Small-scale greenhouse gas emission prediction method based on pre-training model

Abstract

The invention discloses a small-scale greenhouse gas emission prediction method based on a pre-training model, and belongs to the field of greenhouse gas emission prediction. According to the method, a greenhouse gas monitoring micro station is arranged in a park, data such as greenhouse gas concentration and meteorological factors are collected in real time, and high-precision traceability of a greenhouse gas emission source and accurate accounting of greenhouse gas emission can be achieved in combination with a high-precision meteorological numerical value mode, a backward trajectory tracking algorithm, a receptor model and a diffusion model; the greenhouse gas emission amount is predicted in real time by calling the pre-training model, and the prediction efficiency and accuracy are remarkably improved. Different from a mode that a traditional model depends on a large amount of historical data training, the pre-training model can be finely adjusted on the basis of a small amount of existing data, so that effective learning and prediction are realized. According to the method, a large amount of time and computing resources are saved, the environmental change can be quickly responded, and real-time and accurate emission prediction is provided.

Description

Technical Field

[0001] The invention relates to a small-scale greenhouse gas emission prediction method based on a pre-training model, and belongs to the field of greenhouse gas emission prediction. Background Art

[0002] As the global climate change problem becomes increasingly serious, greenhouse gas emissions have become an important factor affecting the ecological environment and human life. The increase in the concentration of greenhouse gases (such as carbon dioxide, methane, etc.) in the atmosphere has led to a series of problems such as global warming, frequent extreme weather events, and imbalance of ecosystems. These hazards not only affect the natural environment, but also pose a serious threat to human health and economic development. Therefore, accurate monitoring and prediction of greenhouse gas emissions are crucial to formulating effective emission reduction policies and taking corresponding measures.

[0003] At present, the measurement and prediction of greenhouse gas emissions face many uncertainties. First, the sources of greenhouse gas emissions are complex and diverse, covering multiple fields such as industry, transportation, and agriculture. The emission characteristics and intensity of various sources vary greatly, which increases the difficulty of monitoring. Secondly, meteorological conditions (such as wind speed, wind direction, temperature, etc.) have an important impact on the diffusion and transfer of gases, and the changes in these meteorological factors are random, which further exacerbates the complexity of emission prediction.

[0004] Traditional emission calculation methods often rely on historical data and hypothetical models, which makes it difficult to reflect current emission dynamics in real time, resulting in lags and uncertainties in prediction results. In recent years, with the development of data science and artificial intelligence technology, the use of machine learning models to predict greenhouse gas emissions has gradually become a research hotspot. However, this technology still faces multiple challenges. First, the accuracy and frequency of real-time data collection directly affect the performance of the prediction model, and the spatial and temporal coverage of the existing monitoring network still needs to be further improved. Second, existing prediction models usually rely on a large amount of historical data for training, which often faces challenges in small-scale greenhouse gas monitoring within the park: the number of monitoring stations within the park is limited, and there is often a lack of long-term and stable historical emission monitoring data, resulting in a serious lack of available data foundation, affecting the training quality and prediction ability of the model. Summary of the invention

[0005] In order to improve the accuracy and real-time performance of predicting small-scale greenhouse gas emissions in a park, the present invention provides a small-scale greenhouse gas emissions prediction method based on a pre-training model. The technical solution is as follows:

[0006] The greenhouse gas emission prediction method of the present invention comprises:

[0007] Step 1: Set up greenhouse gas monitoring microstations in the park to collect greenhouse gas concentration data and meteorological data in real time;

[0008] Step 2: Based on the real-time greenhouse gas monitoring data, trace the source of greenhouse gas emissions and obtain the distribution data of potential greenhouse gas emission sources;

[0009] Step 3: Based on the greenhouse gas emission source location information obtained in step 2 and the greenhouse gas receptor site monitoring data, the greenhouse gas emissions in the park are calculated;

[0010] Step 4: Based on the time series data of greenhouse gas emissions in the park obtained in step 3, the pre-trained model is called to predict the future greenhouse gas emissions in the park.

[0011] Optionally, step 4 includes:

[0012] Step 41: Obtain the greenhouse gas emission time series data in the park obtained in step 3, obtain the timestamp corresponding to the data, sort the data in chronological order, perform interpolation processing on the missing data, and sort the greenhouse gas emission data into a time series form;

[0013] Step 42: Input the time series data of greenhouse gas emissions into the Transformer pre-training model to obtain the predicted data of greenhouse gas emissions in the future.

[0014] Optionally, the step 1 includes:

[0015] Step 11: Deploy greenhouse gas monitoring microstations in the park according to the grid distribution, keep the microstation spacing consistent and evenly distributed, and each microstation has the ability to monitor greenhouse gas concentrations and is equipped with sensors to measure meteorological factors;

[0016] Step 12: The microstation collects data in real time and transmits the data to the central database via a wireless network;

[0017] Step 13: Calibrate and quality control the data collected by the microstation, remove outliers, and standardize the data format.

[0018] Optionally, the step 2 includes:

[0019] Step 21: Use the meteorological numerical model to assimilate the meteorological factor monitoring data of the micro-stations in the park to obtain high-precision meteorological grid data covering the entire park;

[0020] Step 22: Based on the location of the greenhouse gas receptor site in the park, the high-precision meteorological grid data is used, and a backward trajectory tracking algorithm is adopted to obtain a refined transmission path of the greenhouse gas;

[0021] Step 23: Based on the greenhouse gas monitoring data and refined backward trajectory data, a variety of receptor models are used in combination to obtain the distribution data of potential greenhouse gas emission sources in the park.

[0022] Optionally, step 3 includes:

[0023] Step 31: Based on the location information of the greenhouse gas emission sources obtained in step 2 and the high-precision meteorological data, a small-scale diffusion model based on the Lagrangian method is used to perform forward diffusion simulation calculations on the greenhouse gas emission sources;

[0024] Step 32: Based on the location information of the greenhouse gas receptor site and the forward diffusion simulation results, calculate the greenhouse gas simulation concentration data at the receptor site location;

[0025] Step 33: Obtain monitoring concentration data of greenhouse gas receptor sites, calculate the ratio of monitoring data to simulation data, and adjust the source strength of greenhouse gas emission sources according to the size of the ratio;

[0026] Step 34: Use the adjusted source strength data to perform diffusion simulation again, and repeat the above steps to repeatedly fit and calculate the emission source strength until the set threshold conditions are met, and finally obtain the emission data of greenhouse gas emission sources in the park.

[0027] Optionally, the condition for terminating the fitting in step 34 is:

[0028]

[0029] Among them, R represents the simulated concentration of greenhouse gases at the receptor site, O represents the greenhouse gas monitoring concentration at the greenhouse gas receptor site in the park, and M represents the judgment threshold.

[0030] Optionally, the updating formula of the source strength data in step 34 is:

[0031]

[0032] Where S n represents the simulated source strength after the nth fitting update, R n-1 Represents the simulated greenhouse gas concentration at the receptor site calculated during the n-1th fitting process.

[0033] The present invention provides a greenhouse gas emission prediction device, comprising a memory and a processor;

[0034] The memory is used to store computer programs;

[0035] The processor is used to implement the greenhouse gas emission prediction method as described in any one of the above items when executing the computer program.

[0036] The present invention provides a computer-readable storage medium, characterized in that a computer program is stored on the storage medium, and when the computer program is executed by a processor, the greenhouse gas emission prediction method as described in any one of the above items is implemented.

[0037] The beneficial effects of the present invention are:

[0038] By deploying greenhouse gas monitoring microstations in the park, the present invention can collect data such as greenhouse gas concentrations and meteorological factors in real time. By using these real-time data, combined with high-precision meteorological numerical models, backward trajectory tracking algorithms, receptor models and diffusion models, high-precision tracing of greenhouse gas emission sources and accurate calculation of greenhouse gas emissions can be achieved.

[0039] The present invention significantly improves the efficiency and accuracy of prediction by calling pre-trained models to perform real-time prediction of greenhouse gas emissions. Unlike traditional models that rely on a large amount of historical data for training, pre-trained models can be fine-tuned based on a small amount of existing data to achieve effective learning and prediction. This method not only saves a lot of time and computing resources, but also can quickly respond to environmental changes and provide real-time and accurate emission predictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0041] Figure 1 It is a schematic diagram of the overall process of the small-scale greenhouse gas emissions prediction method based on the pre-training model of the present invention.

[0042] Figure 2 It is a schematic diagram of the layout of the greenhouse gas monitoring microstation in the park of the present invention.

[0043] Figure 3 It is the time series trend line of CO2 emission in the next 12 hours predicted by the model of the present invention, and a comparison chart with actual data.

[0044] Figure 4 It is the time series trend line of CH4 emission in the next 12 hours predicted by the model of the present invention, and the comparison with the actual monitoring data. DETAILED DESCRIPTION

[0045] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0046] Embodiment 1:

[0047] This embodiment provides a method for predicting small-scale greenhouse gas emissions based on a pre-trained model, which specifically includes the following steps:

[0048] Step S1: deploy greenhouse gas monitoring microstations in the park to collect greenhouse gas concentration data and meteorological data in real time and perform quality control on the data. Specifically, the following steps are included:

[0049] Step S1.1: Deploy greenhouse gas monitoring microstations in the park according to the grid distribution, keep the spacing between microstations consistent and evenly distributed to ensure the spatial coverage of the monitoring points and the uniformity of the data.

[0050] Greenhouse gas emission sources have complex and diverse characteristics, covering multiple fields such as industry, transportation, and agriculture. The emission characteristics, emission intensity, and emission cycle of each field are different. Traditional monitoring methods often find it difficult to effectively capture these diverse emission sources. There may be significant differences in the gas release patterns and time variation patterns of different emission sources. For example, industrial emissions are usually released in a concentrated manner during working hours, while traffic emissions are affected by vehicle flow and have a strong timeliness. At the same time, environmental conditions within the park (such as wind speed, temperature, humidity, etc.) will affect the diffusion and transmission of gases. Changes in these environmental factors make greenhouse gas monitoring more difficult. Therefore, it is necessary to reasonably deploy monitoring microstations so that changes in greenhouse gas concentrations can be comprehensively and effectively monitored.

[0051] Evenly distributing greenhouse gas monitoring microstations in the park according to grid distribution can help deal with the complex and diverse emission source characteristics and monitoring difficulties. Evenly distributed microstations can ensure that the greenhouse gas concentration of the entire park is fully monitored, avoid monitoring blind spots, and help eliminate data deviations caused by improper location selection, ensuring that the monitoring results are more representative. At the same time, by reducing the distance between microstations, more fine-grained monitoring data can be obtained, thereby providing richer information for subsequent data analysis and model prediction.

[0052] The establishment and calculation of numerical models often have high requirements for data consistency. High-quality data is usually very helpful in improving the prediction performance of the model. The uniform layout of greenhouse gas monitoring microstations helps to use consistent data formats and distributions in the analysis and model building process, making subsequent models more stable and accurate. High-precision data obtained through multi-point monitoring can better reflect the specific sources and diffusion paths of greenhouse gases, and improve the accuracy and generalization ability of the prediction model.

[0053] In this embodiment, greenhouse gas monitoring micro-stations are evenly distributed in the park according to the standard north-south grid distribution. The spacing between each micro-station is consistent, and the distance between micro-stations is one kilometer, so that the greenhouse gas emissions and propagation in the park can be fully monitored, and the granularity of the monitoring data can be ensured to reflect the diffusion process of greenhouse gases at a small scale. Figure 2 The layout of greenhouse gas monitoring microstations in the park is demonstrated.

[0054] Step S1.2: Each microstation should have the ability to monitor the concentration of greenhouse gases such as CO2 and CH4, and be equipped with sensors to measure meteorological factors such as wind speed, wind direction, temperature, and humidity. The microstation collects data in real time and transmits the data to the central database through a wireless network.

[0055] The emission and spread of greenhouse gases are characterized by dynamic changes, so real-time monitoring data collection is particularly important. Greenhouse gas concentrations are affected by meteorological conditions and human activities, and have obvious short-term fluctuations. Greenhouse gas emissions may increase sharply due to emergencies such as industrial accidents, traffic congestion or natural disasters. Therefore, real-time monitoring and data collection of greenhouse gases can capture these dynamic changes in a timely manner, facilitating rapid response and countermeasures.

[0056] Greenhouse gas monitoring microstations can provide high-precision greenhouse gas (CO2, CH4, etc.) concentration data. Accurate measurements not only help improve the monitoring accuracy of gas concentrations, but also help better identify the specific sources and diffusion paths of greenhouse gases. Real-time monitoring microstations can collect data at a high frequency and promptly reflect the dynamic changes of greenhouse gases in the park. This real-time performance is crucial for capturing short-term fluctuations and emergencies (such as accidental emissions). At the same time, real-time monitoring data provides rich information for establishing more accurate prediction models, enabling the models to better reflect actual emission conditions and improve prediction accuracy.

[0057] In this embodiment, each greenhouse gas monitoring microstation is equipped with a high-precision sensor to monitor the concentration of major greenhouse gases such as CO2 and CH4 in real time. At the same time, the sensor has good sensitivity and accuracy, can work stably in different concentration ranges, cover the required concentration range, adapt to the characteristics of different emission sources in the park, and meet the monitoring needs.

[0058] At the same time, each monitoring microstation is equipped with a variety of meteorological factor sensors, including wind speed and direction sensors, temperature sensors, humidity sensors and air pressure sensors. This enables real-time acquisition of key meteorological information that affects the diffusion of greenhouse gases, which not only helps calculate the movement path and concentration changes of greenhouse gases in the air, but also provides key observation data for establishing a meteorological grid data field covering the park.

[0059] Each micro-station sets a unified data collection frequency (monitoring once every hour) to ensure that changes in greenhouse gas concentrations and meteorological factors can be obtained in real time. Each micro-station has local data recording capabilities to ensure that data can continue to be collected when the network is interrupted to prevent data loss. At the same time, the micro-station transmits the collected data to the central database in real time through wireless networks (such as Wi-Fi, LoRa, Zigbee, etc.). The transmitted data uses a unified format to facilitate the reception of the central database and subsequent data processing. The standardized data format helps to improve the efficiency of data integration and analysis.

[0060] Step S1.3: Calibrate and quality control the data collected by the microstation, remove outliers, and standardize the data format.

[0061] The standardization and accuracy of observation data are important guarantees for the model's predictive ability. Data accuracy directly affects the input quality of the model, thereby affecting the reliability of the prediction results. Using high-quality and accurate data for model training and calculation can improve the model's adaptability and generalization ability to data, and enhance the credibility of the prediction results. At the same time, data standardization ensures the consistency of data in time, unit and format. Standardization makes subsequent data processing, analysis and model building simpler and more efficient. A unified data format can reduce errors and confusion caused by inconsistent data, and reduce calculation deviations caused by inconsistent data formats, thereby improving the efficiency of model training.

[0062] In this embodiment, in order to ensure the accuracy of the data, the greenhouse gas sensor needs to be calibrated regularly to eliminate measurement errors caused by instrument drift or environmental changes. Similarly, meteorological sensors (such as temperature, humidity, wind speed and wind direction sensors) also need to be calibrated regularly to ensure that the acquired data reflects the actual environmental conditions.

[0063] The quality control process of the collected real-time data includes data integrity check and outlier monitoring: data integrity check must ensure that the data of each micro-station is continuously recorded, and missing data must be marked and processed by linear interpolation; statistical methods (such as standard deviation method, box plot method, etc.) or machine learning (isolation forest, etc.) are used to identify and eliminate outliers in the data to ensure the rationality and accuracy of the data.

[0064] At the same time, real-time monitoring data needs to be standardized, including data format standardization and data labeling: the data collected by different micro-stations are standardized to ensure the consistency of the data in terms of timestamp, unit and data type. This process can simplify subsequent data analysis and integration; necessary metadata (such as collection time, micro-station location, etc.) is added to each data record to facilitate subsequent data calls and query analysis.

[0065] Store calibrated and quality-controlled data in a central database to ensure data security and easy access. Back up data regularly to prevent data loss due to system failure or other reasons, and ensure data durability and reliability.

[0066] Step S2: Based on the real-time greenhouse gas monitoring data, the greenhouse gas emission sources are traced and calculated to obtain the distribution data of potential greenhouse gas emission sources. Specifically, the following steps are included:

[0067] Step S2.1: Use the meteorological numerical model to assimilate the meteorological factor monitoring data of the micro-stations in the park to obtain high-precision meteorological grid data covering the entire park.

[0068] Obtaining accurate and high-resolution three-dimensional meteorological grid data is essential for high-precision tracing of small-scale greenhouse gases. High-resolution three-dimensional meteorological grid data can provide more detailed wind speed and direction information, which helps to accurately simulate the diffusion path of greenhouse gases, thereby revealing how gases propagate under different terrain and meteorological conditions. At the same time, real-time three-dimensional meteorological data can reflect the rapid changes in meteorological conditions and enhance the model's ability to respond to instantaneous emission events. In the park, the environmental conditions are very complex due to the presence of various buildings. The three-dimensional meteorological data provides a comprehensive perspective at different heights and spaces, making the location and intensity estimation of emission sources more accurate. In addition, accurate meteorological conditions can also reduce errors in the tracing model and improve the accuracy of greenhouse gas emission source identification.

[0069] High-precision meteorological numerical models are currently a common method for obtaining high-resolution three-dimensional meteorological grid data. Meteorological numerical models read the current meteorological data, perform differential numerical operations based on dynamic and thermal meteorological equations, and combine various parameterization schemes to simulate physical and chemical processes in the atmosphere. This method can effectively predict the changes in various meteorological factors in the future and generate high-resolution three-dimensional meteorological grid data over the park.

[0070] In this process, accurate and real-time meteorological factor monitoring data are crucial to meteorological numerical simulation. They provide the real data basis required for model initialization and updating. By combining the monitoring data of micro-stations in the park with numerical models through the data assimilation system, the accuracy of meteorological numerical simulation can be significantly improved. This assimilation process can help the meteorological numerical simulation process more accurately reflect the real-time meteorological changes near the ground within the park, making the output three-dimensional meteorological grid data closer to the actual changes in meteorological conditions.

[0071] In this embodiment, the WRF-ARW numerical forecast model is used, with global meteorological reanalysis data as the initial data, and a multi-layer nested grid is used. The innermost horizontal resolution reaches 100m×100m, covering the entire park and its surrounding areas, providing high-precision meteorological data over the park. At the same time, real-time observation data from ground monitoring stations in the park are collected, including wind speed, wind direction, temperature, relative humidity, and air pressure. Based on the 3DVAR three-dimensional variational data assimilation model, a meteorological data assimilation system is established to cyclically assimilate the ground meteorological data of the monitoring stations in the park, and integrate the continuously updated observation data into the meteorological model to generate high-precision real-time meteorological data near the ground in the park. The model automatically makes a forecast once a day, and each forecast obtains the weather data for 24 hours on the day, and archives the historical data.

[0072] Step S2.2: Based on the location of greenhouse gas receptor sites in the park, high-precision meteorological grid data is used and a backward trajectory tracking algorithm is adopted to obtain the refined transmission path of greenhouse gases.

[0073] Greenhouse gas receptor sites refer to monitoring sites in the greenhouse gas transmission path. The backward trajectory tracking algorithm simulates the reverse propagation process of gases from the receptor site to the emission source, combined with meteorological factors, to identify potential emission sources and propagation paths of gases. Using the location of the greenhouse gas receptor site as the starting point of the trajectory, the backward trajectory tracking is started. The model will perform backward integration of the airflow at the receptor site based on meteorological information such as wind direction and speed, track the possible source path of the airflow, and obtain refined backward trajectory data of greenhouse gas emissions. In this process, accurate and high-precision meteorological grid data is very important to the accuracy of the backward trajectory tracking algorithm.

[0074] In this embodiment, the specific monitoring data of the greenhouse gas monitoring station is first analyzed, which includes the concentration and composition of greenhouse gases at different time points. By carefully studying these data, it is possible to determine the time periods of significance. For example, if there is a significant change or abnormal value in the greenhouse gas concentration during a specific time period, then this time period will be selected as the key analysis period.

[0075] Next, high-precision real-time meteorological data for the required area and time period are obtained based on the location information of the monitoring station. According to the latitude and longitude range of the study area, combined with the near-surface and high-altitude meteorological data obtained in step S2.1, high-precision meteorological data for a specific park are extracted, and the time period corresponds to the previously determined backward trajectory analysis period. Meteorological data includes evenly distributed wind speed, wind direction, temperature, and air pressure.

[0076] Based on the above data, a high-precision backward trajectory tracking system was established using the HYSPLIT (The Hybrid Single-Particle Lagrangian Integrated Trajectory model) model. The HYSPLIT model is a computational model used for airflow trajectory simulation and air quality research. It is based on the Lagrangian method and combines the physical and chemical mechanisms of atmospheric transport and diffusion processes. It takes into account the impact of factors such as wind fields, topography, and meteorological conditions on airflow movement. Using the location of the monitoring site as the starting point of the trajectory, the backward trajectory tracking is started. The model will perform backward integration of greenhouse gases at the receptor site based on meteorological information, track the possible source path of the airflow, and obtain refined backward trajectory data.

[0077] Step S2.3: Based on the greenhouse gas monitoring data and the refined backward trajectory data, a variety of receptor models are used in combination to obtain the distribution data of potential greenhouse gas emission sources in the park.

[0078] The receptor model is a mathematical model used to identify and quantify air pollution emission sources that affect the location of the receptor site. It plays an important role in environmental science, meteorology and air quality research, especially in greenhouse gas monitoring and tracing. By analyzing the gas concentration data at the monitoring site and the gas transmission path, the receptor model can infer the location and intensity of potential gas emission sources and identify key emission sources. At the same time, the model can calculate the contribution of different potential emission sources to the gas concentration at the receptor site, helping to determine the main emission sources and their impact.

[0079] Previous studies have shown that different receptor model tracing algorithms have their own advantages and disadvantages in emission source identification. By comprehensively using a variety of different receptor model algorithms, we can make up for each other's shortcomings, improve the accuracy and stability of the tracing results, reduce the errors caused by a single algorithm, make the tracing results more robust and reliable, maintain high accuracy and robustness in complex atmospheric environments, and be suitable for various greenhouse gas tracing needs. It improves the positioning capability and tracing accuracy of greenhouse gas emission sources within a small scale, and enhances the accuracy and reliability of the tracing results.

[0080] In this embodiment, based on the refined backward trajectory data calculated in step S2.2, the time series of greenhouse gas concentration data of the receptor site in the corresponding period is obtained. According to the start time of the backward trajectory, the greenhouse gas concentration data of the receptor site is selected and presented in the form of a time series. In this process, the data is preprocessed, including quality control and format standardization, to ensure that it corresponds one-to-one with the trajectory data, so as to facilitate subsequent calculation processing.

[0081] Subsequently, five receptor model algorithms, namely PSCF (Potential Source Contribution Function), CWT (Concentration Weighted Trajectory), CFA (Concentration Field Analysis), RTWC (Residence Time Weighted Concentration) and SQTBA (Simplified QuantitativeTransport Bias Analysis), were applied to establish a high-precision greenhouse gas source tracing calculation model to obtain the distribution data of potential greenhouse gas emission sources.

[0082] Finally, the Max-Min normalization algorithm was used to normalize the distribution data of five potential greenhouse gas emission sources, and then the data were averaged to finally generate comprehensive greenhouse gas potential emission source distribution data.

[0083] Step S3: Based on the location information of greenhouse gas emission sources and the monitoring data of greenhouse gas receptor sites obtained in step S2, the greenhouse gas emissions in the park are calculated. Specifically, the following steps are included:

[0084] Step S3.1: Based on the location information of the greenhouse gas emission sources obtained in step S2 and the high-precision meteorological data, a small-scale diffusion model based on the Lagrangian method is used to perform forward diffusion simulation calculations on the greenhouse gas emission sources.

[0085] Accurate forward diffusion simulation of greenhouse gas transport is the basis for greenhouse gas emission accounting. The Lagrangian diffusion model is a commonly used numerical model designed to simulate the diffusion, transport and transformation of atmospheric pollutants. The model predicts the distribution of pollutants in space and time by tracking the movement paths of air masses or particles.

[0086] Compared with the Euler method, the Lagrangian method is more suitable for dealing with complex flow fields and uneven distribution of pollution sources, especially in small-scale environments. The core concept of the Lagrangian method is to regard pollutants in the atmosphere as consisting of a large number of tiny particles or air masses, which move, diffuse and settle in the atmospheric motion field over time. By simulating the trajectories of these particles, the concentration distribution of pollutants at different times and locations can be derived.

[0087] Compared with the Gaussian diffusion model, the particle diffusion model based on the Lagrangian method makes full use of the advantages of high-precision meteorological data and can calculate the movement of particles at various times and locations, thereby more accurately simulating the diffusion trajectory of pollutants in the atmosphere. Common diffusion models based on the Lagrangian method include CALPUFF, FLEXPART, and HYSPLIT.

[0088] In this embodiment, the HYSPLIT model is used to simulate the diffusion of greenhouse gases. The HYSPLIT model integrates a three-dimensional particle diffusion module based on the Lagrangian method. Its basic principle is to release a large number of particles at a specified emission source location to simulate the emission process of pollutants. By calling high-precision meteorological data, the model can calculate the precise motion trajectory of each particle and superimpose the atmospheric turbulent motion component to simulate the diffusion process of pollutants in the atmosphere.

[0089] For turbulent processes in the atmosphere, the model can automatically calculate the atmospheric stratification conditions at that time based on the input meteorological data, thereby obtaining the turbulent mixing coefficients in the horizontal and vertical directions, and on this basis calculate and simulate the atmospheric turbulent motion. This process ensures the accuracy and reliability of the simulation results, providing strong support for the calculation of greenhouse gas emissions.

[0090] The high-precision meteorological grid data and greenhouse gas emission source location information obtained in step S2 are input into the HYSPLIT diffusion model, and these data will be used as the initial conditions of the model. Set the concentration S of the initial emission source to represent the initial concentration of greenhouse gases at the emission source. Start the HYSPLIT diffusion model to simulate the diffusion process of greenhouse gases from the emission source to the surrounding area. The model will track the movement trajectory of gas particles based on meteorological conditions and terrain characteristics, and calculate the concentration distribution of gases at different time and space locations.

[0091] Step S3.2: Based on the location information of the greenhouse gas receptor site and the forward diffusion simulation results, calculate the greenhouse gas simulation concentration data at the receptor site location.

[0092] It is an important step to obtain the precise location information of all greenhouse gas receptor sites in the park, including the latitude and longitude coordinates and altitude (usually relative to the ground) of each receptor site. This location information is input into the model, and the location of the receptor site is matched with the simulated concentration distribution obtained in step S3.1 to determine the model grid or coordinate point corresponding to each receptor site. Using the simulation results of the diffusion model, the simulated greenhouse gas concentration value at the corresponding position of the receptor site is calculated, which represents the concentration of greenhouse gases diffused from the emission source to the receptor site under specific time and meteorological conditions.

[0093] When calculating the simulated concentration values ​​of greenhouse gas receptor sites, it is usually necessary to use an interpolation algorithm to estimate the concentration at the site location. This is because the location of the receptor site may not be completely located on the nodes of the simulation grid, so the concentration at the site location needs to be inferred from the concentration values ​​of the neighboring grid points through an interpolation algorithm. Commonly used interpolation algorithms include bilinear interpolation, kriging difference, and nearest neighbor interpolation. According to the specific data characteristics, select an appropriate interpolation algorithm to ensure that the calculation results of the simulated concentration values ​​are accurate and reliable.

[0094] In this embodiment, based on the greenhouse gas receptor site and its location information, and the simulated concentration distribution obtained in step S3.1, the simulated concentration value of the greenhouse gas at the corresponding position of the receptor site is calculated, which is recorded as R. Considering the data characteristics in this embodiment, the bilinear interpolation method is selected to calculate the simulated concentration value of the greenhouse gas receptor site.

[0095] The bilinear interpolation method is applicable to two-dimensional grids. It linearly interpolates four grid points adjacent to the receptor site, first in one direction (usually the x direction) and then in the other direction (y direction) to obtain the concentration value of the receptor site. The advantage of this algorithm is that it is easy to calculate and has high accuracy in high-resolution grids. Since the horizontal resolution of the grid in this embodiment reaches 100m×100m, the bilinear interpolation method can meet the requirements for interpolation accuracy and effectively save computing resources.

[0096] Step S3.3: Obtain monitoring concentration data of greenhouse gas receptor sites, calculate the ratio of monitoring data to simulation data, and adjust the source strength of greenhouse gas emission sources according to the size of the ratio.

[0097] According to the simulated concentration data obtained in step S3.2, the actual monitoring concentration of the receptor site at the corresponding time is obtained, and a time consistency check is performed to ensure that the monitoring data and the simulated data match in time and space. Specifically, the monitoring data and the simulated data need to be compared at the same time and place. The consistency of the time dimension can be achieved through time series interpolation or aggregation operations.

[0098] After confirming the temporal and spatial consistency, the ratio of the simulated concentration to the monitored concentration is calculated for each receptor site. This ratio reflects the difference between the model simulation results and the actual monitoring results, and indicates the proportional relationship between the model simulated concentration and the actual monitored concentration. It can be reasonably inferred that this ratio has the same changing trend as the ratio of the actual source strength of the greenhouse gas emission source to the simulated source strength in the model. If the ratio is greater than 1, it means that the initial emission source strength in the model is overestimated and the source strength needs to be reduced accordingly; if the ratio is less than 1, it means that the initial emission source strength is underestimated and the source strength needs to be increased. The initial source strength is corrected according to the size of the ratio to obtain a new emission source strength. Through this adjustment, the updated emission source strength data will be closer to the actual source strength data. This process is an iterative correction based on the "model-observation" comparison, aiming to gradually optimize the fitting results of the source strength.

[0099] In this embodiment, the receptor monitoring station records the concentration data of greenhouse gases once an hour. After calibration and quality control, the data is uniformly stored in a central database. After completing the forward diffusion simulation of greenhouse gases, the monitored concentration O at the corresponding moment of the simulated concentration R is obtained, and the ratio of the simulated concentration to the monitored concentration Q=R / O is calculated. Subsequently, the initial emission source strength S is adjusted according to the ratio Q to obtain a new emission source strength S=S / Q.

[0100] Step S3.4: Use the adjusted source strength data to perform diffusion simulation again, and repeat the above steps to repeatedly fit and calculate the emission source strength until the set threshold conditions are met, and finally obtain the emission data of greenhouse gas emission sources in the park.

[0101] Based on the adjusted source strength data, the diffusion model is rerun to evaluate the greenhouse gas concentration at the receptor site and provide a basis for further iterative corrections. After the simulation is completed, the new simulated concentration data for each receptor site are calculated, which should be closer to the actual monitored concentration than the initial simulated concentration. Subsequently, the threshold condition is judged: if the condition is met, the fitting calculation process is exited; if not, the next fitting iteration is continued.

[0102] In each iteration, the source strength of the emission source of this iteration is updated based on the ratio of the simulated concentration of the previous iteration to the monitored concentration. Then the diffusion model is started again to calculate the updated simulated concentration of the receptor site and its ratio to the monitored concentration for threshold judgment and the next iteration calculation. By continuously adjusting the source strength of the emission source, the gap between the model simulation value and the actual monitored value is gradually narrowed, thereby improving the accuracy of the model until the set threshold conditions are met, and finally the emission data of the greenhouse gas emission sources in the park are obtained.

[0103] The convergence threshold is the termination condition of the iterative fitting process, which is used to judge the consistency between the model calculation results and the actual monitoring data. Due to the accuracy of measuring instruments, model calculation errors, and fluctuations in environmental conditions, even highly accurate models are difficult to fully match the actual observation data. Setting a small but non-zero threshold can accommodate these uncertainties. At the same time, diffusion models and meteorological models usually make certain simplifications during the simulation process, especially in small-scale simulations, and these simplifications may cause the output to differ from the actual situation. Therefore, overly strict convergence requirements may lead to too many iterations, increasing computing time and resource consumption.

[0104] In this embodiment, the threshold M for judging whether the iterative process converges is set to 0.01, indicating that the allowed error range of the model result is 1% of the actual monitored concentration. This threshold can ensure that the error between the model fitting result and the actual monitored concentration is within a reasonable range, while balancing the requirements of calculation accuracy and monitoring error, model simplification, and computing resources.

[0105] In each fitting iteration, compare |1-Q| with the convergence threshold M to decide whether to continue fitting: if |1-Q|>M, there is still a large gap between the model and the actual monitoring data, and it is necessary to continue to adjust the emission source intensity and re-run the diffusion model; if |1-Q|≤M, it means that the model has achieved the expected accuracy, and the difference between the simulated concentration and the actual monitoring data is within an acceptable range. At this time, you can exit the fitting process and output the final emission source intensity S.

[0106] Step S4: Based on the greenhouse gas emission time series data in the park obtained in step S3, the pre-trained model is called to predict the future greenhouse gas emissions in the park. Specifically, the following steps are included:

[0107] Step S4.1: Obtain the emission data of greenhouse gas emission sources in the park obtained in step S3, and obtain the timestamp corresponding to the data, organize the data in chronological order, interpolate the missing data, ensure the timeliness and integrity of the data, and finally organize the greenhouse gas emission data into a time series form.

[0108] A time series is a sequence of data points arranged in chronological order. Usually, the time interval of a time series is a constant value, so the time series can be analyzed and processed as discrete time data. In order to carry out subsequent model predictions, it is necessary to organize the greenhouse gas emissions data into a time series form, obtain the timestamps corresponding to the data, organize the data in chronological order, and interpolate the missing data to ensure the timeliness and integrity of the data.

[0109] In this embodiment, first, the greenhouse gas emission data and the corresponding timestamps in the park need to be extracted from step S3. These data are usually stored in a table format, including the corresponding relationship between emissions and time. Next, the data is sorted in chronological order to ensure the continuity and integrity of the data sequence.

[0110] For missing data, interpolation is required to fill the gaps. Common interpolation methods include linear interpolation, spline interpolation or other appropriate time series interpolation methods. After interpolation, the timeliness of the data is ensured so that the emission data can reflect the actual trend of change.

[0111] Finally, the processed greenhouse gas emissions data is organized into a time series format, usually stored in the form of an array or data frame, so that it can be subsequently input into the pre-trained model. This process ensures the quality and consistency of the data, laying the foundation for the subsequent prediction step.

[0112] Step S4.2: Select a suitable pre-trained model, such as a Transformer-based deep learning model for processing time series data, call a suitable deep learning framework (such as TensorFlow, PyTorch, etc.) to load the pre-trained model, adjust the model parameters, and establish a greenhouse gas emissions prediction system.

[0113] Transformer-based pre-training models have made significant progress in time series prediction tasks in recent years, especially in zero-shot prediction, showing their unique advantages. Transformer uses a self-attention mechanism to enable the model to focus on information at different positions in the input sequence, which is especially important for capturing long-term and short-term dependencies. Compared with traditional RNN or LSTM, Transformer can better handle long sequence data. Transformer-based models usually adopt pre-training strategies, such as using a large amount of historical data for training. In this way, the model can learn rich feature representations, thereby achieving zero-shot prediction on specific tasks, that is, it can still effectively predict new data in the absence of relevant samples.

[0114] Zero-shot prediction refers to predicting new time series data when the model has never seen it before. This ability usually relies on the data and pattern recognition ability obtained by the model during the pre-training stage. Transformer-based models can achieve zero-shot prediction in the following ways: Knowledge transfer: In the pre-training stage, the time series features and patterns learned by the model can be transferred to unseen data, and the generality of the model can be used for prediction; Contextual understanding: Transformer can understand the relationship between different time points in the time series through the attention mechanism. Even if the new data is not exactly the same as the training data, the model can still make reasonable inferences through contextual information; Diversity and flexibility: The Transformer architecture supports the input of multiple types of features (for example, seasonality, trend, etc.), which can be captured through training. Even in the absence of direct samples, predictions can be made through existing knowledge.

[0115] Zero-shot prediction based on the Transformer pre-trained model is widely used in a variety of time series forecasting tasks, such as environmental monitoring (such as greenhouse gas emission forecasting), financial data analysis (such as stock market price forecasting) and demand forecasting (such as energy demand). By making full use of the powerful capabilities of the Transformer framework, efficient time series forecasting can be achieved in different scenarios, especially in the case of data scarcity.

[0116] In this embodiment, a time series prediction model based on Transformer pre-training is selected, the PyTorch deep learning framework is called, and the pre-trained model is loaded using the API of the deep learning framework. According to the specific task requirements and data characteristics, the hyperparameters of the model are adjusted, such as learning rate, batch size, training rounds, etc. The selection of these parameters directly affects the training effect and prediction accuracy of the model. Combined with the input data and the adjusted model parameters, a greenhouse gas emissions prediction system is established. If the pre-trained model needs further training, historical greenhouse gas emissions data can be used for fine-tuning to improve the adaptability and prediction ability of the model.

[0117] Step S4.3: Input the greenhouse gas emission time series data into the pre-trained model to obtain the greenhouse gas emission forecast data for the future time, and visualize the forecast results.

[0118] Input the processed time series data in step S4.1 into the loaded Transformer pre-trained model. The model performs inference based on the patterns learned during its training process to predict future greenhouse gas emissions. The model will output predicted values ​​of greenhouse gas emissions at multiple future time points based on the set parameters. Depending on the model settings, these predicted values ​​include means, confidence intervals, or other statistical information, reflecting different possibilities for future emission trends. In order to better understand and display the prediction results, use data visualization tools (such as Matplotlib, Seaborn, etc.) to generate time series trend lines, intuitively display the comparison between historical data and prediction results, and combine the visualization results to analyze the changing trends and potential impacts of greenhouse gas emissions.

[0119] In this embodiment, the greenhouse gas emission time series data of the past 24 hours is input into the pre-training model to obtain the greenhouse gas emission forecast data for the next 12 hours, and the forecast results are visualized. Figure 3 The time series trend line of CO2 emissions predicted by the model for the next 12 hours is shown, as well as the comparison with the actual monitoring data. Figure 4 The time series trend line of CH4 emissions predicted by the model for the next 12 hours is shown, as well as a comparison with the actual monitoring data.

[0120] Some steps in the embodiments of the present invention may be implemented using software, and the corresponding software program may be stored in a readable storage medium, such as a CD or a hard disk.

[0121] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting greenhouse gas emissions, characterized in that: The method comprises: Step 1: Set up greenhouse gas monitoring microstations in the park to collect greenhouse gas concentration data and meteorological data in real time; Step 2: Based on the real-time greenhouse gas monitoring data, trace the source of greenhouse gas emissions and obtain the distribution data of potential greenhouse gas emission sources; Step 3: Based on the greenhouse gas emission source location information obtained in step 2 and the greenhouse gas receptor site monitoring data, the greenhouse gas emissions in the park are calculated; Step 4: Based on the time series data of greenhouse gas emissions in the park obtained in step 3, the pre-trained model is called to predict the future greenhouse gas emissions in the park.

2. The greenhouse gas emission prediction method according to claim 1, characterized in that: The step 4 comprises: Step 41: Obtain the greenhouse gas emission time series data in the park obtained in step 3, obtain the timestamp corresponding to the data, sort the data in chronological order, perform interpolation processing on the missing data, and sort the greenhouse gas emission data into a time series form; Step 42: Input the time series data of greenhouse gas emissions into the Transformer pre-training model to obtain the predicted data of greenhouse gas emissions in the future.

3. The greenhouse gas emission prediction method according to claim 1, characterized in that: The step 1 comprises: Step 11: Deploy greenhouse gas monitoring microstations in the park according to the grid distribution, keep the microstation spacing consistent and evenly distributed, and each microstation has the ability to monitor greenhouse gas concentrations and is equipped with sensors to measure meteorological factors; Step 12: The microstation collects data in real time and transmits the data to the central database via a wireless network; Step 13: Calibrate and quality control the data collected by the microstation, remove outliers, and standardize the data format.

4. The greenhouse gas emission prediction method according to claim 1, characterized in that: The step 2 comprises: Step 21: Use the meteorological numerical model to assimilate the meteorological factor monitoring data of the micro-stations in the park to obtain high-precision meteorological grid data covering the entire park; Step 22: Based on the location of the greenhouse gas receptor site in the park, the high-precision meteorological grid data is used, and a backward trajectory tracking algorithm is adopted to obtain a refined transmission path of the greenhouse gas; Step 23: Based on the greenhouse gas monitoring data and refined backward trajectory data, a variety of receptor models are used in combination to obtain the distribution data of potential greenhouse gas emission sources in the park.

5. The greenhouse gas emission prediction method according to claim 1, characterized in that: The step 3 comprises: Step 31: Based on the location information of the greenhouse gas emission sources obtained in step 2 and the high-precision meteorological data, a small-scale diffusion model based on the Lagrangian method is used to perform forward diffusion simulation calculations on the greenhouse gas emission sources; Step 32: Based on the location information of the greenhouse gas receptor site and the forward diffusion simulation results, calculate the greenhouse gas simulation concentration data at the receptor site location; Step 33: Obtain monitoring concentration data of greenhouse gas receptor sites, calculate the ratio of monitoring data to simulation data, and adjust the source strength of greenhouse gas emission sources according to the size of the ratio; Step 34: Use the adjusted source strength data to perform diffusion simulation again, and repeat the above steps to repeatedly fit and calculate the emission source strength until the set threshold conditions are met, and finally obtain the emission data of greenhouse gas emission sources in the park.

6. The greenhouse gas emission prediction method according to claim 5, characterized in that: The condition for terminating the fitting in step 34 is: Among them, R represents the simulated concentration of greenhouse gases at the receptor site, O represents the greenhouse gas monitoring concentration at the greenhouse gas receptor site in the park, and M represents the judgment threshold.

7. The greenhouse gas emission prediction method according to claim 6, characterized in that: The updating formula of the source strength data in step 34 is: Where S n represents the simulated source strength after the nth fitting update, R n-1 Represents the simulated greenhouse gas concentration at the receptor site calculated during the n-1th fitting process.

8. A greenhouse gas emission prediction device, characterized in that: including memory and processor; The memory is used to store computer programs; The processor is used to implement the greenhouse gas emission prediction method according to any one of claims 1 to 7 when executing the computer program.

9. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the greenhouse gas emission prediction method according to any one of claims 1 to 7 is implemented.

Images