Deep learning-based traffic environment pollutant concentration change scenario analysis method

Abstract

The present application relates to a kind of traffic environmental pollutant concentration variation scenario analysis method based on deep learning, by obtaining meteorological data, traffic data and atmospheric pollutant concentration data, data set is constructed, and traffic environmental pollutant concentration variation scenario analysis model based on LSTM and CNN is trained, according to scenario analysis traffic flow attenuation gradient set, obtain the label atmospheric pollutant concentration prediction value vector of each attenuation coefficient, finally obtain the label pollutant concentration reduction rate corresponding to each attenuation coefficient.LSTM and CNN are used to build a deep learning model that can accurately predict the hourly concentration of pollutants in the traffic environment;Using model to predict the concentration of pollutants in the traffic environment after traffic flow changes, the response relationship between traffic environmental pollutant concentration and traffic flow is determined;By predicting the traffic environmental pollutant concentration reduction rate under different traffic flow, effective technical support is provided for traffic environmental air quality management.

Description

Technical Field

[0001] This invention relates to a deep learning-based method for analyzing traffic environment pollutant concentration changes. Background Technology

[0002] Motor vehicles emit various pollutants during operation, resulting in higher concentrations of air pollutants in traffic environments compared to urban areas. This poses direct or potential health risks to those exposed. To better monitor pollutant concentrations in traffic environments, roadside air quality monitoring stations are becoming increasingly common. These stations monitor typical pollutants such as PM2.5. 2.5 PM 10 CO, SO2, NO2, NO x The data includes pollutant concentrations (such as oxygen and nitrogen oxides), meteorological factors (wind speed, wind direction, temperature, humidity, etc.), and traffic flow (total traffic flow, large vehicle flow, heavy vehicle flow, small vehicle flow, etc.). Since pollutants in the traffic environment mainly originate from motor vehicle emissions, establishing the response relationship between pollutant concentrations and traffic flow is crucial for formulating and evaluating the effectiveness of motor vehicle control measures. Currently, data on pollutant concentrations in the traffic environment are mostly used for comparative analysis with data from ordinary urban atmospheric monitoring stations, but the response relationship between traffic flow and pollutant concentrations in the traffic environment remains unclear. Summary of the Invention

[0003] To address the above challenges, this invention provides a deep learning-based method for analyzing traffic environment pollutant concentration changes.

[0004] The present invention adopts the following technical solution:

[0005] A deep learning-based method for analyzing traffic environment pollutant concentration changes includes:

[0006] Meteorological data, traffic flow data, and air pollutant data from monitoring points are acquired, and a fused dataset is constructed, which includes a training set and a test set.

[0007] A scenario analysis model for traffic environmental pollutant concentration changes is constructed based on LSTM long short-term memory network and CNN convolutional neural network. The single input tensor of the model is obtained according to the fused dataset, and the scenario analysis model for traffic environmental pollutant concentration changes is trained.

[0008] Construct a scenario analysis dataset based on a preset time starting point;

[0009] Determine the traffic flow attenuation gradient set for scenario analysis, adjust the traffic flow data after the preset time starting point in the scenario analysis dataset, and obtain the label air pollutant concentration prediction value vector corresponding to each attenuation coefficient.

[0010] Based on the predicted atmospheric pollutant concentration vector corresponding to each attenuation coefficient, the reduction rate of the pollutant concentration corresponding to each attenuation coefficient is calculated.

[0011] Furthermore, the model's single-input tensor includes: an atmospheric pollutant concentration data vector excluding the predicted label, a historical atmospheric pollutant concentration vector for the predicted label, a traffic flow data vector, and a meteorological data vector.

[0012] Furthermore, the scenario analysis model for changes in traffic environmental pollutant concentrations is trained, including:

[0013] The model forward propagation calculates the predicted values ​​of the labeled pollutant concentrations in the training set;

[0014] The mean squared error between the predicted and observed values ​​is calculated as the model's loss value;

[0015] The model backpropagates in the direction where the loss value decreases to update the model parameters until the loss value is less than the preset accuracy.

[0016] Output the trained scenario analysis model for changes in traffic environmental pollutant concentrations;

[0017] The model's predicted values ​​were calculated by substituting test set data into the trained scenario analysis model for changes in traffic environmental pollutant concentrations. The prediction performance of the model on the test set was evaluated using two evaluation metrics: RMSE and Pearson's r.

[0018] Furthermore, the scenario analysis dataset is as follows:

[0019]

[0020] Where X′ represents an atmospheric pollutant concentration data vector containing n atmospheric pollutant concentration data points for the 24 hours before the preset time start t, the preset time start t, and the time after the preset time start t; Y his,i Z′ represents the historical concentration vector of predicted labeled air pollutants, which includes the predicted concentration of air pollutants 24 hours before the preset time start point t; Z′ represents the traffic flow data vector, which includes n traffic flow data points 24 hours before the preset time start point t, the preset time start point t, and the preset time start point t after the preset time start point t; U′ represents the meteorological data vector, which includes n meteorological data points 24 hours before the preset time start point t, the preset time start point t, and the preset time start point t after the preset time start point t.

[0021] Further, a traffic flow attenuation gradient set for scenario analysis is determined, and the traffic flow data after the preset time starting point in the scenario analysis dataset is adjusted. The predicted value vectors of labeled air pollutant concentrations corresponding to each attenuation coefficient are then obtained, including:

[0022] The first round of the loop is defined, including:

[0023] Read the data input tensor corresponding to the preset time node t from the scenario analysis dataset, including the preset time node t and the data 24 hours before the preset time node t in the scenario analysis dataset;

[0024] The data input tensor corresponding to the preset time node t is input into the trained traffic environmental pollutant concentration change scenario analysis model to obtain the label air pollutant concentration prediction value corresponding to the preset time node t, and the label air pollutant concentration prediction value corresponding to the preset time node t is filled into the scenario analysis dataset to update the scenario analysis dataset.

[0025] Read the data input tensor corresponding to time node t+1 from the updated scenario analysis dataset, including time node t+1 and the data of 24 hours before time node t+1 in the updated scenario analysis dataset;

[0026] The data input tensor corresponding to time node t+1 is input into the trained traffic environmental pollutant concentration change scenario analysis model to obtain the label air pollutant concentration prediction value corresponding to time node t+1. The label air pollutant concentration prediction value corresponding to time node t+1 is then filled into the scenario analysis dataset and the scenario analysis dataset is updated again.

[0027] Read the data input tensor corresponding to time node t+2 from the updated scenario analysis dataset, including time node t+2 and the data of 24 hours before time node t+2 in the updated scenario analysis dataset;

[0028] The data input tensor corresponding to time node t+2 is input into the trained traffic environmental pollutant concentration change scenario analysis model to obtain the label air pollutant concentration prediction value corresponding to time node t+2, and the label air pollutant concentration prediction value corresponding to time node t+2 is filled into the scenario analysis dataset and the scenario analysis dataset is updated again.

[0029] By analogy, n predicted values ​​of air pollutant concentrations for each label are obtained, forming a vector of predicted air pollutant concentrations for each label; where n represents the n time nodes following the preset time node t.

[0030] Starting from a preset time point t, the following second round of looping process is repeated, including:

[0031] Obtain each attenuation coefficient in the scenario analysis traffic flow attenuation gradient set, and adjust the traffic flow data after the preset time starting point t in the scenario analysis dataset according to each attenuation coefficient to obtain the adjusted scenario analysis dataset. Then, execute the first round of loop process to obtain the label air pollutant concentration prediction value vector corresponding to each attenuation coefficient.

[0032] Furthermore, the traffic flow data after the preset time starting point t in the scenario analysis dataset is adjusted according to various attenuation coefficients, including:

[0033] Calculate the difference between value 1 and each attenuation coefficient to obtain the adjustment coefficient corresponding to each attenuation coefficient;

[0034] Each adjustment coefficient is multiplied by the traffic flow data after the preset time starting point t in the scenario analysis dataset to obtain the adjusted scenario analysis dataset.

[0035] Furthermore, based on the predicted atmospheric pollutant concentration vector corresponding to each attenuation coefficient, the reduction rate of the labeled pollutant concentration corresponding to each attenuation coefficient is calculated, including:

[0036] For the j-th attenuation coefficient, the formula for calculating the corresponding reduction rate of the labeled pollutant concentration is as follows:

[0037]

[0038] Where, λ j Let y be the rate of decrease in the concentration of the labeled pollutant corresponding to the j-th attenuation coefficient. i For each time node i in the first cycle, y represents the predicted concentration of air pollutants in the label. i,j This represents the predicted concentration of atmospheric pollutants at the time node i corresponding to the j-th decay coefficient during the second cycle.

[0039] The beneficial effects of this invention include: using LSTM and CNN to build a deep learning model that can accurately predict hourly concentrations of traffic environmental pollutants; using the trained model to predict the concentrations of traffic environmental pollutants after changes in traffic flow, clarifying the response relationship between traffic environmental pollutant concentrations and traffic flow; and providing effective technical support for traffic environmental pollutant management and traffic flow management by predicting the concentrations of traffic environmental pollutants under different traffic flow conditions. Attached Figure Description

[0040] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below:

[0041] Figure 1This is a flowchart of a traffic environment pollutant concentration change scenario analysis method based on deep learning provided by the present invention;

[0042] Figure 2 This is a schematic diagram of the CNN-LSTM structure;

[0043] Figure 3 It is a curve showing the relationship between the reduction rate of traffic flow and the reduction rate of NO2 corresponding to the average NO2 concentration over the next four days;

[0044] Figure 4 This is a curve showing the relationship between the reduction rate of traffic flow and the reduction rate of NO2 corresponding to the average NO2 concentration over two representative days. Detailed Implementation

[0045] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0046] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0047] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0048] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0049] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0050] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0051] To illustrate the technical solution described in this application, specific embodiments will be described below.

[0052] Given the autocorrelation between traffic pollutant concentrations and traffic flow, meteorological factors, and pollutants, this embodiment employs deep learning to establish the response relationship between traffic air pollutant concentrations and traffic flow. Therefore, this embodiment provides a deep learning-based method for traffic environment scenario analysis. Figure 1 As shown, it includes:

[0053] Step 1: Acquire meteorological data, traffic flow data, and air pollutant data at monitoring points, and construct a fused dataset, which includes a training set and a test set:

[0054] Acquire meteorological data, traffic flow data, and air pollutant data at monitoring points. These monitoring points can be traffic stations. Meteorological data may include wind speed, wind direction, temperature, humidity, and pressure. Traffic flow data may include total traffic flow, small vehicle flow, medium vehicle flow, large vehicle flow, and vehicle speed. Air pollutant data may include NO, NO2, NOx, and PM2.5. 10 PM 2.5 O3, CO, BC, etc.

[0055] Constructing a scenario analysis model for changes in traffic environmental pollutant concentrations and training a fusion dataset The training set is then divided according to a ratio θ = 0.2. and test set

[0056] To facilitate subsequent processing, the collected data will be arranged according to... Preprocess and normalize to the range (0, 1).

[0057] Step 2: Construct a scenario analysis model for traffic environmental pollutant concentration changes based on an LSTM (Long Short-Term Memory) network and a CNN (Convolutional Neural Network). Obtain the single-input tensor of the model from the training set and train the model accordingly.

[0058] The traffic environmental pollutant concentration change scenario analysis model proposed in this embodiment employs deep learning to establish the model. A major drawback of traditional neural networks is their inability to persist information, i.e., to utilize previously learned information to understand and learn new information. Recurrent neural networks solve this problem. The key to these successes is the use of Long Short-Term Memory (LSTM) networks, a very special type of recurrent neural network capable of learning long-term dependencies, such as... Figure 2 The diagram includes an LSTM structure.

[0059] All recurrent neural networks (RNNs) take the form of a chain of repeating neural network modules. In a standard RNN, this repeating module has a very simple structure, such as a single Tanh layer. LSTMs also have this chain-like structure, but the repeating modules have different structures. Unlike standard RNNs, LSTMs have four neural network layers that interact in a very specific way. In this embodiment, the LSTM model is used as the base model to extract long-term data dependencies from meteorological data, traffic flow data, and traffic pollutant data. Simultaneously, a convolutional neural network (CNN) layer is added to the model to extract short-term variation features in the data, achieving more accurate changes in traffic environmental pollutant concentrations, thereby providing better technical support for traffic management departments.

[0060] Design a scenario analysis model structure for traffic environment pollutant concentration changes based on LSTM-CNN. The model structure includes an input layer, an LSTM layer, a CNN layer, a dropout layer, and a fully connected layer for outputting prediction results.

[0061] Based on the training set Obtain the model's single-input tensor, which includes: an atmospheric pollutant concentration data vector excluding the predicted labels, a historical atmospheric pollutant concentration vector for the predicted labels, a traffic flow data vector, and a meteorological data vector. As a specific implementation, the model's single-input tensor Q is shown below:

[0062]

[0063] If we take time node t as the current time node, then X represents a vector of atmospheric pollutant concentration data, excluding the prediction label, containing the current time node; Y his,i Y represents the vector of historical atmospheric pollutant concentrations predicted by labels over a 24-hour period.t,i Z represents the predicted atmospheric pollutant concentration at the current time point, U represents the traffic flow data vector, and p represents the backtracking time. This model is 24 hours. In this embodiment, each hour corresponds to one time point, so 24 hours corresponds to 24 time points.

[0064] In this embodiment, the dimension of the model's single input tensor Q is 18*24, including 18 input features and historical data of 24 hours each time the model backtracks.

[0065] Training a scenario analysis model for changes in traffic environmental pollutant concentrations includes the following steps:

[0066] The model forward propagation calculates the predicted values ​​of labeled pollutant concentrations in the training set.

[0067] The mean squared error between the predicted and observed values ​​is calculated as the model's loss value.

[0068] Backpropagate in the direction where the loss value decreases to update the model parameters Θ, Θ = Θ + ΔΘ, until the loss value Loss is less than the preset accuracy σ;

[0069] Output the trained traffic environment pollutant concentration change scenario analysis model REPAM (Road Environment Pollution Analysis Model);

[0070] Using the test set Substitute the values ​​into the trained scenario analysis model for changes in traffic environmental pollutant concentrations to calculate the model's predicted values. The predictive performance of the traffic environmental pollutant concentration change scenario analysis model on the test set was evaluated using two evaluation metrics: RMSE and Pearson's r.

[0071] The formulas for calculating RMSE and Pearson's r are as follows:

[0072]

[0073]

[0074] in These represent the observed value, predicted value, observed average value, and predicted average value of the pollutant, respectively.

[0075] Repeat the above steps until the model predictions achieve good accuracy before performing scenario analysis and prediction.

[0076] Step 3: Construct a scenario analysis dataset based on a preset time starting point:

[0077] Set the preset time start point as the preset time start point t.

[0078] The scenario analysis dataset Q′ is:

[0079]

[0080] Where X′ represents an atmospheric pollutant concentration data vector containing n atmospheric pollutant concentration data points for the 24 hours before the preset time start t, the preset time start t, and the time after the preset time start t; Y his,i Z′ represents the historical concentration vector of predicted labeled air pollutants, which includes the predicted concentration of air pollutants 24 hours before the preset time start point t; Z′ represents the traffic flow data vector, which includes n traffic flow data points 24 hours before the preset time start point t, the preset time start point t, and the preset time start point t after the preset time start point t; U′ represents the meteorological data vector, which includes n meteorological data points 24 hours before the preset time start point t, the preset time start point t, and the preset time start point t after the preset time start point t.

[0081] In this embodiment, the scenario analysis dataset Q′ includes all parameters of 18 model input features for 24 hours before the preset time start t, and parameter values ​​of 17 features (excluding the predicted label) for n consecutive time nodes after the preset time start t.

[0082] Step 4: Determine the traffic flow attenuation gradient set for scenario analysis, adjust the traffic flow data after the preset time starting point in the scenario analysis dataset, and obtain the predicted value vector of the labeled air pollutant concentration corresponding to each attenuation coefficient:

[0083] The first round of the loop is defined, including:

[0084] Read the data input tensor Q1 corresponding to the preset time node t from the scenario analysis dataset. This includes the preset time node t and the data from the 24 hours prior to the preset time node t in the scenario analysis dataset. The specific details of the data input tensor Q1 are as follows:

[0085]

[0086] The data input tensor Q1 is fed into the trained traffic environmental pollutant concentration change scenario analysis model to obtain the predicted value y of the labeled air pollutant concentration corresponding to the preset time node t. t And label the predicted value of air pollutant concentration y t Fill in the scenario analysis dataset Q′ and update the scenario analysis dataset Q′.

[0087] Read the data input tensor Q2 corresponding to time node t+1 from the updated scenario analysis dataset Q′. The data input tensor Q2 includes the data at time node t+1 and the data in the 24 hours before time node t+1 from the updated scenario analysis dataset Q′. The specific details of the data input tensor Q2 are as follows:

[0088]

[0089] The data input tensor Q2 is fed into the trained traffic environmental pollutant concentration change scenario analysis model to obtain the predicted value y of the labeled air pollutant concentration corresponding to time node t+1. t+1 And label the predicted value of air pollutant concentration y t+1 The scenario analysis dataset Q′ is then updated again by filling it into the scenario analysis dataset.

[0090] Read the data input tensor Q3 corresponding to time node t+2 from the updated scenario analysis dataset Q′. The data input tensor Q3 includes the data of time node t+2 and the data of 24 hours before time node t+2 in the updated scenario analysis dataset Q′.

[0091] The data input tensor Q3 is fed into the trained traffic environmental pollutant concentration change scenario analysis model to obtain the predicted value y of the labeled air pollutant concentration corresponding to the time node t+2. t+2 And label the predicted value of air pollutant concentration y t+2 The scenario analysis dataset Q′ is then populated and updated again.

[0092] By analogy, n predicted values ​​of air pollutant concentrations for each label are obtained, which constitutes a vector of predicted air pollutant concentrations from time node t to time node t+n. This vector is called the original predicted value vector. Here, n represents the n time nodes after the preset time node t.

[0093] Starting from a preset time point t, the following second round of looping process is repeated, including:

[0094] Obtain the set of traffic flow attenuation gradients Ω = {ω1, ω2, ..., ω} for scenario analysis. k}, where ω is the attenuation coefficient corresponding to different scenarios, thus obtaining k attenuation coefficients in the traffic flow attenuation gradient set Ω for scenario analysis. Each attenuation coefficient is a value between (0, 1).

[0095] The scenario analysis dataset Q′ is reconstructed, i.e., the scenario analysis dataset Q′ before the update is obtained. The traffic flow data after the preset time starting point t in the scenario analysis dataset Q′ is adjusted according to each attenuation coefficient to obtain the adjusted scenario analysis dataset. The first round of the above-mentioned loop process is then executed to obtain the label air pollutant concentration prediction value vector corresponding to each attenuation coefficient. Specifically, adjusting the traffic flow data after the preset time starting point t in the scenario analysis dataset Q′ according to each attenuation coefficient involves: calculating the difference between the value 1 and each attenuation coefficient to obtain the adjustment coefficient corresponding to each attenuation coefficient; then multiplying each adjustment coefficient by the traffic flow data after the preset time starting point t in the scenario analysis dataset Q′ to obtain the adjusted scenario analysis dataset. This is one specific implementation method.

[0096] Reconstruct the scenario analysis dataset Q′, keeping other data unchanged, and combine the traffic flow data after the preset time start t in the traffic flow dataset Z of the scenario analysis dataset Q′ with the attenuation coefficient ω1 to obtain Z′={z t-23 , z t-22 ,…,z t-1 , z t},(1-ω1)×{z t+1 …, z t+n Then, following the first round of iterations described above, the adjusted scenario analysis dataset Q′ is used to obtain the vector of predicted atmospheric pollutant concentrations Y1 corresponding to the attenuation coefficient ω1. t 1 y t+1 1 , ..., y t+n 1}

[0097] Then, keeping other data unchanged, the traffic flow data after the preset time starting point t in the traffic flow dataset Z from the scenario analysis dataset Q′ is combined with the attenuation coefficient ω2 to obtain Z′={z t-23 , z t-22 ,…,z t-1 , z t},(1-ω2)×{z t+1 …, z t+n Then, following the first round of iterations, the adjusted scenario analysis dataset Q′ is used to obtain the vector of predicted atmospheric pollutant concentrations corresponding to the attenuation coefficient ω2, Y2={y t 2 y t+1 2 , ..., y t+n 2}

[0098] Following the above pattern, each attenuation coefficient of the attenuation gradient set Ω is substituted into the matrix to obtain the predicted value vector of labeled air pollutant concentrations under all traffic flow attenuation scenarios from time node t to time node t+n, thus obtaining the labeled air pollutant concentration matrix:

[0099]

[0100] Step 5: Based on the predicted atmospheric pollutant concentration vector corresponding to each attenuation coefficient, calculate the reduction rate of the pollutant concentration corresponding to each attenuation coefficient.

[0101] For the j-th attenuation coefficient, the formula for calculating the corresponding reduction rate of the labeled pollutant concentration is as follows:

[0102]

[0103] Where, λ j Let y be the rate of decrease in the concentration of the labeled pollutant corresponding to the j-th attenuation coefficient, where j = 1…k. i For the first cycle (i.e., in the original predicted value vector), y represents the predicted concentration of air pollutants at time node i corresponding to the label. i，j This represents the predicted concentration of atmospheric pollutants at the time node i corresponding to the j-th decay coefficient during the second cycle.

[0104] Finally, the traffic flow attenuation rates ω1, ω2, ..., ω are obtained. k The following is a set of label pollutant concentration reduction rates:

[0105] Λ={λ1,…,λ j , …, λ k}

[0106] Taking NO2 in the traffic environment as an example, the NO2 reduction scenario under the traffic flow attenuation rate Ω = {5%, 10%, ..., 95%} was analyzed when selecting the scenario analysis. n = 96h (i.e., within four days of traffic flow change).

[0107] Figure 3 The curve shows the relationship between the reduction rate of traffic flow and the reduction rate of NO2 corresponding to the average NO2 concentration over the next four days. Figure 4 The curves showing the relationship between the reduction rate of traffic flow and the reduction rate of NO2 corresponding to the average NO2 concentration over two representative days.

[0108] Therefore, the deep learning-based traffic environment pollutant concentration change scenario analysis method provided in this embodiment can realize the analysis of pollutant concentration scenarios in the traffic environment after traffic flow control, and provide a technical basis for policy formulation in scenarios such as heavy pollution weather warnings, which has good practical significance.

[0109] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for analyzing traffic environment pollutant concentration changes based on deep learning, characterized in that, include: Meteorological data, traffic flow data, and air pollutant data from monitoring points are acquired, and a fused dataset is constructed, which includes a training set and a test set. A scenario analysis model for traffic environmental pollutant concentration changes is constructed based on LSTM long short-term memory network and CNN convolutional neural network. The single input tensor of the model is obtained according to the fused dataset, and the scenario analysis model for traffic environmental pollutant concentration changes is trained. Construct a scenario analysis dataset based on a preset time starting point; Determine the traffic flow attenuation gradient set for scenario analysis, adjust the traffic flow data after the preset time starting point in the scenario analysis dataset, and obtain the label air pollutant concentration prediction value vector corresponding to each attenuation coefficient. Based on the predicted value vector of the air pollutant concentration on the label corresponding to each attenuation coefficient, the reduction rate of the pollutant concentration on the label corresponding to each attenuation coefficient is calculated. Determine the traffic flow attenuation gradient set for scenario analysis, adjust the traffic flow data after the preset time starting point in the scenario analysis dataset, and obtain the label air pollutant concentration prediction value vector corresponding to each attenuation coefficient, including: The first round of the loop is defined as follows: Read from the scenario analysis dataset and the preset time nodes. t The corresponding data input tensor includes the preset time nodes in the scenario analysis dataset. t And data from 24 hours prior to the preset time point t; Will be in accordance with the preset time node t The corresponding data input tensors are input into the trained traffic environmental pollutant concentration change scenario analysis model to obtain the data at the preset time nodes. t The corresponding predicted values ​​of air pollutant concentrations will be displayed and compared with the preset time points. t The corresponding predicted values ​​of atmospheric pollutant concentrations are filled into the scenario analysis dataset to update the scenario analysis dataset; Read the time nodes from the updated scenario analysis dataset. t +1 corresponds to the data input tensor, including the time nodes in the updated scenario analysis dataset. t +1 and time node t +1 data from the previous 24 hours; Will be related to time nodes t The data input tensor corresponding to +1 is input into the trained traffic environmental pollutant concentration change scenario analysis model to obtain the time node. t +1 corresponds to the predicted concentration of air pollutants in the label, and will be compared with the time node. t The predicted atmospheric pollutant concentration value corresponding to +1 is filled into the scenario analysis dataset and the scenario analysis dataset is updated again. Read the time nodes from the updated scenario analysis dataset. t +2 corresponds to the data input tensor, including the time nodes in the updated scenario analysis dataset. t +2 and time nodes t +2 data from the previous 24 hours; Will be related to time nodes t The data input tensor corresponding to +2 is input into the trained traffic environmental pollutant concentration change scenario analysis model to obtain the time node. t +2 corresponds to the predicted concentration of air pollutants in the label, and will be compared with the time node. t The predicted atmospheric pollutant concentration values ​​corresponding to +2 are filled into the scenario analysis dataset and the scenario analysis dataset is updated again. And so on, we get n The predicted concentrations of air pollutants from each label constitute a vector of predicted air pollutant concentrations from each label; where... n Preset time nodes t After n Each time point; From the preset time starting point t To begin, the second round of the loop will proceed as follows: Obtain each attenuation coefficient from the traffic flow attenuation gradient set in the scenario analysis, and apply each attenuation coefficient to the preset time starting point in the scenario analysis dataset. t The subsequent traffic flow data is adjusted to obtain the adjusted scenario analysis dataset, and the first round of the loop process is executed to obtain the vector of predicted atmospheric pollutant concentrations corresponding to each attenuation coefficient.

2. The method for analyzing traffic environmental pollutant concentration changes based on deep learning according to claim 1, characterized in that, The model's single-input tensor includes: atmospheric pollutant concentration data vectors excluding prediction labels, historical atmospheric pollutant concentration vectors for prediction labels, traffic flow data vectors, and meteorological data vectors.

3. The method for analyzing traffic environmental pollutant concentration changes based on deep learning according to claim 1, characterized in that, Training the scenario analysis model for changes in traffic environmental pollutant concentrations includes: The model forward propagation calculates the predicted values ​​of the labeled pollutant concentrations in the training set; The mean squared error between the predicted and observed values ​​is calculated as the model's loss value; The model backpropagates in the direction where the loss value decreases to update the model parameters until the loss value is less than the preset accuracy. Output the trained scenario analysis model for changes in traffic environmental pollutant concentrations; The model's predicted values ​​were calculated by substituting test set data into the trained scenario analysis model for changes in traffic environmental pollutant concentrations. The prediction performance of the model on the test set was evaluated using two evaluation metrics: RMSE and Pearson's r.

4. The method for analyzing traffic environmental pollutant concentration changes based on deep learning according to claim 1, characterized in that, The scenario analysis dataset is as follows: ； in, This indicates that a preset time start point is included. t 24 hours prior, preset start time t and preset time start t After n A vector of atmospheric pollutant concentration data; This indicates that a preset time start point is included. t The vector of historical concentrations of predicted labeled air pollutants in the previous 24 hours. This indicates that a preset time start point is included. t 24 hours prior, preset start time t and preset time start t After n A traffic flow data vector of traffic flow data; This indicates that a preset time start point is included. t 24 hours prior, preset start time t and preset time start t After n A meteorological data vector of meteorological data.

5. The method for analyzing traffic environmental pollutant concentration changes based on deep learning according to claim 1, characterized in that, Based on each attenuation coefficient, the preset time starting point in the scenario analysis dataset is... t Subsequent traffic flow data will be adjusted, including: Calculate the difference between value 1 and each attenuation coefficient to obtain the adjustment coefficient corresponding to each attenuation coefficient; Each adjustment coefficient is multiplied by the preset time starting point in the scenario analysis dataset. t The subsequent traffic flow data yielded an adjusted scenario analysis dataset.

6. The method for analyzing traffic environmental pollutant concentration changes based on deep learning according to claim 1, characterized in that, Based on the predicted atmospheric pollutant concentration vectors corresponding to each attenuation coefficient, the reduction rate of pollutant concentration corresponding to each attenuation coefficient is calculated, including: For the j The formula for calculating the reduction rate of the label pollutant concentration corresponding to each attenuation coefficient is as follows: ； in, For the first j The rate of reduction in the concentration of the labeled pollutant corresponding to each attenuation coefficient For the time nodes in the first cycle i The corresponding predicted values ​​of air pollutant concentrations on the label, For the second round of the cycle, the first j The time points corresponding to each attenuation coefficient i The corresponding predicted values ​​of atmospheric pollutant concentrations.

Images