OBD data uncertainty inference method based on vehicle inspection data optimized distribution
By using OBD data processing methods based on vehicle inspection data, and leveraging generative adversarial networks and temporal embedding techniques, the emission model was optimized, solving the problem of emission inventory uncertainty. This enabled high-precision emission data estimation and uncertainty analysis, thereby improving the scientific nature and management efficiency of the emission inventory.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- UNIV OF SCI & TECH OF CHINA
- Filing Date
- 2022-05-06
- Publication Date
- 2026-06-12
Smart Images
Figure CN115186728B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of urban road traffic environment detection technology, specifically an OBD data uncertainty inference method based on optimized distribution of vehicle inspection data. Background Technology
[0002] Mobile source pollution has become a significant source of air pollution in large and medium-sized cities, and is a major cause of photochemical smog and fine particulate matter pollution. The prevention and control of motor vehicle pollution is urgent, and precise control of ambient air based on high-quality emission inventories has become a key breakthrough.
[0003] Emission inventories are summaries of the types and quantities of air pollutants emitted from different emission sources at a specific spatiotemporal scale, providing important reference value for scientific research and decision-making. The quality of emission inventories is crucial to their effectiveness in practical air quality management, and uncertainty analysis is a key method for assessing this quality. Missing key data, poor data representativeness, and unreliable data sources all introduce uncertainty into emission inventories. Quantifying emission inventory uncertainty is essential for formulating normal air pollution control measures. Therefore, this invention provides a method for inferring uncertainty in OBD data based on the optimized distribution of vehicle inspection data. Summary of the Invention
[0004] To address the problems of the above solutions, this invention provides an OBD data uncertainty inference method based on optimized distribution of vehicle inspection data.
[0005] The objective of this invention can be achieved through the following technical solutions:
[0006] A method for inferring uncertainty in OBD data based on optimized distribution of vehicle inspection data; the specific method is as follows:
[0007] Step 1: Utilize the characteristic of OBD data to record values second by second to construct instantaneous emission datasets for different pollution sources;
[0008] Step 2: Use a temporal embedding network to extract the instantaneous emission characteristics and the dependencies between emissions of the vehicle's OBD. At the same time, introduce Gaussian distribution and Wiener process into the generation of emission data, and use a generative adversarial mechanism to estimate the characteristics and distribution of OBD emission data.
[0009] Step 3: Based on vehicle inspection data measured by high-precision instruments, the error of the emission model is corrected by utilizing the road driving information from the OBD data; achieving high-precision OBD data distribution optimization that integrates actual road driving information;
[0010] Step 4: By measuring the distribution estimation difficulty of the features of OBD data samples and introducing Dropout into the temporal generative adversarial model, the inherent random uncertainty of OBD data and the perceived uncertainty of the temporal generative adversarial network model are respectively inferred.
[0011] Methods for constructing instantaneous emission datasets for different pollution sources by utilizing the second-by-second recording characteristic of OBD data include:
[0012] Step SA1: Perform data cleaning;
[0013] Obtain second-by-second driving trajectory data from the OBD system of a specific vehicle model, and obtain the instantaneous speed V over a day. i =[v i1 v i2 , ..., v im ], calculate the acceleration Ai = [ai1, ai2, ..., a im ], where m is the time length; data is cleaned using set velocity and acceleration thresholds, invalid data is deleted, and cubic spline interpolation is performed to obtain a complete and continuous velocity V. i =[v i1 v i2 , ..., v im Acceleration data A i =[a i1 a i2 , ..., a im ], where i = 1, 2, ..., n, and n represents the number of car models.
[0014] Step SA2: Perform data classification; obtain the continuous and valid speed dataset V for the i-th vehicle type. i =[v i1 v i2 , ..., v im The COPERT model was used to estimate the pollution sources. The emissions were categorized by pollution source into carbon monoxide emissions, hydrocarbon emissions, and nitrogen oxide emissions. The total emissions were calculated every five minutes throughout the day, resulting in 144 emission values, including carbon monoxide emissions. hydrocarbon emissions Nitrogen oxide emissions
[0015] The specific steps of step SA2 are as follows:
[0016] The emission factor of the w-th pollution source for the i-th vehicle type is calculated using the COPERT model, as shown in the formula:
[0017]
[0018]
[0019]
[0020] Where a, b, c, d, and e represent the parameters of each pollution source, as shown in Table 1; v i V represents the continuously valid speed dataset of the i-th vehicle type. i =[v i1 v i2 , ..., v im The average value of ], where m represents the time length.
[0021] Further, based on the calculated emission factors of different vehicle models and different emission sources, the emissions of different pollutants are calculated. As shown in the formula:
[0022] E w =Ef iw ×R len ×f
[0023] Among them, Ef iw Let R represent the emission factor of the w-th pollution source for the i-th vehicle type. len Expressed as the distance traveled, that is Where 300 represents the 5-minute data length of the time interval for calculating emissions from continuous driving data second by second, and f is the traffic flow. Here, emissions are calculated for a specific vehicle model, so the value is 1.
[0024] The specific implementation steps for step two are as follows:
[0025] Step SB1: Extract features from vehicle OBD emission data, construct a training dataset, and input it into an embedding network to provide an invertible mapping between input data features and latent representations;
[0026] Step SB2: Design a generative adversarial network model for vehicle OBD emission data, which consists of a sequence generator and a discriminator.
[0027] Step SB3: The discriminator also operates from the embedding space;
[0028] Step SB4: Train the temporal generative adversarial network.
[0029] The specific steps of step SB2 include:
[0030] make Let the vector space of static features and the vector space of temporal attributes be represented respectively, and random vectors be extracted from them as input to generate...
[0031] Generating function g: Use tuples of static and temporal random vectors to generate latent codes. Implementing the generating function using a recursive network:
[0032]
[0033]
[0034] in, Generative networks representing static features Recurrent generative networks representing temporal features.
[0035] random vector Sample from the distribution, z t It follows a stochastic process, taking a Gaussian distribution and a Wiener process respectively.
[0036] The specific steps of step SB3 include:
[0037] Discriminant function d: Retrieve vehicle information codes and instantaneous emission codes, and return the classification. The symbol represents the true embedding (h) * ) or synthetic embedding Similarly, Represents the actual data (y) * ) or synthetic data The classification.
[0038] The method for implementing the discriminant function is as follows:
[0039] The discriminant function is implemented using a bidirectional recursive network (RNN) with a feedforward output layer.
[0040]
[0041]
[0042] in, This represents the positive hidden state of the instantaneous emission sequence. This represents the reverse hidden state of the instantaneous emission sequence. It is a recursive function. d T These are the output layer classification functions for vehicle information features and instantaneous emission features, respectively.
[0043] The specific implementation steps for step four are as follows:
[0044] Step SD1: Infer the inherent random uncertainty of OBD data by measuring the distributional difficulty of the features of the OBD data samples;
[0045]
[0046] Wherein, σ(T) i () represents the instantaneous emission value T i To mitigate the random uncertainties, we use Mean Average Precision (MAP) to obtain a single set of model parameter values. N represents the number of emission data samples from different vehicles of the same model on different days. σ(T) i ) 2 This is equivalent to an adaptive weight. For emission samples where the data distribution is difficult to estimate and there is a lot of inherent noise in the data, this weight is relatively small.
[0047] Step SD2: Introduce Dropout into the temporal generative adversarial model to infer the perceptual uncertainty of the OBD data temporal generative adversarial network model. Use dropout in the emission time series embedding network and emission time series generator, and take the average value after generating emission data several times. Neurons are randomly deactivated with Bernoulli probability, therefore the total uncertainty is:
[0048]
[0049] Compared with the prior art, the beneficial effects of the present invention are: by utilizing the characteristic of OBD data to record values second by second, an instantaneous emission dataset of different pollution sources is constructed; the vehicle OBD emission dataset obtained based on the actual road driving measurement of the whole vehicle fully integrates information such as road traffic conditions, driving behavior, and road type, and is more suitable for reflecting the emission status of the whole vehicle in the road driving environment.
[0050] A generative adversarial mechanism is employed to better estimate the characteristics and distribution of OBD emission data. By fully utilizing the road driving information in OBD data while correcting errors in the emission model, high-precision OBD data distribution optimization that incorporates actual road driving information is achieved. By measuring the difficulty of estimating the distribution of features in OBD data samples and introducing Dropout into the temporal generative adversarial model, the inherent random uncertainty of OBD data and the perceived uncertainty of the temporal generative adversarial network model are inferred, respectively. Attached Figure Description
[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0052] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation
[0053] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.
[0054] like Figure 1 As shown, an OBD data uncertainty inference method based on optimized distribution of vehicle inspection data is presented; the specific method involves the following steps:
[0055] Step 1: Utilize the characteristic of OBD data to record values second by second to construct instantaneous emission datasets for different pollution sources;
[0056] Step 2: Use a temporal embedding network to extract the instantaneous emission characteristics and the dependencies between emissions of the vehicle's OBD. At the same time, introduce Gaussian distribution and Wiener process into the generation of emission data, and use a generative adversarial mechanism to estimate the characteristics and distribution of OBD emission data.
[0057] Step 3: Based on vehicle inspection data measured by high-precision instruments, the error of the emission model is corrected by utilizing the road driving information from the OBD data; achieving high-precision OBD data distribution optimization that integrates actual road driving information;
[0058] Step 4: By measuring the distribution estimation difficulty of the features of OBD data samples and introducing Dropout into the temporal generative adversarial model, the inherent random uncertainty of OBD data and the perceived uncertainty of the temporal generative adversarial network model are respectively inferred.
[0059] Methods for constructing instantaneous emission datasets for different pollution sources by utilizing the second-by-second recording characteristic of OBD data include:
[0060] Step SA1: Perform data cleaning;
[0061] Obtain second-by-second driving trajectory data from the OBD system of a specific vehicle model, and obtain the instantaneous speed V over a day. i =[v i1 v i2 , ..., v im ] Calculate acceleration A i =[a i1 a i2 , ..., a im ], where m is the time length; data is cleaned using set velocity and acceleration thresholds, invalid data is deleted, and cubic spline interpolation is performed to obtain a complete and continuous velocity V. i =[v i1 v i2 , ..., v imAcceleration data A i =[a i1 a i2 , ..., a im ], where i = 1, 2, ..., n, and n represents the number of car models.
[0062] The speed and acceleration thresholds can be set according to the actual situation; during the selection process, the instantaneous speed of a day with a large amount of data can be selected.
[0063] Step SA2: Perform data classification; obtain the continuous and valid speed dataset V for the i-th vehicle type. i =[v i1 v i2 , ..., v im The COPERT model was used to estimate the pollution sources. The emissions were categorized by pollution source into carbon monoxide emissions, hydrocarbon emissions, and nitrogen oxide emissions. The total emissions were calculated every five minutes throughout the day, resulting in 144 emission values, including carbon monoxide emissions. hydrocarbon emissions Nitrogen oxide emissions
[0064] The specific steps of step SA2 are as follows:
[0065] The emission factor of the w-th pollution source for the i-th vehicle type is calculated using the COPERT model, as shown in the formula:
[0066]
[0067]
[0068]
[0069] Where a, b, c, d, and e represent the parameters of each pollution source, as shown in Table 1; v i V represents the continuously valid speed dataset of the i-th vehicle type. i =[v i1 v i2 , ..., v im The average value of ], where m represents the time length.
[0070] Table 1: Calculation parameters of COPERT model for different pollutants
[0071]
[0072]
[0073] Further, based on the calculated emission factors of different vehicle models and different emission sources, the emissions of different pollutants are calculated. As shown in the formula:
[0074] E w =Ef iw ×R len ×f
[0075] Among them, Ef iw Let R represent the emission factor of the w-th pollution source for the i-th vehicle type. len Expressed as the distance traveled, that is Where 300 represents the 5-minute data length of the time interval for calculating emissions from continuous driving data second by second, and f is the traffic flow. Here, emissions are calculated for a specific vehicle model, so the value is 1.
[0076] The vehicle OBD emission dataset, obtained from actual road driving measurements, fully integrates information such as road traffic conditions, driving behavior, and road type, making it more suitable for reflecting the vehicle's emission status under road driving conditions.
[0077] The specific implementation steps for step two are as follows:
[0078] Step SB1: Feature extraction of vehicle OBD emission data, including emission data from different pollution sources for different vehicle models, possessing both static features (such as vehicle model and pollution source) and time-series attributes (such as carbon monoxide emissions). hydrocarbon emissions This allows for the construction of a training dataset, which is then fed into an embedding network to provide an invertible mapping between input data features and latent representations, thereby reducing the high dimensionality of features and effectively learning time dependencies.
[0079] Step SB2: Design of the generative adversarial network model for vehicle OBD emission data, consisting of a sequence generator and a discriminator. The discriminator performs binary classification, separating real data from generated data. Output 0 indicates data significantly different from real emission data, and output 1 indicates data slightly different from real emission data. The emission sequence generator first outputs to the embedding space, and the discriminator also performs calculations from the embedding space.
[0080] Step SB3: The discriminator also operates from the embedding space;
[0081] Step SB4: Temporal Generative Adversarial Network Training. Vehicle model data as static data and emissions data as temporal data are input together into the generator. In pure open-loop mode, the autoregressive generator receives the synthetic embeddings. (i.e., its previous output) to generate the next composite vector. Then, the gradient is calculated based on the unsupervised loss. The discriminator is maximized, and the generator is minimized using the training data. h 1：144 and the synthesized output from the generator Provide correct classification Possibility:
[0082]
[0083] Training is performed in closed-loop mode, where the generator receives the embedded sequence h of the actual data. 1：t-1 (Computed by the embedded network) to generate the next potential vector. The gradient can be obtained by capturing the distribution. and The supervised loss is calculated by using the previous difference loss and applying maximum likelihood.
[0084]
[0085] in, Approximate to Using a sample z t As a standard for stochastic gradient descent, the difference between the actual next-step latent vector (from the embedding function) and the synthesized next-step latent vector (from a generator based on the actual historical sequence of latent vectors) is evaluated. When the generator is driven to create realistic sequences (evaluated by an imperfect adversary), This further ensures that it generates a similar gradual transition (evaluated by the real target). For the discriminator, T represents the actual data. y t The closer to 1, the better, that is to say ∑ t logy t The larger the better, for the generated data. The smaller the output, the better; that is to say, The larger the better, but for the generator, the goal is to fool the discriminator, i.e., to output... The closer to 1, the better.
[0086] The specific steps of step SB1 include:
[0087] Static characteristics S (e.g., vehicle type, pollution source) and time-series attributes T (e.g., carbon monoxide emissions) hydrocarbon emissions Therefore, the training dataset is constructed. N represents emission data and vehicle model data for different days of the same type of vehicle. The following information describes emission data and other information for a certain type of vehicle, and the subscript i is omitted.
[0088] Using the training dataset Learning distribution As close as possible to the distribution p(S, T) of the original data 1：144 Embedding and recovery functions realize the mapping between features and latent space, and learn the latent temporal dynamics of data through low-dimensional representation.
[0089] make The latent vector space corresponding to vehicle model data and emission data S and T is represented by the embedding function e: Encoding vehicle information and instantaneous emission characteristics, this method uses a recursive network to implement the embedding function:
[0090] h S =e S (S)
[0091] h t =e T (h S h t-1 T t )
[0092] Where e: It is an embedded network of vehicle information features, e T : A recurrent embedding network representing instantaneous emission characteristics.
[0093] Conversely, the recovery function r: The static and temporal features are encoded to restore their feature representations, and r is implemented at each step through a feedforward network:
[0094]
[0095]
[0096] Where, r S : and r T : Recovery networks for static and temporal embedding.
[0097] The first objective function is the reconstruction loss. The embedding function and the recovery function realize the invertible mapping between static input and temporal input. Therefore, the errors between static input data and reconstructed static input data and between temporal input data and reconstructed temporal input data are calculated separately:
[0098]
[0099] The specific steps of step SB2 include:
[0100] make Let the vector space of static features and the vector space of temporal attributes be represented respectively, and random vectors be extracted from them as input to generate...
[0101] Generating function g: Use tuples of static and temporal random vectors to generate latent codes. Implementing the generating function using a recursive network:
[0102]
[0103]
[0104] in, Generative networks representing static features, g T : Recurrent generative networks representing temporal features.
[0105] random vector Sampling can be performed from the distribution, z t Following a stochastic process, we take a Gaussian distribution (i.e., a normal distribution) and a Wiener process (i.e., a Brownian motion process, where the change at any finite time follows a normal distribution, and its variance increases linearly with the length of the time interval; the probability distribution of the change at any time interval is independent of the probability of the change at any other time interval).
[0106] The specific steps of step SB3 include:
[0107] Discriminant function d: Retrieve vehicle information codes and instantaneous emission codes, and return the classification. The symbol represents the true embedding (h) * ) or synthetic embedding Similarly, Represents the actual data (y) * ) or synthetic data The classification. This method implements the discriminant function through a bidirectional recurrent network (RNN) with a feedforward output layer:
[0108]
[0109]
[0110] in, This represents the positive hidden state of the instantaneous emission sequence. This represents the reverse hidden state of the instantaneous emission sequence. It is a recursive function. d T These are the output layer classification functions for vehicle information features and instantaneous emission features, respectively.
[0111] The discriminator can take either real emissions data or generated emissions data as input. The discriminator uses a bidirectional RNN, so it needs to evaluate the RNN outputs from both directions before finally outputting either 0 or 1.
[0112] The specific implementation steps for step three are as follows:
[0113] Step SC1: Select a continuous segment of OBD data, and based on vehicle speed and vehicle type information, use a high-precision pollutant measuring instrument to collect vehicle emission data under the same vehicle speed conditions.
[0114] Step SC2: The vehicle inspection data is obtained from high-precision instrument measurements. The inversion of OBD emission data is limited by the emission model and low measurement accuracy. Therefore, an error term ΔE is introduced to correct the distribution of OBD emission data E and vehicle inspection data. The error between them serves as the basis for the conditional constraint loss function of the temporal generative adversarial network.
[0115]
[0116] Step SC3: Vehicle information generation faces constraints. Vehicle information and other data are inherent attributes, and the space for correcting vehicle inspection data is limited, so they are ignored.
[0117] Step SC4: Generation of instantaneous emission data to counteract constraints in order to optimize the distribution of OBD emission data.
[0118]
[0119] Where α is the coefficient of the vehicle inspection data constraint term, which is used to penalize the error between the generated emission sample and the vehicle inspection emission sample.
[0120] The specific implementation steps for step four are as follows:
[0121] Step SD1: Infer the inherent random uncertainty of OBD data by measuring the difficulty of estimating the distribution of features of OBD data samples.
[0122]
[0123] Wherein, σ(T) i () represents the instantaneous emission value T i To mitigate the random uncertainties, we use Mean Average Precision (MAP) to obtain a single set of model parameter values. N represents the number of emission data samples from different vehicles of the same model on different days. σ(T) i ) 2 This is equivalent to an adaptive weight. For emission samples where the data distribution is difficult to estimate and there is a lot of inherent noise in the data, this weight is relatively small.
[0124] Step SD2: Introduce Dropout into the temporal generative adversarial model to infer the perceptual uncertainty of the OBD data temporal generative adversarial network model. Use dropout in the emission time series embedding network and emission time series generator, and take the average value after generating emission data several times. Neurons are randomly deactivated with Bernoulli probability, therefore the total uncertainty is:
[0125]
[0126] By adopting the above technical solution and utilizing the characteristic of OBD data recording values second by second, instantaneous emission datasets of different pollution sources are constructed. The vehicle OBD emission dataset obtained from actual road driving measurements of the whole vehicle fully integrates information such as road traffic conditions, driving behavior, and road type, and is more suitable for reflecting the emission status of the whole vehicle in the road driving environment.
[0127] A generative adversarial mechanism is employed to better estimate the characteristics and distribution of OBD emission data. By fully utilizing the road driving information in OBD data while correcting errors in the emission model, high-precision OBD data distribution optimization that incorporates actual road driving information is achieved. By measuring the difficulty of estimating the distribution of features in OBD data samples and introducing Dropout into the temporal generative adversarial model, the inherent random uncertainty of OBD data and the perceived uncertainty of the temporal generative adversarial network model are inferred, respectively.
[0128] This invention overcomes the model risks associated with analyzing the distribution characteristics of input parameters and determining the probability distribution model describing the uncertainty of model inputs, as well as the computational challenges of random sampling of model input variables. This method offers significant computational advantages and model robustness. Existing methods only consider static total emissions and neglect the time dependence of instantaneous emissions. To fully learn the characteristics of instantaneous emissions, this study uses OBD data from actual road driving and vehicle inspection data measured by high-precision instruments to achieve high-precision OBD data distribution optimization that integrates actual road driving information. Finally, an uncertainty boundary inference method based on the optimized OBD distribution model is proposed, inferring the inherent random uncertainty of OBD data and the perceived uncertainty of the time-series generative adversarial network model, respectively.
[0129] The above formulas are all numerical calculations after removing dimensions. The formulas are obtained by software simulation based on a large amount of data and are closest to the real situation. The preset parameters and preset thresholds in the formulas are set by those skilled in the art according to the actual situation or obtained by simulation based on a large amount of data.
[0130] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.
Claims
1. A method for inferring uncertainty in OBD data based on optimized distribution of vehicle inspection data, characterized in that, The specific method is as follows: Step 1: Utilize the characteristic of OBD data to record values second by second to construct instantaneous emission datasets for different pollution sources; Step 2: Use a temporal embedding network to extract the instantaneous emission characteristics and the dependencies between emissions of the vehicle's OBD. At the same time, introduce Gaussian distribution and Wiener process into the generation of emission data, and use a generative adversarial mechanism to estimate the characteristics and distribution of OBD emission data. Step 3: Based on vehicle inspection data measured by high-precision instruments, the road driving information from OBD data is used to correct the error of the emission model. Step 4: By measuring the distribution estimation difficulty of the features of OBD data samples and introducing Dropout into the temporal generative adversarial model, the inherent random uncertainty of OBD data and the perceived uncertainty of the temporal generative adversarial network model are respectively inferred. The specific implementation steps for step two are as follows: Step SB1: Extract features from vehicle OBD emission data, construct a training dataset, and input it into an embedding network to provide an invertible mapping between input data features and latent representations; Step SB2: Design a generative adversarial network model for vehicle OBD emission data, which consists of a sequence generator and a discriminator. Step SB3: The discriminator also operates from the embedding space; Step SB4: Train the temporal generative adversarial network; The detailed steps of step SB2 are as follows: make , Let the vector space of static features and the vector space of temporal attributes be represented respectively, and random vectors be extracted from them as input to generate... , ; Generating functions Use tuples of static and temporal random vectors to generate latent codes. The generating function is implemented through a recursive network: in, Generative networks representing static features Recurrent generative networks representing temporal features; random vector Sampling from the distribution Following a stochastic process, it takes a Gaussian distribution and a Wiener process, respectively; The detailed steps of step SB3 are as follows: Discriminant function Retrieve vehicle information codes and instantaneous emission codes, and return the classification. , The symbol represents the true embedding ( ) or synthetic embedding ( Similarly, Represents real data ( ) or synthetic data ( The classification of ) The method for implementing the discriminant function is as follows: The discriminant function is implemented using a bidirectional recursive network (RNN) with a feedforward output layer. in, This represents the positive hidden state of the instantaneous emission sequence. This represents the reverse hidden state of the instantaneous emission sequence. , It is a recursive function. , These are the output layer classification functions for vehicle information features and instantaneous emission features, respectively.
2. The OBD data uncertainty inference method based on optimized distribution of vehicle inspection data according to claim 1, characterized in that, Methods for constructing instantaneous emission datasets for different pollution sources by utilizing the second-by-second recording characteristic of OBD data include: Step SA1: Perform data cleaning; obtain the second-by-second driving trajectory data of the vehicle's OBD system for the model to be analyzed, and obtain the instantaneous speed for one day. Calculate acceleration ,in The time length is specified; data is cleaned using set velocity and acceleration thresholds, invalid data is deleted, and cubic spline interpolation is performed to obtain a complete and continuous velocity. Acceleration data ,in , Indicates the number of vehicle models; Step SA2: Perform data classification; obtain the first... Continuous and valid speed datasets for various vehicle models The COPERT model was used to estimate the pollution sources. The emissions were categorized by pollution source into carbon monoxide emissions, hydrocarbon emissions, and nitrogen oxide emissions. The total emissions were calculated every five minutes throughout the day, resulting in 144 emission values, including carbon monoxide emissions. hydrocarbon emissions Nitrogen oxide emissions .
3. The OBD data uncertainty inference method based on optimized distribution of vehicle inspection data according to claim 1, characterized in that, The specific implementation steps for step four are as follows: Step SD1: Infer the inherent random uncertainty of OBD data by measuring the distributional difficulty of the features of the OBD data samples; in, Indicates instantaneous emission value The randomness and uncertainty, This indicates the number of emission data samples from different vehicles of the same model on different days. Step SD2: Introduce Dropout into the temporal generative adversarial model to infer the perceptual uncertainty of the OBD data temporal generative adversarial network model. Use dropout in the emission time series embedding network and emission time series generator, and take the average value after generating emission data several times. The total uncertainty is obtained by randomly deactivating neurons with Bernoulli probability.
4. The OBD data uncertainty inference method based on optimized distribution of vehicle inspection data according to claim 3, characterized in that, The formula for calculating the total uncertainty is: 。