Carbon emission estimation methods, devices, computer equipment and storage media

Abstract

This application relates to a carbon emission estimation method, apparatus, computer equipment, storage medium, and computer program product. The method includes: acquiring raw multimodal carbon data; performing matching processing on the raw multimodal carbon data to establish correlations between the raw multimodal carbon data, obtaining target multimodal carbon data; using the target multimodal carbon data as input, calling a trained carbon emission estimation model to estimate carbon emissions, obtaining a carbon emission estimation result, wherein the carbon emission estimation model is trained based on historical target multimodal carbon data. This method can obtain carbon emission estimation results with higher reliability.

Description

Technical Field

[0001] This application relates to the field of carbon emission technology, and in particular to a carbon emission estimation method, apparatus, computer equipment, storage medium, and computer program product. Background Technology

[0002] Building an accurate and objective carbon measurement system is of paramount importance.

[0003] Existing carbon measurement methods mainly rely on data calculated and reported by enterprises or individuals themselves. For example, macro-level measurement is carried out through carbon data provided by authoritative organizations, or lower-level units calculate their own carbon emissions and report and summarize them to higher-level units, such as enterprises' self-monitoring of carbon emissions.

[0004] However, due to the lack of standardization in existing carbon emission accounting methods and the absence of a unified calculation standard, the accuracy and completeness of such data are often questioned, and it often fails to accurately reflect the actual carbon emission situation. In other words, the carbon emission results obtained through existing methods have low reliability. Summary of the Invention

[0005] Therefore, it is necessary to provide a multimodal carbon data fusion analysis method, apparatus, computer equipment, computer-readable storage medium, and computer program product that can improve the reliability of carbon emission results, addressing the aforementioned technical problems.

[0006] Firstly, this application provides a method for estimating carbon emissions. The method includes:

[0007] Obtain raw multimodal carbon data;

[0008] The original multimodal carbon data is matched to establish the correlation between the original multimodal carbon data, thereby obtaining the target multimodal carbon data;

[0009] Using the target multimodal carbon data as input, a trained carbon emission estimation model is invoked to estimate carbon emissions, and the carbon emission estimation result is obtained. The carbon emission estimation model is trained based on historical target multimodal carbon data.

[0010] In one embodiment, the multimodal carbon data includes carbon satellite data and ground-based carbon data, and the target multimodal carbon data includes a first multimodal dataset;

[0011] The matching process for the original multimodal carbon data, establishing the correlation between the original multimodal carbon data, and obtaining the target multimodal carbon data includes:

[0012] Centered on the carbon emission sources of the ground-based carbon data, carbon satellite data within a preset distance range is matched to obtain a first multimodal carbon dataset, which includes carbon satellite data and ground-based carbon data that are geographically matched.

[0013] In one embodiment, the method further comprises:

[0014] Remove anomalous carbon satellite data from the first multimodal carbon dataset;

[0015] Carbon satellite data is considered anomalous if it meets at least one of the following conditions:

[0016] The first item is that the number of carbon emission sources matched with the carbon satellite data is greater than a preset threshold for the number of carbon emission sources;

[0017] The second item is that the width of the satellite strip corresponding to the carbon satellite data is less than a preset strip width threshold;

[0018] Thirdly, the number of satellite strip recording regions corresponding to the carbon satellite data is less than a preset threshold for the number of regions;

[0019] Fourthly, the carbon satellite data mentioned are outliers.

[0020] In one embodiment, the multimodal carbon data further includes environmental data, and the target multimodal carbon data further includes a second multimodal carbon dataset;

[0021] After obtaining the first multimodal carbon dataset, the process also includes:

[0022] Based on the collection time period of carbon satellite data in the first multimodal carbon dataset, environmental data collected within a preset time range are matched to obtain the second multimodal carbon dataset;

[0023] The preset time range is determined based on the collection period, and the second multimodal carbon dataset includes time-matched carbon satellite data and environmental data.

[0024] In one embodiment, before calling the trained carbon emission estimation model to estimate carbon emissions using the target multimodal carbon data as input, the method further includes:

[0025] Acquire historical multimodal carbon data;

[0026] The historical multimodal carbon data is matched to establish the correlation between the historical multimodal carbon data, thereby obtaining the historical target multimodal carbon data.

[0027] The historical target multimodal carbon data is encoded to obtain encoded historical target multimodal carbon data;

[0028] Using the encoded historical target multimodal carbon data as training data, a pre-built deep learning model is trained to obtain a trained carbon emission estimation model.

[0029] In one embodiment, the encoded historical target multimodal carbon data includes a first historical multimodal carbon dataset without carbon emission labels and a second historical multimodal carbon dataset with carbon emission labels.

[0030] The step of training a pre-built deep learning model using the encoded historical target multimodal carbon data as training data to obtain a trained carbon emission estimation model includes:

[0031] Using the encoded first historical multimodal carbon dataset as training data, a self-supervised carbon emission pre-training sub-model in the pre-constructed deep learning model is trained. The self-supervised carbon emission pre-training sub-model is used to extract carbon emission feature data from the target multimodal carbon data.

[0032] Using the encoded second historical multimodal carbon dataset as training data, the linear regression sub-model in the pre-constructed deep learning model is trained to obtain the trained carbon emission estimation model.

[0033] The first historical multimodal carbon dataset includes geographically matched historical carbon satellite data and historical ground carbon data, while the second historical multimodal carbon dataset includes time-matched historical carbon satellite data and historical environmental data with carbon emission labels. The linear regression model is used to fit the carbon emission characteristic data and estimate the carbon emission results based on the carbon emission characteristic data.

[0034] Secondly, this application also provides a carbon emission estimation device. The device includes:

[0035] The data acquisition module is used to acquire raw multimodal carbon data;

[0036] The data preprocessing module is used to perform matching processing on the original multimodal carbon data, establish the correlation between the original multimodal carbon data, and obtain the target multimodal carbon data.

[0037] The data analysis module is used to take the target multimodal carbon data as input, call the trained carbon emission estimation model to estimate carbon emissions, and obtain carbon emission estimation results. The carbon emission estimation model is trained based on historical target multimodal carbon data.

[0038] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method described above.

[0039] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the method described above.

[0040] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the steps of the method described above.

[0041] The aforementioned carbon emission estimation methods, devices, computer equipment, storage media, and computer program products differ from traditional, single-modal, and poorly standardized carbon emission estimation methods. They combine multimodal carbon data with a trained carbon emission estimation model. Specifically, the carbon emission estimation model is pre-trained based on historical target multimodal carbon data. The model captures the carbon emission characteristics of different modes from this historical data and, combined with the experience and patterns observed in the historical target multimodal carbon data, can quickly and accurately estimate reliable carbon data. In the actual carbon emission estimation process, the original multimodal carbon data is acquired and matched to obtain the target multimodal carbon data. This establishes the correlation between the original multimodal carbon data, revealing the mutual influence and relationships between different modes, which helps improve the accuracy and reliability of carbon emission estimation. Then, using the target multimodal carbon data as input, the trained carbon emission estimation model is invoked to perform carbon emission estimation, resulting in a rapid, comprehensive, and accurate carbon emission estimation result. Therefore, compared to traditional methods of carbon emission estimation, this approach yields more reliable carbon emission estimation results. Attached Figure Description

[0042] Figure 1 This is a diagram illustrating the application environment of a carbon emission estimation method in one embodiment;

[0043] Figure 2 This is a flowchart illustrating a carbon emission estimation method in one embodiment;

[0044] Figure 3 This is a schematic diagram of the carbon data matching process in one embodiment;

[0045] Figure 4 This is a schematic diagram of the carbon data matching process in another embodiment;

[0046] Figure 5 This is a schematic diagram of the process for constructing a carbon emission estimation model in one embodiment;

[0047] Figure 6 This is a schematic diagram illustrating the process of constructing a carbon emission estimation model in another embodiment;

[0048] Figure 7 This is a structural block diagram of a carbon emission estimation device in one embodiment;

[0049] Figure 8 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. Detailed Implementation

[0050] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0051] The carbon emission estimation method provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or located in the cloud or on other network servers. The data storage system of server 104 has a pre-stored trained carbon emission estimation model, which is trained based on historical target multimodal carbon data.

[0052] Specifically, an operator can upload the collected raw multimodal carbon data to server 104 via a communication network through terminal 102 and send a carbon emission estimation request to server 104. Server 104 responds to the request, performs matching processing on the raw multimodal carbon data, establishes the correlation between the raw multimodal carbon data, and obtains the target multimodal carbon data. Then, server 104 calls a trained carbon emission estimation model to estimate the carbon emissions from the target multimodal carbon data, thus obtaining the carbon emission estimation result. Terminal 102 can be, but is not limited to, various carbon data acquisition devices, and server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0053] In one embodiment, such as Figure 2 As shown, a carbon emission estimation method is provided, which is then applied to... Figure 1 Taking server 104 as an example, the following steps are included:

[0054] S200, acquire raw multimodal carbon data.

[0055] Carbon data can be carbon dioxide emissions per unit time or carbon dioxide column concentration in the atmosphere. Raw multimodal carbon data can include carbon data monitored by remote sensing carbon satellites, carbon data monitored by ground carbon monitoring equipment, environmental data, or data provided by other relevant carbon emission monitoring agencies.

[0056] For example, carbon data monitored by remote sensing carbon satellites (hereinafter referred to as carbon satellite data) can be obtained by accessing a dedicated satellite data platform, environmental data for the corresponding time period can be obtained through environmental monitoring agencies, and corresponding carbon data (hereinafter referred to as ground carbon data) can be obtained by accessing the monitoring platform corresponding to ground carbon monitoring equipment. Raw multimodal carbon data can be obtained directly through various means, and the carbon data between different modes has not yet been processed.

[0057] S400 performs matching processing on the original multimodal carbon data, establishes the correlation between the original multimodal carbon data, and obtains the target multimodal carbon data.

[0058] Following the steps above, since the obtained raw multimodal carbon data may be isolated and no correlation has been established between carbon data of different modes, it is necessary to perform matching processing on the raw multimodal carbon data to establish the correlation between the raw multimodal carbon data and obtain target multimodal carbon data for subsequent carbon emission estimation.

[0059] Specifically, the matching process between raw multimodal carbon data can involve training machine learning models or statistical models to establish a matching association model for the raw multimodal carbon data, and then using this matching association model to match the raw multimodal carbon data to obtain the target multimodal carbon data. The matching process can include temporal and spatial matching of the raw multimodal carbon data, as well as constructing other associations between different modes of carbon data. For example, correlation analysis and time-series analysis can be performed on carbon satellite data and ground-based carbon data to establish associations between different modes of carbon data.

[0060] In addition, in other embodiments, temporal resolution matching, spatial resolution matching, and data format matching can be performed on the raw multimodal carbon data. The specific matching process performed on the raw multimodal carbon data can be determined based on actual needs and the acquired data modes.

[0061] S600 takes target multimodal carbon data as input, calls a trained carbon emission estimation model to estimate carbon emissions, and obtains carbon emission estimation results. The carbon emission estimation model is trained based on historical target multimodal carbon data.

[0062] The trained carbon emission estimation model is based on historical target multimodal carbon data. This means the model can extract carbon emission-related feature data from the target multimodal carbon data, analyze this feature data, and output carbon emission estimation results. The carbon emission estimation results can include the carbon dioxide emissions from a specific emission source within a certain time period.

[0063] For example, in practical applications, the target multimodal carbon data includes carbon satellite data, ground-based monitoring data, and environmental data such as wind speed. Before inputting the target multimodal carbon data into a trained carbon emission estimation model, it may be necessary to standardize it to ensure it conforms to the format of the training data used during model training. The trained carbon emission estimation model estimates carbon emissions from the target multimodal data, outputting the carbon emission estimation result. Furthermore, this result can be visualized, for example, using charts or graphs, to help businesses or individuals better understand the carbon emission estimation results.

[0064] The aforementioned carbon emission estimation method differs from traditional, single-modal, and poorly standardized approaches. It combines multimodal carbon data with a trained carbon emission estimation model. Specifically, the model is pre-trained based on historical target multimodal carbon data. This model captures the carbon emission characteristics of different modes from the historical target multimodal carbon data and, combined with the experience and patterns observed in that data, can quickly and accurately estimate reliable carbon data. In the actual carbon emission estimation process, the original multimodal carbon data is acquired and matched to obtain the target multimodal carbon data. This establishes the correlations between the original multimodal carbon data, revealing the mutual influences and relationships between different modes, which improves the accuracy and reliability of carbon emission estimation. Then, using the target multimodal carbon data as input, the trained carbon emission estimation model is invoked to perform carbon emission estimation, resulting in a rapid, comprehensive, and accurate estimate. Therefore, compared to traditional methods, this approach yields more reliable carbon emission estimation results.

[0065] In one embodiment, the multimodal carbon data includes carbon satellite data and ground-based carbon data, and the target multimodal carbon data includes a first multimodal dataset.

[0066] like Figure 3 As shown, S400 includes: S420, taking the carbon emission source of the ground carbon data as the center, matching carbon satellite data within a preset distance range to obtain a first multimodal carbon dataset, the first multimodal carbon dataset including carbon satellite data and ground carbon data that are geographically matched.

[0067] In practical applications, carbon satellite data is carbon data collected by remote sensing satellites. As these satellites move in orbit, they periodically record carbon data at different geographical locations, which is then recorded on satellite strips. Each time a remote sensing satellite passes through the same orbit, it generates a new strip representing the carbon data recorded within that time period. By scanning the data acquisition area of ​​these satellite strips, carbon satellite data can be obtained, including carbon dioxide column concentration, latitude and longitude, carbon dioxide column concentration mass label, carbon dioxide column concentration uncertainty, solar azimuth angle, and satellite azimuth angle.

[0068] In this embodiment, matching carbon satellite data can include geographic location matching from a spatial dimension. Specifically, the carbon emission sources of the ground-based carbon data can be located first to confirm their location information, which can be represented by longitude and latitude and can be obtained directly from ground monitoring stations. Centered on the carbon emission source, an area can be determined, which can be a square area or a circular area. The carbon satellite data within this area will be matched with the ground-based carbon data corresponding to the carbon emission source to obtain a first multimodal carbon dataset that includes geographically matched carbon satellite data and ground-based carbon data.

[0069] For example, carbon satellite data is filtered based on the geographical location of the carbon emission sources. Considering the spatial resolution of carbon satellite data, nearby carbon emission sources can be considered as a single source. For instance, two carbon emission sources less than 3 km apart can be treated as one source to reduce the complexity of data processing and facilitate carbon data analysis. Then, using the merged carbon emission source as the center, the ground-based carbon data of that source is matched with carbon satellite data for all time periods within a certain area. For example, matching the carbon satellite data for all time periods within a 1° longitude × 1° latitude region around the source yields corresponding satellite strips. The data acquisition area of ​​these satellite strips is then scanned to obtain the carbon satellite data. Furthermore, to reduce the impact of carbon emissions from the source on atmospheric propagation, carbon satellite data that is more than a preset distance from the satellite strips can be deleted, concentrating the carbon satellite data within the area closer to the carbon emission source. Finally, a first multimodal carbon dataset is obtained, comprising geographically matched carbon satellite data and ground-based carbon data.

[0070] In this embodiment, a first multimodal carbon dataset is obtained by matching carbon satellite data within a preset distance range with the carbon emission source of ground-based carbon data as the center. The first multimodal carbon dataset includes carbon satellite data and ground-based carbon data that are geographically matched, which can provide accurate and relevant multimodal carbon data for subsequent carbon emission estimation.

[0071] Following the above embodiments, in one embodiment, the carbon emission estimation method further includes removing anomalous carbon satellite data from the first multimodal carbon dataset. Anomalous carbon satellite data may include carbon satellite data with a number of matching carbon emission sources greater than a preset threshold for the number of carbon emission sources, carbon satellite data with a corresponding satellite strip width less than a preset threshold for the strip width, carbon satellite data with a corresponding satellite strip recording area less than a preset threshold for the number of areas, as well as outliers and carbon satellite data that meet other anomalous conditions.

[0072] For example, when the number of carbon emission sources matched by carbon satellite data exceeds a preset threshold, it may indicate the existence of multiple carbon emission sources in the same area, such as more than 10 carbon emission sources in the same area. This makes it impossible to determine which carbon emission source's ground-based carbon emission monitoring data should be matched with the carbon satellite data. Therefore, carbon satellite data that meets this condition is identified as abnormal carbon satellite data and removed from the first multimodal carbon dataset to reduce the confusion and uncertainty of carbon satellite data, and also to reduce the complexity of subsequent data processing.

[0073] Furthermore, satellite stripes are strip-shaped regions observed by remote sensing satellite sensors. As the remote sensing satellite travels along its orbit, due to factors such as the accuracy of the remote sensing satellite sensor itself or the observation angle, there may be situations where the width of the satellite strip is less than the preset strip width threshold, such as a satellite strip width of less than 8 km. This may result in insufficient spatial resolution of the carbon satellite data, thus affecting the accuracy of subsequent data analysis. Therefore, carbon satellite data that meets this condition is judged as abnormal carbon satellite data and removed from the first multimodal carbon dataset.

[0074] Among them, the satellite strip recording area refers to the smallest scanning area unit of the satellite strip scan. When the number of satellite strip recording areas is less than the preset area number threshold, such as less than 200, it may indicate that the area coverage is insufficient during the carbon satellite acquisition process, such as being blocked by clouds, resulting in too little carbon satellite data acquired by the satellite strip, which is insufficient to support subsequent carbon emission estimation. Therefore, carbon satellite data that meets this condition needs to be judged as abnormal carbon satellite data and removed from the first multimodal carbon dataset.

[0075] When remote sensing satellite sensors malfunction, the monitoring environment changes, or other anomalies occur, some carbon satellite data may deviate from the overall data distribution. For example, the carbon dioxide column concentration value may be far above the average. If this data is retained and used for subsequent carbon emission estimation, the accuracy of the carbon emission estimation results may be reduced. Therefore, carbon satellite data that meets this condition needs to be identified as anomalous carbon satellite data and removed from the first multimodal carbon dataset. In this embodiment, after obtaining the first multimodal carbon dataset, anomalous carbon satellite data is also identified and removed from the first multimodal carbon dataset to improve the quality and reliability of the carbon satellite data, thereby improving the reliability of subsequent carbon emission estimation results.

[0076] In one embodiment, such as Figure 4 As shown, multimodal carbon data also includes environmental data, and the target multimodal carbon data also includes a second multimodal carbon dataset. Following S420, it also includes:

[0077] S440, based on the collection time period of carbon satellite data in the first multimodal carbon dataset, match environmental data collected within a preset time range to obtain the second multimodal carbon dataset, wherein the preset time range is determined based on the collection time period, and the second multimodal carbon dataset includes time-matched carbon satellite data and environmental data.

[0078] Following the previous embodiment, in practical applications, carbon satellite data carries timestamps, representing the time period during which the carbon satellite data was collected. Therefore, after matching the carbon satellite data and ground-based carbon data geographically based on the spatial dimension, it is possible to further match the carbon satellite data and environmental data temporally. Specifically, for the carbon satellite data in the first multimodal carbon dataset, environmental data collected within a preset time range is matched, with this preset time corresponding to the time period during which the carbon satellite data was collected. For example, wind speed data within a specific area and at the same time period is matched with the carbon satellite data to consider the impact of wind direction and speed on carbon emission propagation, thereby gaining a more accurate understanding of the carbon data. After matching the carbon satellite data with the environmental data, a second multimodal carbon dataset carrying carbon emission labels can be obtained. At this point, the ground-based carbon data and carbon satellite data are matched geographically, and the carbon satellite data and environmental data are matched temporally. Furthermore, during the matching process, some carbon satellite data lacking hourly carbon emission data can be removed to ensure the degree of matching between the carbon satellite data and environmental data in the dataset.

[0079] In this embodiment, by matching environmental data collected within a preset time range with the collection time period of carbon satellite data in the first multimodal carbon dataset, a second multimodal carbon dataset is obtained. In this way, ground carbon data, carbon satellite data, and environmental data are linked spatially and temporally to estimate carbon emissions by combining multimodal data, thereby improving the reliability of carbon emission estimation results.

[0080] In one embodiment, such as Figure 5 As shown, prior to S600, it also includes:

[0081] S520, acquire historical multimodal carbon data.

[0082] S540 performs matching processing on historical multimodal carbon data, establishes the correlation between historical multimodal carbon data, and obtains historical target multimodal carbon data.

[0083] S560 encodes the historical target multimodal carbon data to obtain the encoded historical target multimodal carbon data.

[0084] S580 uses encoded historical target multimodal carbon data as training data to train a pre-built deep learning model, resulting in a trained carbon emission estimation model.

[0085] Historical multimodal carbon data can include carbon data monitored by historical remote sensing carbon satellites, carbon data monitored by historical ground-based carbon monitoring equipment, historical environmental data, or historical data provided by other relevant carbon emission monitoring agencies. Matching processing can include temporal and spatial matching of historical multimodal carbon data, and can also involve constructing other correlations between various modalities of carbon data. For example, correlation analysis and time-series analysis can be performed on historical carbon satellite data and historical ground-based carbon data to establish correlations between carbon data from different modalities. In another embodiment, temporal resolution matching, spatial resolution matching, and data format matching can also be performed on historical multimodal carbon data. The specific matching processing method used for historical multimodal carbon data can be determined based on actual needs and the acquired data modalities.

[0086] Specifically, the carbon emission estimation model can be trained in the following way:

[0087] Carbon data can be obtained by accessing dedicated satellite data platforms (i.e., carbon satellite data), historical environmental data for the corresponding time period from environmental monitoring agencies, and historical carbon data (i.e., ground-based carbon data) from monitoring platforms corresponding to ground-based carbon monitoring equipment. However, the acquired historical multimodal carbon data may be isolated, with no correlation established between different modalities. Therefore, matching processing is needed to establish correlations between historical multimodal carbon data, yielding historical target multimodal carbon data, which can then be used as training data for carbon emission estimation models. For example, a matching and correlation model for historical multimodal carbon data can be established by training machine learning models or statistical models, and this model can then be used to match historical multimodal carbon data to obtain historical target multimodal carbon data.

[0088] When using historical target multimodal carbon data as training data into a model, it is necessary to first encode the data to ensure that the format of the historical target multimodal carbon data is consistent with a format that the model can recognize and process. Data encoding can involve normalizing the historical target multimodal carbon data to ensure that it improves the training effect of the model on the same scale. Alternatively, it can involve data dimensionality reduction or data transformation, depending on the actual needs.

[0089] For example, the j-th satellite strip data is represented as: ,in and This represents the center coordinates of the i-th observation region in the satellite strip. This indicates the carbon dioxide column concentration within the observation area. This represents the number of areas observed by the j-th satellite strip, i.e., the length of that satellite strip. Wind speeds below 10 meters are expressed as... ,in and These represent wind speeds along the longitude and latitude directions, respectively. Combining satellite strip data and wind speed data, the strip shape is obtained: s i =[ lon i , lat i , x i , u i , v i ] This serves as the encoded historical target multimodal carbon data.

[0090] Encoded historical target multimodal carbon data is used as training data. An appropriate deep learning model architecture is selected. In each training iteration, the model performs forward propagation based on the input data, calculates the loss, backpropagates the error, and continuously updates the parameters until training is complete, resulting in a trained carbon emission estimation model. Furthermore, validation data can be used to evaluate the trained carbon emission estimation model to check its performance in handling unseen target multimodal carbon data, and the model parameters can be continuously adjusted until the carbon emission estimation model can adequately meet the requirements for carbon emission estimation.

[0091] In this embodiment, historical multimodal carbon data is acquired and matched to establish correlations between them, resulting in historical target multimodal carbon data. This historical target multimodal carbon data is then encoded to obtain encoded historical target multimodal carbon data. Using this encoded data as training data, a pre-built deep learning model is trained to obtain a trained carbon emission estimation model. This trained carbon emission estimation model can quickly and accurately analyze the corresponding carbon emission estimation results based on the target multimodal carbon data. Compared to traditional carbon emission estimation methods, the carbon emission estimation results obtained using the model constructed in this embodiment are more reliable.

[0092] In one embodiment, the encoded historical target multimodal carbon data includes a first historical multimodal carbon dataset without carbon emission labels and a second historical multimodal carbon dataset with carbon emission labels.

[0093] like Figure 6 As shown, S580 includes:

[0094] S582 uses the encoded first historical multimodal carbon dataset as training data to train a self-supervised carbon emission pre-training sub-model in a pre-built deep learning model. The self-supervised carbon emission pre-training sub-model is used to extract carbon emission feature data from the target multimodal carbon data.

[0095] S584 uses the encoded second historical multimodal carbon dataset as training data to train the linear regression sub-model in the pre-built deep learning model, thus obtaining the trained carbon emission estimation model.

[0096] The self-supervised carbon emission pre-training sub-model can be used to extract carbon emission feature data from the target multimodal carbon data. Therefore, the encoded first multimodal carbon dataset can be used as the training dataset to allow the self-supervised carbon emission pre-training sub-model to learn from rich carbon data. The first historical multimodal carbon dataset includes geographically matched historical carbon satellite data and historical ground carbon data. The second historical multimodal carbon dataset includes time-matched historical carbon satellite data and historical environmental data with carbon emission labels. A linear regression model is used to fit the carbon emission feature data, and the carbon emission results are estimated based on the carbon emission feature data.

[0097] For example, a self-supervised carbon emission pre-training sub-model is used to extract carbon emission feature data from target multimodal carbon data. The feature extraction process can be, given the input strips... Then, it first goes through a fully connected embedding layer. Obtain features :

[0098]

[0099] in The dimension representing the features can be 256. The network ultimately extracts deep features. :

[0100]

[0101] in It is the deep feature extraction module, consisting of... It consists of a residual module based on a self-attention mechanism

[18] . Specifically, the intermediate features It is obtained through iterative steps using residual modules. ,in Indicates the first One residual module. F K =[ f 0 , f 1 ,…, f N ] For the first The output of each deep feature extraction module, where For dimension The vector. To reduce the dimensionality of deep features, intermediate features can be taken. The average value across the feature dimensions represents the final deep features extracted by the model. :

[0102]

[0103] Among them, depth features , where the length is equal to the feature dimension eigenvectors.

[0104] In a scenario relying solely on a first historical multimodal carbon dataset without carbon emission labels, training a self-supervised carbon emission pre-trained sub-model can be achieved by using zero-probability... ρ ∈[0,1] Randomly generate a binary mask This causes some input data to be obscured, so a binary mask is used. Multiply input Get mask input :

[0105]

[0106] Then, through the intermediate features obtained above Then use another fully connected layer. Dimension is intermediate features Mapping back to the original 5-dimensional feature stripe symbol:

[0107]

[0108] in, , and represent the weights and biases of the fully connected layer, respectively. During pre-training, the carbon emission estimation model takes the masked first historical multimodal carbon dataset as input and outputs the complete first historical multimodal carbon dataset.

[0109] After training the self-supervised carbon emission pre-training sub-model, a linear regression sub-model needs to be constructed based on the second historical multimodal carbon dataset to obtain the carbon emission estimation model, thus completing the construction of the carbon emission estimation model. For example, after obtaining the aforementioned trained self-supervised carbon emission pre-training sub-model, this model is called to extract carbon emission feature data from the encoded first multimodal carbon dataset. This carbon emission feature data, along with the second historical multimodal carbon dataset carrying carbon emission labels, is used as input to the linear regression model for supervised learning. The loss function optimizes the model output and the unmasked input. The distance. After training the linear regression sub-model, the carbon emission estimation model is complete, and this model can be used to analyze and estimate carbon emissions from target multimodal carbon data.

[0110] In this embodiment, first historical multimodal carbon data without carbon emission labels and second historical multimodal carbon datasets with carbon emission labels are used as training data for the self-supervised carbon emission pre-training sub-model and the linear regression sub-model in the carbon emission estimation model, respectively, ultimately training the carbon emission estimation model. Because this model uses multimodal carbon data as training data, it can accurately estimate carbon emissions. Combined with the fast and efficient characteristics of deep learning technology, it can effectively improve the efficiency and accuracy of carbon emission estimation.

[0111] To provide a clearer explanation of the carbon emission estimation method provided in this application, a specific embodiment and appendix are described below. Figure 6 The specific embodiment includes the following steps:

[0112] S200, acquire raw multimodal carbon data.

[0113] S420 uses carbon emission sources from ground-based carbon data as the center and matches them with carbon satellite data within a preset distance range to obtain a first multimodal carbon dataset without carbon emission labels. The first multimodal carbon dataset includes carbon satellite data and ground-based carbon data that are geographically matched.

[0114] S440, based on the collection time period of the carbon satellite data in the first multimodal carbon dataset, match the environmental data collected within a preset time range to obtain the second multimodal carbon dataset carrying carbon emission labels, wherein the preset time range is determined based on the collection time period and the carbon emission labels are determined based on the environmental data.

[0115] S520, acquire historical multimodal carbon data.

[0116] S540, perform matching processing on the historical multimodal carbon data, establish the correlation between the historical multimodal carbon data, and obtain historical target multimodal carbon data.

[0117] S560, the historical target multimodal carbon data is encoded to obtain encoded historical target multimodal carbon data.

[0118] S582 uses the encoded first historical multimodal carbon dataset as training data to train a self-supervised carbon emission pre-training sub-model in a pre-built deep learning model. The self-supervised carbon emission pre-training sub-model is used to extract carbon emission feature data from the target multimodal carbon data.

[0119] S584 uses the encoded second historical multimodal carbon dataset as training data to train the linear regression sub-model in the pre-built deep learning model, thus obtaining the trained carbon emission estimation model.

[0120] S600 takes target multimodal carbon data as input, calls a trained carbon emission estimation model to estimate carbon emissions, and obtains carbon emission estimation results. The carbon emission estimation model is trained based on historical target multimodal carbon data.

[0121] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0122] Based on the same inventive concept, this application also provides a carbon emission estimation device for implementing the carbon emission estimation method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more carbon emission estimation device embodiments provided below can be found in the limitations of the carbon emission estimation method described above, and will not be repeated here.

[0123] In one embodiment, such as Figure 7 As shown, a carbon emission estimation device 700 is provided, including: a data acquisition module 720, a data preprocessing module 740, and a data analysis module 760, wherein:

[0124] The data acquisition module 720 is used to acquire raw multimodal carbon data.

[0125] The data preprocessing module 740 is used to perform matching processing on the raw multimodal carbon data, establish the correlation between the raw multimodal carbon data, and obtain the target multimodal carbon data.

[0126] The data analysis module 760 is used to take the target multimodal carbon data as input, call the trained carbon emission estimation model to estimate carbon emissions, and obtain the carbon emission estimation results. The carbon emission estimation model is trained based on historical target multimodal carbon data.

[0127] In one embodiment, the multimodal carbon data includes carbon satellite data and ground-based carbon data, and the target multimodal carbon data includes a first multimodal dataset. The data preprocessing module 740 is further configured to match carbon satellite data within a preset distance range with the carbon emission source of the ground-based carbon data as the center to obtain the first multimodal carbon dataset. The first multimodal carbon dataset includes carbon satellite data and ground-based carbon data that are geographically matched.

[0128] In one embodiment, the data preprocessing module 740 is further configured to remove anomalous carbon satellite data from the first multimodal carbon dataset. Carbon satellite data is classified as anomalous carbon satellite data when it meets at least one of the following conditions:

[0129] The first condition is that the number of carbon emission sources matched with carbon satellite data is greater than the preset threshold for the number of carbon emission sources.

[0130] The second item is that the width of the satellite stripe corresponding to the carbon satellite data is less than the preset stripe width threshold;

[0131] Thirdly, the number of satellite strip record areas corresponding to carbon satellite data is less than the preset threshold for the number of areas;

[0132] Fourthly, carbon satellite data shows outliers.

[0133] In one embodiment, the multimodal carbon data also includes environmental data, and the target multimodal carbon data also includes a second multimodal carbon dataset.

[0134] The data preprocessing module 740 is further configured to match environmental data collected within a preset time range with the carbon satellite data collected in the first multimodal carbon dataset to obtain a second multimodal carbon dataset. The preset time range is determined based on the collection time period, and the second multimodal carbon dataset includes time-matched carbon satellite data and environmental data.

[0135] In one embodiment, the carbon emission estimation device 700 is further configured to acquire historical multimodal carbon data. The historical multimodal carbon data is matched to establish correlations between different data points, resulting in historical target multimodal carbon data. The historical target multimodal carbon data is then encoded to obtain encoded historical target multimodal carbon data. Using the encoded historical target multimodal carbon data as training data, a pre-built deep learning model is trained to obtain a trained carbon emission estimation model.

[0136] In one embodiment, the encoded historical target multimodal carbon data includes a first historical multimodal carbon dataset without carbon emission labels and a second historical multimodal carbon dataset with carbon emission labels.

[0137] The carbon emission estimation device 700 is also used to train a self-supervised carbon emission pre-training sub-model in a pre-built deep learning model using an encoded first historical multimodal carbon dataset as training data. This pre-trained sub-model extracts carbon emission feature data from the target multimodal carbon data. Using an encoded second historical multimodal carbon dataset as training data, a linear regression sub-model in the pre-built deep learning model is trained to obtain the trained carbon emission estimation model. The first historical multimodal carbon dataset includes geographically matched historical carbon satellite data and historical ground-based carbon data, while the second historical multimodal carbon dataset includes time-matched historical carbon satellite data and historical environmental data with carbon emission labels. The linear regression model is used to fit the carbon emission feature data, and the carbon emission results are estimated based on this data.

[0138] Each module in the aforementioned carbon emission estimation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0139] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 8 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores carbon emission estimation data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a carbon emission estimation method.

[0140] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0141] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described embodiments of carbon emission estimation methods.

[0142] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-described carbon emission estimation method embodiments.

[0143] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the carbon emission estimation method embodiments described above.

[0144] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0145] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0146] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0147] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A carbon emission estimation method, characterized in that, The method includes: Acquire raw multimodal carbon data; the raw multimodal carbon data includes carbon satellite data, ground-based carbon data, and environmental data; Centered on the carbon emission sources of the ground carbon data, carbon satellite data within a preset distance range is matched to obtain a first multimodal carbon dataset, which includes carbon satellite data and ground carbon data that are geographically matched. Based on the collection time period of carbon satellite data in the first multimodal carbon dataset, environmental data collected within a preset time range is matched to obtain a second multimodal carbon dataset; wherein, the first multimodal carbon dataset and the second multimodal carbon dataset are collectively referred to as target multimodal carbon data; the preset time range is determined based on the collection time period, and the second multimodal carbon dataset includes time-matched carbon satellite data and environmental data; Using the encoded first historical multimodal carbon dataset as training data, a self-supervised carbon emission pre-training sub-model in a pre-constructed deep learning model is trained. The self-supervised carbon emission pre-training sub-model is used to extract carbon emission feature data from the target multimodal carbon data. Using the encoded second historical multimodal carbon dataset as training data, the linear regression sub-model in the pre-constructed deep learning model is trained to obtain the trained carbon emission estimation model. The first historical multimodal carbon dataset includes geographically matched historical carbon satellite data and historical ground carbon data, and the second historical multimodal carbon dataset includes time-matched historical carbon satellite data and historical environmental data carrying carbon emission labels. The linear regression sub-model is used to fit the carbon emission characteristic data and estimate the carbon emission results based on the carbon emission characteristic data. Using the target multimodal carbon data as input, the trained carbon emission estimation model is invoked to estimate carbon emissions, and the carbon emission estimation results are obtained.

2. The method according to claim 1, characterized in that, The method further includes: Remove anomalous carbon satellite data from the first multimodal carbon dataset; Carbon satellite data is considered anomalous if it meets at least one of the following conditions: The first item is that the number of carbon emission sources matched with the carbon satellite data is greater than a preset threshold for the number of carbon emission sources; The second item is that the width of the satellite strip corresponding to the carbon satellite data is less than a preset strip width threshold; Thirdly, the number of satellite strip recording regions corresponding to the carbon satellite data is less than a preset threshold for the number of regions; Fourthly, the carbon satellite data mentioned are outliers.

3. The method according to claim 2, characterized in that, The satellite strip recording area refers to the smallest scanning area unit of a satellite strip scan.

4. The method according to any one of claims 1 to 2, characterized in that, The first historical multimodal carbon dataset and the second historical multimodal carbon dataset are collectively referred to as the encoded historical target multimodal carbon data; before using the target multimodal carbon data as input to call the trained carbon emission estimation model for carbon emission estimation, the process further includes: Acquire historical multimodal carbon data; The historical multimodal carbon data is matched to establish the correlation between the historical multimodal carbon data, thereby obtaining the historical target multimodal carbon data. The historical target multimodal carbon data is encoded to obtain encoded historical target multimodal carbon data.

5. A carbon emission estimation device, characterized in that, The device includes: The data acquisition module is used to acquire raw multimodal carbon data, which includes carbon satellite data, ground-based carbon data, and environmental data. The data preprocessing module is used to match carbon satellite data within a preset distance range with the carbon emission source of the ground carbon data as the center to obtain a first multimodal carbon dataset, which includes carbon satellite data and ground carbon data that are geographically matched. Based on the collection time period of carbon satellite data in the first multimodal carbon dataset, environmental data collected within a preset time range is matched to obtain a second multimodal carbon dataset; wherein, the first multimodal carbon dataset and the second multimodal carbon dataset are collectively referred to as target multimodal carbon data; the preset time range is determined based on the collection time period, and the second multimodal carbon dataset includes time-matched carbon satellite data and environmental data; Using the encoded first historical multimodal carbon dataset as training data, a self-supervised carbon emission pre-training sub-model in a pre-constructed deep learning model is trained. The self-supervised carbon emission pre-training sub-model is used to extract carbon emission feature data from the target multimodal carbon data. Using the encoded second historical multimodal carbon dataset as training data, the linear regression sub-model in the pre-constructed deep learning model is trained to obtain the trained carbon emission estimation model. The first historical multimodal carbon dataset includes geographically matched historical carbon satellite data and historical ground carbon data, and the second historical multimodal carbon dataset includes time-matched historical carbon satellite data and historical environmental data carrying carbon emission labels. The linear regression sub-model is used to fit the carbon emission characteristic data and estimate the carbon emission results based on the carbon emission characteristic data. The data analysis module is used to take the target multimodal carbon data as input, call the trained carbon emission estimation model to estimate carbon emissions, and obtain the carbon emission estimation results.

6. The apparatus according to claim 5, characterized in that, The data preprocessing module is also used for: Remove anomalous carbon satellite data from the first multimodal carbon dataset; Carbon satellite data is considered anomalous if it meets at least one of the following conditions: The first item is that the number of carbon emission sources matched with the carbon satellite data is greater than a preset threshold for the number of carbon emission sources; The second item is that the width of the satellite strip corresponding to the carbon satellite data is less than a preset strip width threshold; Thirdly, the number of satellite strip recording regions corresponding to the carbon satellite data is less than a preset threshold for the number of regions; Fourthly, the carbon satellite data mentioned are outliers.

7. The apparatus according to any one of claims 5 to 6, characterized in that, The first historical multimodal carbon dataset and the second historical multimodal carbon dataset are collectively referred to as the encoded historical target multimodal carbon data; the apparatus is further configured to: Acquire historical multimodal carbon data; The historical multimodal carbon data is matched to establish the correlation between the historical multimodal carbon data, thereby obtaining the historical target multimodal carbon data. The historical target multimodal carbon data is encoded to obtain encoded historical target multimodal carbon data.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 4.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 4.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 4.

Images