Method for constructing prediction model, method for evaluating sensitivity and related products

Abstract

The invention discloses a method for constructing a prediction model, a method for evaluating sensitivity and a related product, and the method for constructing the prediction model comprises the steps: obtaining a multi-source fusion data set which comprises the emission amount of emissions, a time parameter, a weather parameter, an activity level parameter of an emission source and the concentration of a target object; and training a preset model by using the multi-source fusion data set to obtain a prediction model. The constructed prediction model comprehensively considers emissions and other influence factors, the concentration change of the target object can be accurately captured, and the prediction error is greatly reduced.

Description

Technical Field

[0001] This disclosure generally relates to the field of computer technology. More specifically, this disclosure relates to a method for constructing a prediction model, a method for evaluating the sensitivity of emissions to the concentration of a target, an electronic device, and a computer-readable storage medium. Background Art

[0002] Under the dual effects of the rapid development of urbanization and the sharp increase in the number of motor vehicles, motor vehicle emissions have become a major "culprit" in urban air pollution. Among them, nitrogen oxides (NO x ) as the main pollutants in motor vehicle exhaust not only directly threaten human health but also convert into nitrogen dioxide (NO 2 ), profoundly affecting air quality and being a key precursor for the formation of ozone and fine particulate matter. Therefore, accurately evaluating the sensitivity of motor vehicle emission pollutants to the concentration of atmospheric pollutants is extremely important, which will provide a basis for formulating scientific air quality management strategies and thus effectively guarantee air quality. However, there is currently little research on evaluating the sensitivity of NO x emission to NO 2 concentration, and most of the research mainly focuses on the prediction of NO 2 concentration.

[0003] In view of this, there is an urgent need to provide a method for constructing a prediction model, a method for evaluating the sensitivity of emissions to the concentration of a target, an electronic device, and a computer-readable storage medium, so as to accurately evaluate the sensitivity of emissions such as pollutants released by motor vehicles to the concentration of atmospheric pollutants. Summary of the Invention

[0004] To at least solve one or more of the above-mentioned technical problems, this disclosure proposes a method for constructing a prediction model, a method for evaluating the sensitivity of emissions to the concentration of a target, an electronic device, and a computer-readable storage medium in multiple aspects.

[0005] In a first aspect, this disclosure provides a method for constructing a prediction model, including: obtaining a multi-source fusion data set, where the multi-source fusion data set includes emissions, time parameters, weather parameters, emission source activity level parameters, and target concentration; using the multi-source fusion data set to train a preset model to obtain a prediction model.

[0006] In some embodiments, training a preset model using the multi-source fusion dataset to obtain a prediction model includes: dividing the multi-source fusion dataset into a training set and a test set; based on the training set, training the preset model using the k-fold cross-validation method to obtain target hyperparameters; setting the target hyperparameters as the hyperparameters of the preset model; and training the preset model using the training set and the test set to obtain the prediction model, wherein the hyperparameters of the preset model remain constant.

[0007] In some embodiments, training the preset model using the training set and the test set to obtain the prediction model further includes: training the preset model using the training set and the test set; determining a first performance value of the preset model for the training set and a second performance value for the test set; if at least one of the first performance value and the second performance value does not meet a preset usage standard, jumping to execute the step of training the preset model using the k-fold cross-validation method based on the training set to obtain target hyperparameters, or continuing to execute the step of training the preset model using the training set and the test set; if both the first performance value and the second performance value meet the preset usage standard, determining the trained preset model as the prediction model.

[0008] In some embodiments, the preset model includes a random forest model.

[0009] In some embodiments, the emissions include NO x , and the target substance includes NO 2 .

[0010] In a second aspect, the present disclosure provides a method for evaluating the sensitivity of emissions to the concentration of a target substance using the prediction model described in the first aspect and any of its embodiments above, including: constructing a target multi-source fusion dataset of emissions under different emission scenarios, wherein the target multi-source fusion dataset includes target time parameters, target weather parameters, target emission source activity level parameters, and target emissions, and the target emissions in the target multi-source fusion dataset are constant; inputting the target multi-source fusion dataset into the prediction model to obtain the predicted target substance concentration of emissions under different emission scenarios;

[0011] Determining the sensitivity of emissions to the concentration of the target substance under different emission scenarios according to the target emissions and the predicted target substance concentration under different emission scenarios.

[0012] In some embodiments, determining the sensitivity of emissions to the concentration of the target substance under different emission scenarios based on the target emissions and the predicted target substance concentration under different emission scenarios includes: constructing a linear regression equation for different emission scenarios based on the target emissions and the predicted target substance concentration under different emission scenarios; and determining the sensitivity of emissions to the concentration of the target substance under different emission scenarios according to the slope of the linear regression equation.

[0013] In some embodiments, constructing a target multi-source fusion dataset of emissions under different emission scenarios includes: randomly selecting a first preset number of first target multi-source fusion data subsets from the multi-source fusion dataset of a target historical period in a target area to be measured; determining the target emissions according to the maximum emissions in the first target multi-source fusion data subsets; keeping the target emissions constant, and determining the target time parameter, the target weather parameter, and the target emission source activity level parameter based on the first target multi-source data subset.

[0014] In some embodiments, keeping the target emissions constant and determining the target time parameter, the target weather parameter, and the target emission source activity level parameter based on the first target multi-source data subset includes: randomly selecting a second preset number of second target multi-source fusion data subsets from the first target multi-source fusion data subsets; and using the mean or median of the time parameter, the weather parameter, and the emission source activity level parameter in the second target multi-source fusion data subsets as the target time parameter, the target weather parameter, and the target emission source activity level parameter.

[0015] In some embodiments, the emissions include NO x , and the target substance includes NO 2 .

[0016] In a third aspect, the present disclosure provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the methods for constructing a prediction model described in the first aspect and any of its embodiments and the methods for evaluating the sensitivity of emissions to the concentration of a target substance described in the second aspect and any of its embodiments are implemented.

[0017] In a fourth aspect, the present disclosure provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the methods for constructing a prediction model described in the first aspect and any of its embodiments and the methods for evaluating the sensitivity of emissions to the concentration of a target substance described in the second aspect and any of its embodiments are implemented.

[0018] Through the method provided above for constructing a prediction model that can be used to evaluate the sensitivity of emissions to the concentration of a target, embodiments of the present disclosure obtain a multi-source fusion dataset including emissions, time parameters, weather parameters, emission source activity level parameters, and the concentration of the target, so that the multi-source fusion dataset covers various factors affecting the concentration of the target. Then, the preset model is trained using the multi-source fusion dataset to obtain a prediction model. Based on this, the prediction model constructed by the present disclosure comprehensively considers emissions and other influencing factors, can accurately capture changes in the concentration of the target, and significantly reduce the prediction error. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features, and advantages of the exemplary embodiments of the present disclosure will become readily understood. In the drawings, several embodiments of the present disclosure are shown by way of example and not limitation, and like or corresponding reference numerals represent like or corresponding parts, wherein:

[0020] Figure 1 An exemplary flowchart showing a method for constructing a prediction model according to some embodiments of the present disclosure;

[0021] Figure 2 An exemplary flowchart showing a method for constructing a prediction model according to some embodiments of the present disclosure;

[0022] Figure 3 An exemplary flowchart showing a method for constructing a prediction model according to some embodiments of the present disclosure;

[0023] Figure 4 An exemplary flowchart showing a method for evaluating the sensitivity of emissions to the concentration of a target according to some embodiments of the present disclosure;

[0024] Figure 5 An exemplary flowchart showing a method for evaluating the sensitivity of emissions to the concentration of a target according to some embodiments of the present disclosure;

[0025] Figure 6 An exemplary overall flowchart showing a method for constructing a prediction model and a method for evaluating the sensitivity of emissions to the concentration of a target using the prediction model according to some embodiments of the present disclosure;

[0026] Figure 7 A schematic structural diagram of an electronic device according to some embodiments of the present disclosure. DETAILED DESCRIPTION

[0027] The technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present disclosure. Obviously, the described embodiments are some, but not all, of the embodiments of the present disclosure. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present disclosure without creative efforts shall fall within the protection scope of the present disclosure.

[0028] It should be understood that the terms "including" and "comprising" used in the specification and claims of the present disclosure indicate the presence of the described features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or their combinations.

[0029] It should also be understood that the terms used in the specification of the present disclosure are only for the purpose of describing specific embodiments and are not intended to limit the present disclosure. As used in the specification and claims of the present disclosure, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include the plural forms. It should be further understood that the term "and / or" used in the specification and claims of the present disclosure refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0030] As used in this specification and the claims, the term "if" can be interpreted as "when", "once", "in response to determining", or "in response to detecting" depending on the context. Similarly, the phrase "if determined" or "if [the described condition or event] is detected" can be interpreted as meaning "once determined", "in response to determining", "once [the described condition or event] is detected", or "in response to detecting [the described condition or event]" depending on the context.

[0031] The specific implementation manners of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0032] Exemplary Application Scenarios

[0033] As one of the main causes of urban air pollution, motor vehicle emissions produce a wide variety of pollutants and cause great harm. Among them, nitrogen oxides (NOx), as the main pollutants in motor vehicle exhaust, not only directly endanger human health, but also further affect air quality by being converted into nitrogen dioxide (NO2), becoming an important precursor for the formation of ozone and fine particulate matter. Therefore, it is crucial to accurately evaluate the sensitivity of motor vehicle emission pollutants to atmospheric pollutant concentrations. Only by formulating air quality management strategies based on accurate evaluation results can the excellent air quality of the city be effectively guaranteed. However, at present, there is little research on evaluating the sensitivity of motor vehicle NOx emissions to NO2 concentration, and most of the research mainly focuses on the prediction of NO2 concentration.

[0034] Exemplary application scenarios

[0035] In view of this, embodiments of the present disclosure provide a method for constructing a prediction model, and the prediction model constructed by this method can be used in a method for evaluating the sensitivity of emissions to the concentration of a target. By obtaining a multi-source fusion data set including emissions, time parameters, weather parameters, emission source activity level parameters, and the concentration of the target, the multi-source fusion data set covers various factors affecting the concentration of the target. Then, the preset model is trained using the multi-source fusion data set to obtain a prediction model. Based on this, the prediction model constructed by the present disclosure comprehensively considers emissions and other influencing factors, can accurately capture changes in the concentration of the target, and significantly reduces prediction errors.

[0036] Figure 1 An exemplary flowchart of method 100 for constructing a prediction model according to some embodiments of the present disclosure is shown. As Figure 1 shown, first, at step S101, a multi-source fusion data set is obtained. It can be understood that the multi-source fusion data set refers to a data set that integrates data from multiple different sources and of various types. In the embodiments of the present disclosure, the multi-source fusion data set includes emissions, time parameters, weather parameters, emission source activity level parameters, and the concentration of the target. That is, each sample in the multi-source fusion data set is composed of emissions, time parameters, weather parameters, emission source activity level data, and the concentration of the target. It can be understood that the emission source here refers to the source or place that generates and releases emissions. In addition, the multi-source fusion data set in the embodiments of the present disclosure is a data set that covers various situations in the real environment to ensure that the constructed prediction model can fully learn various complex scenarios, has wide applicability, and can thus meet the detection requirements in different application scenarios.

[0037] It should be noted that the target here refers to a substance that is widely distributed in a dispersed and suspended state in the atmospheric environment and has no fixed boundary or shape. Correspondingly, emissions are substances directly or indirectly affecting changes in the concentration of the target released by the emission source. For example, the target can be an atmospheric pollutant such as NO 2 (nitrogen dioxide), and the corresponding emissions can be NO x (nitrogen oxides) and other exhaust gases released by motor vehicles. Of course, the above emissions can also be other emissions such as carbon oxides and sulfur dioxide, and the embodiments of the present disclosure do not make specific limitations on this.

[0038] It has been found through in-depth research in this disclosure that the concentration of the target object is not only affected by emissions, but also fluctuates over time and with weather changes, and the emissions of the emissions will change due to changes in the activity level parameters of the emission sources. Therefore, as described above, the multi-source fusion dataset of the embodiments of this disclosure has extremely high comprehensiveness. This dataset not only covers the emissions of emissions and the concentration data of the target object, but also accurately incorporates time parameters, weather parameters, and the activity level parameters of the emission sources. This design aims to provide sufficient data support for the construction of subsequent prediction models, enabling the models to comprehensively learn the comprehensive impact of time, weather, the activity level parameters of the emission sources, and the emissions of emissions on the concentration of the target object. In this way, when predicting the concentration of the target object, the constructed prediction model can not only accurately analyze the changes in the concentration of the target object dominated by the emissions of emissions, but also skillfully integrate the time periodic characteristics and complex and variable weather factors, greatly improving the accuracy and reliability of the prediction results.

[0039] In step S102, a preset model is trained using the multi-source fusion dataset to obtain a prediction model. In some embodiments, the preset model can be a Random Forest Model (RF). Furthermore, in this step S102, the random forest model is trained using the multi-source fusion dataset to obtain a prediction model.

[0040] In addition, in other embodiments, before performing step S102, the multi-source fusion dataset can be standardized using the minimum-maximum normalization method to scale all data to values between 0 and 1, thereby eliminating the influence of the dimension (unit) between different data.

[0041] The above combination Figure 1 The method 100 for constructing a prediction model in some embodiments of this disclosure has been specifically described overall. It obtains a multi-source fusion dataset including the emissions of emissions, time parameters, weather parameters, the activity level parameters of the emission sources, and the concentration of the target object, so that the multi-source fusion dataset covers various factors affecting the concentration of the target object. Then, the preset model is trained using this multi-source fusion dataset to obtain a prediction model. Based on this, the prediction model constructed in this disclosure comprehensively considers emissions and other influencing factors, can accurately capture the changes in the concentration of the target object, and greatly reduces the prediction error. At the same time, the model also has excellent generalization performance and can flexibly adapt to various complex scenarios, providing an efficient and accurate solution for the detection of the concentration of the target object.

[0042] However, those skilled in the art should understand that Figure 1 The method 100 shown is exemplary rather than restrictive, and those skilled in the art can make corresponding modifications as needed. As an example, a specific implementation manner of obtaining the multi-source fusion dataset, that is, step S101, will be further given next.

[0043] In some embodiments, the multi-source fusion dataset can be constructed using real data that has historically existed. The real data can truly reflect the impact of emissions, time, weather, and emission source activity level parameters on the target concentration, thereby facilitating the construction of a reliable and highly accurate prediction model.

[0044] Specifically, given the vast geographical area of the Earth, it is extremely difficult and highly infeasible to obtain data over a large area and a long time span in actual operation. Moreover, the preset model training does not rely on a large amount of data, and too much data may even increase the training burden and reduce efficiency. Therefore, in the embodiments of this disclosure, first, a continuous time range in the past can be selected. This time range is called the historical period, and one or more areas can be selected as the areas to be measured. A multi-source fusion dataset that meets the preset model training can be constructed by using the real data of the areas to be measured during this historical period.

[0045] In some embodiments, when selecting the areas to be measured and the historical period, the principles of comprehensiveness and typicality can be adhered to. That is, the areas to be measured should cover as diverse geographical features as possible, such as rural areas, urban areas, etc., in order to include different environmental conditions brought about by terrain differences, uneven population density, and diverse industrial activities. For the selection of the historical period, various weather conditions can be selected as much as possible, such as extreme weather like heavy rain, sandstorms, high temperatures, cold snaps, etc., as well as the alternation of conventional weather such as sunny, rainy, and cloudy days, while taking into account different social activity periods such as weekdays, holidays, day and night. In this way, the finally obtained multi-source fusion dataset can cover datasets under various different scenarios, ensuring that the constructed prediction model can fully learn various complex scenarios, thereby having wide applicability to easily meet the detection requirements under different application scenarios.

[0046] After selecting the areas to be measured and the historical period, datasets of time parameters, weather parameters, emission source activity level parameters, emissions, and target concentration can be constructed synchronously or separately. Next, specific descriptions will be made on the datasets of time parameters, weather parameters, emission source activity level parameters, emissions, and target concentration constructed respectively.

[0047] In some embodiments, the timestamp information of this historical period can be obtained, and a dataset of time parameters can be constructed based on this timestamp information. The timestamp information can be automatically generated by existing software. Specifically, if the accuracy of the timestamp information of the historical period is different from the accuracy of the time parameters, the target timestamp information with the same accuracy as the time parameters can be extracted from the timestamp information of the historical period first, and then this target timestamp information can be used as the dataset of time parameters. If the accuracy of the timestamp information of the historical period is the same as the accuracy of the time parameters, then this timestamp information can be directly used as the dataset of time parameters.

[0048] In some further embodiments, if the data format of the timestamp information of the historical period is different from the data format of the time parameter, it is further necessary to convert the data format of the timestamp information of the historical period to be consistent with the data format of the time parameter.

[0049] In the embodiments of the present disclosure, the precision of the time parameter may include one or more of year, month, day, hour, weekend / holiday, and day / night. For example, the precision of the time parameter may include year, month, day, hour, weekend / holiday, and day / night. That is, the constructed dataset of time parameters includes each dataset with time precision of year, month, day, hour, weekend / holiday, and day / night. Of course, those skilled in the art can also adjust according to their own experience and actual needs, and the embodiments of the present disclosure do not make specific limitations in this regard.

[0050] In some embodiments, the emission source activity level parameter is a parameter used to quantify the activity degree of the emission source, and it reflects the relevant activity amount of the emission source generating emissions within a certain period. Therefore, the emission source activity level parameter is also a key variable affecting the concentration of the target object. In some embodiments, the dataset of the emission source activity level parameter can be collected by relevant monitoring devices.

[0051] As an example, when the emission source is a motor vehicle and the emission is NO x the NOx emissions released by the motor vehicle are related to the average speed of the motor vehicle and the motor vehicle flow. It can be understood that the motor vehicle flow refers to the number of motor vehicles passing through a specific location, section, or area of a road within a certain time. Therefore, the average speed of the motor vehicle and the motor vehicle flow can be used as the activity level parameters of the motor vehicle as an emission source. And the datasets of the average speed of the motor vehicle and the motor vehicle flow can be obtained from the traffic checkpoints of each road. A traffic checkpoint refers to a specific location point set on the road for monitoring and recording information of passing vehicles, usually equipped with relevant devices and systems.

[0052] It should be noted that the time precision of the emission source activity level parameter obtained here is the same as the precision of the time parameter. Therefore, the calculated emissions are also the same as the precision of the time parameter. Similarly, the time precision of other parameters in the multi-source fusion dataset is also like this, and it will not be repeated below.

[0053] In some embodiments, the emissions can be calculated by the empirical coefficient method. For example, the emissions can be calculated using the emission factor and the target activity level parameter of the emission source.

[0054] As an example, in some embodiments, the emission source is a motor vehicle and the emission is NO xWhen obtaining the target activity level parameter of a motor vehicle, which corresponds to the driving mileage per hour. In some embodiments, the driving mileage per hour of motor vehicles on the main road can be obtained through traffic checkpoints configured in the area to be measured, and then substituted into the calculation expression of emissions to obtain the emissions of motor vehicles on the main road.

[0055] The calculation expression of emissions can be the following formula (1):

[0056] E NOx = L r,i * EF NOx (1)

[0057] Wherein, E NOx is the NO x emissions (g) of motor vehicles per hour on the road; i is the hour; h is the total number of hours in the historical period; Lr,i is the driving mileage per hour (km) of motor vehicles on road r, calculated from traffic checkpoint data; EF NOx is the motor vehicle emission factor (g / km), which can be obtained from the "Technical Guidelines for Compiling the Road Motor Vehicle Atmospheric Pollutant Emission Inventory", and the embodiments of this disclosure do not limit this.

[0058] In some embodiments, the weather parameters corresponding to the time stamp information of the historical period can be extracted from the weather parameter dataset of the area to be measured provided by official or authoritative institutions according to the time stamp information of the historical period. The set of these extracted weather parameters constitutes the weather parameter dataset in the multi-source fusion dataset. For example, the weather parameters within the area to be measured and corresponding to the time stamp information of the historical period can be extracted from the ERA5 dataset (the fifth-generation global climate atmospheric reanalysis dataset) of the European Centre for Medium-Range Weather Forecasts (ECMWF) according to the time stamp information of the historical period.

[0059] As an example, in some embodiments, the weather parameters may include any weather variables that affect the change in the concentration of the target substance, such as the boundary layer height (abbreviated as blh), relative humidity (abbreviated as r), surface pressure (abbreviated as sp), surface solar radiation downward, clear-sky (abbreviated as ssrdc), temperature at 2 meters above ground (abbreviated as t2m), total cloud cover (abbreviated as tcc), total precipitation (abbreviated as tp), wind speed at altitude a (abbreviated as Va), and wind direction (abbreviated as wd).

[0060] In some embodiments, similarly, the concentration of the target substance corresponding to the timestamp information of the historical period can be extracted from the dataset of the concentration of the target substance in the area to be measured provided by an official or authoritative institution according to the timestamp information of the historical period. The set of the extracted concentrations of the target substance constitutes the dataset of the concentration of the target substance in the multi-source fusion dataset.

[0061] After successfully obtaining the datasets of the time parameters, weather parameters, emission source activity level parameters, emissions of pollutants, and concentrations of the target substance in the area to be measured during the historical period, the time parameters, weather parameters, emissions of pollutants, and concentrations of the target substance with consistent time in each of the above datasets can be combined into a data record. The set of each combined data record constitutes the multi-source fusion dataset.

[0062] As an alternative embodiment, if the amount of data in the multi-source fusion dataset constructed using real data cannot meet the data volume requirements for preset model training, it can also be expanded based on the real data. The specific expansion method can refer to the prior art and will not be elaborated in the embodiments of this disclosure.

[0063] The above generally further describes the acquisition of the multi-source fusion dataset. It collects the datasets of the time parameters, weather parameters, emission source level parameters, emissions of pollutants, and concentrations of the target substance in the area to be measured during the historical period, and then combines the time parameters, weather parameters, emissions of pollutants, and concentrations of the target substance with consistent time in each dataset into a data record. Each data record belongs to a sample, and the set of each combined data record is the multi-source fusion dataset. However, those skilled in the art should understand that the method for acquiring the multi-source fusion dataset described above is exemplary rather than restrictive, and those skilled in the art can make corresponding adjustments as needed.

[0064] Figure 2 An exemplary flowchart of method 200 for constructing a prediction model showing some embodiments of the present disclosure. Method 200 may be a specific execution process of step S102 in method 100 above, that is, the detailed steps of training a preset model using a multi-source fusion dataset to obtain a prediction model. As Figure 2 shown, method 200 includes steps S201 to S204. First, in step S201, the multi-source fusion dataset is divided into a training set and a test set. In some embodiments, the multi-source fusion dataset can be divided into a training set and a test set according to a preset ratio. As an example, the preset ratio can be 7:3.

[0065] Furthermore, in step S202, based on the training set, the preset model is trained using the k-fold cross-validation method to obtain target hyperparameters. In the embodiments of the present disclosure, the purpose of training the preset model using the k-fold cross-validation method is to obtain a set of target hyperparameters that enable the preset model to have the best performance. The specific value of k can be 5, and of course, it can also be other values such as 6, etc. The present disclosure does not make specific limitations on this, and those skilled in the art can adjust it flexibly.

[0066] The k-fold cross-validation method refers to randomly dividing the training set into k subsets of roughly equal size, and each subset is called a fold. Subsequently, the preset model is trained and validated k times to obtain the performance values of the preset model after each training. It can be understood that after k times of training and validation operations, k performance values of the preset model for different sub-validation sets can be obtained. It should be noted that each time the preset model completes a training, a set of hyperparameters determined based on that training process will be generated. Finally, by comparing the performance values of the preset model on different sub-validation sets, a set of target hyperparameters that enable the preset model to have the best performance is selected.

[0067] Specifically, each training and validation operation first selects one fold as the sub-validation set, and the remaining k - 1 folds as the sub-training set. Then the sub-training set is input into the preset model for training to enable the preset model to learn the data features and rules in the sub-training set. After that, the sub-validation set is input into the preset model to determine the performance value of the preset model for the sub-validation set.

[0068] In some embodiments, RMSE (Root Mean Square Error) can be used to represent the performance value of the preset model. Considering that the number of samples in the sub-validation set is large, if the RMSE of each sample in the sub-validation set is compared individually, the operation is cumbersome and it is difficult to intuitively reflect the overall performance of the model. Therefore, in this embodiment, the mean value of the RMSE of all samples in the sub-validation set can be statistically calculated to accurately represent the performance value of the pre-trained model for the sub-validation set.

[0069] It can be understood that RMSE is an index for measuring the degree of difference between observed values and predicted values. The smaller its value, the smaller the average error between the model prediction result and the true value, the better the fitting effect of the model on the data, and the higher the performance. On the contrary, the larger the RMSE, the greater the deviation between the predicted value and the true value of the model, and the worse the model performance. Based on this, in the embodiments of the present disclosure, the hyperparameters of the preset model with the minimum performance value (RMSE) are selected as the target hyperparameters.

[0070] Then, in step S203, the target hyperparameters are set as the hyperparameters of the preset model.

[0071] Finally, in step S204, the preset model is trained using the training set and the test set to obtain a prediction model. In the embodiments of the present disclosure, based on the foregoing description, it can be known that the hyperparameters set for the preset model are already the optimal hyperparameters. Therefore, during the entire training process of the preset model here, the hyperparameters set need to be kept constant. And this training is different from the previous one and does not require the use of the K-fold cross-validation method for training. Instead, the entire training set is first input into the preset model for training, so that the preset model can learn features and rules through a large amount of data and initially build the corresponding prediction or analysis ability. Then the test set is input into the trained preset model for secondary training to test and optimize the performance of the preset model. The preset model after being trained by the training set and the test set is the prediction model.

[0072] The above combination Figure 2 has generally described the method 200 for constructing a prediction model in some embodiments of the present disclosure. It divides the multi-source fusion data set into a training set and a test set. Then, based on the training set, the preset model is trained using the K-fold cross-validation method to obtain the optimal target hyperparameters. Furthermore, after setting the target hyperparameters as the hyperparameters of the preset model, the preset model with constant hyperparameters is trained using the training set and the test set to obtain a prediction model with excellent performance. However, those skilled in the art should understand that Figure 2 the method 200 shown is exemplary rather than restrictive, and those skilled in the art can make corresponding modifications as needed.

[0073] Next, in combination with Figure 3 further describe the method 300 for constructing a prediction model in some embodiments of the present disclosure, which is another specific implementation manner of step S204 in the foregoing method 200. As Figure 3As shown, method 300 includes steps S301 to S304. First, in step S301, a preset model is trained using a training set and a test set. In the embodiments of this disclosure, the complete training set is directly input into the preset model first for comprehensive training, enabling the preset model to fully learn the data features and rules in the training set. After the training is completed, the entire test set is input into the trained preset model for secondary training to test and optimize the performance of the preset model.

[0074] Next, in step S302, a first performance value of the preset model for the training set and a second performance value for the test set are determined. In the embodiments of this disclosure, R 2 (coefficient of determination) is used to represent the first performance value and the second performance value of the preset model for the training set and the test set. Considering the large number of samples in the training set and the test set, if each sample in the dataset is compared individually for R 2 , the operation is cumbersome and it is difficult to intuitively reflect the overall performance of the model. Therefore, in this embodiment, the mean value of R 2 for all samples in the training set can be statistically calculated to accurately represent the first performance value of the pre-trained model for the training set. Similarly, for the test set, the mean value of R 2 is also calculated in the same way to represent the second performance value of the pre-trained model for the test set.

[0075] It can be understood that the value range of R 2 is between 0 and 1. The closer R 2 is to 1, the higher the proportion of the variation part of the dependent variable that the model can explain in the total variation, that is, the higher the fitting degree of the model to the data and the better the performance. When R 2 = 1, it means that the model perfectly fits all data points and there is no error between the observed value and the predicted value, which is an ideal perfect fitting state. On the contrary, the closer R 2 is to 0, the less variation part the model can explain and the worse the model performance. If R 2 is negative, it means that the prediction effect of the model is not as good as using the mean for prediction, and in this case, the model is invalid.

[0076] Furthermore, in step S303, if at least one of the first performance value and the second performance value does not meet the preset usage standard, jump to execute the step of training the preset model based on the training set by using the k-fold cross-validation method to obtain the target hyperparameters, or continue to execute the step of training the preset model by using the training set and the test set. In the embodiments of the present disclosure, that the preset usage standard is met indicates that the performance of the trained preset model is not good enough. Therefore, to continuously improve the performance of the preset model, by jumping to execute the step of training the preset model based on the training set by using the k-fold cross-validation method to obtain the target hyperparameters, the hyperparameters of the preset model are readjusted. Or continue to execute the step of training the preset model by using the training set and the test set to continue training the preset model and optimize the performance of the preset model. Repeat this process until the first performance value of the preset model for the training set and the second performance value for the test set meet the usage standard, and a prediction model is successfully obtained.

[0077] In step S304, if both the first performance value and the second performance value meet the preset usage standard, the trained preset model is determined as the prediction model. In the embodiments of the present disclosure, if both the first performance value and the second performance value reach the preset usage standard, it indicates that the performance of the preset model meets the actual application requirements. Immediately, the preset model that has been trained and tested is determined as the prediction model.

[0078] In some embodiments, the usage standard may be a preset threshold. And the preset threshold may include a first preset threshold and a second preset threshold. If the first performance value does not meet the first preset threshold and / or the second performance value does not meet the second preset threshold, it indicates that the currently trained preset model does not meet the usage standard. If the first performance value meets the first preset threshold and the second performance value meets the second preset threshold, it indicates that the currently trained preset model meets the usage standard and can be used as a prediction model, that is, the currently trained preset model is the prediction model.

[0079] The above combines Figure 3 Generally, the method 300 for constructing a prediction model in some embodiments of the present disclosure is specifically described. It trains a preset model through a training set and a test set to obtain the first performance value of the preset model for the training set and the second performance value for the test set. If at least one of these two performance values does not meet the preset usage standard, then it can jump to train the preset model by using the k-fold cross-validation method to obtain the target hyperparameters, or continue to train the preset model with the training set and the test set; if both performance values meet the preset usage standard, the trained preset model is determined as the prediction model. Through the above method, the embodiments of the present disclosure can obtain a prediction model with the best performance. However, those skilled in the art should understand that Figure 2The method 300 shown is exemplary and not restrictive, and those skilled in the art can make corresponding modifications as needed.

[0080] In addition, in some embodiments, the preset model is a random forest model. Therefore, the prediction model obtained by training the preset model also belongs to the random forest model. It can be known that during the training process of the random forest model, it can automatically calculate the importance of each input parameter for the prediction output of the prediction model (i.e., the concentration of the predicted target). The basic idea is to measure the importance of a feature by calculating the average decrease in impurity at all decision tree nodes of the random forest for each feature. The greater the importance, the more important the feature is for the classification or regression task of the model. Therefore, when training the random forest model to obtain the prediction model, or after that, it is also possible to obtain the importance data of various input parameters output by the prediction model. Furthermore, based on the importance of various input parameters, the input parameters with low importance can be removed to reduce the dimension of the input parameters and improve the performance of the prediction model.

[0081] In other embodiments, if the preset model is a model that cannot automatically calculate the importance of input parameters, it can be converted to analyze the importance of various input parameters through the correlation coefficient.

[0082] Further, in some embodiments, after obtaining the prediction model, the SHAP (SHapley Additive exPlanations) analysis method can also be used to evaluate the interpretability of the prediction model. The SHAP analysis method is a technique for explaining the prediction results of machine learning models. It decomposes the model prediction results into the contributions of each feature to the prediction results. Therefore, for a given prediction model, the SHAP analysis can calculate the importance score of each input parameter for the model output, that is, the SHAP value. Through these scores, the role played by each input parameter when the prediction model makes a decision can be understood, thereby evaluating the interpretability of the prediction model. This provides a basis for evaluating the sensitivity of the target concentration to the emissions through the subsequent use of this prediction model.

[0083] Based on the prediction model constructed above, to further expand its application scope and value, some embodiments of this disclosure also provide a method for evaluating the sensitivity of emissions to the concentration of a target object by using the prediction model constructed in any of the previous embodiments. This method constructs a target multi-source fusion dataset of emissions under different emission scenarios, where the target multi-source fusion dataset includes target time parameters, target weather parameters, target emission source activity level parameters, and target emission amounts. Then, the target multi-source fusion dataset is input into the prediction model to obtain the predicted target object concentration under different emission scenarios. Thus, according to the target emission amounts and the predicted target object concentrations under different emission scenarios, the sensitivity of emissions to the target object concentration under different emission scenarios is determined. Through the above method provided by the embodiments of this disclosure, the sensitivity of emissions to the target object concentration under different emission scenarios can be accurately evaluated, thereby providing solid data support and a scientific basis for formulating air quality management strategies in the future, and helping relevant departments to carry out air pollution control work targeted.

[0084] The following will describe in detail the specific implementation manners of the method for evaluating the sensitivity of emissions to the concentration of a target object in some embodiments of this disclosure with reference to the accompanying drawings.

[0085] Figure 4 An exemplary flowchart of a method 400 for evaluating the sensitivity of emissions to the concentration of a target object in some embodiments of this disclosure is shown, including step S401 to step S403. First, in step S401, a target multi-source fusion dataset of emissions under different emission scenarios is constructed. In the embodiments of this disclosure, the target multi-source fusion dataset includes target time parameters, target weather parameters, target emission source activity level parameters, and target emission amounts. Additionally, it should be particularly noted that the emission scenario here refers to the scenario of the emission amount of this emission released in different regions. That is to say, this step aims to construct a target multi-source fusion dataset of the emission amounts of this emission released in different regions, so as to accurately evaluate the sensitivity of the emission amounts of emissions released in different regions to the concentration of the target object in the future.

[0086] Based on the previous description, it can be further known that the target object refers to a substance that is widely distributed in a dispersed and suspended state in the atmospheric environment and has no fixed boundary or shape. Correspondingly, the emission is a substance that directly or indirectly affects the change in the concentration of the target object. For example, the target object can be an atmospheric pollutant such as NO 2 (nitrogen dioxide), and correspondingly, the emission can be the exhaust gas released by vehicles such as NO x (nitrogen oxides). Of course, the above emissions can also be other emissions such as carbon oxides and sulfur dioxide, and the embodiments of this disclosure do not make specific limitations on this.

[0087] In addition, in some embodiments, the accuracy of the target time parameter may include one or more of year, month, day, hour, weekend / holiday, and day / night. For example, the accuracy of the target time parameter may include year, month, day, hour, weekend / holiday, and day / night. The target weather parameter may be weather variables that affect the change in the concentration of the target substance, including boundary layer height, relative humidity, air pressure, surface solar radiation, temperature, total cloud cover, total precipitation, wind speed, wind direction, etc. The embodiments of the present disclosure do not make specific limitations in this regard, and those skilled in the art can make flexible adjustments according to relevant experience and professional knowledge.

[0088] In some embodiments, the target emission source activity level parameter may be a parameter used to quantify the activity degree of the emission source, which reflects the relevant activity amount of the emission source generating emissions within a certain period. As an example, when the emission source is a motor vehicle and the emission is NO x x, the NOx emissions released by the motor vehicle are related to the average speed of the motor vehicle and the motor vehicle flow (Passenger Vehicle Flow, PV). It can be understood that the motor vehicle flow refers to the number of motor vehicles passing through a specific location, section, or area of a road within a certain period. Therefore, the average speed of the motor vehicle and the target motor vehicle activity level parameter can be determined.

[0089] Then, in step S402, the target multi-source fusion dataset is input into the prediction model to obtain the predicted target substance concentration under different emission scenarios. It can be understood that the prediction model is a model that has been pre-trained and has converged or meets the usage standard, so in the embodiments of the present disclosure, a predicted target substance concentration with high accuracy can be obtained.

[0090] Finally, in step S403, according to the target emissions and the predicted target substance concentration under different emission scenarios, the sensitivity of the emissions to the target substance concentration under different emission scenarios is determined. In the embodiments of the present disclosure, a linear regression equation under different emission scenarios can be constructed according to the target emissions and the predicted target substance concentration under different emission scenarios. It can be understood that the linear regression equation refers to the linear regression equation between the emissions and the target substance concentration. Furthermore, according to the slope of the linear regression equation, the sensitivity of the emissions to the target substance concentration under different emission scenarios is determined. The slope of the linear regression equation represents the sensitivity of the emissions to the target substance concentration under the corresponding emission scenario. The larger the slope, the higher the sensitivity, and the smaller the slope, the lower the sensitivity.

[0091] In some embodiments, existing software such as the scikit-learn library in Python can be used to construct the linear regression equation under different emission scenarios. The embodiments of the present disclosure do not make specific limitations on the method used to construct the linear regression equation, and those skilled in the art can make flexible selections.

[0092] In some embodiments, after the linear regression equation is constructed, the R 2 (Coefficient of Determination) and p-value (probability value) can be further used to evaluate the quality and effectiveness of the linear regression equation. If the quality and effectiveness of the linear regression equation meet the requirements, the linear regression equation can be used to evaluate the sensitivity of the emissions of pollutants to the concentration of the target substance. If the quality and effectiveness of the linear regression equation do not meet the requirements, the step of constructing the linear regression equation according to the target emissions of pollutants and the predicted concentration of the target substance is re-executed to reconstruct the linear regression equation.

[0093] Specifically, if the R 2 is close to 1 and the p-value is less than the significance level, it is determined to meet the requirements. If the R 2 is far from 1 and / or the p-value is greater than the significance level, it is determined not to meet the requirements. It can be understood that if the R 2 is close to 1, it indicates that the linear regression equation can explain most of the variation of the dependent variable and has a good fitting effect. If the p-value is less than the significance level, it can be determined that there is a significant linear relationship between the independent variable and the dependent variable, and the constructed linear regression equation is meaningful. The significance level can be 0.05 or 0.01.

[0094] For example, taking the emission source as a motor vehicle, the pollutant as NO x , the concentration of the target substance as the NO 2 concentration, and it is required to evaluate the sensitivity of the NO x emissions released in regions a, b, and c to the NO 2 concentration in each region. Therefore, the target multi-source fusion datasets in regions a, b, and c are constructed respectively, and then the target multi-source fusion datasets in regions a, b, and c are input into the prediction model to predict the predicted NO 2 concentration in regions a, b, and c. Further, according to the NO x emissions of each sample in the target multi-source fusion datasets in regions a, b, and c and their corresponding predicted NO 2 concentrations, the linear regression equations between the NO x emissions and the NO 2 concentrations in regions a, b, and c as shown in Table 1 are constructed. In the linear regression equation, C NO2 is the NO 2 concentration level of the corresponding region, and E NOx is the NO x emissions of the corresponding region. As shown in Table 1, the R2 values of the linear regression equations corresponding to regions a, b, and c are 0.78, 0.84, and 0.89 respectively, and the p-values are all <0.01. Therefore, it meets the requirements and can be used to evaluate the NO xSensitivity to NO 2 concentration.

[0095] Furthermore, it can be seen from Table 1 that the slopes k of the linear regression equations corresponding to regions a, b, and c are 0.0003, 0.0003, and 0.0001 respectively, indicating that the sensitivity of the NO 2 concentration to changes in NO x emission levels is higher in regions a and b than in region c.

[0096] Table 1

[0097] Region Linear regression equation <![CDATA[R 2 > p-value a <![CDATA[C NO2 = 0.0003E NOx + 8.6935]]> 0.7813 1.9341e-02 b <![CDATA[C NO2 = 0.0003E NOx + 9.6283]]> 0.8352 1.0799e-02 c <![CDATA[C NO2 = 0.0001E NOx + 9.7166]]> 0.8943 4.3500e-03

[0098] The above combination Figure 4 has specifically described Method 400 for evaluating the sensitivity of emissions of some embodiments of the present disclosure to the concentration of the target substance. By constructing a target multi-source fusion dataset of emissions under different emission scenarios, where the target multi-source fusion dataset includes target time parameters, target weather parameters, target emission source activity level parameters, and target emission amounts. Then, the target multi-source fusion dataset is input into a prediction model to obtain the predicted target substance concentration under different emission scenarios. Thus, according to the target emission amounts and the predicted target substance concentrations under different emission scenarios, the sensitivity of the emissions to the target substance concentration under different emission scenarios is determined. Through the above method provided by the embodiments of the present disclosure, the sensitivity of emissions to the target substance concentration under different emission scenarios can be accurately evaluated, thereby providing solid data support and a scientific basis for formulating air quality management strategies in the future, and assisting relevant departments to carry out air pollution control work in a targeted manner. However, those skilled in the art should understand that Figure 4 the shown Method 400 is exemplary rather than restrictive, and those skilled in the art can make corresponding modifications as needed.

[0099] Next, in combination with Figure 5 Method 500 for evaluating the sensitivity of emissions of some embodiments of the present disclosure to the concentration of the target substance will be described. This Method 500 corresponds to a specific implementation manner of step S401 in the foregoing Method 400. As Figure 5 shown, Method 500 includes steps S501 to S503.

[0100] First, in step S501, a first preset number of first target multi-source fusion data subsets are randomly selected from the multi-source fusion data set of the target historical period of the target area to be tested. In the disclosed embodiment, the target area to be tested is the area to be actually evaluated, and the target historical period refers to a continuous time range in the past. In other embodiments, the target area to be tested and the target historical period may also be the same as the area to be tested and the historical period selected when the prediction model was constructed in the previous text, but it can be understood by those skilled in the art that if the target area to be tested and the target historical period here are the same as the area to be tested and the historical period selected when the prediction model was constructed in the previous text, it indicates that the area to be tested described in the previous text is evaluated.

[0101] It should be noted that the method for obtaining the multi-source fusion data set in the embodiment of the present disclosure can refer to the corresponding description in the previous text, and will not be repeated here.

[0102] In some embodiments, the first preset number may be 10,000. That is, in step S501, 10,000 samples may be randomly selected from the multi-source fusion data set as the first target multi-source fusion data subset. It can be understood that random sampling of a large number of samples takes into account the actual distribution of all variables, thereby improving statistical reliability.

[0103] Then, in step S502, a target emission amount is determined according to the maximum emission amount of the emission in the first target multi-source fusion data subset.

[0104] In the disclosed embodiment, in order to ensure that the target multi-source fusion data set finally constructed can fully cover the minimum to maximum range of emission amounts, in order to maintain the efficiency of calculation and the feasibility of the subsequent construction of the linear regression equation. A step length can be obtained by dividing the maximum emission amount in the above-mentioned first target multi-source fusion data subset by the target value. The step length represents the fixed difference in emission amounts between two adjacent sampling points within the emission amount value range. The target value can be set based on experience, for example, the target value is set to 5. Then, the step length is used to calculate the various target emission amounts covering different value ranges.

[0105] As an example, the step size can be calculated using the following formula (2):

[0106] S=E i,max =max(E i ) / 5(2)

[0107] Where S is the step size (g) and Ei,max is the maximum emission of the emission in region i.

[0108] As an example, the target emission amount can be calculated using the following formula (3):

[0109] Nj = j × S(3)

[0110] where Nj is the emission amount of the j-th target emission, and j = 0, 1, 2, 3, 4, 5.

[0111] Thus, through the above formula (3), the emission amounts of all target emissions can be calculated, which is expressed by the following formula (4):

[0112] {N j} = {0, S, 2S, 3S, 4S, 5S}(4)

[0113] Based on formula (4), it can be known that the emission amounts of the target emissions are 0, S, 1S, 2S, 3S, 4S, and 5S respectively, covering the minimum and maximum values of the emission amounts and all values in between.

[0114] In step S503, while keeping the emission amount of the target emission constant, the target time parameter, the target weather parameter, and the target emission source activity level parameter are determined based on the first target multi-source data subset. In the embodiments of the present disclosure, while keeping the emission amount of the target emission constant, the target time parameter, the target weather parameter, and the target emission source activity level parameter corresponding to each target emission amount covering the minimum and maximum values of the emission amount and all values in between as described above and having different situations are determined, so as to obtain the target multi-source fusion data set. It can be understood that the target multi-source fusion data set constructed in this way can evaluate the stability and variation law of the sensitivity of the emission amount to the target substance concentration under different environmental conditions, making the research results more realistic and universal.

[0115] Next, a specific implementation manner for determining the target time parameter, the target weather parameter, and the target emission source activity level parameter is given. Specifically, step S503 may include the following steps: First, randomly select a second preset number of second target multi-source fusion data subsets from the first target multi-source fusion data subset. Wherein the second preset number may be less than the first preset number. For example, it may be 50 or 100 or even larger. The embodiments of the present disclosure do not limit the specific value of the second preset number, and those skilled in the art can flexibly set it according to the actual situation. Then, the mean or median of the time parameter, the weather parameter, and the emission source activity level parameter in the second target multi-source fusion data subset is used as the target time parameter, the target weather parameter, and the target emission source activity level parameter.

[0116] Thus, by repeatedly performing the above operations, multiple samples can be constructed. The emission amount of the target emission in these samples is constant at the value in the above formula (4), while the target time parameter, the target weather parameter, and the target emission source activity level parameter are randomly obtained from the first target multi-source data subset, covering various situations in the real environment.

[0117] It should be specifically noted that the reason for selecting the mean or median of each parameter in the second target multi-source fusion data subset as the target parameter in the embodiments of this disclosure is that the mean and median can effectively converge the central tendency of the data. By this means, the interference of various variability factors in this data subset on the final result can be greatly reduced, so as to more clearly and purely highlight the independent influence of the key factor of the emissions of the emissions on the concentration of the target, making the sensitivity result of the finally obtained emissions of the emissions to the concentration of the target more accurate and reliable.

[0118] The above is combined overall Figure 5 A method 500 for evaluating the sensitivity of emissions to the concentration of a target in some embodiments of this disclosure has been specifically described. It randomly selects a preset number of first target multi-source fusion data subsets from the multi-source fusion data set of the target historical period in the target area to be measured. Then, further determines the target emissions based on the maximum emissions in the first target multi-source fusion data subset. Finally, by keeping the target emissions constant, the target time parameter, the target weather parameter, and the target emission source activity level parameter are determined based on the first target multi-source data subset. However, those skilled in the art should understand that Figure 5 The method 500 shown is exemplary rather than restrictive, and those skilled in the art can make corresponding modifications as needed.

[0119] Next, combined with Figure 6 Describe the overall system process of the method for constructing a prediction model in this disclosure and the method for evaluating the sensitivity of emissions to the concentration of a target using the prediction model. As Figure 5 It can be known from the shown process that this disclosure first obtains a multi-source fusion data set including time parameters, weather parameters, emission source activity level parameters, emissions, and the concentration of the target. Then, the preset model is trained using the multi-source fusion data set to construct a prediction model. After obtaining the prediction model, a target multi-source fusion data set of emissions under different emission scenarios is further constructed. Thus, the target multi-source fusion data set is input into the selected prediction model, and the predicted target concentration of emissions under different emission scenarios is predicted. Finally, by according to the predicted target concentration and the target emissions, the sensitivity of emissions under different emission scenarios to the concentration of the target is determined.

[0120] In order to implement the method steps described in this disclosure in combination with the accompanying drawings at the software and hardware levels, embodiments of this disclosure also provide a processing device, and this processing device can be an Figure 7 electronic device as shown. Figure 7 An exemplary structural block diagram of the electronic device 70 in the embodiments of this disclosure is shown. As Figure 6As shown, the electronic device 70 of the present disclosure may include a processor 710 and a memory 720. Among them, an executable program is stored on the memory 720, and the processor 710 can load and execute the executable program, so that the electronic device 70 implements any of the method steps described above.

[0121] In an example scenario, the processor 710 can be used to control the memory 720. Further, the processor 710 may be a central processing unit (CPU), an application processor (AP), etc. integrated in the electronic device 70; and the memory 720, as the hardware for implementing the storage function, may be a read-only memory (ROM), a dynamic RAM (DRAM), etc.

[0122] The present disclosure can also be implemented as a computer-readable storage medium, in which program instructions are stored. When the program instructions are executed by the processor of the electronic device, the processor executes at least part of the steps of the above methods of constructing a prediction model and evaluating the sensitivity of emissions to the concentration of the target object in the present disclosure.

[0123] In the embodiments of the present disclosure, a computer program product is also provided, including a computer program or instruction. When the computer program or instruction is executed by the processor, the methods of constructing a prediction model and evaluating the sensitivity of emissions to the concentration of the target object described in the embodiments of the present disclosure are implemented.

[0124] Although multiple embodiments of the present disclosure have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art can think of many changes, alterations, and alternative ways without departing from the spirit and scope of the present disclosure. It should be understood that various alternatives to the embodiments of the present disclosure described herein may be employed in practicing the present disclosure. The appended claims are intended to define the scope of protection of the present disclosure and thus cover equivalents or alternatives within the scope of these claims.

Claims

1. A method for constructing a prediction model, characterized in that: include: Acquire a multi-source fusion data set, wherein the multi-source fusion data set includes emission amount, time parameter, weather parameter, emission source activity level parameter and target object concentration; The multi-source fusion data set is used to train a preset model to obtain a prediction model.

2. The method for constructing a prediction model according to claim 1, characterized in that: The using the multi-source fusion data set to train a preset model to obtain a prediction model includes: Dividing the multi-source fusion data set into a training set and a test set; Based on the training set, the preset model is trained using a k-fold cross-validation method to obtain target hyperparameters; Setting the target hyperparameter as the hyperparameter of the preset model; The preset model is trained using the training set and the test set to obtain the prediction model, wherein the hyperparameters of the preset model remain constant.

3. The method for constructing a prediction model according to claim 2, characterized in that: The using the training set and the test set to train the preset model to obtain the prediction model further includes: Using the training set and the test set to train the preset model; Determine a first performance value of the preset model for the training set and a second performance value of the preset model for the test set; If at least one of the first performance value and the second performance value does not meet the preset usage standard, jump to the step of training the preset model based on the training set using the k-fold cross-validation method to obtain the target hyperparameter, or continue to execute the step of training the preset model using the training set and the test set; If both the first performance value and the second performance value meet the preset usage standard, the trained preset model is determined as the prediction model.

4. The method for constructing a prediction model according to claim 1, characterized in that: The preset model includes a random forest model.

5. The method for constructing a prediction model according to claim 1, characterized in that: The emissions include NO x , the target includes NO2.

6. A method for evaluating the sensitivity of emissions to target concentrations using the prediction model described in any one of claims 1 to 5, characterized in that: include: Constructing a target multi-source fusion data set of emissions under different emission scenarios, wherein the target multi-source fusion data set includes a target time parameter, a target weather parameter, a target emission source activity level parameter and a target emission amount, and the target emission amount in the target multi-source fusion data set is constant; Inputting the target multi-source fusion data set into the prediction model to obtain the predicted target concentration of the emission under different emission scenarios; The sensitivity of the emission under different emission scenarios to the target object concentration is determined according to the emission amount of the target emission under the different emission scenarios and the predicted target object concentration.

7. The method for evaluating the sensitivity of emissions to target concentration according to claim 6, characterized in that: Determining the sensitivity of the emission under different emission scenarios to the target object concentration according to the target emission under the different emission scenarios and the predicted target object concentration includes: Constructing linear regression equations under different emission scenarios according to the target emission amount and the predicted target concentration under the different emission scenarios; According to the slope of the linear regression equation, the sensitivity of the emissions to the target concentration under different emission scenarios is determined.

8. The method for evaluating the sensitivity of emissions to target concentrations according to claim 6, characterized in that: The target multi-source fusion dataset of emissions under different emission scenarios includes: Randomly selecting a first preset number of first target multi-source fusion data subsets from the multi-source fusion data set of the target historical period of the target area to be tested; Determining the target emission amount according to the maximum emission amount of the emission in the first target multi-source fusion data subset; The target emission amount of the emission substance is kept constant, and the target time parameter, the target weather parameter and the target emission source activity level parameter are determined based on the first target multi-source data subset.

9. The method for evaluating the sensitivity of emissions to target concentration according to claim 8, characterized in that: The step of maintaining the target emission amount constant and determining the target time parameter, the target weather parameter and the target emission source activity level parameter based on the first target multi-source data subset comprises: Randomly selecting a second preset number of second target multi-source fusion data subsets from the first target multi-source fusion data subset; The mean or median of the time parameter, weather parameter and emission source activity level parameter in the second target multi-source fusion data subset is used as the target time parameter, the target weather parameter and the target emission source activity level parameter.

10. The method for evaluating the sensitivity of emissions to target concentration according to claim 6, characterized in that: The emissions include NO x , the target includes NO2.

Images