Method and system for dynamically monitoring carbon emission of Yellow River basin by using big data

Abstract

The invention discloses a method and a system for dynamically monitoring carbon emission of a Yellow River basin by using big data. The method comprises the following steps of: constructing a multi-source data acquisition network, cleaning acquired multi-source data, then performing data standardization, and finally performing data fusion to form a multi-dimensional fusion feature vector representing a relationship among the multi-source data; a neural network and a support vector machine are fused to construct a carbon emission accounting model, and the carbon emission of the Yellow River basin is dynamically monitored in real time by using the constructed carbon emission accounting model; and continuously updating various data in the data acquisition network, optimizing the multi-dimensional fusion feature vector and the carbon emission accounting model by using new data, and then monitoring the carbon emission of the Yellow River basin by using the optimized carbon emission accounting model. According to the method and the system, the carbon emission of the Yellow River basin can be monitored timely, comprehensively and accurately.

Description

Technical Field

[0001] The present invention relates to the technical field of carbon emission monitoring methods, and in particular to a dynamic carbon emission monitoring system for the Yellow River basin using big data. Background Art

[0002] As the global climate change problem becomes increasingly serious, carbon emission monitoring is crucial for environmental protection and sustainable development. The Yellow River Basin covers many industries and densely populated areas, with large energy consumption and complex and diverse sources of carbon emissions, including industrial production, energy extraction, transportation, residents' lives and other fields. Traditional carbon emission monitoring methods often rely on regular field surveys, sampling tests and self-reporting of data by enterprises. These methods have many limitations: First, the data acquisition cycle is long, and it is difficult to reflect the dynamic changes of carbon emissions in real time; second, the sampling range is limited, and it is impossible to accurately cover various emission sources in the entire basin, which is easy to cause data deviation; third, the data format and standards between different departments and regions are not unified, and integration is difficult, which hinders the accurate grasp of the full picture of carbon emissions in the basin. Summary of the invention

[0003] The technical problem to be solved by the present invention is how to provide a method and system for timely, comprehensive and accurate monitoring of carbon emissions in the Yellow River Basin.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for dynamic monitoring of carbon emissions in the Yellow River Basin using big data, comprising the following steps: S1, Data Collection: Build a multi-source data collection network to integrate carbon emission-related data from satellite remote sensing, ground monitoring stations, IoT sensors, and various government affairs and enterprise information systems in the Yellow River Basin; S2, data preprocessing: First, clean the collected multi-source data, then standardize the data, and unify the units and dimensions of data from different sources based on a unified data dictionary; finally, fuse the data to form a multi-dimensional fusion feature vector that represents the relationship between multi-source data; S3, carbon emission accounting model construction: by fusing neural networks and support vector machines, a carbon emission accounting model is constructed, with the processed multi-dimensional fusion feature vector as input and carbon emission information as output, the carbon emission accounting model is trained, and the random forest regression algorithm is used to dynamically adjust the accounting parameters of the carbon emission accounting model and optimize the model parameters; S4, dynamic monitoring and early warning: using the constructed carbon emission accounting model, real-time dynamic monitoring of carbon emissions in the Yellow River Basin is carried out. When it is detected that the carbon emission growth rate in a certain region or industry is too fast or the intensity exceeds the standard, the system automatically triggers an early warning; S5, data update and model optimization: continuously update various types of data in the data collection network, and use the new data to optimize the multi-dimensional fusion feature vector and carbon emission accounting model, and then use the optimized carbon emission accounting model to monitor the carbon emissions in the Yellow River Basin.

[0005] Correspondingly, the present invention also discloses a dynamic monitoring system for carbon emissions in the Yellow River Basin using big data, including: Data collection module: used to build a multi-source data collection network to integrate carbon emission-related data from satellite remote sensing, ground monitoring stations, IoT sensors, and various government affairs and enterprise information systems in the Yellow River Basin; Data preprocessing module: used to first clean the collected multi-source data, then standardize the data, and unify the units and dimensions of data from different sources based on a unified data dictionary; finally, perform data fusion to form a multi-dimensional fusion feature vector that represents the relationship between multi-source data; Carbon emission accounting model construction module: used to construct a carbon emission accounting model by fusing a neural network and a support vector machine, taking the processed multi-dimensional fusion feature vector as input and carbon emission information as output, training the carbon emission accounting model, and using a random forest regression algorithm to dynamically adjust the accounting parameters of the carbon emission accounting model and optimize the model parameters; Dynamic monitoring and early warning module: It is used to use the constructed carbon emission accounting model to conduct real-time dynamic monitoring of carbon emissions in the Yellow River Basin. When it is detected that the carbon emission growth rate in a certain region or industry is too fast or the intensity exceeds the standard, the system automatically triggers an early warning; Data update and model optimization module: used to continuously update various types of data in the data collection network, and use the new data to optimize the multi-dimensional fusion feature vector and carbon emission accounting model, and then use the optimized carbon emission accounting model to monitor the carbon emissions in the Yellow River Basin.

[0006] The beneficial effects of adopting the above technical solution are: 1) It integrates various types of data sources such as satellite remote sensing, ground monitoring stations, Internet of Things sensors, government affairs, and enterprise information systems, greatly expanding the breadth and depth of carbon emission monitoring data. Different data sources each have a rapid update frequency that suits their characteristics. It can track the dynamic changes of carbon emissions in the Yellow River Basin, capture instantaneous fluctuations such as sudden equipment failures in factories leading to a sharp increase in energy consumption and soaring carbon emissions during urban traffic rush hours, making carbon emission monitoring data more realistic.

[0007] 2) Using the geographic information system (GIS) and the Yellow River basin map as the basis for data fusion, we can accurately align multi-source data in space, integrate carbon emission-related information at various geographical locations in the basin, and present it in three-dimensional space and time, intuitively showing the evolution of carbon emissions over time in the basin space, thus providing high-quality input data for subsequent carbon emission accounting models.

[0008] 3) Accounting models are set up according to the characteristics of different industries and fields in the Yellow River Basin, which can fit the actual situation of the basin and greatly improve the accuracy of carbon emission accounting. Machine learning algorithms are introduced to explore hidden influencing factors, dynamically adjust the accounting parameters of the model, and adaptively learn the changes in carbon emission patterns, optimize accuracy in real time, and accurately reflect the actual carbon emissions in the Yellow River Basin.

[0009] 4) The constructed carbon emission accounting model can be used to regularly calculate the overall carbon emission intensity of the basin and the distribution of each region and industry, and display it on the monitoring screen in the form of visual charts. High-frequency calculation updates and intuitive chart presentations convert the dynamic changes of carbon emissions in the Yellow River Basin into visual information in real time; multi-level carbon emission warning thresholds are set. Once the carbon emission growth rate of a region or industry is detected to be too fast or the intensity exceeds the standard, the system will automatically trigger a warning immediately. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0011] Figure 1 is a main flow chart of the method described in the embodiment of the present invention; Figure 2 is a flow chart of forming a multi-dimensional fusion feature vector in the method according to an embodiment of the present invention; Figure 3 is a flow chart of attribute fusion and feature extraction in the method described in an embodiment of the present invention; Figure 4 is a flow chart of the construction of the carbon emission accounting model in the method according to the embodiment of the present invention; Figure 5 is a flow chart of dynamic monitoring and early warning in the method described in an embodiment of the present invention; Figure 6 It is a principle block diagram of the system described in the embodiment of the present invention. DETAILED DESCRIPTION

[0012] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0013] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0014] like Figure 1 As shown, the embodiment of the present invention discloses a method for dynamic monitoring of carbon emissions in the Yellow River Basin using big data, comprising the following steps: S1, Data Collection: Build a multi-source data collection network to integrate carbon emission-related data from satellite remote sensing, ground monitoring stations, Internet of Things sensors, and various government affairs and enterprise information systems in the Yellow River Basin; satellite remote sensing data is used to obtain large-scale land use changes and vegetation coverage information in the Yellow River Basin that indirectly reflect carbon emission factors; ground monitoring stations are set up in cities, industrial parks and energy production areas in the Yellow River Basin to monitor the concentration of atmospheric pollutants and meteorological parameter information in real time; Internet of Things sensors are deployed at carbon emission micro-sources, including factory workshops, traffic arteries and residential communities, to collect the operating power of energy-consuming equipment of carbon emission micro-sources, vehicle mileage and fuel consumption, and household electricity and gas consumption data, and transmit them to the data center in real time; at the same time, connect with the information systems of environmental protection, energy and industry and information technology departments and large enterprises in the basin to obtain enterprise production scale, energy consumption structure and product output data, and update them synchronously every day.

[0015] S2, data preprocessing: first, the collected multi-source data is cleaned, and then the data is standardized. According to a unified data dictionary, the units and dimensions of data from different sources are unified; finally, data fusion is performed to form a multi-dimensional fusion feature vector that represents the relationship between multi-source data; further, after receiving the collected multi-source data, the data is first cleaned. For abnormal values ​​caused by sensor failure and transmission interruption, an anomaly detection algorithm based on time series analysis is used. The threshold is set by combining the mean and standard deviation of data in adjacent time periods to identify and eliminate abnormal data points; for data collected by satellite remote sensing, clear image data is restored through image restoration and filtering algorithms; secondly, data standardization is performed. According to a unified data dictionary, the units and dimensions of data from different sources are unified; finally, data fusion is performed. Based on the geographic information system and the Yellow River Basin map, the satellite remote sensing data, ground monitoring data, Internet of Things data and industry data that have been cleaned and standardized are matched and fused according to geographical location coordinates to form a multi-dimensional fusion feature vector.

[0016] S3, carbon emission accounting model construction: by fusing neural networks and support vector machines, a carbon emission accounting model is constructed, with the processed multi-dimensional fusion feature vector as input and carbon emission information as output, the carbon emission accounting model is trained, and the random forest regression algorithm is used to dynamically adjust the accounting parameters of the carbon emission accounting model and optimize the model parameters; S4, dynamic monitoring and early warning: using the constructed carbon emission accounting model, real-time dynamic monitoring of carbon emissions in the Yellow River Basin is carried out. When it is detected that the carbon emission growth rate in a certain region or industry is too fast or the intensity exceeds the standard, the system automatically triggers an early warning; Furthermore, the constructed carbon emission accounting model is used to conduct real-time dynamic monitoring of carbon emissions in the Yellow River Basin, and the collected multi-source data is input into the carbon emission accounting model in real time. The overall carbon emission intensity of the Yellow River Basin and the carbon emission distribution of each region and industry are calculated once at a set time interval, and displayed in the form of a visual chart on the display module; multi-level carbon emission warning thresholds are set. When the growth rate or increase of carbon emissions in a certain region or industry exceeds the set threshold, the system automatically triggers a warning, and the warning information is pushed to environmental protection regulatory departments and related companies via SMS and / or email, and a pop-up reminder is displayed on the mobile APP.

[0017] S5, data update and model optimization: continuously update various types of data in the data collection network, and use the new data to optimize the multi-dimensional fusion feature vector and carbon emission accounting model, and then use the optimized carbon emission accounting model to monitor the carbon emissions in the Yellow River Basin.

[0018] Further, such as Figure 2 As shown, the method for forming a multi-dimensional fusion feature vector in S2 includes the following steps: S2-1) Geographic coordinate unification and calibration: 1) Satellite remote sensing data coordinate processing: First, coordinate registration of satellite remote sensing images is performed through the remote sensing analysis module in the remote sensing image processing software ArcGIS. Multiple control points with known precise geographical locations in the Yellow River Basin are selected (including iconic peaks, large bridges and other fixed landmarks in the basin, whose accurate coordinates can be obtained through high-precision geodetic methods). The coordinate system of the remote sensing images is calibrated one by one to a coordinate system consistent with the selected Yellow River Basin map benchmark through the least squares method, so that the geographical location of the remote sensing data is accurately matched to the actual geographical space of the Yellow River Basin, and the error is controlled within the allowable range (sub-meter level or accuracy standards set according to actual needs); 2) Ground monitoring site data coordinate matching: When constructing and arranging ground monitoring sites, high-precision GPS positioning equipment is used to record their precise longitude and latitude coordinates, and the longitude and latitude coordinates are converted to the Yellow River Basin. Figure 1 In the data collection system, the corresponding geographical location coordinate information is associated with each type of monitoring data (atmospheric pollutant concentration, meteorological parameters, etc.) of each ground monitoring station in a consistent coordinate system; 3) IoT data coordinate association: When installing IoT sensors at micro-sources of carbon emissions in factory workshops, traffic arteries and residential areas, obtain their geographic location coordinates with the help of on-site mapping tools or mobile positioning methods, and embed these coordinate information into the data transmission protocol collected by the sensors, so that each set of collected operating power of energy-consuming equipment, vehicle mileage and fuel consumption, and household electricity and gas consumption data are accompanied by geographic coordinates, so that IoT data has a corresponding location in the geographic space of the Yellow River Basin; 4) Industry data coordinate positioning: Through the geographical location information of the enterprise's registered address and production facilities, combined with geographic coding, the corresponding geographical location coordinates are assigned to the industry data in the government affairs and enterprise information systems, so that the industry data can be located at the geographical space level of the Yellow River Basin and participate in the subsequent data fusion process; S2-2) Data stratification and spatial processing 1) Data hierarchical organization Using GIS software, data from different data sources that have been uniformly processed with coordinates are managed in layers according to categories; satellite remote sensing image data is used as a separate data layer, ground monitoring site data is constructed into different data layers according to site categories (including urban sites, industrial park sites, energy production area sites, etc.), IoT data is layered according to the area where the sensor is located (factory, transportation, residential and other different categories of areas), and industry data is layered according to the industry to which the enterprise belongs or the area where it is located; 2) Spatial representation The data of each layer is spatially represented and converted so that the data can be presented in the form of intuitive geographic spatial elements. The ground monitoring site data is visualized on the map in the form of point elements. Each point represents a monitoring site, and its attribute information includes the real-time monitoring of atmospheric pollutant concentrations and meteorological parameters. The satellite remote sensing data is spread over the corresponding areas of the Yellow River Basin map (factory workshop range, traffic section range, residential area range, etc.) in the form of raster data. Each raster pixel carries information on land use changes and vegetation coverage. The Internet of Things data represents the corresponding area as surface elements or point elements according to its deployment location, and associates the corresponding energy consumption and operation attributes. Industry data is represented by point elements or regional surface elements (for large enterprise parks, etc.) of the enterprise location, and its attributes include production scale information. S2-3) Matching and fusion based on geospatial location Based on the spatial overlay analysis function of GIS, the data layers of each layer are overlaid according to the geographic location coordinates. First, based on the Yellow River Basin map, the point feature data layer of the ground monitoring site is overlaid with the raster data layer of satellite remote sensing. At each monitoring site, the real-time monitoring data of the site and the surrounding land use and vegetation coverage information reflected by the corresponding remote sensing image are simultaneously obtained; then the regional or point feature layer corresponding to the IoT data is overlaid on the above layer, so that the energy consumption data of factories, traffic arteries and residential areas at the microscopic location are integrated with the site monitoring data and remote sensing macro data; finally, the industry data layer is overlaid, so that the enterprise's industry-related data is combined with other carbon emission-related data at a specific geographical location, so as to achieve the matching and fusion of multi-source data in the geographic space location; S2-4) Attribute fusion and feature extraction: For the multi-source data at each spatial location after superposition, attribute fusion operation is performed. By writing GIS scripts, relevant attribute data from different data sources but corresponding to the same location are integrated. The energy consumption data collected by IoT sensors in a factory area, the surrounding atmospheric pollutant concentration data obtained by ground monitoring stations, and the production scale data of the factory in the enterprise information system are fused according to methods such as weighted average to form a set of new attribute data that comprehensively reflects the carbon emission-related characteristics of the location; then, from these fused attribute data, representative features are extracted based on the key factors that affect carbon emissions to form a multi-dimensional fusion feature vector. The vector elements include the fused comprehensive energy consumption index, comprehensive pollutant emission index and industrial scale index. The feature vector is used to comprehensively and comprehensively reflect the carbon emission-related characteristics of the corresponding geographical location.

[0019] Further, such as Figure 3 As shown, the step S2-4) attribute fusion and feature extraction includes the following steps: S2-4-1, determine the weight coefficient of weighted average: Analyze the relative importance of data from different data sources in reflecting the carbon emission characteristics of the location, and set the weight coefficients of energy consumption data, atmospheric pollutant concentration data obtained from ground monitoring stations, and enterprise production scale data according to their importance; refer to past carbon emission accounting cases of the data source, relevant research data of similar data sources, and consult industry experts to obtain the actual proportion of the impact of each data on carbon emissions in similar scenarios. Based on this information, further calibrate the weight coefficients preliminarily set above, with a correction range of ±0.1; S2-4-2, write GIS script to realize weighted average fusion: The scripting language supported by GIS software is used to implement attribute fusion operations. According to the specific data table structure, field names and real weights, adjustments and improvements are made. The relevant attribute data of different data sources at the same location are fused according to the weighted average method, and new attribute data related to energy consumption, atmospheric pollutant concentration and production scale are generated for each data source element. S2-4-3, multi-dimensional fusion feature vector construction: Based on the understanding of the carbon emission mechanism and the industrial and environmental characteristics of the Yellow River Basin, the key factors affecting carbon emissions in this location are determined, including energy consumption, atmospheric pollutant concentrations and production scale factors. The energy consumption data are processed comprehensively into comprehensive energy consumption indicators that reflect the overall energy consumption level of the region, including energy consumption per unit area and energy consumption per unit output value. For pollutant emission data, a weighted comprehensive index based on the environmental impact weights of different pollutants is constructed. For production scale data, the proportion of carbon-intensive product output that reflects the relationship between industrial scale and carbon emissions is extracted. In the factory area, IoT sensors collect the operating power and energy consumption time data of multiple devices. First, the total energy consumption of the area is calculated by adding the energy consumption of each device, and then the energy consumption per unit area is calculated in combination with the area of ​​the area. At the same time, the output value data of the factory is obtained to calculate the energy consumption per unit output value. These two indicators are combined by assigning different weights to form the final energy consumption comprehensive index, which is included in the multi-dimensional fusion feature vector. For the concentration data of various atmospheric pollutants obtained by ground monitoring stations, the concentrations of each pollutant are weighted and summed according to the weight of each pollutant's impact on the greenhouse effect and air quality, and a comprehensive pollutant emission index that comprehensively reflects the overall intensity of atmospheric pollutant emissions in the area and the degree of impact on the environment is obtained, which is added to the multi-dimensional fusion feature vector. The data related to the production scale of the enterprise and the industrial structure data are combined, and the enterprise product classification and output share are integrated to obtain the industrial scale index, which is added to the multi-dimensional fusion feature vector. The fusion energy consumption comprehensive index, pollutant emission comprehensive index and industrial scale index are combined in an orderly manner to form a multi-dimensional fusion feature vector [ ],in Represents the comprehensive index of energy consumption, It represents the comprehensive index of pollutant emissions. Represents an industry scale indicator.

[0020] Furthermore, the analysis of the relative importance of data from different data sources in reflecting the carbon emission characteristics of the location includes the following steps: 1) Get the corresponding data sequence The reference sequence is the carbon emission characteristic value sequence: ; in, Indicated in The carbon emission value obtained at the moment, is the number of elements in the carbon emission characteristic value sequence; The comparison series include energy consumption data series, air pollutant concentration data series and enterprise production scale data series; Energy consumption data series: ; in, Indicated in Energy consumption data corresponding to each moment; Atmospheric pollutant concentration data series: ; in, Indicated in The atmospheric pollutant concentration data monitored at each moment; Enterprise production scale data series: ; in, Indicates that the enterprise Quantitative indicators of production scale corresponding to each moment; is the number of elements in the comparison sequence; 2) Calculate the correlation coefficient Reference sequence and The sequences after initialization are , ; Calculate the correlation coefficient between each comparison sequence and the reference sequence at each time : Correlation coefficient The value range is from 0 to 1, and the closer it is to 1, the closer the association is. Indicates the reference sequence after initialization at the kth moment Compare sequence with the i-th The absolute value of the corresponding data difference is used to indicate the degree of difference between the two at that moment; It indicates the minimum absolute value of the difference between the reference sequence and the comparison sequence in all n moments, that is, for a specific comparison sequence , find its difference with the reference sequence The absolute value of the difference at the moment with the smallest difference among n moments; It means that among all i comparison sequences, the minimum value among the absolute values ​​of the minimum differences is found, that is, the minimum value among all comparison sequences and the reference sequence with the minimum difference at each moment, indicating the closest degree between the overall data sequences; , represents the situation where all comparison sequences have the largest difference with the reference sequence at each moment, that is, first find the maximum absolute value of the difference between each comparison sequence and the reference sequence at n moments, and then select the largest value from these maximum values, which represents the maximum degree of difference between the overall data sequences; Among them, the reference sequence is initialized: ; It represents the carbon emission value at the kth moment divided by the carbon emission value at the first moment, converting the carbon emission characteristic value sequence into a relative change sequence based on the first value; Comparison sequence initialization: , ; It represents the energy consumption, air pollutant concentration or enterprise production scale data at the kth moment divided by the energy consumption, air pollutant concentration or enterprise production scale data at the first moment, and converts the energy consumption characteristic sequence, air pollutant concentration characteristic sequence and enterprise production scale characteristic sequence into a relative change sequence based on the first value; is the resolution coefficient, which ranges from 0 to 1 and is used to adjust the size of the correlation coefficient; 3) Calculate the grey correlation After obtaining the correlation coefficient at each moment, the grey correlation degree between each comparison sequence and the reference sequence is calculated by the following formula: , ; Grey relational degree For comparing sequences At n moments with the reference sequence The average value of the correlation coefficient reflects the closeness of the correlation between the comparison sequence and the reference sequence in the entire data interval. The value range is between 0 and 1. The larger the value, the stronger the correlation between the corresponding energy consumption, atmospheric pollutant concentration or enterprise production scale data and the carbon emission characteristic value, and the higher the relative importance in reflecting the carbon emission characteristics.

[0021] Further, such as Figure 4 As shown, S3 specifically includes the following steps: S3-1, Neural network model selection and structure design According to the characteristics of the carbon emission data in the Yellow River Basin, the multi-layer perceptron MLP neural network is selected as the basic architecture. MLP has strong nonlinear mapping capabilities and can handle complex input-output relationships. It is suitable for mining the carbon emission laws hidden in the multi-dimensional fusion feature vector. The number of input layer nodes corresponds to the dimension of the multi-dimensional fusion feature vector, so that it can fully receive the processed multi-source data feature information. The number of hidden layer neurons is set in a decreasing manner. The number of neurons in the first hidden layer is 2 / 3 of the number of input layer nodes, and the number of subsequent hidden layers decreases in turn. The commonly used ReLU function is used as the hidden layer activation function. The number of output layer nodes is determined according to the dimension of the carbon emission information to be predicted, and the output layer activation function uses a linear activation function. S3-2, Support Vector Machine (SVM) model configuration The radial basis kernel function RBF is used as the kernel function of the SVM model to map the original input space to the high-dimensional feature space, and the penalty parameters related to the RBF kernel function are and kernel width parameter , the cross-validation method is used to optimize the selection, a part of the labeled training data is divided into multiple subsets, one of which is used as the validation set in turn, and the remaining subsets are used as the training set, and the cross-validation method is used to optimize the selection. and , find the parameter combination that makes the model have the best prediction performance on the validation set; S3-3, model fusion: First, the multi-dimensional fusion feature vector is input into the designed neural network. After nonlinear transformation of each layer of the neural network, high-level abstract features are extracted and an intermediate feature representation is output. Then, this intermediate feature is used as the input of the support vector machine, and the support vector machine performs the final carbon emission information prediction. S3-4, Training of carbon emission accounting model The collected multi-dimensional fusion feature vector data set with carbon emission labeling information is divided into training set, validation set and test set according to a certain ratio. The training set is used for learning and adjusting model parameters, and the validation set is used to evaluate the generalization ability of the model during the training process. The mean square error MSE is selected as the loss function. For the neural network part, the Adam optimization algorithm is used for parameter update training. For the support vector machine part, the parameters are solved according to the training algorithm provided by the selected optimization software library, and the optimal classification or regression hyperplane is found based on algorithms such as quadratic programming. According to the set training algorithm and parameters, the training set data is repeatedly input into the integrated carbon emission accounting model, and the weight and bias parameters of the model are continuously adjusted to gradually reduce the loss function value. At the same time, after each round of training, the model is evaluated with the validation set data, and the loss value and mean absolute error (MAE) on the validation set are recorded to observe whether the model is overfitting. S3-5, use the random forest regression algorithm to dynamically adjust the calculation parameters: By comparing the changes in the prediction error of the model on the validation set with different numbers of decision trees, the number of decision trees that can make the model performance stable is selected; the maximum depth of the decision tree and the minimum number of samples required for node splitting are configured to build a random forest regression model for parameter adjustment; each feature in the multidimensional fusion feature vector is input into the constructed random forest regression model, and the relative importance of each feature to the carbon emission prediction results is analyzed based on the average reduction in node impurity to determine which features are key factors affecting carbon emissions and which are relatively minor factors; at the same time, the random forest regression model is used to analyze the hidden influencing factors. Mining is carried out to obtain the influencing mechanism by analyzing the branch conditions of the decision tree and the combination of each feature in different decision trees; new labeled carbon emission data and the corresponding multi-dimensional fusion feature vectors are regularly collected, and input into the trained carbon emission accounting model to obtain the prediction results, and the deviation between the predicted value and the true value is calculated; according to the key features and influencing factors analyzed by the random forest regression model, the parameters in the carbon emission accounting model related to these factors are determined, and the quantitative relationship between the features obtained by the random forest regression model and carbon emissions is used to dynamically adjust these key parameters to optimize the model's prediction ability for new data.

[0022] Further, such as Figure 5 As shown, the S4 specifically includes the following steps: S4-1, real-time dynamic monitoring and calculation: First, the collected multi-source data (satellite remote sensing, ground monitoring sites, IoT sensors, government affairs, enterprise information systems, etc.) are transmitted in real time to the computing server or cloud computing platform environment where the model is located in accordance with the established data format through the data interface and network communication protocol. During the data transmission process, the real-time incoming data is preprocessed; the carbon emission accounting model is started for calculation at the set time interval (for example, every 1 hour, half an hour, etc., which can be flexibly adjusted according to actual needs and computing resources, monitoring accuracy and other factors). The model uses the processed multi-dimensional fusion feature vector input in real time as input, runs the internal neural network and support vector machine fusion calculation logic, and outputs the corresponding carbon emission information; When calculating the overall carbon emission intensity of the Yellow River Basin, the estimated carbon emissions of all regions and all industries are summarized, and combined with the geographical scope information of the Yellow River Basin, the carbon emission intensity per unit area is obtained by dividing the total carbon emissions by the area; for the calculation of carbon emission distribution in each region, the carbon emission estimation results are divided into corresponding different regions (such as administrative divisions within the Yellow River Basin, different ecological functional areas or pre-set monitoring grid areas) according to the geographical location coordinate information carried in the data, and the carbon emission values ​​and proportions of each region are counted; for the carbon emission distribution of each industry, the carbon emissions of different industries (such as electricity, steel, chemical industry, transportation industry, service industry, etc. in industry) are obtained according to the industry classification information of the enterprise, and the total carbon emissions of each industry and its proportion in the overall emissions are calculated.

[0023] S4-2, visual chart display: Select the type of visualization chart based on the characteristics of the carbon emission-related indicators to be displayed. For the overall carbon emission intensity change trend, use a line chart with time (hours, days, months, etc., determined according to the actual display period) as the horizontal axis and the carbon emission intensity value as the vertical axis; use a heat map to display the carbon emission distribution in each region, using the Yellow River Basin map as the base, and rendering different regions in different shades of color according to their carbon emission values ​​(regions with high carbon emissions are red, and the darker the color, the more emissions; regions with low carbon emissions are green, and the lighter the color, the less emissions); use a pie chart to display the proportion of carbon emissions in each industry, with the entire circle representing the total carbon emissions in the Yellow River Basin, and each sector corresponding to a different industry, and the size of the sector is proportional to the proportion of carbon emissions in that industry; In addition to the above-mentioned main chart types, you can also combine bar charts (comparing the absolute values ​​of carbon emissions in different regions or industries in the same period of time, etc.), stacked charts (showing the composition of different types of emission sources in carbon emissions in each region or industry, etc.) and other chart forms to comprehensively display information related to carbon emissions in the Yellow River Basin from multiple angles, meeting the needs of different users (environmental protection regulatory personnel, corporate managers, scientific researchers, etc.) to view and analyze data.

[0024] Build a visualization display module and integrate it into the user interface of the monitoring system. By writing code, bind the indicator data calculated by the carbon emission accounting model to the corresponding visualization chart elements in real time. When using Echarts to draw a line chart to display the overall carbon emission intensity, configure the basic style of the chart through JavaScript code, and then use the data interface to obtain the intensity value output by each round of calculation of the carbon emission accounting model. Dynamically add these data to the data series of the line chart according to the time series to achieve real-time update and display of the line chart; for the heat map to display the carbon emission distribution of each region, the carbon emission data of each region output by the model is also associated with the corresponding regional elements on the map through code, and the regional color rendering effect is adjusted according to the value to reflect the changes in regional carbon emissions in real time.

[0025] S4-3, multi-level carbon emission warning threshold setting and warning triggering: Setting thresholds based on historical data and policy standards: Analyze the historical data of carbon emissions in the Yellow River Basin over the past 10 years to obtain the normal range and fluctuation of carbon emissions in different regions and industries in the Yellow River Basin under normal production and operation and different seasons, and set basic warning thresholds based on this; at the same time, refer to national and local carbon emission-related policies and regulations and industry standards, and combine with the ecological protection goals of the Yellow River Basin to comprehensively determine the specific values ​​of warning thresholds at different levels; Early warning trigger mechanism and information push: The carbon emission growth rate and increment of each region and industry calculated by the carbon emission accounting model are compared and judged in real time with the set multi-level early warning thresholds. By writing program logic, once the carbon emission growth rate or increment of a certain region or industry exceeds the corresponding threshold level, the early warning mechanism is immediately triggered, and the SMS gateway service, email sending service and mobile APP development interface functions are integrated into the monitoring system to realize early warning of different information.

[0026] Generally, it can be divided into three or four levels of warning: Level 1 warning (mild warning): triggered when the growth rate or increase of carbon emissions in a region or industry is slightly higher than the normal fluctuation range, but has not yet reached the level that may have a significant impact on the environment or industrial development. For example, if the growth rate of carbon emissions exceeds the historical average growth rate for the same period by 10% - 20% (the specific value is adjusted according to the actual situation), it can be set as the level 1 warning threshold range to remind relevant parties to pay attention to changes in carbon emissions and appropriately investigate possible causes.

[0027] Level 2 warning (moderate warning): triggered when the growth rate or increase of carbon emissions further exceeds the reasonable range and may begin to exert certain pressure on regional environmental quality, industrial low-carbon transformation, etc. For example, if the growth rate exceeds the historical period by 20% - 50%, relevant departments and enterprises need to pay more attention, analyze the reasons and prepare to take some preliminary regulatory measures.

[0028] Level 3 warning (severe warning): triggered when the growth rate or increase of carbon emissions seriously exceeds the normal level, posing a major risk to the ecological environment and economic sustainable development of the Yellow River Basin. For example, if the growth rate exceeds 50% or even higher, strong emission reduction and control measures must be taken immediately. Environmental protection regulatory authorities may need to intervene to conduct strict supervision and urge rectification.

[0029] Correspondingly, such as Figure 6 As shown, the embodiment of the present invention also discloses a dynamic monitoring system for carbon emissions in the Yellow River Basin using big data, including: Data collection module 101: used to build a multi-source data collection network to integrate carbon emission-related data from satellite remote sensing, ground monitoring stations, IoT sensors, and various government affairs and enterprise information systems in the Yellow River Basin; Data preprocessing module 102: used to first clean the collected multi-source data, then standardize the data, and unify the units and dimensions of data from different sources according to a unified data dictionary; finally, perform data fusion to form a multi-dimensional fusion feature vector that represents the relationship between the multi-source data; Carbon emission accounting model construction module 103: used to construct a carbon emission accounting model by fusing a neural network and a support vector machine, taking the processed multi-dimensional fusion feature vector as input and carbon emission information as output, training the carbon emission accounting model, and using a random forest regression algorithm to dynamically adjust the accounting parameters of the carbon emission accounting model and optimize the model parameters; Dynamic monitoring and early warning module 104: used to use the constructed carbon emission accounting model to conduct real-time dynamic monitoring of carbon emissions in the Yellow River Basin. When it is detected that the carbon emission growth rate in a certain region or industry is too fast or the intensity exceeds the standard, the system automatically triggers an early warning; Data updating and model optimization module 105: used to continuously update various types of data in the data collection network, and use the new data to optimize the multi-dimensional fusion feature vector and the carbon emission accounting model, and then use the optimized carbon emission accounting model to monitor the carbon emissions in the Yellow River Basin.

[0030] It should be noted that the implementation method of the corresponding modules in the system can refer to the aforementioned dynamic monitoring method, which will not be elaborated here.

[0031] The method and system can monitor carbon emissions in the Yellow River Basin in a timely, comprehensive and accurate manner.

Claims

1. A dynamic monitoring method for carbon emissions in the Yellow River Basin using big data, characterized in that The steps include: S1, Data Collection: Build a multi-source data collection network to integrate carbon emission-related data from satellite remote sensing, ground monitoring stations, IoT sensors, and various government affairs and enterprise information systems in the Yellow River Basin; S2, data preprocessing: First, the collected multi-source data is cleaned, then the data is standardized, and the units and dimensions of data from different sources are unified according to a unified data dictionary. Finally, the data is fused to form a multi-dimensional fusion feature vector that represents the relationship between multi-source data; S3, construction of carbon emission accounting model: by fusing neural network and support vector machine, a carbon emission accounting model is constructed, with the processed multi-dimensional fusion feature vector as input and carbon emission information as output, the carbon emission accounting model is trained, and the random forest regression algorithm is used to dynamically adjust the accounting parameters of the carbon emission accounting model and optimize the model parameters; S4, dynamic monitoring and early warning: using the constructed carbon emission accounting model, real-time dynamic monitoring of carbon emissions in the Yellow River Basin is carried out. When it is detected that the carbon emission growth rate in a certain region or industry is too fast or the intensity exceeds the standard, the system automatically triggers an early warning; S5, data update and model optimization: continuously update various types of data in the data collection network, and use the new data to optimize the multi-dimensional fusion feature vector and carbon emission accounting model, and then use the optimized carbon emission accounting model to monitor the carbon emissions in the Yellow River Basin.

2. The method for dynamic monitoring of carbon emissions in the Yellow River Basin using big data according to claim 1, characterized in that: In the S1 data collection step: Satellite remote sensing data is used to obtain information on large-scale land use changes in the Yellow River Basin that indirectly reflects carbon emission factors and vegetation cover information Ground monitoring stations are set up in cities, industrial parks and energy production areas in the Yellow River Basin to monitor the concentration of air pollutants in real time And meteorological parameter information IoT sensors are deployed at micro-sources of carbon emissions, including factory workshops, traffic arteries, and residential areas, to collect operating power data of energy consumption equipment at micro-sources of carbon emissions. , vehicle mileage and fuel consumption data , Household electricity and gas consumption data Collect data and transmit it to the data center in real time; at the same time, connect with the information systems of environmental protection, energy and industry and information technology departments and large enterprises in the basin to obtain data on enterprise production scale, energy consumption structure and product output .

3. The method for dynamic monitoring of carbon emissions in the Yellow River Basin using big data according to claim 1, characterized in that: The step S2 data preprocessing includes: After receiving the collected multi-source data, the data is first cleaned. For outliers caused by sensor failure and transmission interruption, an anomaly detection algorithm based on time series analysis is used. The threshold is set in combination with the mean and standard deviation of the data in adjacent time periods to identify and remove abnormal data points. For data collected by satellite remote sensing, the image data is restored through image restoration and filtering algorithms. Secondly, data standardization is performed. According to a unified data dictionary, the units and dimensions of data from different sources are unified. Finally, data fusion is performed. Based on the geographic information system GIS and the Yellow River basin map, the satellite remote sensing data, ground monitoring data, Internet of Things data and industry data that have been cleaned and standardized are matched and fused according to the geographic location coordinates to form a multi-dimensional fusion feature vector.

4. The method for dynamic monitoring of carbon emissions in the Yellow River Basin using big data as claimed in claim 3, characterized in that: The method for forming a multi-dimensional fusion feature vector comprises the following steps: S2-1) Geographic coordinate unification and calibration: 1) Satellite remote sensing data coordinate processing: First, the remote sensing analysis module in ArcGIS, a remote sensing image processing software, was used to perform coordinate registration on satellite remote sensing images. Multiple control points with known precise geographical locations in the Yellow River Basin were selected, and the coordinate system of the remote sensing images was calibrated one by one to a coordinate system consistent with the selected Yellow River Basin map benchmark through the least squares method, so that the geographical location of the remote sensing data corresponded to the actual geographical space of the Yellow River Basin. 2) Ground monitoring site data coordinate matching: When constructing and laying out ground monitoring sites, the precise longitude and latitude coordinates are recorded by GPS positioning equipment, and the longitude and latitude coordinates are converted to a coordinate system consistent with the Yellow River Basin map. During the data collection process, the corresponding geographic location coordinate information is associated with each type of monitoring data for each ground monitoring site; 3) IoT data coordinate association: Install and deploy IoT sensors for micro-sources of carbon emissions, use field mapping tools or mobile positioning methods to obtain their geographic coordinates, and embed these coordinate information into the data transmission protocol collected by the sensors, so that each set of collected data on the operating power of energy-consuming equipment, vehicle mileage and fuel consumption, and household electricity and gas consumption is accompanied by geographic coordinates, so that IoT data has a corresponding location in the geographic space of the Yellow River Basin; 4) Industry data coordinate positioning: By combining the geographical location information of the enterprise's registered address and the location of its production facilities with geographic coding, the corresponding geographical location coordinates are assigned to the industry data in the government affairs and enterprise information systems, so that the industry data can be located at the geographical space level of the Yellow River Basin; S2-2) Data stratification and spatial processing: 1) Data hierarchical organization Using GIS software, data from different data sources that have been uniformly processed with coordinates are managed in layers according to categories; satellite remote sensing image data is used as a separate data layer, ground monitoring site data is constructed into different data layers according to site categories, IoT data is layered according to the area where the sensor is located, and industry data is layered according to the industry or area to which the enterprise belongs; 2) Spatial representation The data of each layer is spatially represented and converted so that the data can be presented in the form of intuitive geographic spatial elements. The ground monitoring station data is visualized on the map in the form of point elements. Each point represents a monitoring station, and its attribute information includes real-time monitored atmospheric pollutant concentrations and meteorological parameters. The satellite remote sensing data is spread over the corresponding area of ​​the Yellow River Basin map in the form of raster data. Each raster pixel carries land use change information and vegetation coverage information. IoT data is represented by area elements or point elements according to its deployment location, and the corresponding energy consumption and operation attributes are associated; industry data is represented by point elements or regional area elements of the enterprise location, and its attributes include production scale information; S2-3) Matching and fusion based on geospatial location Based on the spatial overlay analysis function of GIS, the data layers of each layer are overlaid according to the geographic location coordinates. First, based on the Yellow River Basin map, the point feature data layer of the ground monitoring site is overlaid with the raster data layer of satellite remote sensing. At each monitoring site, the real-time monitoring data of the site and the surrounding land use and vegetation coverage information reflected by the corresponding remote sensing image are simultaneously obtained; then the regional or point feature layer corresponding to the IoT data is overlaid on the above layer, so that the energy consumption data of factories, traffic arteries and residential areas at the microscopic location are integrated with the site monitoring data and remote sensing macro data; finally, the industry data layer is overlaid, so that the enterprise's industry-related data is combined with other carbon emission-related data at a specific geographical location, so as to achieve the matching and fusion of multi-source data in the geographic space location; S2-4) Attribute fusion and feature extraction: For the multi-source data at each spatial location after superposition, attribute fusion operation is performed. By writing GIS scripts, relevant attribute data from different data sources but corresponding to the same location are integrated and fused according to the weighted average method to form a set of new attribute data that comprehensively reflects the carbon emission-related characteristics of the location; Then, from these fused attribute data, representative features are extracted based on the key factors that affect carbon emissions and combined into a multi-dimensional fused feature vector. The vector elements include the fused comprehensive energy consumption index, comprehensive pollutant emission index and industrial scale index.

5. The method for dynamic monitoring of carbon emissions in the Yellow River Basin using big data according to claim 1, characterized in that: The step S2-4) attribute fusion and feature extraction includes the following steps: S2-4-1, determine the weight coefficient of weighted average: Analyze the relative importance of data from different data sources in reflecting the carbon emission characteristics of the location, and set the weight coefficients of energy consumption data, atmospheric pollutant concentration data obtained from ground monitoring stations, and enterprise production scale data according to their importance; Refer to the past carbon emission accounting cases of the data source, relevant research data of the same type of data sources, and consult industry experts to obtain the actual proportion of the impact of each data on carbon emissions in similar scenarios. Based on this information, further calibrate the weight coefficients preliminarily set above, with a correction range of ±0.1; S2-4-2, write GIS script to realize weighted average fusion: The scripting language supported by GIS software is used to implement attribute fusion operations. According to the specific data table structure, field names and real weights, adjustments and improvements are made. The relevant attribute data of different data sources at the same location are fused according to the weighted average method, and new attribute data related to energy consumption, atmospheric pollutant concentration and production scale are generated for each data source element. S2-4-3, multi-dimensional fusion feature vector construction: Based on the understanding of carbon emission mechanism and the industrial and environmental characteristics of the Yellow River Basin, the key factors affecting carbon emissions in this location are determined, including energy consumption, atmospheric pollutant concentration and production scale factors, and the energy consumption data is processed into comprehensive energy consumption indicators that reflect the overall energy consumption level of the region, including energy consumption per unit area and energy consumption per unit output value; For pollutant emission data, a weighted comprehensive index based on the environmental impact weights of different pollutants is constructed; for production scale data, the output ratio of carbon-intensive products that reflects the correlation between industrial scale and carbon emissions is extracted; In the factory area, IoT sensors collect the operating power and energy consumption time data of multiple devices. First, the total energy consumption of the area is calculated by adding the energy consumption of each device, and then the energy consumption per unit area is calculated based on the area of ​​the area. At the same time, the output value data of the factory is obtained to calculate the energy consumption per unit output value. These two indicators are combined by assigning different weights to form the final comprehensive energy consumption indicator, which is included in the multi-dimensional fusion feature vector. For the concentration data of various air pollutants obtained by ground monitoring stations, the concentrations of various pollutants are weighted and summed according to the weight of each pollutant's impact on the greenhouse effect and air quality, and a comprehensive pollutant emission index that comprehensively reflects the overall intensity of air pollutant emissions in the region and the degree of impact on the environment is obtained, and added to the multidimensional fusion feature vector; Combine the data related to the enterprise's production scale and industrial structure data, and integrate the enterprise's product classification and output share to obtain the industrial scale index, and add it to the multi-dimensional fusion feature vector; The comprehensive energy consumption index, pollutant emission index and industrial scale index are combined in an orderly manner to form a multi-dimensional fusion feature vector [ ],in Represents the comprehensive index of energy consumption, It represents the comprehensive index of pollutant emissions. Represents an industry scale indicator.

6. The method for dynamic monitoring of carbon emissions in the Yellow River Basin using big data as claimed in claim 5, characterized in that: The analysis of the relative importance of data from different data sources in reflecting the carbon emission characteristics of the location includes the following steps: 1) Get the corresponding data sequence The reference sequence is the carbon emission characteristic value sequence: ； in, Indicated in The carbon emission value obtained at the moment, is the number of elements in the carbon emission characteristic value sequence; The comparison series include energy consumption data series, air pollutant concentration data series and enterprise production scale data series; Energy consumption data series: ； in, Indicated in Energy consumption data corresponding to each moment; Atmospheric pollutant concentration data series: ； in, Indicated in The atmospheric pollutant concentration data monitored at each moment; Enterprise production scale data series: ； in, Indicates that the enterprise Quantitative indicators of production scale corresponding to each moment; is the number of elements in the comparison sequence; 2) Calculate the correlation coefficient Reference sequence and compare sequences The sequences after initialization are , ; Calculate the correlation coefficient between each comparison sequence and the reference sequence at each time : Correlation coefficient The value range is from 0 to 1, and the closer it is to 1, the closer the association is. Indicates the reference sequence after initialization at the kth moment Compare sequence with the i-th The absolute value of the corresponding data difference is used to indicate the degree of difference between the two at that moment; It indicates the minimum absolute value of the difference between the reference sequence and the comparison sequence in all n moments, that is, for a specific comparison sequence , find its difference with the reference sequence The absolute value of the difference at the moment with the smallest difference among n moments; It means that among all i comparison sequences, the minimum value among the absolute values ​​of the minimum differences is found, that is, the minimum value among all comparison sequences and the reference sequence with the minimum difference at each moment, indicating the closest degree between the overall data sequences; , represents the situation where all comparison sequences have the largest difference with the reference sequence at each moment, that is, first find the maximum absolute value of the difference between each comparison sequence and the reference sequence at n moments, and then select the largest value from these maximum values, which represents the maximum degree of difference between the overall data sequences; Among them, the reference sequence is initialized: ； It represents the carbon emission value at the kth moment divided by the carbon emission value at the first moment, converting the carbon emission characteristic value sequence into a relative change sequence based on the first value; Comparison sequence initialization: ， ； It represents the energy consumption, air pollutant concentration or enterprise production scale data at the kth moment divided by the energy consumption, air pollutant concentration or enterprise production scale data at the first moment, and converts the energy consumption characteristic sequence, air pollutant concentration characteristic sequence and enterprise production scale characteristic sequence into a relative change sequence based on the first value; is the resolution coefficient, which ranges from 0 to 1 and is used to adjust the size of the correlation coefficient; 3) Calculate the grey relational degree After obtaining the correlation coefficient at each moment, the grey correlation degree between each comparison sequence and the reference sequence is calculated by the following formula: ， ； Grey relational degree For comparing sequences At n moments with the reference sequence The average value of the correlation coefficient reflects the closeness of the correlation between the comparison sequence and the reference sequence in the entire data interval. The value range is between 0 and 1. The larger the value, the stronger the correlation between the corresponding energy consumption, atmospheric pollutant concentration or enterprise production scale data and the carbon emission characteristic value, and the higher the relative importance in reflecting the carbon emission characteristics.

7. The method for dynamic monitoring of carbon emissions in the Yellow River Basin using big data according to claim 1, characterized in that: The S3 specifically includes the following steps: S3-1, Neural network model selection and structure design: According to the characteristics of the carbon emission data in the Yellow River Basin, a multi-layer perceptron MLP neural network is selected as the basic architecture, in which the number of input layer nodes corresponds to the dimension of the multi-dimensional fusion feature vector, so that it can fully receive the processed multi-source data feature information; the number of hidden layer neurons is set in a decreasing manner, the number of neurons in the first hidden layer is 2 / 3 of the number of input layer nodes, and the subsequent hidden layers decrease in turn, and the hidden layer activation function uses the ReLU function; the number of output layer nodes is determined according to the dimension of the carbon emission information to be predicted, and the output layer activation function uses the linear activation function; S3-2, support vector machine SVM model configuration: The radial basis kernel function RBF is used as the kernel function of the SVM model to map the original input space to the high-dimensional feature space, and the penalty parameters related to the RBF kernel function are and kernel width parameter The cross-validation method is used to optimize the selection. A part of the labeled training data is divided into multiple subsets, and one of the subsets is used as the validation set in turn, and the remaining subsets are used as the training set. and , find the parameter combination that makes the model have the best prediction performance on the validation set; S3-3, model fusion: First, the multi-dimensional fusion feature vector is input into the designed neural network. After nonlinear transformation of each layer of the neural network, high-level abstract features are extracted and an intermediate feature representation is output. Then, this intermediate feature is used as the input of the support vector machine, and the support vector machine performs the final carbon emission information prediction. S3-4, training of carbon emission accounting model: The collected multi-dimensional fusion feature vector data set with carbon emission labeling information is divided into training set, validation set and test set according to a certain ratio. The training set is used for learning and adjusting model parameters, and the validation set is used to evaluate the generalization ability of the model during the training process. The mean square error MSE is selected as the loss function. For the neural network part, the Adam optimization algorithm is used for parameter update training. For the support vector machine part, the parameters are solved according to the training algorithm provided by the selected optimization software library, and the optimal classification or regression hyperplane is found based on the quadratic programming algorithm. According to the set training algorithm and parameters, the training set data is repeatedly input into the integrated carbon emission accounting model, and the weight and bias parameters of the model are continuously adjusted to gradually reduce the loss function value. At the same time, after each round of training, the model is evaluated with the validation set data, and the loss value and mean absolute error (MAE) on the validation set are recorded to observe whether the model is overfitting. S3-5, use the random forest regression algorithm to dynamically adjust the calculation parameters: Through experiments, the prediction error changes of the model on the validation set when different numbers of decision trees are used are compared, and the number of decision trees that can achieve stable model performance is selected; the maximum depth of the decision tree and the minimum number of samples required for node splitting are configured to build a random forest regression model for parameter adjustment; each feature in the multidimensional fusion feature vector is input into the constructed random forest regression model, and the relative importance of each feature to the carbon emission prediction results is analyzed based on the average reduction of node impurity to determine which features are key factors affecting carbon emissions and which are relatively minor factors; at the same time, the random forest regression model mines the implicit influencing factors, and obtains the influencing mechanism by analyzing the branching conditions of the decision tree and the combination of each feature in different decision trees; regularly collect new labeled carbon emission data and the corresponding multidimensional fusion feature vectors, input them into the trained carbon emission accounting model to obtain the prediction results, and calculate the deviation between the predicted value and the true value; according to the key features and influencing factors analyzed by the random forest regression model, the parameters in the carbon emission accounting model related to these factors are determined, and the quantitative relationship between the features and carbon emissions obtained by the random forest regression model is used to adjust these key parameters to optimize the model's prediction ability for new data.

8. A dynamic monitoring system for carbon emissions in the Yellow River basin using big data, characterized in that include: Data collection module: used to build a multi-source data collection network to integrate carbon emission-related data from satellite remote sensing, ground monitoring stations, IoT sensors, and various government affairs and enterprise information systems in the Yellow River Basin; Data preprocessing module: used to first clean the collected multi-source data, then standardize the data, and unify the units and dimensions of data from different sources based on a unified data dictionary; finally, perform data fusion to form a multi-dimensional fusion feature vector that represents the relationship between multi-source data; Carbon emission accounting model construction module: used to construct a carbon emission accounting model by fusing a neural network and a support vector machine, taking the processed multi-dimensional fusion feature vector as input and carbon emission information as output, training the carbon emission accounting model, and using a random forest regression algorithm to dynamically adjust the accounting parameters of the carbon emission accounting model and optimize the model parameters; Dynamic monitoring and early warning module: It is used to use the constructed carbon emission accounting model to conduct real-time dynamic monitoring of carbon emissions in the Yellow River Basin. When it is detected that the carbon emission growth rate in a certain region or industry is too fast or the intensity exceeds the standard, the system automatically triggers an early warning; Data update and model optimization module: used to continuously update various types of data in the data collection network, and use the new data to optimize the multi-dimensional fusion feature vector and carbon emission accounting model, and then use the optimized carbon emission accounting model to monitor the carbon emissions in the Yellow River Basin.

Images