Vehicle emission estimation method and system based on sparse trajectory

Abstract

The invention discloses a sparse trajectory-based vehicle emission estimation method and system. The method comprises the steps of obtaining a vehicle driving trajectory; vehicle specific power and engine stress are calculated, and corresponding driving working conditions are determined; classifying the vehicle driving track according to a sampling frequency to obtain a sparse track fragment data set; a data driving model is constructed, and training is carried out by using the sparse trajectory fragment data set; inputting a sparse trajectory to be measured into the trained driving condition time distribution prediction model to obtain predicted driving condition time distribution; gas emissions for the travel segment are estimated based on an IVE physical model. The system comprises a data acquisition module, a data preprocessing module, a model training module, a driving condition time distribution prediction module and an emission estimation module. By using the method and the device, the emission data of the current track fragment can be accurately predicted. The method can be widely applied to the field of traffic emission estimation.

Description

Technical Field

[0001] The present invention relates to the field of quantitative estimation of traffic emissions, and in particular to a vehicle emission estimation method and system based on sparse trajectories. Background Art

[0002] Motor vehicle emissions have greatly aggravated air pollution and climate change. The atmospheric pollutants such as nitrogen oxides, hydrocarbons, and carbon monoxide contained in them have directly worsened the air quality of cities. Therefore, it is necessary to pay close attention to vehicle emission data. At present, it is still difficult to accurately quantify individual vehicle emissions on a large scale. The reason is that the traffic emission models that quantify vehicle emissions are mainly divided into two categories, and they both have certain limitations. The two types of models are divided into models based on driving conditions and models based on average speed according to the main parameters: the model based on driving conditions divides the driving conditions and determines the corresponding emission factors by combining the vehicle driving trajectory recorded second by second with the past twenty seconds and the current micro-driving state. The model based on average speed takes the average speed as the core input parameter, and establishes a mapping relationship between the average speed and the emission factor under different circumstances according to the vehicle type and road type. For vehicle emission prediction, although the model based on average speed is widely used due to its low input data acquisition cost and wide time coverage, it cannot accurately quantify the emissions under the micro-driving state, and its accuracy is lower than that of the model based on driving conditions. It can be seen from this that the key to accurately quantifying individual vehicle emissions on a large scale is to be able to obtain a large amount of vehicle driving data that meets the input requirements of the model based on driving conditions.

[0003] With the development of intelligent transportation systems (ITS), the acquisition of vehicle driving data has become more convenient. However, the traffic data collected by various data collection devices are often sparse driving trajectories of vehicles. Sparse trajectories refer to vehicle driving trajectory data from which the vehicle's second-by-second driving status cannot be known due to the low data sampling frequency or sparse recording points during vehicle driving. The instantaneous data of the on-board equipment is recorded and uploaded at certain time intervals. Although it has a large time and space coverage compared to the second-by-second driving trajectory, and the acquisition equipment, data storage and transmission costs are low, it is impossible to know the second-by-second driving status of the vehicle, which makes it difficult to accurately estimate the driving conditions and the emission estimation has a large deviation. If massive sparse trajectories can be used to accurately quantify emissions, it will be of great significance to the compilation and updating of road emission inventories. Summary of the invention

[0004] In view of this, in order to solve the technical problem that the driving conditions are difficult to estimate in the existing vehicle emission estimation method based on sparse trajectories, which leads to a large deviation in emission estimation, the present invention proposes a vehicle emission estimation method based on sparse trajectories, the method comprising the following steps:

[0005] Acquire a vehicle driving trajectory and define a vehicle type corresponding to the vehicle driving trajectory;

[0006] In this step, if the vehicle's driving trajectory is different from the source vehicle type of the sparse trajectory to be tested later, or the driving state is not similar, the prediction result of the model will be affected, so the corresponding vehicle type needs to be marked.

[0007] Based on the vehicle driving trajectory, the vehicle specific power VSP (vehicle specific power) and the engine stress ES (engine stress) are calculated, and the corresponding driving conditions are determined;

[0008] This step also includes removing abnormal samples in the vehicle's driving trajectory and recalculating the second-by-second acceleration, speed, and slope.

[0009] Classifying the vehicle driving trajectory according to sampling frequency to obtain a sparse trajectory segment data set;

[0010] In this step, the second-by-second trajectory is divided into sparse trajectory segments with different sampling frequencies, and the average speed and average slope of each sparse trajectory segment under different sampling frequencies are calculated. The processed data are summarized into sparse trajectory segment data sets with different sampling frequencies.

[0011] Constructing a data-driven model and using the sparse trajectory segment data set for training to obtain a trained driving condition time distribution prediction model;

[0012] In this step, the sparse trajectory segment dataset is processed by sliding the time window to generate multiple time windows, each of which contains features and label matrices of multiple time steps.

[0013] Inputting the sparse trajectory to be tested into the trained driving condition time distribution prediction model to obtain the predicted driving condition time distribution;

[0014] Based on the IVE physical model, the gas emission of the driving segment is estimated according to the predicted driving cycle time distribution.

[0015] In some embodiments, the data-driven model is a data-driven model of a seq2seq framework based on an attention mechanism, including an encoder, a decoder, an attention mechanism, and a fully connected layer.

[0016] First, the encoder encodes the average speed and average slope of the sparse trajectory segments, and then passes the encoder output to the decoder for decoding. The output of the decoder is weighted combined through the attention layer, and finally the output of the attention layer is passed to the fully connected layer to generate the predicted driving time distribution.

[0017] The present invention also proposes a vehicle emission estimation system based on sparse trajectories, the system comprising:

[0018] A data acquisition module, used to acquire the vehicle's driving trajectory;

[0019] A data preprocessing module, based on classifying the vehicle driving trajectory according to the sampling frequency, to obtain a sparse trajectory segment data set;

[0020] A model training module, used to construct a data-driven model and use the sparse trajectory segment data set for training to obtain a trained driving condition time distribution prediction model;

[0021] A driving condition time distribution prediction module, used for inputting the sparse trajectory to be tested into the trained driving condition time distribution prediction model to obtain a predicted driving condition time distribution;

[0022] The emission estimation module estimates the gas emission of the driving segment according to the predicted driving cycle time distribution based on the IVE physical model.

[0023] Based on the above scheme, the present invention provides a vehicle emission estimation method and system based on sparse trajectories, which quantifies sparse trajectory emissions through a data-physics hybrid driven model. The data-driven model establishes a mapping relationship between the driving characteristics contained in the sparse trajectory and the vehicle driving conditions. The model adopts a seq2seq model based on an attention mechanism, in which a sliding window reconstructs the input features of the model so that the input features contain multiple time steps, so that the model can consider the influence of historical segment driving characteristics in the process of predicting the current segment driving conditions; the physical model adopts a microscopic traffic emission model IVE, and the model receives the driving conditions containing the vehicle's microscopic driving state information and then quantifies multi-gas emissions, thereby incorporating the vehicle's historical driving characteristics and microscopic driving states into the influencing factors of emission quantification, thereby achieving the purpose of accurately predicting the current segment emissions. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 is a flow chart of the steps of a vehicle emission estimation method based on sparse trajectories of the present invention;

[0025] Figure 2 It is a schematic diagram of the feature and label matrix set of multiple time windows of the present invention;

[0026] Figure 3 It is a data flow diagram of the vehicle emission estimation method of the present invention. ;

[0027] Figure 4 It is a structural block diagram of a vehicle emission estimation system based on sparse trajectories of the present invention; DETAILED DESCRIPTION

[0028] At present, the mainstream solutions to the problem of quantifying individual vehicle emissions using sparse trajectories include dividing the sparse trajectory fragments into different average speed categories, using the known second-by-second trajectory and taking the median driving condition of each average speed category as the representative condition of the category to match the driving condition for the sparse trajectory; dividing the known second-by-second driving trajectory into different average speed intervals, and estimating the driving condition distribution of the sparse trajectory fragments by fitting the driving condition distribution function of each interval; or combining the cubic spline interpolation method with the random acceleration interpolation method to restore the driving condition of the sparse trajectory. However, the above methods ignore the impact of the historical driving characteristics of the vehicle during driving on the current driving condition and even emissions.

[0029] At the same time, although the traffic emission model based on average speed can calculate emissions through sparse trajectories, it ignores the impact of the vehicle's microscopic driving state on emissions. The defects of the above sparse trajectory emission quantification method lead to poor vehicle emission estimation results.

[0030] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0031] It should be noted that, for the convenience of description, only the parts related to the relevant invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other.

[0032] It should be understood that the "system", "device", "unit" and / or "module" used in this application is a method for distinguishing different components, elements, parts, portions or assemblies at different levels. However, if other words can achieve the same purpose, the word can be replaced by other expressions.

[0033] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "a kind" and / or "the" do not refer to the singular, but also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of clearly identified steps and elements, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements. The elements defined by the sentence "includes a..." do not exclude the existence of other identical elements in the process, method, commodity or device that includes the elements.

[0034] In the description of the embodiments of the present application, "plurality" means two or more than two. The following terms "first" and "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features.

[0035] In addition, flow charts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, the various steps may be processed in reverse order or simultaneously. At the same time, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0036] Reference Figure 1 , which is a flow chart of an optional example of a vehicle emission estimation method based on sparse trajectories proposed in the present invention, which method can be applied to a computer device. The imaging method proposed in this embodiment may include but is not limited to the following steps:

[0037] Step S1, obtaining the vehicle driving trajectory;

[0038] Step S2, based on the vehicle driving trajectory, calculating the vehicle specific power and engine stress, and determining the corresponding driving condition;

[0039] Step S3, classifying the vehicle driving trajectory according to the sampling frequency to obtain a sparse trajectory segment data set;

[0040] Step S4, constructing a data-driven model and using the sparse trajectory segment data set for training to obtain a trained driving condition time distribution prediction model;

[0041] Step S5, inputting the sparse trajectory to be measured into the trained driving condition time distribution prediction model to obtain the predicted driving condition time distribution;

[0042] Step S6: based on the IVE physical model, estimate the gas emission of the driving segment according to the predicted driving condition time distribution.

[0043] The present invention proposes an individual vehicle emission estimation method for sparse trajectories. The data comes from heavy fuel vehicles in Chongqing. The data is GPS trajectory data with a sampling frequency of one second. Chongqing has special geographical conditions. The road slope in the city is large and the variation is significant. Therefore, the data in the city can be used to deeply explore the impact of road slope on vehicle driving conditions and even emissions. The specific method of the present invention is to divide the second-by-second driving trajectory data into driving segments, and calculate the average speed, average slope, and time distribution of the driving condition of the vehicle in each driving segment through the driving characteristics in the second-by-second state. The average speed and average slope are used as the features of the model, and the time distribution of the driving condition is used as the label. The seq2seq data-driven model based on the attention mechanism is trained. The model can use the features available in the sparse trajectory: the average speed of the segment, the average slope of the segment, to predict the time distribution of the driving condition, and finally input it into the emission estimation physical model to obtain the segment emissions of individual vehicles.

[0044] In some feasible embodiments, the step S1 specifically includes:

[0045] Obtain vehicle driving trajectories with a sampling frequency of one second at a time, clarify the vehicle type or driving status of the source of the trajectory data, and the second-by-second driving trajectory is processed as model training data. If the second-by-second trajectory is different from the source vehicle type of the sparse trajectory to be tested later, or the driving status is not similar, the prediction result of the model will be affected. The differences and similarities between vehicle types and driving status refer to the following classification standards. Vehicle types are divided into two categories: passenger cars and trucks. Further, passenger cars are subdivided into micro, small, taxis, medium, large, and buses, and trucks are subdivided into micro, light, medium, heavy, low-speed trucks, and three-wheeled vehicles. If the vehicle types are different, the two trajectories must come from vehicles with similar driving status. Specifically, the vehicle's driving speed threshold, acceleration threshold, and vehicle activity area slope threshold are similar.

[0046] In some feasible embodiments, the step S2 specifically includes:

[0047] The speed threshold [1,120] km / h and the slope threshold [-0.14,0.14]° are used as the characteristic normal thresholds to remove samples with abnormal speed and slope in the second-by-second trajectory. Then, the vehicle specific power VSP (vehicle specific power) and engine stress ES (engine stress) are obtained by calculating the second-by-second speed, acceleration, and second-by-second slope of the second-by-second trajectory, and the driving condition of the second-by-second trajectory is determined by VSP and ES. The method of determining the driving condition of the second-by-second trajectory refers to the International Motor Vehicle Emission Model IVE. The second-by-second driving state of the vehicle is divided into 60 driving conditions, i.e., 60 Bins, according to VSP and ES. Each Bin uses a different condition correction factor, which reflects the impact of different driving conditions on vehicle emissions. In each driving segment, the time proportion of different second-by-second driving states constitutes the driving condition time distribution. The detailed calculation process and formula are as follows:

[0048] The vehicle specific power VSP of each second-by-second sample is calculated by the vehicle's second-by-second driving speed, slope, and acceleration. The formula is as follows:

[0049] VSP=v[1.1a+9.81(atan(sin(grade)))+0.132]+0.000302v 3

[0050] Where VSP is the vehicle specific power, kw / ton; v is the speed per second, m / s; a is the acceleration, m / s 2 ; h is the altitude, m; grade is the slope per second, calculated as (h t＝0 –h t＝-1 ) / v;h t＝0 is the altitude at this second, h t＝-1 is the altitude in the last second, m.

[0051] When dividing the driving conditions into second-by-second segment data sets, the first 20 samples do not participate in the calculation of the operating condition classification, because the acquisition of the engine stress ES requires the calculation of parameters such as the vehicle specific power and engine speed per minute in the first 20 seconds. The calculation formula is as follows:

[0052] ES=RPMIndex+0.08*PreaveragePower

[0053] Wherein, ES is the engine stress, dimensionless; PreaveragePower is the average value of VSP in the first 20 seconds, kW / ton; RPMIndex is the engine speed index per minute, calculated as (v / speed classifier (dimensionless)); the minimum value of RPMIndex is 0.9.

[0054] Table 1 Cut points used in RPMIndex calculation

[0055]

[0056] After obtaining VSP and ES, the two parameters jointly determine the Bin category. ES is divided into three intervals: low, medium, and high load. Combined with the 20 intervals divided by VSP, the specific Bin category can be determined for the second-by-second trajectory.

[0057] Table 2 Correspondence between VSP, ES and Bin

[0058]

[0059]

[0060] In some embodiments, the step S3 specifically includes:

[0061] The second-by-second trajectories are divided into sparse driving trajectory segments with six sampling frequencies of 10 seconds, 20 seconds, 30 seconds, 40 seconds, 50 seconds and 60 seconds. The average speed, average slope and time distribution of driving conditions of each sparse driving trajectory segment under the six sampling frequencies are calculated based on the driving characteristics of its predecessor second-by-second trajectory. The processed data are summarized into sparse trajectory segment data sets with different sampling frequencies.

[0062] In some embodiments, the step S4 specifically includes:

[0063] The average speed and average slope in the sparse trajectory segment data set of each sampling frequency are used as features, and the driving condition time distribution is used as a label. To ensure the accuracy of vehicle emission prediction, the current emission of the vehicle is affected by the driving state of the previous 20 seconds in the time window feature engineering. Under the six sampling frequencies selected in step S3, the step size in the time window is set to 2 or 3, so as to better retain the feature information of the historical driving segment and transfer it to the current segment, thereby improving the model prediction performance.

[0064] The dataset is processed by sliding the time window to generate multiple time windows, each of which contains feature and label matrices of multiple time steps. These matrices serve as the basic input units of the model, such as Figure 2 As shown. Next, the matrix set is divided into training data set, validation data set and test data set in a ratio of 8:1:1. In the training and validation stages of the model, the features and labels in the training and validation data sets are input into the model, while in the prediction stage, only the feature matrix is ​​input into the test data set, and the labels are estimated by the physical model through the emission values ​​of the second-by-second driving trajectory fragments, which are used to compare with the prediction results of the model.

[0065] In some embodiments, the step S4 further includes:

[0066] Construct a data-driven model based on the seq2seq framework of the attention mechanism. The model structure is as follows Figure 2 As shown in the figure, it includes an encoder and decoder composed of GRU, as well as an attention mechanism and a fully connected layer. The encoder captures the key information in the historical driving characteristics of the vehicle by encoding the input feature sequence into a hidden state sequence. The decoder receives the label of the previous time step and decodes the encoder's hidden state sequence into a target hidden state sequence. The attention mechanism dynamically assigns the weight of the encoder output hidden state, emphasizing the important information in the feature to assist in the generation of the prediction label. The fully connected layer combines the output of the attention mechanism with the decoder hidden state, and generates the final prediction label through a function after linear transformation.

[0067] For each time step t, the encoder GRU receives the feature vector x of the current time step t and the hidden state h of the previous time step t-1 , calculate the update gate z t and reset gate r t , whose update gate z t Decision t-1 How much information needs to be retained, reset gate r t Determine the hidden state h at the current time step t How much information needs to be reset. The hidden state update process of the encoder is as follows:

[0068] Z t =σ(W z ·x t +U z ·h t-1 +b z )

[0069] r t =σ(W r ·x t +U r ·h t-1 +b r )

[0070]

[0071] In the formula, σ is the sigmoid function, W is the input weight matrix; U is the hidden state weight matrix; b is the bias; ⊙ represents element-by-element multiplication. The overall process can be simplified as:

[0072] h t enc =GRU enc (x t ,h t-1 enc )

[0073] The overall process of the decoder can be simplified as follows:

[0074] h t dec =GRU dec (h T enc / y t-1 ,h t-1 dec )

[0075] In the formula, h t dec is the hidden state of the decoded current time step output; h T enc is the hidden state output by the encoder at the last time step, and is used as the input of the decoder when t = 1; t-1 is the label vector of the previous time step, which serves as the input of the decoder when t>1.

[0076] Since feature information will be lost to a certain extent during the encoding process, the model adds an attention mechanism so that the decoder can dynamically refer to all hidden states of the encoder when generating the output of each time step, thereby alleviating the problem of information loss during compression and transmission. At the same time, in the driving condition prediction, the input features of different time steps have different effects on the current decoder hidden state. The attention mechanism generates weights for the encoder hidden state at each time step, indicating the correlation with the current decoder hidden state, so that the model can more accurately receive feature information of multiple time steps and improve the accuracy of the predicted label. The attention mechanism calculates the current hidden state of the decoder h t dec and the encoder hidden state h at each time step t′ enc The alignment score between t，t′ , and then convert the score into weight α through the softmax function t，t′ , the process is as follows:

[0077] score t，t′ =h t dec ·h t′ enc

[0078]

[0079] The weighted sum of the hidden states of each time step of the encoder is used to obtain the context vector context of the current time step t , the process is as follows:

[0080]

[0081] The decoder’s current hidden state h tdec With context vector context t After splicing, the softmax function of the fully connected layer is used to generate the predicted label of the current time step as follows:

[0082]

[0083] During the model training phase, the feature vectors in the training data set are gradually input into the encoder to generate a hidden state sequence containing the vehicle's driving history feature information. At the same time, the hidden state of the final time step is input into the decoder. When the decoder gradually generates the hidden state sequence, the label vector of the previous time step is passed to the decoder as input to provide a reference for outputting the hidden state of the current time step, helping the model to learn the mapping relationship between input and output faster. Then the attention mechanism dynamically assigns weights to each time step of the decoder. These weights represent the correlation between the current time step of the decoder and each time step of the encoder. The weights and the encoder output sequence are processed by the attention layer to generate a context vector. The context vector and the decoder output sequence are processed through the fully connected layer to generate a predicted label. The predicted label is then compared with the training set label, the loss is calculated, and then backpropagation is performed based on the loss to update the model parameters.

[0084] In the model validation phase, the feature matrix and label matrix in the validation dataset are input into the encoder and decoder, the validation loss is calculated, and the hyperparameters such as the number of GRU layers, the number of GRU neurons, and the learning rate are tuned.

[0085] In some feasible embodiments, the step S6 specifically includes:

[0086] The driving time distribution of the sparse trajectory segments predicted by the data-driven model is input into the emission estimation model IVE, and the driving segment emission calculation formula of an individual vehicle is as follows:

[0087]

[0088] Where, E is the emission in the driving segment; is the average speed of the segment; T is the driving time of the segment; Q B is the basic emission factor of the vehicle; z represents the number of the driving condition category, is the frequency of the vehicle's driving condition category z in the segment (i.e., the proportion of the time when the driving condition is category z to the driving time in the segment); Q z is the correction factor for operating condition category z; Q C It is the correction factor of vehicle attribute C. Attribute C includes temperature, humidity, altitude, government management system, fuel quality, etc. It is determined according to the time and geographical area characteristics corresponding to the road section.

[0089] In the model prediction stage, the test data set is input into the hybrid drive model. The input of the encoder in the data-driven model is the feature vector of the test data set, and a segment of all-zero vectors is used as the initial input of the decoder to prompt the model to start generating output sequences. The prediction sequence generated by the fully connected layer is the driving time distribution of the sparse trajectory segment data set to be processed. Then, in the physical model, the driving time distribution affects the parameters in step S6. The parameter T is the sampling frequency of a specific sparse trajectory segment, and the gas emissions of each driving segment estimated by the sparse trajectory are calculated by the formula.

[0090] The seq2seq structure adopted by the present invention can effectively capture long-term and short-term dependencies through the encoder-decoder mechanism, ensuring that the model can work effectively in both the training and prediction stages and provide accurate prediction results. The seq2seq structure can encode the information of the entire input sequence into the hidden state of the final time step through the encoder, so that the decoder can refer to the global information when generating the output of each time step and capture long-term dependencies. In the training stage, through the teacher forcing method, when the decoder gradually generates the output sequence, the label of the previous time step is passed to the decoder as input, providing a reference for outputting the hidden state of the current time step, helping the model to learn the mapping relationship between input and output faster, making the training process of the model more efficient and stable. In the prediction stage, since there is no actual label, the decoder uses the predicted output of the previous time step as input at each time step, and generates the prediction of the current time step by autoregression.

[0091] Considering the features and labels in practice, the encoder input dimension is 2, the decoder output dimension is 60, and the seq2seq framework allows for mismatches between input and output dimensions. The traditional GRU model assumes that the input and output dimensions are the same, and it is difficult to process input and output sequences with different dimensions. In the seq2seq framework, the decoder output sequence can be flexibly generated based on the hidden state of the final time step generated by the encoder.

[0092] Modular design: The encoder and decoder in the seq2seq structure are independent modules that can be optimized and adjusted separately. The encoder can better extract feature information from the input sequence, and the decoder can better generate accurate prediction sequences. In the process of the encoder generating hidden states, although the information faces certain loss problems during compression and transmission, compared with traditional RNNs that do not use this structure, the independent encoder module is easy to optimize and can focus on how to best represent the input sequence.

[0093] In addition, the attention mechanism is integrated: feature information is inevitably lost in the process of the encoder generating the hidden state of the last time step. The attention mechanism allows the decoder to dynamically refer to all hidden states of the encoder when generating the output of each time step, rather than just the last hidden state, thereby alleviating the problem of information loss during compression and transmission.

[0094] In driving condition prediction, since the driving time of different segments is uncertain, the input features of different time steps may have different effects on the current time step. The attention layer can dynamically assign weights to each time step of the decoder. These weights represent the correlation between the current time step of the decoder and each time step of the encoder. The model can automatically select the most relevant input segment for reference, thereby improving the accuracy of the prediction. Compared with the fixed hidden state transmission method, the attention mechanism can more flexibly utilize the information in the input sequence, thereby capturing richer features.

[0095] Finally, the overall data flow of this method refers to Figure 3 .

[0096] Tables 3, 4, and 5 show the emission estimation effect in this case. Based on the second-by-second driving data of multiple heavy trucks for 97,100 seconds, each of them is divided into sparse trajectory segments with different sampling frequencies. The emission estimation adopts the data-physics hybrid driving method described in the present invention and the method of determining the driving condition by median accumulation described in the reference respectively. Method I is the technology of the present invention, and method II is a method of matching driving conditions with sparse trajectories by dividing the sparse trajectory segments into different average speed categories and taking the median driving condition of each average speed category of the known second-by-second trajectory as the representative condition of the category. The results show that the error evaluation indicators of the emission estimation method of the present invention are basically better than those of the method described in the reference. Only when the sampling frequency of the sparse trajectory is 60 seconds, the RMSE of CO and CO2 emissions of the method of the present invention is slightly inferior to that of the method described in the reference. In the sparse trajectories collected by mobile sensors, samples with a sampling frequency of 10 to 30 seconds are more common, while data with a sampling frequency greater than 30 seconds account for a small proportion. Therefore, the prediction limitations of the technology of the present invention at a sampling frequency of 60 seconds have less impact.

[0097] Table 3 CO emission estimation errors of various methods at different sampling frequencies

[0098]

[0099] Table 4 NOx emission estimation errors of various methods at different sampling frequencies

[0100]

[0101]

[0102] Table 5 CO2 emission estimation errors of various methods at different sampling frequencies

[0103]

[0104] like Figure 4 As shown, a vehicle emission estimation system based on sparse trajectories includes:

[0105] A data acquisition module, used to acquire the vehicle's driving trajectory;

[0106] A data preprocessing module, based on classifying the vehicle driving trajectory according to the sampling frequency, to obtain a sparse trajectory segment data set;

[0107] A model training module, used to construct a data-driven model and use the sparse trajectory segment data set for training to obtain a trained driving condition time distribution prediction model;

[0108] A driving condition time distribution prediction module, used for inputting the sparse trajectory to be tested into the trained driving condition time distribution prediction model to obtain a predicted driving condition time distribution;

[0109] The emission estimation module estimates the gas emission of the driving segment according to the predicted driving cycle time distribution based on the IVE physical model.

[0110] The contents of the above method embodiments are all applicable to the present system embodiments. The functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0111] A vehicle emission estimation device based on sparse trajectories:

[0112] at least one processor;

[0113] at least one memory for storing at least one program;

[0114] When the at least one program is executed by the at least one processor, the at least one processor implements the vehicle emission estimation method based on sparse trajectories as described above.

[0115] The contents of the above method embodiments are all applicable to the present device embodiments. The functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0116] A storage medium stores instructions executable by a processor, wherein the instructions executable by the processor are used to implement a vehicle emission estimation method based on sparse trajectories as described above when executed by the processor.

[0117] The contents of the above method embodiments are all applicable to the present storage medium embodiments. The functions specifically implemented by the present storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0118] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. A vehicle emission estimation method based on sparse trajectories, characterized in that: The following steps are involved: Obtain vehicle driving trajectory; Based on the vehicle driving trajectory, calculating the vehicle specific power and engine stress, and determining the corresponding driving condition; Classifying the vehicle driving trajectory according to sampling frequency to obtain a sparse trajectory segment data set; Constructing a data-driven model and using the sparse trajectory segment data set for training to obtain a trained driving condition time distribution prediction model; Inputting the sparse trajectory to be tested into the trained driving condition time distribution prediction model to obtain the predicted driving condition time distribution; Based on the IVE physical model, the gas emission of the driving segment is estimated according to the predicted driving cycle time distribution.

2. The vehicle emission estimation method based on sparse trajectories according to claim 1, characterized in that: The step of calculating the vehicle specific power and engine stress based on the vehicle driving trajectory and determining the corresponding driving condition specifically includes: Eliminating abnormal samples in the vehicle driving trajectory; Calculating the second-by-second acceleration, speed, and second-by-second slope of the vehicle's travel trajectory; calculating vehicle specific power and engine stress based on the second-by-second acceleration, speed, and second-by-second grade; The driving condition of the second-by-second trajectory is determined based on the vehicle specific power and the engine stress.

3. The vehicle emission estimation method based on sparse trajectories according to claim 1, characterized in that: The step of classifying the vehicle driving trajectory according to the sampling frequency to obtain a sparse trajectory segment data set specifically includes: Classifying the vehicle driving trajectory according to different sampling frequencies to obtain sparse trajectory driving segments; The average speed, average slope, and time distribution of the driving conditions corresponding to the sparse trajectory driving segments are calculated to obtain a sparse trajectory segment data set.

4. The vehicle emission estimation method based on sparse trajectories according to claim 1, characterized in that: The step of constructing a data-driven model and using the sparse trajectory segment data set for training to obtain a trained driving condition time distribution prediction model specifically includes: Processing the sparse trajectory segment dataset through a sliding window, combining features and labels of multiple consecutive time steps into a single time window as a basic input form of model data; The data driven model is trained based on the features and label matrix in the time window to obtain a trained driving condition time distribution prediction model.

5. The vehicle emission estimation method based on sparse trajectories according to claim 4, characterized in that: The data-driven model is a data-driven model of a seq2seq framework based on an attention mechanism, including an encoder, a decoder, an attention mechanism and a fully connected layer.

6. The vehicle emission estimation method based on sparse trajectories according to claim 1, characterized in that: The step of obtaining the vehicle driving trajectory also includes: Define the vehicle type corresponding to the vehicle driving trajectory.

7. The vehicle emission estimation method based on sparse trajectories according to claim 1, characterized in that: The calculation formula of the gas emission of the driving segment is as follows: Among them, E is the emission in the driving segment; is the average speed of the segment; T is the driving time of the segment; Q B is the basic emission factor of the vehicle; z represents the number of the driving condition category; is the frequency of the driving condition category z of the vehicle in the segment; Q z is the correction factor for driving condition category z; Q C is the correction factor of vehicle attribute C.

8. A vehicle emission estimation system based on sparse trajectories, characterized in that: include: A data acquisition module, used to acquire the vehicle's driving trajectory; A data preprocessing module, based on classifying the vehicle driving trajectory according to the sampling frequency, to obtain a sparse trajectory segment data set; A model training module, used to construct a data-driven model and use the sparse trajectory segment data set for training to obtain a trained driving condition time distribution prediction model; A driving condition time distribution prediction module, used for inputting the sparse trajectory to be tested into the trained driving condition time distribution prediction model to obtain a predicted driving condition time distribution; The emission estimation module estimates the gas emission of the driving segment according to the predicted driving cycle time distribution based on the IVE physical model.

9. A vehicle emission estimation device based on sparse trajectories, characterized in that: include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the vehicle emission estimation method based on sparse trajectories as described in any one of claims 1 to 7.

Images