation.

[0105] The purpose of this step is to correct for deviations in the results that may be caused by the superposition of multi-source measurement errors, systematic biases, and dynamic disturbances, based on the consistency relationship between the fusion results and each data source. The specific process is as follows:

[0106] Step S5-1: Based on the fusion valuation obtained in step S4 In addition to the corrected historical sequences of each data source obtained in step S2, a deviation estimate is constructed to characterize the overall deviation between the fusion result and the multi-source observation data. :

[0107]

[0108] In the above formula, Indicates the current moment The overall deviation; Indicates the corrected number Data sources at the current moment The carbon metering value is taken from the corrected historical series. The latest sampling point; The result calculated in step S4 is the same as the first... Current time of each data source The corresponding fusion weights. This bias estimate essentially reflects the overall direction and magnitude of the deviation of the current fusion result from the observations of each data source.

[0109] Step S5-2: Introduce deviation compensation coefficient The fusion valuation is then corrected to obtain the final output. :

[0110]

[0111] In the above formula, The current moment after compensation The results of the electricity carbon metering can be directly used in scenarios such as electricity billing, carbon emission accounting, and energy consumption monitoring; This is the deviation compensation coefficient, used to adjust the strength of the compensation. Its value is determined using existing least squares calibration methods, and its range is [range missing]. .when A smaller value indicates a smaller correction to the original fusion result, suitable for stable systems with high data consistency; when... A larger value indicates a stronger reliance on multi-source consistency, thereby increasing the correction force for deviations.

[0112] Optionally, to achieve adaptive optimization of relevant parameters in the scheme, a loss function is constructed based on the total global deviation potential energy intensity obtained in step S3-3, combined with the curvature and estimation accuracy of the fused estimate. This loss function is then used to automatically adjust key parameters in the scheme, such as the stability penalty coefficient, through an iterative optimization method. Variance response sensitivity coefficient Periodic disturbance compensation amplitude control factor The method optimizes the fusion and bias compensation effects by minimizing the global bias potential energy while ensuring the smoothness and accuracy of the fused estimate. Through this parameter optimization process, the method can adaptively achieve optimal metrological performance based on the actual data characteristics.

[0113] It should be noted that, as will be apparent to those skilled in the art, the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics thereof. The scope of the present invention is defined by the claims rather than the foregoing description.

Claims

1. A method for fusing and compensating for heterogeneous data in carbon metering, comprising step S1: acquiring data from... The carbon metering data of the same object collected by three heterogeneous carbon metering sources are characterized by: In step S1, the carbon metering data of each data source are preprocessed according to time sequence, and a multi-source historical input sequence under a unified sliding window is constructed. Also includes: Step S2: Construct a nonlinear compensation function that integrates the hyperbolic tangent variance response suppression term and the sinusoidal periodic modulation term, correct each sample value in the historical input sequence of each data source point by point to obtain the corrected historical sequence, and construct the local residual trajectory tensor of each point of each data source based on the corrected historical sequence. Step S3: Calculate the residual resonance mapping value between any two data sources based on the local residual trajectory tensor; then calculate the total deviation potential energy based on the residual resonance mapping value to quantify the position of each data source in the overall deviation field; Step S4: Use the exponential normalization suppression model to convert the total bias potential of each data source into fusion weights; construct the residual covariance matrix based on all corrected historical sequences, and perform singular value decomposition to extract the first principal eigenvector as the principal fluctuation direction of the residual trajectory to obtain the principal projection results of each data source; then perform weighted summation based on the fusion weights and principal projection results of each data source to obtain the fusion estimate. Step S5: Construct a weighted deviation estimate between the fused estimate and the corrected values ​​from each data source, and introduce a deviation compensation coefficient to perform deviation compensation on the fused estimate, and output the final carbon metering result after compensation.

2. The method for fusing and compensating for heterogeneous data in carbon metering as described in claim 1, characterized in that: In step S1, for the first A data source, at the current moment The length of a sliding window is defined as follows: The historical input sequence is represented as: ； In the above formula, Indicates the first Data sources at time The historical input sequence; Indicates the first Data sources at time Preprocessed carbon metering data obtained from sampling ; It is the length of the sliding window; Indicates transpose; Step S2 specifically includes step S2-1: introducing a stability penalty coefficient. Variance response sensitivity coefficient and periodic disturbance compensation amplitude control factor A nonlinear compensation function is constructed to correct the historical input sequence, resulting in the corrected historical sequence. For the Historical input sequences of data sources At any time in raw value The correction calculation method is as follows: ； In the above formula, It is the revised version of the first Data source at time Carbon metering data; For the first Stability penalty coefficient of the data source; It is the variance response sensitivity coefficient; Represents the historical input sequence The variance; It is the hyperbolic tangent function; It is the periodic disturbance compensation amplitude control factor; The period of environmental disturbance; The corrected carbon metering data corresponding to each element in the historical input sequence constitute the corrected historical sequence. .

3. The method for fusing and compensating for heterogeneous data in carbon metering as described in claim 2, characterized in that, Step S2 also includes step S2-2: calculating the first... The corrected historical series mean of each data source And calculate the local residual trajectory tensor of each point in the corrected historical sequence one by one. : ； In the above formula, Indicates the first In the corrected historical sequence of the data source, the first... The local residual trajectory tensor of a point.

4. The method for fusing and compensating for heterogeneous data in carbon metering as described in claim 1, characterized in that, In step S3, for any two data sources and Calculate its residual resonance mapping value : ； In the above formula, It is the first The data source and the first Data sources at time The residual resonance mapping value; and They are the first The and the first The first data source The local residual trajectory tensor of a point; It is a preset stability constant; and They are the first The and the first The mean of the corrected historical series from each data source; It is the cosine frequency modulation coefficient; This is the size of the sliding window.

5. The method for fusing and compensating for heterogeneous data in carbon metering as described in claim 1, characterized in that, In step S3, the first Total deviation potential of each data source The calculation method is as follows: ； In the above formula, It is the first Total deviation potential of each data source; It is the first The data source and the first Data sources at the current moment The residual resonance mapping value; It is the natural logarithm function; and They are the first The and the first The mean of the corrected historical series from each data source.

6. The method for fusing and compensating for heterogeneous data in carbon metering as described in claim 1, characterized in that, The method for calculating the fusion weights in step S4 is as follows: ； In the above formula, For the first Data sources at time The fusion weight; It is a natural constant; It is the first The total deviation potential of each data source.

7. The method for fusing and compensating for heterogeneous data in carbon metering as described in claim 1, characterized in that, In step S4, the residual covariance matrix is ​​constructed. The method is as follows: ； In the above formula, It is the first The corrected historical sequence from a data source; It is a sequence of the mean values ​​from all data sources; This indicates transpose.

8. The method for fusing and compensating for heterogeneous data in carbon metering as described in claim 1, characterized in that, In step S4, the covariance matrix is... Perform singular value decomposition: ； In the above formula, It is a singular vector matrix. Given a singular value diagonal matrix, extract the first principal eigenvector corresponding to the largest singular value from it. The corrected historical sequence of each data source is projected onto the main fluctuation direction to obtain its main direction projection result. : ； In the above formula, Indicates the first The principal projection results of the data sources; It is the first The corrected historical sequence from a data source; It is the mean sequence of all data sources.

9. The method for fusing and compensating for heterogeneous data in carbon metering as described in claim 1, characterized in that, In step S4, the fusion valuation is calculated. The method is as follows: ； In the above formula, For a moment The integrated valuation, For the first Data sources at time The fusion weight, Indicates the first The main projection results of the data source.

10. The method for fusing and compensating for heterogeneous data in carbon metering as described in claim 1, characterized in that, The specific process of step S5 is as follows: Step S5-1: Based on the fusion valuation obtained in step S4 In addition to the corrected historical sequences of each data source obtained in step S2, a deviation estimate is constructed to characterize the overall deviation between the fusion result and the multi-source observation data. : ； In the above formula, Indicates the current moment The overall deviation; Indicates the first In the corrected historical sequence of data sources, at the current moment The carbon metering value is taken from the corrected historical series. The latest sampling point; The result calculated in step S4 is the same as the first... Current time of each data source The corresponding fusion weights; Step S5-2: Introduce deviation compensation coefficient The fusion valuation is then corrected to obtain the final output. : ； In the above formula, The current moment after compensation The results of the carbon metering; This is the deviation compensation coefficient.

Images