High-precision carbon emission online monitoring method and system based on multi-source data fusion

Abstract

The invention discloses a high-precision carbon emission online monitoring method and system based on multi-source data fusion, and relates to the technical field of carbon emission monitoring. The method comprises the following steps: constructing a multi-source heterogeneous carbon emission data acquisition layer, and acquiring carbon emission original data of a target area from different sources in real time; performing data preprocessing and feature extraction on the carbon emission original data to obtain a standard carbon emission feature data set; establishing a dynamic adaptive weight distribution mechanism, and distributing target fusion weights for standard carbon emission characteristic data sets from different sources; inputting the standard carbon emission characteristic data set configured with the target fusion weight into a pre-trained carbon emission inversion model, and calculating and generating the spatial gridding carbon emission intensity of the target area; and carrying out trend analysis on the spatial gridding carbon emission intensity, and executing abnormity alarm according to an analysis result. The technical problem of insufficient reliability of the carbon emission monitoring result in the prior art is solved, and the technical effect of improving the reliability of the carbon emission monitoring result is achieved.

Description

Technical Field

[0001] This invention relates to the field of carbon emission monitoring technology, specifically to a high-precision online carbon emission monitoring method and system based on multi-source data fusion. Background Technology

[0002] Carbon emission monitoring has become a crucial foundation for supporting regional emission reduction decisions and carbon asset management. However, existing carbon emission monitoring technologies largely rely on single data sources (such as remote sensing observations, ground monitoring stations, or statistical data). Data from different sources exhibit significant differences in temporal resolution, spatial resolution, and measurement accuracy, making it difficult to form a unified and highly reliable characterization of carbon emissions. Furthermore, due to the substantial differences in acquisition methods, noise characteristics, and spatiotemporal matching relationships among multi-source heterogeneous data, existing monitoring systems are prone to biases during data fusion, leading to instability in the prediction results of carbon emission inversion models. In addition, traditional monitoring methods rely heavily on manual or single-dimensional threshold judgments for the identification and response to abnormal emission events, making it difficult to reflect the dynamic trends of regional carbon emissions in a timely and accurate manner. Summary of the Invention

[0003] This application provides a high-precision online carbon emission monitoring method and system based on multi-source data fusion, which solves the technical problem of insufficient reliability of carbon emission monitoring results in the prior art.

[0004] The first aspect of this application provides a high-precision online carbon emission monitoring method based on multi-source data fusion, the method comprising: A multi-source heterogeneous carbon emission data acquisition layer is constructed to collect raw carbon emission data of the target area in real time from different sources; the raw carbon emission data is preprocessed and feature extracted to obtain a standard carbon emission feature dataset; a dynamic adaptive weight allocation mechanism is established to assign target fusion weights to the standard carbon emission feature datasets from different sources; the standard carbon emission feature datasets configured with target fusion weights are input into a pre-trained carbon emission inversion model to calculate and generate the spatial gridded carbon emission intensity of the target area; trend analysis is performed on the spatial gridded carbon emission intensity, and anomaly alarms are executed based on the analysis results.

[0005] A second aspect of this application provides a high-precision online carbon emission monitoring system based on multi-source data fusion, the system comprising: Data Acquisition Module: Constructs a multi-source heterogeneous carbon emission data acquisition layer to collect raw carbon emission data of the target area in real time from different sources; Data Preprocessing Module: Performs data preprocessing and feature extraction on the raw carbon emission data to obtain a standard carbon emission feature dataset; Weight Allocation Module: Establishes a dynamic adaptive weight allocation mechanism to assign target fusion weights to the standard carbon emission feature datasets from different sources; Carbon Emission Inversion Module: Inputs the standard carbon emission feature datasets configured with target fusion weights into a pre-trained carbon emission inversion model to calculate and generate the spatial gridded carbon emission intensity of the target area; Alarm Module: Performs trend analysis on the spatial gridded carbon emission intensity and executes anomaly alarms based on the analysis results.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: First, a multi-source heterogeneous carbon emission data acquisition layer is constructed to collect raw carbon emission data of the target area in real time from different sources. Next, the raw carbon emission data undergoes data preprocessing and feature extraction to obtain a standard carbon emission feature dataset. Furthermore, a dynamic adaptive weight allocation mechanism is established to assign target fusion weights to the standard carbon emission feature datasets from different sources. Then, the standard carbon emission feature datasets configured with target fusion weights are input into a pre-trained carbon emission inversion model to calculate and generate the spatially gridded carbon emission intensity of the target area. Finally, trend analysis is performed on the spatially gridded carbon emission intensity, and anomaly alarms are executed based on the analysis results. This solves the technical problem of insufficient reliability of carbon emission monitoring results in existing technologies, achieving the technical effect of improving the reliability of carbon emission monitoring results. Attached Figure Description

[0007] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 A schematic diagram of the process for a high-precision online carbon emission monitoring method based on multi-source data fusion provided in this application embodiment; Figure 2 A schematic diagram of the structure of a high-precision online carbon emission monitoring system based on multi-source data fusion provided in this application embodiment.

[0009] Figure labeling: Data acquisition module 11, data preprocessing module 12, weight allocation module 13, carbon emission inversion module 14, alarm module 15. Detailed Implementation

[0010] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0011] Example 1, as Figure 1 As shown, this application provides a high-precision online carbon emission monitoring method based on multi-source data fusion, wherein the method includes: Construct a multi-source heterogeneous carbon emission data acquisition layer to collect raw carbon emission data of the target area in real time from different sources.

[0012] In this embodiment, the multi-source data collection range is determined based on the administrative boundaries, energy structure, and distribution of major emission sources of the target area. Within this range, data channels from different departments, industries, and sensor networks are deployed or connected to form a multi-source heterogeneous carbon emission data collection layer. This layer includes enterprise-level direct monitoring terminals (such as online flue gas monitoring equipment and industrial emission sensors), activity level statistics interfaces (such as data collection interfaces for indicators like energy consumption, traffic activity, and industrial output), remote sensing observation channels (such as satellite remote sensing imagery and atmospheric composition inversion data access modules), and socio-economic data interfaces (such as regional GDP, population density, and climate and meteorological element data interfaces). Each data channel transmits data in real time through a unified acquisition protocol and time synchronization mechanism, forming raw carbon emission data.

[0013] Furthermore, the multi-source heterogeneous carbon emission data acquisition layer includes at least: The first data source is enterprise-level direct monitoring data; the second data source is activity level data; the third data source is environmental remote sensing observation data; and the fourth data source is socio-economic and macroeconomic situation data.

[0014] The primary data source is enterprise-level direct monitoring data, which mainly comes from continuous emission monitoring systems deployed in key emission facilities (such as thermal power plants, steel plants, cement plants, chemical plants, etc.). These systems collect data on carbon dioxide concentration, flow rate, temperature, humidity, and other accompanying gas concentrations in flue gas in real time and transmit them to the monitoring center server through industrial communication interfaces.

[0015] The second data source is activity level data, which reflects the intensity of carbon emission activities. Activity level data primarily comes from energy consumption statistics reports, enterprise production process databases, and traffic flow monitoring systems. Energy consumption statistics reports provide data on energy consumption such as coal, oil, gas, and electricity; production process databases record unit product output and process energy consumption; and traffic flow monitoring systems provide data on road traffic volume, vehicle type, and average speed.

[0016] The third data source is environmental remote sensing observation data, mainly from satellite remote sensing platforms or airborne monitoring systems. Environmental remote sensing observation data includes atmospheric carbon dioxide column concentration, nitrogen dioxide concentration, methane concentration, and nighttime light remote sensing data, etc. After being collected by optical, infrared, or laser detection payloads, high spatial resolution gas concentration distribution maps are generated through inversion algorithms.

[0017] The fourth data source is socioeconomic and macroeconomic data, used to depict the background trends in regional carbon emissions. This data originates from power grid load monitoring systems, industrial output statistics systems, and meteorological monitoring systems. Power grid load data reflects the time-varying patterns of regional energy consumption; industrial output data reflects the intensity of production activities; and meteorological data, including temperature, wind speed, humidity, and boundary layer height, can be used to explain the impact of atmospheric dilution and diffusion conditions on carbon concentration.

[0018] The raw carbon emission data is preprocessed and features are extracted to obtain a standard carbon emission feature dataset.

[0019] Data standardization and cleaning were performed on raw carbon emission data from various data sources. For enterprise-level direct monitoring data, communication anomalies, drifting data, and invalid zero values ​​were removed. For activity-level data, energy consumption, production, and traffic data from different sources were matched and corrected based on timestamps and spatial location. For remote sensing observation data, resampling and radiometric correction were performed based on observation time and geographic coordinates. For socioeconomic and meteorological data, unit unification, missing data interpolation, and noise filtering were performed. Through these processes, a standardized carbon emission dataset with consistent structure and spatiotemporal alignment was generated.

[0020] Based on a standard carbon emission dataset, a combined air-to-ground correction operation was performed on remote sensing observation data. By selecting satellite pixels corresponding to ground monitoring points, a regression model was established between carbon dioxide concentration and ground-based emission flux to correct biases and scale the remote sensing data. The corrected remote sensing data was then re-injected into the standard carbon emission dataset to improve consistency and comparability among multi-source data.

[0021] Multi-level feature indicators are extracted from the corrected standard carbon emission dataset, including: time-domain features reflecting emission change trends (such as daily change rate, seasonality coefficient, time series gradient, etc.), spatial features reflecting spatial distribution characteristics (such as emission hotspot clustering index, spatial heterogeneity coefficient, spatial autocorrelation coefficient, etc.), and spatiotemporal interaction features reflecting the spatiotemporal coupling relationship of emissions (such as cross-time lag correlation matrix, spatiotemporal synergy index, etc.). Finally, the extracted time-domain features, spatial features, and spatiotemporal coupling features are uniformly encoded into feature vectors with standard time identifiers and spatial grid identifiers, forming the standard carbon emission feature dataset.

[0022] Furthermore, the raw carbon emission data undergoes data preprocessing and feature extraction to obtain a standard carbon emission feature dataset, including: The raw carbon emission data is cleaned using a standardization process to generate a standard carbon emission dataset. Based on the standard carbon emission dataset, environmental remote sensing observation data is extracted and subjected to air-to-ground co-correction before being reintroduced into the standard carbon emission dataset. Temporal features, spatial features, and spatiotemporal coupling features are extracted from the corrected standard carbon emission dataset to form a standard carbon emission feature dataset with a unified spatiotemporal identifier.

[0023] First, standardized cleaning operations are performed on the raw carbon emission data. For raw data from different sources, cleaning strategies adapted to their data characteristics are adopted: for enterprise-level direct monitoring data, noise removal, drift correction, and outlier removal are performed; for activity-level data, data consistency processing is performed based on time alignment and industry code matching; for socio-economic and meteorological data, missing data interpolation, unit conversion, and numerical smoothing are performed; and for remote sensing observation data, geometric correction and radiometric calibration are performed. After unified cleaning, a structured and comparable standard carbon emission dataset is generated. Second, based on the standard carbon emission dataset, environmental remote sensing observation data is extracted and subjected to air-to-ground co-correction operations. These air-to-ground co-correction operations include at least: radiometric calibration, atmospheric profile correction, georegistration, and spatial downscaling. Radiometric calibration is used to correct observation response errors of satellite sensors; atmospheric profile correction eliminates deviations caused by different observation conditions by introducing atmospheric temperature, humidity, and aerosol parameters; georegistration establishes spatial matching relationships between ground monitoring points and remote sensing pixels to achieve air-to-ground spatial alignment; spatial downscaling uses interpolation or machine learning methods to reduce the spatial resolution of remote sensing data to match the grid scale of the target area. The remote sensing data, after air-to-ground co-correction, is then reintroduced into the standard carbon emission dataset to enhance consistency across multiple data sources. Finally, multidimensional information indicators characterizing the dynamic features of carbon emissions are extracted from the corrected standard carbon emission dataset, including: temporal features (such as daily variation rate, peak-to-valley difference, and fluctuation frequency), spatial features (such as emission hotspot intensity, spatial gradient distribution, and local spatial autocorrelation coefficient), and spatiotemporal coupling features (such as spatiotemporal correlation matrix, time lag propagation coefficient, and cross-regional synergy index). By uniformly encoding these features, a standard carbon emission feature dataset with a unified timestamp and spatial grid identifier is generated.

[0024] A dynamic adaptive weight allocation mechanism is established to assign target fusion weights to the standard carbon emission feature datasets from different sources.

[0025] Furthermore, a dynamic adaptive weight allocation mechanism is established to assign target fusion weights to the standard carbon emission feature datasets from different sources, including: Based on the quality of the data source, a multi-dimensional evaluation index system is constructed for the standard carbon emission characteristic dataset; based on the multi-dimensional evaluation index system, the basic weights of each data source are calculated and generated; a time decay factor is introduced, and the basic weights are dynamically adjusted according to the quality changes of each data source in the recent monitoring period, and the target fusion weights are output.

[0026] A multi-dimensional evaluation index system for the standard carbon emission characteristic dataset is constructed based on data source quality. This evaluation index system comprehensively characterizes the performance of each data source in terms of data accuracy, stability, and representativeness, and includes at least the following indicators: data integrity index (reflecting the missing data rate and outlier ratio), timeliness index (measuring data update delay and real-time performance), spatial coverage index (describing the spatial distribution balance of data within the target area), noise level index (assessing the degree of acquisition error and interference), and historical stability index (reflecting the degree of data fluctuation over multiple monitoring periods). All indicators are dimensionless after calculation using quantization functions to ensure comparability. Based on the multi-dimensional evaluation index system, weighted summation is used to score the standard carbon emission characteristic datasets from different sources, generating basic weights for each data source. These basic weights represent the fusion reliability of the data source under static conditions; higher weight values ​​indicate a greater contribution to the overall carbon emission characterization results. A time decay factor is introduced to dynamically adjust the basic weights. The time decay factor is calculated based on the quality change trend of each data source within the recent monitoring period. When a data source experiences data anomalies, packet loss, or high noise, the system automatically decreases its weight; conversely, when data quality stabilizes or improves, its weight is automatically increased. The time decay factor can be implemented using an exponential decay function or a sliding time window function to ensure the continuity and smoothness of the weight update process. The target fusion weight is calculated by combining the basic weight and the time decay factor, forming a dynamically adaptive multi-source weight allocation result.

[0027] The standard carbon emission feature dataset configured with target fusion weights is input into a pre-trained carbon emission inversion model to calculate the spatial gridded carbon emission intensity of the target region.

[0028] First, based on preset geographical boundaries and monitoring accuracy requirements, the target area is divided into several fixed or variable-sized geographical grid units. Each grid unit contains multi-source carbon emission characteristic data collected within that area. For the standard carbon emission characteristic data within each grid unit, weighted fusion processing is performed on the data from different sources according to the target fusion weights generated by the aforementioned dynamic adaptive weight allocation mechanism, calculating the weighted fusion feature vector for each grid unit. The weighted fusion feature vectors of each grid are then sequentially input into a pre-trained carbon emission inversion model. The carbon emission inversion model is a deep learning model jointly trained based on historical standard carbon emission characteristic datasets and carbon emission inventory data. It can employ a structure combining convolutional neural networks and long short-term memory networks to simultaneously capture the spatial dependence and temporal dynamic characteristics of carbon emissions. The model calculates and outputs the quantified carbon emission intensity value of each grid unit through a nonlinear mapping relationship, i.e., the amount of carbon dioxide emitted per unit time or equivalent emission intensity. The quantified carbon emission intensity values ​​output by each grid unit are spatiotemporally reorganized and visualized to generate a spatially gridded carbon emission intensity map covering the entire target area.

[0029] Furthermore, the construction of the carbon emission inversion model includes: A joint modeling approach based on spatial and temporal dependencies is used to build an initial network architecture. The initial network architecture is iteratively trained using historical standard carbon emission feature datasets as training samples and contemporaneous carbon emission inventory data as supervision signals until the model converges. Repeated sampling validation is used to evaluate the performance stability of the converged model, and a carbon emission inversion model is selected accordingly.

[0030] First, an initial network architecture is built by jointly modeling spatial and temporal dependencies. The model structure adopts a spatiotemporal fusion design framework, where the spatial dependency part uses a convolutional neural network to extract spatial correlation features between different grid regions, and the temporal dependency part uses a long short-term memory network or a gated recurrent singleton to capture the dynamic trend of carbon emissions over time. The model input receives standard carbon emission feature datasets from each time slice, and the output generates the predicted value of regional carbon emission intensity at the corresponding time point. Second, the initial network architecture is iteratively trained using historical standard carbon emission feature datasets as training samples and carbon emission inventory data verified by authoritative statistical agencies or inventory accounting systems during the same period as supervision signals. During training, the Adam or SGD optimization algorithm is used, defining the mean squared error or mean absolute error as the loss function, and updating the model parameters through backpropagation; the training set and validation set are randomly divided in each training round, and an early stopping strategy is used to prevent overfitting. The training process continues until the model converges, that is, the rate of change of the loss function is lower than a preset threshold over several consecutive rounds. After the model converges, repeated sampling validation (K-fold cross-validation) is used to evaluate the stability of the model performance. By repeatedly dividing the sample set and training the model repeatedly, the mean and standard deviation of the prediction error under different data partitions are calculated to evaluate its generalization ability and robustness. If the performance indicators meet the set convergence conditions (e.g., the standard deviation of the prediction error is lower than a preset threshold), the trained model is selected as the final carbon emission inversion model.

[0031] Furthermore, the standard carbon emission feature dataset configured with target fusion weights is input into a pre-trained carbon emission inversion model to calculate and generate the spatially gridded carbon emission intensity of the target region, including: According to the preset geographic grid, the standard carbon emission feature dataset configured with target fusion weights is spatially divided, and a weighted fusion feature vector is generated in each grid. The weighted fusion feature vector in each grid is then sequentially input into the carbon emission inversion model. The carbon emission inversion model outputs the carbon emission intensity quantification value of the corresponding region grid by grid, and the data is aggregated to generate a spatial gridded carbon emission intensity map covering the target region.

[0032] According to preset geographic grid division rules, the standard carbon emission feature dataset configured with target fusion weights is spatially divided. The geographic grid can be generated based on latitude and longitude or administrative boundaries. The grid resolution is adaptively set according to the area of ​​the target region, data density, and monitoring accuracy requirements; for example, a spatial division unit of 1km×1km or a finer scale can be used. After spatial division, multi-source feature data belonging to the same grid area are aggregated. Based on the target fusion weights determined by the aforementioned dynamic adaptive weight allocation mechanism, weighted fusion calculations are performed on data from different sources to form a weighted fusion feature vector for each grid unit. This feature vector includes emission activity intensity characteristics, environmental remote sensing observation characteristics, meteorological condition characteristics, and macro-socioeconomic characteristics, used to comprehensively describe the carbon emission characteristic state of the grid unit. The weighted fusion feature vectors within each grid are sequentially input into a pre-trained carbon emission inversion model. The carbon emission inversion model is built based on a deep learning structure, which can simultaneously capture the spatial correlation and temporal dynamic characteristics of carbon emissions, and outputs the quantified value of carbon emission intensity corresponding to the grid unit through a nonlinear mapping relationship. The quantified carbon emission intensity values ​​output by each grid cell are spatially stitched and temporally processed to generate a spatially gridded carbon emission intensity map covering the target area. The spatially gridded carbon emission intensity map uses a unified geographic coordinate system as a reference and reflects the distribution of carbon emission intensity in different spatial locations of the target area.

[0033] The spatial gridded carbon emission intensity is analyzed for trends, and anomaly alarms are executed based on the analysis results.

[0034] Furthermore, trend analysis is performed on the spatially gridded carbon emission intensity, and anomaly alerts are executed based on the analysis results, including: Historical spatial gridded carbon emission intensity data is acquired, and a dynamic baseline model representing its normal fluctuation range is constructed. The real-time generated spatial gridded carbon emission intensity is compared with the dynamic baseline model. When a persistent deviation is detected, it is diagnosed as an emission anomaly event. Based on the impact range of the emission anomaly event, a graded early warning is triggered.

[0035] First, historical spatial gridded carbon emission intensity data is acquired to construct a dynamic baseline model characterizing its normal fluctuation range. This dynamic baseline model is based on the historical carbon emission intensity time series of each spatial grid unit. Through sliding time window calculation, seasonal trend decomposition, and fluctuation range estimation, the average variation trend, periodic fluctuation characteristics, and standard deviation range of each grid are extracted. To adapt to changes at different time scales, a multi-time scale fusion mechanism can be introduced into the model, simultaneously considering daily, weekly, and seasonal fluctuation patterns, thereby establishing a dynamic baseline model that reflects the normal fluctuation range of carbon emission intensity in the region. Second, the real-time generated spatial gridded carbon emission intensity is compared with the dynamic baseline model. At each time step, the system performs a difference analysis between the real-time monitoring results and the baseline model output, calculating deviation indicators such as relative deviation rate, Z-score, or standardized residuals. When the carbon emission intensity of a certain grid unit deviates from the normal fluctuation range of the baseline model for multiple consecutive monitoring periods and the deviation exceeds a preset threshold, it is determined that the grid has a persistent abnormal change trend and is diagnosed as an emission anomaly event. The identification process for abnormal events can employ sliding window detection algorithms or sequence anomaly detection models to ensure the timeliness and accuracy of anomaly identification. Finally, based on the impact range of the emission anomaly event, a tiered early warning system is triggered. The system automatically calculates the anomaly impact index based on the number of abnormal grids, their spatial clustering, and the magnitude of deviation, and responds in three levels: mild, moderate, and severe. Mild warnings are used to indicate localized short-term anomalies, moderate warnings to indicate regional emission anomaly trends, and severe warnings to identify major emission anomalies and trigger emergency response mechanisms. The warning information includes the time of anomaly occurrence, coordinates of the anomaly area, degree of deviation, impact range, and suggested response measures, and is pushed in real-time to regulatory or decision-making authorities through the monitoring and management platform, enabling rapid identification and dynamic alerts for regional carbon emission anomalies.

[0036] In summary, the embodiments of this application have at least the following technical effects: First, a multi-source heterogeneous carbon emission data acquisition layer is constructed to collect raw carbon emission data of the target area in real time from different sources. Next, the raw carbon emission data undergoes data preprocessing and feature extraction to obtain a standard carbon emission feature dataset. Furthermore, a dynamic adaptive weight allocation mechanism is established to assign target fusion weights to the standard carbon emission feature datasets from different sources. Then, the standard carbon emission feature datasets configured with target fusion weights are input into a pre-trained carbon emission inversion model to calculate and generate the spatially gridded carbon emission intensity of the target area. Finally, trend analysis is performed on the spatially gridded carbon emission intensity, and anomaly alarms are executed based on the analysis results. This solves the technical problem of insufficient reliability of carbon emission monitoring results in existing technologies, achieving the technical effect of improving the reliability of carbon emission monitoring results.

[0037] Example 2 is based on the same inventive concept as the high-precision online carbon emission monitoring method based on multi-source data fusion in the previous examples, such as... Figure 2 As shown, this application provides a high-precision online carbon emission monitoring system based on multi-source data fusion, wherein the system includes: Data acquisition module 11: Constructs a multi-source heterogeneous carbon emission data acquisition layer to collect raw carbon emission data of the target area in real time from different sources; Data preprocessing module 12: Performs data preprocessing and feature extraction on the raw carbon emission data to obtain a standard carbon emission feature dataset; Weight allocation module 13: Establishes a dynamic adaptive weight allocation mechanism to assign target fusion weights to the standard carbon emission feature datasets from different sources; Carbon emission inversion module 14: Inputs the standard carbon emission feature datasets configured with target fusion weights into a pre-trained carbon emission inversion model to calculate and generate the spatial gridded carbon emission intensity of the target area; Alarm module 15: Performs trend analysis on the spatial gridded carbon emission intensity and executes anomaly alarms based on the analysis results.

[0038] Furthermore, the data acquisition module 11 is used to perform the following methods: The first data source is enterprise-level direct monitoring data; the second data source is activity level data; the third data source is environmental remote sensing observation data; and the fourth data source is socio-economic and macroeconomic situation data.

[0039] Furthermore, the data preprocessing module 12 is used to perform the following methods: The raw carbon emission data is cleaned using a standardization process to generate a standard carbon emission dataset. Based on the standard carbon emission dataset, environmental remote sensing observation data is extracted and subjected to air-to-ground co-correction before being reintroduced into the standard carbon emission dataset. Temporal features, spatial features, and spatiotemporal coupling features are extracted from the corrected standard carbon emission dataset to form a standard carbon emission feature dataset with a unified spatiotemporal identifier.

[0040] Furthermore, the weight allocation module 13 is used to perform the following method: Based on the quality of the data source, a multi-dimensional evaluation index system is constructed for the standard carbon emission characteristic dataset; based on the multi-dimensional evaluation index system, the basic weights of each data source are calculated and generated; a time decay factor is introduced, and the basic weights are dynamically adjusted according to the quality changes of each data source in the recent monitoring period, and the target fusion weights are output.

[0041] Furthermore, the carbon emission inversion module 14 is used to perform the following method: A joint modeling approach based on spatial and temporal dependencies is used to build an initial network architecture. The initial network architecture is iteratively trained using historical standard carbon emission feature datasets as training samples and contemporaneous carbon emission inventory data as supervision signals until the model converges. Repeated sampling validation is used to evaluate the performance stability of the converged model, and a carbon emission inversion model is selected accordingly.

[0042] Furthermore, the carbon emission inversion module 14 is used to perform the following method: According to the preset geographic grid, the standard carbon emission feature dataset configured with target fusion weights is spatially divided, and a weighted fusion feature vector is generated in each grid. The weighted fusion feature vector in each grid is then sequentially input into the carbon emission inversion model. The carbon emission inversion model outputs the carbon emission intensity quantification value of the corresponding region grid by grid, and the data is aggregated to generate a spatial gridded carbon emission intensity map covering the target region.

[0043] Furthermore, the alarm module 15 is used to perform the following method: Historical spatial gridded carbon emission intensity data is acquired, and a dynamic baseline model representing its normal fluctuation range is constructed. The real-time generated spatial gridded carbon emission intensity is compared with the dynamic baseline model. When a persistent deviation is detected, it is diagnosed as an emission anomaly event. Based on the impact range of the emission anomaly event, a graded early warning is triggered.

[0044] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A high-precision carbon emission online monitoring method based on multi-source data fusion, characterized in that, The method includes: Construct a multi-source heterogeneous carbon emission data acquisition layer to collect raw carbon emission data of the target area in real time from different sources; The raw carbon emission data is preprocessed and features are extracted to obtain a standard carbon emission feature dataset; A dynamic adaptive weight allocation mechanism is established to assign target fusion weights to the standard carbon emission feature datasets from different sources; The standard carbon emission feature dataset configured with target fusion weights is input into the pre-trained carbon emission inversion model to calculate the spatial gridded carbon emission intensity of the target region. The spatial gridded carbon emission intensity is analyzed for trends, and anomaly alarms are executed based on the analysis results.

2. The high-precision carbon emission online monitoring method based on multi-source data fusion according to claim 1, characterized in that, The multi-source heterogeneous carbon emission data acquisition layer includes at least: The primary data source is enterprise-level direct monitoring data. The second data source is activity level data; The third data source is environmental remote sensing observation data; The fourth data source is socioeconomic and macroeconomic data.

3. The high-precision online carbon emission monitoring method based on multi-source data fusion as described in claim 1, characterized in that, The raw carbon emission data is preprocessed and features are extracted to obtain a standard carbon emission feature dataset, including: Perform a standardization cleaning operation on the raw carbon emission data to generate a standard carbon emission dataset; Based on the standard carbon emission dataset, environmental remote sensing observation data is extracted, and after performing air-to-ground collaborative correction, it is put back into the standard carbon emission dataset. From the corrected standard carbon emission dataset, temporal features, spatial features, and spatiotemporal coupling features are extracted to form a standard carbon emission feature dataset with unified spatiotemporal identification.

4. The high-precision online carbon emission monitoring method based on multi-source data fusion as described in claim 1, characterized in that, A dynamic adaptive weight allocation mechanism is established to assign target fusion weights to the standard carbon emission feature datasets from different sources, including: Based on the quality of the data source, a multi-dimensional evaluation index system is constructed for the standard carbon emission characteristic dataset. Based on the aforementioned multidimensional evaluation index system, the basic weights of each data source are calculated and generated. A time decay factor is introduced to dynamically adjust the basic weights based on the quality changes of each data source within the recent monitoring period, and the target fusion weights are output.

5. The high-precision carbon emission online monitoring method based on multi-source data fusion according to claim 1, characterized in that, The construction of the carbon emission inversion model includes: A joint model is performed based on spatial and temporal dependencies to build the initial network architecture; Using historical standard carbon emission feature datasets as training samples and carbon emission inventory data from the same period as supervision signals, the initial network architecture is iteratively trained until the model converges. Repeated sampling verification was used to evaluate the performance stability of the converged model, and the carbon emission inversion model was selected accordingly.

6. The high-precision carbon emission online monitoring method based on multi-source data fusion according to claim 1, characterized in that, The standard carbon emission feature dataset configured with target fusion weights is input into a pre-trained carbon emission inversion model to calculate and generate the spatially gridded carbon emission intensity of the target region, including: According to the preset geographic grid, the standard carbon emission feature dataset configured with target fusion weights is spatially divided, and a weighted fusion feature vector is generated in each grid. The weighted fusion feature vectors within each grid are sequentially input into the carbon emission inversion model. The carbon emission inversion model outputs the quantified carbon emission intensity values ​​for the corresponding region grid by grid, and then aggregates them to generate a spatially gridded carbon emission intensity map covering the target region.

7. The high-precision carbon emission online monitoring method based on multi-source data fusion according to claim 1, characterized in that, Perform trend analysis on the spatially gridded carbon emission intensity and execute anomaly alarms based on the analysis results, including: Acquire historical spatial gridded carbon emission intensity data and construct a dynamic baseline model characterizing its normal fluctuation range; The spatially gridded carbon emission intensity generated in real time is compared with the dynamic baseline model. When a persistent deviation is detected, it is diagnosed as an emission anomaly event. A tiered early warning system is triggered based on the scope of impact of the aforementioned abnormal emission events.

8. A high-precision carbon emission online monitoring system based on multi-source data fusion, characterized in that, The system is used to implement the high-precision online carbon emission monitoring method based on multi-source data fusion as described in any one of claims 1-7, the system comprising: Data acquisition module: Constructs a multi-source heterogeneous carbon emission data acquisition layer to collect raw carbon emission data of the target area in real time from different sources; Data preprocessing module: performs data preprocessing and feature extraction on the raw carbon emission data to obtain a standard carbon emission feature dataset; Weight allocation module: Establishes a dynamic adaptive weight allocation mechanism to assign target fusion weights to the standard carbon emission feature datasets from different sources; Carbon emission inversion module: Input the standard carbon emission feature dataset configured with target fusion weights into the pre-trained carbon emission inversion model to calculate and generate the spatial gridded carbon emission intensity of the target region; Alarm module: Performs trend analysis on the spatial gridded carbon emission intensity and executes anomaly alarms based on the analysis results.

Images