A multi-source data fusion method and device for monitoring road carbon emissions

Abstract

This invention discloses a multi-source data fusion method and apparatus for monitoring carbon emissions from roads. The method includes: collecting multi-source carbon emission data based on monitoring equipment on a mobile monitoring vehicle; preprocessing the multi-source carbon emission data; constructing an environmental model based on the preprocessed multi-source carbon emission data and obtaining the collection time of the multi-source carbon emission data; obtaining spatial information based on the constructed environmental model; performing time matching and spatial matching on the multi-source carbon emission data based on the collection time and spatial information; obtaining the spatiotemporal distribution results of carbon emissions based on the time-matched and spatially matched multi-source carbon emission data; and achieving high-precision spatiotemporal matching by fusing data from greenhouse gas concentration sensors, GPS, meteorological sensors, panoramic cameras, lidar measurement equipment, etc., effectively eliminating differences in spatial distribution and time points of single data sources, and ensuring the accuracy of urban road traffic greenhouse gas monitoring.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring technology, and in particular to a multi-source data fusion method and apparatus for monitoring road carbon emissions. Background Technology

[0002] With the growing severity of global climate change, monitoring and managing urban carbon emissions has become a crucial task for reducing greenhouse gas emissions and promoting sustainable development. This is especially true in the urban transportation sector, where transportation is a major source of carbon emissions; therefore, accurate and real-time monitoring of urban road carbon emissions is of paramount importance.

[0003] Currently, existing road traffic greenhouse gas monitoring methods mostly employ fixed monitoring stations. These stations are set up at fixed points along specific road sections or in urban areas to collect data such as atmospheric CO2 concentration, which is then combined with traffic flow and conditions to estimate emissions. This method provides some carbon concentration monitoring data. However, its drawbacks are also significant. For example, fixed monitoring stations can only monitor atmospheric CO2 concentration levels at designated locations, failing to comprehensively cover all roads, potentially leading to the omission of data from some high-traffic sections or specific road segments.

[0004] In addition, carbon emissions can also be calculated theoretically. For example, patent CN117727185A discloses a carbon emission monitoring method based on traffic flow data. This method involves acquiring images of vehicles entering and exiting a designated road, both of which carry corresponding image generation times. Based on the entry and exit images and their generation times, the vehicle type and total travel time are determined. Then, based on the vehicle type, the air-fuel ratio and its associated pollutant emission concentration-volume ratio, pollutant volumetric conversion factor, and pollutant carbon content are determined. Furthermore, by combining the total travel time with pre-stored lengths of flat roads, uphill sections, and downhill sections on the designated road, the vehicle's pollutant emissions are calculated. However, this method relies on experience and its accuracy is not high.

[0005] With the development of sensor and mobile computing technologies, vehicle-mounted mobile monitoring systems are gradually becoming a new means of carbon emission monitoring. By installing greenhouse gas monitoring equipment on vehicles, dynamic real-time monitoring of greenhouse gases such as CO2 and other air pollutants on roads can be achieved. This method can provide a wider coverage area than fixed monitoring stations, obtain accurate carbon emission data, and can be flexibly monitored according to real-time road conditions. For example, it can be used to monitor detours to heavily polluting enterprises. However, since different types of data have different collection frequencies and spatial distributions, how to effectively match these data from different sensors in time and space and extract accurate spatiotemporal carbon concentration information remains a problem that urgently needs to be solved. Summary of the Invention

[0006] This invention provides a multi-source data fusion method and apparatus for monitoring road carbon emissions, which can obtain accurate spatiotemporal carbon concentration information.

[0007] A multi-source data fusion method for monitoring road carbon emissions includes:

[0008] Carbon emission data from multiple sources are collected using monitoring equipment on mobile monitoring vehicles;

[0009] The multi-source carbon emission data are preprocessed.

[0010] An environmental model is constructed based on the preprocessed multi-source carbon emission data, and the collection time of the multi-source carbon emission data is obtained. Spatial information is obtained based on the constructed environmental model, and time matching and spatial matching of the multi-source carbon emission data are performed based on the collection time and the spatial information.

[0011] The spatiotemporal distribution of carbon emissions was obtained from multi-source carbon emission data based on time and spatial matching.

[0012] Furthermore, the multi-source carbon emission data includes greenhouse gas data, mobile vehicle location data, meteorological data, panoramic environmental image data, and environmental point cloud data.

[0013] Furthermore, the multi-source carbon emission data is preprocessed, including:

[0014] For the same type of data in the multi-source carbon emission data, calculate the standard deviation and mean;

[0015] Based on the standard deviation and mean, calculate the Z score for each data point, determine whether the data point is an outlier based on the Z score, and delete the outlier data point.

[0016] The time-series data in the multi-source carbon emission data are interpolated and filled using linear interpolation.

[0017] Furthermore, the time-series data in the multi-source carbon emission data are interpolated and filled using linear interpolation, including:

[0018] For each type of time-series data, sort them according to the collection time;

[0019] Based on the collection frequency of various types of data, identify the missing points in the time series data;

[0020] The value of the missing point is calculated based on the value of the previous data point and the value of the subsequent data points.

[0021] Furthermore, an environmental model is constructed based on the preprocessed multi-source carbon emission data, and the acquisition time of the multi-source carbon emission data is obtained. Spatial information is obtained based on the constructed environmental model, and time and spatial matching is performed on the multi-source carbon emission data based on the acquisition time and the spatial information, including:

[0022] A neural network model is used to identify vehicles in the panoramic environment image data to obtain vehicle type and quantity;

[0023] Based on the environmental point cloud data, a three-dimensional environmental point cloud model is constructed, and based on the three-dimensional environmental point cloud model, road vector data is extracted to form a first dataset, wherein the road vector data includes the road point cloud acquisition time and road location coordinates;

[0024] According to a preset time window, the greenhouse gas data and meteorological data are resampled to obtain greenhouse gas resampled data and meteorological resampled data under the time window.

[0025] Based on the acquisition time of the panoramic environmental image data, greenhouse gas resampling data, and meteorological resampling data, and the corresponding mobile vehicle location data at that acquisition time, timestamps and location coordinates are set for vehicle type and quantity, greenhouse gas resampling data, and meteorological resampling data to form a second dataset.

[0026] Match the first and second datasets using nearest neighbor interpolation or inverse distance weighting.

[0027] Further, based on the nearest neighbor interpolation method, matching is performed on the first dataset and the second dataset, including:

[0028] Select data points to be matched from the first dataset as target points, and calculate the Euclidean distance and time difference between the target points and each data point in the second dataset;

[0029] In the second dataset where the Euclidean distance and time difference meet preset conditions, the optimal data point is selected and associated with the target point to obtain a matching result.

[0030] Furthermore, based on the aforementioned inverse distance weighting method, matching is performed on the first and second datasets, including:

[0031] Select data points to be matched from the first dataset as target points, and calculate the Euclidean distance and time difference between the target points and each data point in the second dataset;

[0032] Set an initial search radius, and for the target point, count the number of data points belonging to the second dataset within the initial search radius whose Euclidean distance and time difference satisfy the preset conditions, and calculate the local point density;

[0033] The search radius is adjusted according to the preset density of the local points.

[0034] Within the adjusted search radius, select data points belonging to the second dataset that meet the preset conditions. Perform distance-weighted calculations based on the attribute values ​​of the selected data points and their Euclidean distance to the target point to obtain attribute estimates.

[0035] The attribute estimates are associated with the target point to obtain a matching result.

[0036] Furthermore, the preset conditions are that the Euclidean distance is less than a preset distance and the time difference is less than a preset duration.

[0037] A multi-source data fusion device for monitoring road carbon emissions includes:

[0038] The data acquisition module is used to collect multi-source carbon emission data based on the monitoring equipment on the mobile monitoring vehicle;

[0039] The preprocessing module is used to preprocess the multi-source carbon emission data;

[0040] The matching module is used to construct an environmental model based on the preprocessed carbon emission multi-source data, obtain the collection time of the carbon emission multi-source data, obtain spatial information based on the constructed environmental model, and perform time matching and spatial matching of the carbon emission multi-source data based on the collection time and the spatial information.

[0041] The results processing module is used to obtain the spatiotemporal distribution results of carbon emissions based on multi-source carbon emission data with time and spatial matching.

[0042] Furthermore, the multi-source carbon emission data includes greenhouse gas data, mobile vehicle location data, meteorological data, panoramic environmental image data, and environmental point cloud data.

[0043] Furthermore, the preprocessing module preprocesses the multi-source carbon emission data, including:

[0044] For the same type of data in the multi-source carbon emission data, calculate the standard deviation and mean;

[0045] Based on the standard deviation and mean, calculate the Z score for each data point, determine whether the data point is an outlier based on the Z score, and delete the outlier data point.

[0046] The time-series data in the multi-source carbon emission data are interpolated and filled using linear interpolation.

[0047] Furthermore, the preprocessing module performs interpolation and imputation on the time-series data in the multi-source carbon emission data based on linear interpolation, including:

[0048] For each type of time-series data, sort them according to the collection time;

[0049] Based on the collection frequency of various types of data, identify the missing points in the time series data;

[0050] The value of the missing point is calculated based on the value of the previous data point and the value of the subsequent data points.

[0051] Furthermore, the matching module constructs an environmental model based on the preprocessed multi-source carbon emission data, obtains the acquisition time of the multi-source carbon emission data, acquires spatial information based on the constructed environmental model, and performs time matching and spatial matching of the multi-source carbon emission data based on the acquisition time and the spatial information, including:

[0052] A neural network model is used to identify vehicles in the panoramic environment image data to obtain vehicle type and quantity;

[0053] Based on the environmental point cloud data, a three-dimensional environmental point cloud model is constructed, and based on the three-dimensional environmental point cloud model, road vector data is extracted to form a first dataset, wherein the road vector data includes the road point cloud acquisition time and road location coordinates;

[0054] According to a preset time window, the greenhouse gas data and meteorological data are resampled to obtain greenhouse gas resampled data and meteorological resampled data under the time window.

[0055] Based on the acquisition time of the panoramic environmental image data, greenhouse gas resampling data, and meteorological resampling data, and the corresponding mobile vehicle location data at that acquisition time, timestamps and location coordinates are set for vehicle type and quantity, greenhouse gas resampling data, and meteorological resampling data to form a second dataset.

[0056] Match the first and second datasets using nearest neighbor interpolation or inverse distance weighting.

[0057] Further, the matching module matches the first dataset and the second dataset based on the nearest neighbor interpolation method, including:

[0058] Select data points to be matched from the first dataset as target points, and calculate the Euclidean distance and time difference between the target points and each data point in the second dataset;

[0059] In the second dataset where the Euclidean distance and time difference meet preset conditions, the optimal data point is selected and associated with the target point to obtain a matching result.

[0060] Furthermore, the matching module matches the first dataset and the second dataset based on the inverse distance weighting method, including:

[0061] Select data points to be matched from the first dataset as target points, and calculate the Euclidean distance and time difference between the target points and each data point in the second dataset;

[0062] Set an initial search radius, and for the target point, count the number of data points belonging to the second dataset within the initial search radius whose Euclidean distance and time difference satisfy the preset conditions, and calculate the local point density;

[0063] The search radius is adjusted according to the preset density of the local points.

[0064] Within the adjusted search radius, select data points belonging to the second dataset that meet the preset conditions. Perform distance-weighted calculations based on the attribute values ​​of the selected data points and their Euclidean distance to the target point to obtain attribute estimates.

[0065] The attribute estimates are associated with the target point to obtain a matching result.

[0066] Furthermore, the preset conditions are that the Euclidean distance is less than a preset distance and the time difference is less than a preset duration.

[0067] An electronic device is characterized by comprising a processor and a storage device, wherein the storage device stores a plurality of instructions, and the processor is configured to read the instructions and execute the method described above.

[0068] The multi-source data fusion method and apparatus for monitoring road carbon emissions provided by this invention have at least the following beneficial effects:

[0069] (1) Improve the richness of greenhouse gas monitoring information fusion. By fusing greenhouse gas data, mobile vehicle location data, meteorological data, panoramic environmental image data, and environmental point cloud data, high-precision spatiotemporal matching can be achieved, effectively eliminating the differences in spatial distribution and time points of single data sources, and ensuring the accuracy of urban road traffic greenhouse gas monitoring;

[0070] (2) The spatiotemporal matching algorithm and multi-source data processing framework can effectively process data from different sensors and sources, providing reliable data support for further carbon emission estimation and promoting the realization of intelligent transportation systems and green and low-carbon transportation policies;

[0071] (3) Data is fused in time and space based on nearest neighbor interpolation and inverse distance weighting to improve the accuracy of data fusion. Attached Figure Description

[0072] Figure 1 This is a flowchart of one embodiment of a multi-source data fusion method for monitoring road carbon emissions provided by the present invention.

[0073] Figure 2 This is a flowchart of an embodiment of data preprocessing in a multi-source data fusion method for monitoring road carbon emissions provided by the present invention.

[0074] Figure 3 This is a flowchart illustrating one embodiment of data matching in a multi-source data fusion method for monitoring road carbon emissions provided by the present invention.

[0075] Figure 4 This is a flowchart of an embodiment of a multi-source data fusion method for monitoring road carbon emissions provided by the present invention, which uses the nearest neighbor interpolation method for data matching.

[0076] Figure 5 This is a flowchart of an embodiment of a multi-source data fusion method for monitoring road carbon emissions provided by the present invention, which uses an inverse distance weighting method for data matching.

[0077] Figure 6 This is a flowchart of one embodiment of a multi-source data fusion device for monitoring road carbon emissions provided by the present invention. Detailed Implementation

[0078] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0079] refer to Figure 1 In some embodiments, a multi-source data fusion method for monitoring road carbon emissions is provided, comprising:

[0080] S1. Collect multi-source carbon emission data based on monitoring equipment on mobile monitoring vehicles;

[0081] S2. Preprocess the multi-source carbon emission data;

[0082] S3. Construct an environmental model based on the preprocessed carbon emission multi-source data, obtain the collection time of the carbon emission multi-source data, obtain spatial information based on the constructed environmental model, and perform time matching and spatial matching on the carbon emission multi-source data based on the collection time and the spatial information.

[0083] S4. Obtain the spatiotemporal distribution results of carbon emissions based on multi-source carbon emission data with time and spatial matching.

[0084] Specifically, in step S1, a mobile urban road monitoring system is constructed by installing hardware systems such as greenhouse gas monitoring equipment, GPS system, panoramic camera, meteorological sensor, lidar measurement equipment, and data processor on the mobile monitoring vehicle. Based on this system, greenhouse gas data, panoramic environmental image data, mobile vehicle location data, meteorological data, and environmental point cloud data can be obtained. Its main hardware equipment includes:

[0085] Mobile monitoring vehicle: Select large vehicles such as SUVs as dedicated mobile monitoring vehicles, and make simple modifications to them, including: setting up gas sampling pipeline inlets and fixing devices; setting up power system fixing devices, etc.; the vehicle body should be marked with prominent markings such as mobile monitoring test.

[0086] GPS System: The Global Positioning System (GPS) is essential for acquiring the location data of mobile vehicles. At the same time, the system provides accurate time, geographic coordinates, and vehicle driving status information, which is of great significance for subsequent mobile monitoring data analysis.

[0087] Greenhouse gas measurement equipment: used to detect and quantify the concentration of greenhouse gases in the atmosphere and obtain greenhouse gas data. Depending on the requirements, in some embodiments, high-precision gas measurement equipment such as the Picarro G2401 for carbon dioxide (CO2) and methane (CH4) can be used, or medium-precision, low-cost CO2 monitoring equipment (Visara GMP343) can be used. The Picarro G2401 device has a CO2 concentration measurement accuracy of 0.1 ppm, while the Vaisala GMP343 device has a measurement accuracy of approximately 1-2 ppm.

[0088] Meteorological sensors: In addition to greenhouse gas data, it is also necessary to collect meteorological data, such as temperature, humidity, wind speed, and wind direction. This data helps to understand the distribution and changes of greenhouse gases in the atmosphere and to conduct subsequent greenhouse gas concentration analysis and simulation.

[0089] Panoramic camera: To accurately reflect the road vehicle driving and traffic flow conditions during the data collection process, this system is equipped with a panoramic camera. This camera uses 5 fisheye cameras to target the front, rear, left, right and top of the vehicle, and has 40%-60% image overlap. The panoramic image data can be obtained by reconstructing the panorama based on these 5 photos.

[0090] LiDAR measurement equipment: Equipped with AS-900 equipment, the scanning beam is adjusted to 200 revolutions per second, and the scanner frequency is set to 550KHZ to acquire point cloud data of the road surface and environmental features such as roadside objects and roadside trees on both sides.

[0091] Data logging equipment: Data logging equipment is used to collect and store data generated by sensors. In some embodiments, a chip-based Linux system processor such as a Raspberry Pi can be used, along with a data SD memory card, to collect greenhouse gas data, mobile vehicle location data, meteorological data, panoramic environmental image data, and environmental point cloud data every second, facilitating subsequent data processing.

[0092] Communication equipment: In order to enable real-time monitoring and remote control, a wireless network connection is required. In some embodiments, Huawei Cloud's 4G IoT card is used to transmit data from the BeagleBone chip Linux system processor to the laboratory server every hour.

[0093] Power System: Since the monitoring equipment needs to operate for extended periods, and also requires constant temperature heating, air pump pressure reduction, and network data transmission, which consume significant amounts of power, in some embodiments, a large portable power bank (such as a Jackery portable power bank, which stores approximately 2 kWh and can support continuous power supply to the monitoring equipment for about 8 hours) can be selected to provide a stable long-term power supply to the observation equipment and other auxiliary equipment.

[0094] Safety equipment: To ensure the safety of vehicles and equipment, safety cameras, etc., can be installed.

[0095] Based on this, the aforementioned multi-source carbon emission data includes greenhouse gas data, mobile vehicle location data, meteorological data, panoramic environmental image data, and environmental point cloud data.

[0096] Further, refer to Figure 2 In step S2, the multi-source carbon emission data is preprocessed, including:

[0097] S21. For the same type of data in the multi-source carbon emission data, calculate the standard deviation and mean;

[0098] S22. Calculate the Z-score for each data point based on the standard deviation and mean, determine whether the data point is an outlier based on the Z-score, and delete the outlier data point.

[0099] S23. The time series data in the carbon emission multi-source data are interpolated and filled using linear interpolation.

[0100] Specifically, in steps S21 and S22, during the data acquisition process, outliers may occur due to sensor malfunctions, measurement errors, or environmental interference. The presence of outliers can negatively impact subsequent analysis, therefore they need to be filtered out. This embodiment uses the Z-score method for outlier filtering. This method calculates the standard deviation of data points and detects data points with significant deviations from the mean. For the same type of data, i.e., greenhouse gas data or meteorological data, a data point... Its Z-score Z(xi) is calculated using the following formula:

[0101] (1)

[0102] in, The mean of all data in the same category. It represents the standard deviation of all data in the same class.

[0103] In some embodiments, data points with Z scores greater than a first preset value or less than a second preset value are considered outliers and are deleted. For example, the first preset value can be 3 and the second preset value can be -3.

[0104] Furthermore, in step S23, data loss is a common problem in actual data acquisition, which may be caused by sensor malfunction, communication interruption, or external factors. Before performing spatiotemporal matching of data, missing data needs to be filled in. Methods such as linear interpolation can be used for data filling. Specifically, the time-series data in the multi-source carbon emission data is filled in using linear interpolation, including:

[0105] For each type of time-series data, sort them according to the collection time;

[0106] Based on the collection frequency of various types of data, identify the missing points in the time series data;

[0107] The value of the missing point is calculated based on the value of the previous data point and the value of the subsequent data points.

[0108] The time-series data includes greenhouse gas data and meteorological data.

[0109] For time series data, assuming the changes between data points are linear, the following formula is used for interpolation:

[0110] (2)

[0111] Where x(t) is the interpolation result of the missing point, and x(t-1) and x(t+n) are the values ​​of the previous and subsequent data points of the missing point, respectively. n-1 represents the number of missing values.

[0112] Further, refer to Figure 3 In step S3, an environmental model is constructed based on the preprocessed multi-source carbon emission data, and the acquisition time of the multi-source carbon emission data is obtained. Spatial information is obtained based on the constructed environmental model, and time and spatial matching is performed on the multi-source carbon emission data based on the acquisition time and the spatial information, including:

[0113] S31. Use a neural network model to identify vehicles in the panoramic environment image data to obtain vehicle type and quantity;

[0114] S32. Based on the environmental point cloud data, construct a three-dimensional environmental point cloud model, and based on the three-dimensional environmental point cloud model, extract road vector data and form a first dataset, wherein the road vector data includes the road point cloud acquisition time and road location coordinates;

[0115] S33. Resample the greenhouse gas data and meteorological data according to the preset time window to obtain greenhouse gas resampled data and meteorological resampled data under the time window.

[0116] S34. Based on the acquisition time of the panoramic environment image data, greenhouse gas resampling data, and meteorological resampling data, and the corresponding mobile vehicle location data at the acquisition time, timestamps and location coordinates are set for vehicle type and quantity, greenhouse gas resampling data, and meteorological resampling data to form a second dataset.

[0117] S35. Match the first and second datasets based on the nearest neighbor interpolation method or the inverse distance weighting method.

[0118] Specifically, in step S31, a neural network model is used to identify vehicles in the panoramic environment image data, including:

[0119] The environmental image data is input into a pre-trained YOLOv8 algorithm for target recognition to obtain the type and number of target vehicles.

[0120] Specifically, panoramic environment image data is input into the YOLOv8 model and scaled to a preset size, typically 640x640 or 1280x1280, to ensure a consistent input data format to fit the model's structure. Features are then extracted from the input image, from low-level edge and texture information to high-level object structure information, layer by layer. Furthermore, YOLOv8 incorporates a Feature Pyramid Network (FPN) and a Path Aggregation Network (PAN) during this process to ensure effective feature extraction at different scales, adapting to vehicles of varying sizes.

[0121] Furthermore, anchor box generation and prediction: YOLOv8 divides the entire image into a grid, and each grid predicts multiple anchor boxes. Each anchor box contains location information and class probability of possible targets. The position and size of vehicles are estimated using the anchor boxes, and each box is labeled with a class such as "car", "truck", or "bus".

[0122] Furthermore, nonmaximum suppression (NMS) is used to eliminate redundant detection results, retain the bounding box with the highest confidence, and suppress other boxes with overlap exceeding the threshold, ensuring that each vehicle corresponds to only one detection box.

[0123] Furthermore, the classification and localization results are output: a set of recognition results are output, each including the vehicle category, confidence score, and bounding box location. This yields the vehicle type and quantity.

[0124] Furthermore, in step S32, environmental point cloud data is collected using a lidar measurement device to obtain the spatial geometric texture and attribute information of the road, including road name, number of motor vehicle lanes, one-way and two-way traffic conditions, road type, etc.

[0125] Furthermore, the environmental point cloud data is processed to extract the geometric features of the road. The point cloud data is matched to the correct geographic coordinate system using measured correction points. The least squares method is used to adjust the spatial relationship between the point cloud data and the correction points to reduce errors and ensure the accuracy and reliability of the data.

[0126] Vector features of roads are extracted based on the corrected environmental point cloud data, mainly including the spatial coordinates, shape and size of features such as motor vehicle lanes, non-motor vehicle lanes, sidewalks, and green belts.

[0127] The collected 3D vector elements are classified, integrated, and topologically checked to construct a complete 3D point cloud model of the environment, including motor vehicle lanes, non-motor vehicle lanes, sidewalks, green belts, etc. Road vector data is extracted and formed into the first dataset, which includes the road point cloud acquisition time and road location coordinates.

[0128] Further, in step S33, the greenhouse gas data and meteorological data are resampled according to a preset time window, including:

[0129] Calculate the mean concentration of greenhouse gases within the time window to obtain the greenhouse gas resampling data;

[0130] Calculate the mean values ​​of temperature and humidity within the time window to obtain meteorological resampling data.

[0131] Because the sampling times and time intervals of different sensors vary significantly, it is necessary to resample the data from each sensor over time to ensure that all data have the same temporal granularity. For example, the sampling frequency for CO2 concentration data is once per second, while the onboard meteorological sensor has a time resolution of 2 seconds. Meanwhile, the sampling frequency for traffic and meteorological data may be 1 minute later. Therefore, this embodiment sets a fixed time window (e.g., 60 seconds) to resample the original data over time, and obtains the composite data for the current time period using an arithmetic average method. The specific formula (taking atmospheric CO2 concentration monitoring data as an example) is shown below:

[0132] (3)

[0133] in, This represents the CO2 concentration value after resampling at time t. , and This represents the original time resolution (1 second) CO2 concentration monitoring value within the time window at time t, where n is the number of monitoring data points within the time window.

[0134] Further, in step S34, based on the acquisition time of the panoramic environmental image data, greenhouse gas resampling data, and meteorological resampling data, and the corresponding mobile vehicle location data at that acquisition time, timestamps and location coordinates are set for the vehicle type and quantity, greenhouse gas resampling data, and meteorological resampling data to form a second dataset.

[0135] Further, refer to Figure 4 In step S35, based on the nearest neighbor interpolation method, the first dataset and the second dataset are matched, including:

[0136] S35a. Select data points to be matched from the first dataset as target points, and calculate the Euclidean distance and time difference between the target point and each data point in the second dataset.

[0137] S35b: Select the optimal data point from the data points in the second dataset where the Euclidean distance and time difference meet the preset conditions, and associate it with the target point to obtain a matching result.

[0138] Specifically, a first dataset Droad and a second dataset Denv are constructed. The first dataset Droad contains the road point cloud collection time and road location coordinates, while the second dataset Denv contains data that needs to be matched with the road vector data, namely the timestamps and location coordinates at the time of collection corresponding to vehicle type and number, greenhouse gas resampling data, and meteorological resampling data.

[0139] For each element i in the first dataset Droad and the second dataset Denv, define its position coordinates as two-dimensional coordinates (xi, yi), where xi and yi are longitude and latitude, respectively.

[0140] Each location coordinate is converted into a geometric point object Pi=(xi,yi) for use in subsequent calculations.

[0141] Define a preset duration ΔT to limit the maximum time difference between two data points that can be considered a match. For example, data within ±1 minute are considered a valid match.

[0142] Set a preset distance D, representing the maximum distance (in meters) when performing spatial joins. Spatially, two geometric objects are considered a match only if the distance between them is less than or equal to this threshold. Temporally, a proximity condition must also be met.

[0143] Furthermore, the data point Pa to be matched is selected from the first dataset Droad as the target point, and its Euclidean distance to any data point Pb in the second dataset is calculated using the following formula:

[0144] (4)

[0145] Assuming the acquisition time of target point Pa is ta and the acquisition time of data point Pb is tb, when d(Pa, Pb) ≤ D and |ta-tb| ≤ ΔT, Pa and Pb are considered to be a potential matching pair. That is, the preset condition is that the Euclidean distance is less than the preset distance and the time difference is less than the preset duration.

[0146] Specifically, for the target point Pa in the first dataset Droad, all data points in the second dataset Denv that satisfy the conditions d(Pa, Pb) ≤ D and |ta-tb| ≤ ΔT are searched. The data point with the smallest distance is selected as the optimal data point. The attribute values ​​of this optimal data point are then matched with the corresponding target point to obtain the matching result. If no matching point is found, the record is left empty or marked as unmatched. A new dataset is generated based on the matching results, containing all columns of the original road vector data and information about the nearest neighbors found in the second dataset Denv. All unmatched road records (i.e., those roads for which no matching environmental monitoring station was found) are removed to obtain the final valid matching result.

[0147] Further, refer to Figure 5 In step S35, based on the inverse distance weighting method, the first dataset and the second dataset are matched, including:

[0148] S351. Select data points to be matched from the first dataset as target points, and calculate the Euclidean distance and time difference between the target point and each data point in the second dataset.

[0149] S352. Set an initial search radius. For the target point, count the number of data points belonging to the second dataset within the initial search radius whose Euclidean distance and time difference satisfy the preset conditions, and calculate the local point density.

[0150] S353. Adjust the search radius according to the preset density of the local points;

[0151] S354. Within the adjusted search radius, select data points belonging to the second dataset that meet the preset conditions, and perform distance-weighted calculations based on the attribute values ​​of the selected data points and their Euclidean distance to the target point to obtain attribute estimates.

[0152] S355. Associate the attribute estimate with the target point to obtain a matching result.

[0153] Specifically, in step S351, the data point Pa to be matched is selected from the first dataset Droad as the target point, and the Euclidean distance between it and any data point Pb in the second dataset is calculated by formula (4), and the time difference |ta-tb| is calculated.

[0154] Further, in step S352, an initial search radius r0 is set. For the target point Pa, based on the Euclidean distance, the number n of data points belonging to the first dataset that satisfy the preset conditions of the Euclidean distance and time difference within the initial search radius r0 is counted. localIdentify data points that satisfy d(Pa, Pb) ≤ D and |ta-tb| ≤ ΔT, and calculate the local point density. pa .

[0155] The initial search radius r0 can be one to two times the average distance between data points, or set according to empirical values ​​based on specific application scenarios.

[0156] Local point density pa The calculation is performed using the following formula:

[0157] (5)

[0158] Where r0 represents the initial search radius, n local This represents the number of data points belonging to the first dataset within the initial search radius r0 whose Euclidean distance and time difference satisfy the preset conditions.

[0159] Furthermore, in step S353, a preset density d is set. target This preset density can be an empirical value or an optimal density value determined experimentally. It represents the ideal local data point density, that is, the density at which the best interpolation effect can be obtained.

[0160] Based on local point density pa and preset density d target Adjust the search radius:

[0161] (6)

[0162] Where, r a This is the adjusted search radius.

[0163] If the local point density Pa Greater than the preset density d target Then the adjusted search radius r a It will be less than the initial search radius r0; conversely, if the local point density is greater than the initial search radius r0, it will be less than the initial search radius r0. Pa less than the preset density d target Then the adjusted search radius r a It will be greater than the initial search radius r0.

[0164] Further, in step S354, assuming that the number of data points belonging to the first dataset that meet the preset conditions within the adjusted search radius is n, the attribute estimate is obtained by distance-weighted calculation according to the following formula:

[0165] (7)

[0166] Where Z(Pa) represents the attribute estimate matched with the target point Pa, zi represents the attribute value of the i-th data point Pbi in the first dataset that meets the preset conditions within the adjusted search radius, which can be the greenhouse gas concentration value, the number of vehicle types, or the value of meteorological data, n represents the number of data points in the first dataset that meet the preset conditions within the adjusted search radius, p is the weight index, which is usually between 1 and 3 and is used to control the degree of influence of distance on weight, and d(Pa, Pbi) represents the Euclidean distance between the target point Pa and the data point Pbi.

[0167] Specifically, in the above formula, the numerator: These are the attribute values ​​of Pbi for all data points that meet the criteria. Its distance to the target point Pa The sum of ratios raised to powers of 1. This means that the attribute value of each data point Pbi is weighted according to its distance from the target point, with points closer to the target point having a greater impact on the final estimate.

[0168] Denominator: This involves normalizing the weighted values ​​in the numerator to ensure that the final estimate is within a reasonable range. It is the sum of the reciprocals of the p-th powers of the distances from all data points Pbi that meet the criteria to the target point Pa.

[0169] In step S355, the attribute estimate is associated with the target point to obtain a matching result.

[0170] Furthermore, in step S4, a mapping relationship is established between the attribute values ​​of the selected target points and the matched data points through time matching and spatial matching to obtain the spatiotemporal distribution results of carbon emissions.

[0171] refer to Figure 6 In some embodiments, a multi-source data fusion device for monitoring road carbon emissions is also provided, comprising:

[0172] The data acquisition module 201 is used to collect multi-source carbon emission data based on the monitoring equipment on the mobile monitoring vehicle;

[0173] Preprocessing module 202 is used to preprocess the multi-source carbon emission data;

[0174] The matching module 203 is used to construct an environmental model based on the preprocessed carbon emission multi-source data, obtain the collection time of the carbon emission multi-source data, obtain spatial information based on the constructed environmental model, and perform time matching and spatial matching on the carbon emission multi-source data based on the collection time and the spatial information.

[0175] The results processing module 204 is used to obtain the spatiotemporal distribution results of carbon emissions based on multi-source carbon emission data with time matching and spatial matching.

[0176] Furthermore, the multi-source carbon emission data includes greenhouse gas data, mobile vehicle location data, meteorological data, panoramic environmental image data, and environmental point cloud data.

[0177] Furthermore, the preprocessing module 202 preprocesses the multi-source carbon emission data, including:

[0178] For the same type of data in the multi-source carbon emission data, calculate the standard deviation and mean;

[0179] Based on the standard deviation and mean, calculate the Z score for each data point, determine whether the data point is an outlier based on the Z score, and delete the outlier data point.

[0180] The time-series data in the multi-source carbon emission data are interpolated and filled using linear interpolation.

[0181] Furthermore, the preprocessing module 202 performs interpolation and imputation on the time-series data in the multi-source carbon emission data based on linear interpolation, including:

[0182] For each type of time-series data, sort them according to the collection time;

[0183] Based on the collection frequency of various types of data, identify the missing points in the time series data;

[0184] The value of the missing point is calculated based on the value of the previous data point and the value of the subsequent data points.

[0185] Further, the matching module 203 constructs an environmental model based on the preprocessed multi-source carbon emission data, obtains the acquisition time of the multi-source carbon emission data, acquires spatial information based on the constructed environmental model, and performs time matching and spatial matching on the multi-source carbon emission data based on the acquisition time and the spatial information, including:

[0186] A neural network model is used to identify vehicles in the panoramic environment image data to obtain vehicle type and quantity;

[0187] Based on the environmental point cloud data, a three-dimensional environmental point cloud model is constructed, and based on the three-dimensional environmental point cloud model, road vector data is extracted to form a first dataset, wherein the road vector data includes the road point cloud acquisition time and road location coordinates;

[0188] According to a preset time window, the greenhouse gas data and meteorological data are resampled to obtain greenhouse gas resampled data and meteorological resampled data under the time window.

[0189] Based on the acquisition time of the panoramic environmental image data, greenhouse gas resampling data, and meteorological resampling data, and the corresponding mobile vehicle location data at that acquisition time, timestamps and location coordinates are set for vehicle type and quantity, greenhouse gas resampling data, and meteorological resampling data to form a second dataset.

[0190] Match the first and second datasets using nearest neighbor interpolation or inverse distance weighting.

[0191] Further, the matching module 203 matches the first dataset and the second dataset based on the nearest neighbor interpolation method, including:

[0192] Select data points to be matched from the first dataset as target points, and calculate the Euclidean distance and time difference between the target points and each data point in the second dataset;

[0193] In the second dataset where the Euclidean distance and time difference meet preset conditions, the optimal data point is selected and associated with the target point to obtain a matching result.

[0194] Furthermore, the matching module 203 performs matching on the first dataset and the second dataset based on the inverse distance weighting method, including:

[0195] Select data points to be matched from the first dataset as target points, and calculate the Euclidean distance and time difference between the target points and each data point in the second dataset;

[0196] Set an initial search radius, and for the target point, count the number of data points belonging to the second dataset within the initial search radius whose Euclidean distance and time difference satisfy the preset conditions, and calculate the local point density;

[0197] The search radius is adjusted according to the preset density of the local points.

[0198] Within the adjusted search radius, select data points belonging to the second dataset that meet the preset conditions. Perform distance-weighted calculations based on the attribute values ​​of the selected data points and their Euclidean distance to the target point to obtain attribute estimates.

[0199] The attribute estimates are associated with the target point to obtain a matching result.

[0200] Furthermore, the preset conditions are that the Euclidean distance is less than a preset distance and the time difference is less than a preset duration.

[0201] In some embodiments, an electronic device is also provided, including a processor and a storage device, the storage device storing a plurality of instructions, the processor being configured to read the instructions and execute the methods described above.

[0202] The multi-source data fusion method and apparatus for monitoring road carbon emissions provided in the above embodiments have at least the following beneficial effects:

[0203] (1) Improve the richness of greenhouse gas monitoring information fusion. By fusing greenhouse gas data, mobile vehicle location data, meteorological data, panoramic environmental image data, and environmental point cloud data, high-precision spatiotemporal matching can be achieved, effectively eliminating the differences in spatial distribution and time points of single data sources, and ensuring the accuracy of urban road traffic greenhouse gas monitoring;

[0204] (2) The spatiotemporal matching algorithm and multi-source data processing framework can effectively process data from different sensors and sources, providing reliable data support for further carbon emission estimation and promoting the realization of intelligent transportation systems and green and low-carbon transportation policies;

[0205] (3) Data is fused in time and space based on nearest neighbor interpolation and inverse distance weighting to improve the accuracy of data fusion.

[0206] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A multi-source data fusion method for monitoring road carbon emissions, characterized in that, include: Carbon emission data from multiple sources is collected based on monitoring equipment on mobile monitoring vehicles. This multi-source carbon emission data includes greenhouse gas data, mobile vehicle location data, meteorological data, panoramic environmental image data, and environmental point cloud data. The multi-source carbon emission data are preprocessed. An environmental model is constructed based on the preprocessed multi-source carbon emission data, and the acquisition time of the multi-source carbon emission data is obtained. Spatial information is obtained based on the constructed environmental model. Time and spatial matching are performed on the multi-source carbon emission data based on the acquisition time and spatial information. A neural network model is used to identify vehicles in the panoramic environmental image data to obtain vehicle type and quantity. A three-dimensional environmental point cloud model is constructed based on the environmental point cloud data, and road vector data is extracted based on the environmental three-dimensional point cloud model to form a first dataset. The road vector data includes the road point cloud acquisition time and road location coordinates. The greenhouse gas data and meteorological data are resampled according to a preset time window to obtain greenhouse gas resampled data and meteorological resampled data under the time window. Based on the acquisition time of the panoramic environmental image data, greenhouse gas resampled data, and meteorological resampled data, and the corresponding mobile vehicle location data under the acquisition time, timestamps and location coordinates are set for vehicle type and quantity, greenhouse gas resampled data, and meteorological resampled data to form a second dataset. Match the first and second datasets using the nearest neighbor interpolation method or the inverse distance weighting method; Based on the inverse distance weighting method, matching is performed on the first dataset and the second dataset, including: selecting data points to be matched from the first dataset as target points; calculating the Euclidean distance and time difference between the target point and each data point in the second dataset; setting an initial search radius; for the target point, counting the number of data points belonging to the second dataset within the initial search radius whose Euclidean distance and time difference satisfy a preset condition, and calculating the local point density; adjusting the search radius according to the local point density and the preset density; selecting data points belonging to the second dataset that satisfy the preset condition within the adjusted search radius; performing distance weighting calculation based on the attribute values ​​of the selected data points and their Euclidean distances to the target point to obtain attribute estimates; and associating the attribute estimates with the target point to obtain a matching result. Local point density pa The calculation is performed using the following formula: ； Where r0 represents the initial search radius, n local This represents the number of data points belonging to the first dataset within the initial search radius r0 whose Euclidean distance and time difference satisfy the preset conditions. Based on local point density pa and preset density d target Adjust the search radius: ； Where, r a This is the adjusted search radius; The spatiotemporal distribution of carbon emissions was obtained from multi-source carbon emission data based on time and spatial matching.

2. The method according to claim 1, characterized in that, The preprocessing of the multi-source carbon emission data includes: For the same type of data in the multi-source carbon emission data, calculate the standard deviation and mean; Based on the standard deviation and mean, calculate the Z score for each data point, determine whether the data point is an outlier based on the Z score, and delete the outlier data point. The time-series data in the multi-source carbon emission data are interpolated and filled using linear interpolation.

3. The method according to claim 2, characterized in that, The time-series data in the multi-source carbon emission data are interpolated and filled using linear interpolation, including: For each type of time-series data, sort them according to the collection time; Based on the collection frequency of various types of data, identify the missing points in the time series data; The value of the missing point is calculated based on the value of the previous data point and the value of the subsequent data points.

4. The method according to claim 1, characterized in that, Based on the nearest neighbor interpolation method, matching is performed on the first dataset and the second dataset, including: Select data points to be matched from the first dataset as target points, and calculate the Euclidean distance and time difference between the target points and each data point in the second dataset; In the second dataset where the Euclidean distance and time difference meet preset conditions, the optimal data point is selected and associated with the target point to obtain a matching result.

5. The method according to claim 1 or 4, characterized in that, The preset conditions are that the Euclidean distance is less than a preset distance and the time difference is less than a preset duration.

6. A multi-source data fusion device for monitoring road carbon emissions, characterized in that, include: The data acquisition module is used to collect multi-source carbon emission data based on the monitoring equipment on the mobile monitoring vehicle. The multi-source carbon emission data includes greenhouse gas data, mobile vehicle location data, meteorological data, panoramic environmental image data, and environmental point cloud data. The preprocessing module is used to preprocess the multi-source carbon emission data; The matching module is used to construct an environmental model based on preprocessed multi-source carbon emission data, obtain the acquisition time of the multi-source carbon emission data, acquire spatial information based on the constructed environmental model, and perform time and spatial matching on the multi-source carbon emission data based on the acquisition time and spatial information: a neural network model is used to identify vehicles in the panoramic environmental image data to obtain vehicle type and quantity; a three-dimensional environmental point cloud model is constructed based on the environmental point cloud data, and road vector data is extracted based on the environmental three-dimensional point cloud model to form a first dataset, the road vector data including road point cloud acquisition time and road location coordinates; greenhouse gas data and meteorological data are resampled according to a preset time window to obtain greenhouse gas resampled data and meteorological resampled data under the time window; based on the acquisition time of the panoramic environmental image data, greenhouse gas resampled data, and meteorological resampled data, and the corresponding mobile vehicle location data at that acquisition time, timestamps and location coordinates are set for vehicle type and quantity, greenhouse gas resampled data, and meteorological resampled data to form a second dataset; Match the first and second datasets using the nearest neighbor interpolation method or the inverse distance weighting method; Based on the inverse distance weighting method, matching is performed on the first dataset and the second dataset, including: selecting data points to be matched from the first dataset as target points; calculating the Euclidean distance and time difference between the target point and each data point in the second dataset; setting an initial search radius; for the target point, counting the number of data points belonging to the second dataset within the initial search radius whose Euclidean distance and time difference satisfy a preset condition, and calculating the local point density; adjusting the search radius according to the local point density and the preset density; selecting data points belonging to the second dataset that satisfy the preset condition within the adjusted search radius; performing distance weighting calculation based on the attribute values ​​of the selected data points and their Euclidean distances to the target point to obtain attribute estimates; and associating the attribute estimates with the target point to obtain a matching result. Local point density pa The calculation is performed using the following formula: ； Where r0 represents the initial search radius, n local This represents the number of data points belonging to the first dataset within the initial search radius r0 whose Euclidean distance and time difference satisfy the preset conditions. Based on local point density pa and preset density d target Adjust the search radius: ； Where, r a This is the adjusted search radius; The results processing module is used to obtain the spatiotemporal distribution results of carbon emissions based on multi-source carbon emission data with time and spatial matching.

7. An electronic device, characterized in that, The method includes a processor and a storage device, the storage device storing a plurality of instructions, and the processor being configured to read the instructions and execute the method as described in any one of claims 1-5.

Images