DPF regeneration early warning method based on EDO-GRU model

Abstract

The invention provides a DPF regeneration early warning method based on an EDO-GRU model, and the method comprises the following steps: S1, collecting the historical data of a vehicle, and obtaining a training set and a test set; s2, inputting the processed historical data of the vehicle into an EDO-GRU model, and carrying out model training; s3, predicting the carbon load increment per unit kilometer in the future according to the trained EDO-GRU model, and calculating the residual regeneration mileage according to the predicted value; according to the invention, frequent interruption of DPF regeneration of the vehicle can be effectively avoided, the operation efficiency of the vehicle is improved, and the energy consumption of the vehicle is reduced; the problems that in the prior art, the future DPF regeneration mileage cannot be predicted, and the driving style difference is not considered, so that long-term dependence is caused, and the convergence speed is low are solved.

Description

Technical Field

[0001] The invention relates to the field of computer technology, and in particular to a DPF regeneration early warning method based on an EDO-GRU model. Background Art

[0002] In order to reduce the emission of carbon particles in the engine exhaust, diesel locomotives are generally equipped with a diesel particulate filter DPF system. The diesel particulate filter DPF system can effectively capture carbon particles in the vehicle exhaust, thereby significantly reducing harmful substances emitted into the atmosphere; however, as time goes by, the carbon particles accumulated on the surface of the DPF will gradually increase. These accumulated particles will not only cause the DPF's filtering efficiency to decrease, but also cause the exhaust back pressure to increase; increased back pressure means that the engine needs to consume more energy to overcome the exhaust resistance, which will not only affect the engine's economy and increase fuel consumption, but also have a negative impact on the engine's power, causing the vehicle's acceleration performance and maximum output power to decrease.

[0003] To solve this problem, when the carbon particles accumulated on the DPF surface reach a certain level, the system will trigger the DPF regeneration process. The main purpose of the regeneration process is to restore the DPF back pressure to a normal level by burning the carbon particles inside the DPF, thereby ensuring that the engine's economy and power are not affected.

[0004] When the vehicle is undergoing DPF regeneration, if it stops, idles for a long time, starts and stops frequently, or enters a gas station to refuel, the DPF regeneration process may be interrupted, resulting in unsuccessful regeneration; when DPF regeneration is interrupted, the carbon particles inside the DPF are not effectively removed, which will cause DPF blockage, and further affect engine performance and fuel consumption; therefore, the vehicle may frequently trigger DPF regeneration, which will not only affect the vehicle's operating efficiency, but also increase fuel consumption and operating costs.

[0005] To avoid this situation, a reliable early warning mechanism is needed to inform the driver of the future regeneration time in advance, so that the driver can arrange the trip reasonably and avoid frequent regeneration. The early warning methods for DPF regeneration in the prior art include threshold-based early warning methods and neural network-based early warning methods. The above early warnings have the following disadvantages: 1. Threshold-based warning method, for example, a warning is generated when the carbon load density value is greater than a certain threshold. This method cannot predict the future DPF regeneration mileage; if the threshold is set too high, the DPF system may not respond in time, resulting in a delayed warning; if the threshold is set too low, the warning may be triggered too early, causing unnecessary trouble and misjudgment to the driver; 2. The early warning method based on neural network does not consider the differences in driving styles and has problems such as long-term dependence and slow convergence. Summary of the invention

[0006] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a DPF regeneration early warning method based on the EDO-GRU model.

[0007] To achieve the above object, the present invention adopts the following technical solutions: A DPF regeneration warning method based on an EDO-GRU model comprises the following steps: S1: Collect vehicle historical data to obtain training sets and test sets; The following sub-steps are included: S11: Collect vehicle historical data and perform preprocessing; Engineers collect multiple vehicle historical data from vehicle sensors and control units and save them into a dataset; The vehicle history data includes the current DPF carbon load, the current total vehicle mileage, vehicle speed, engine speed, engine torque, post-treatment temperature, intake air flow, current fuel consumption, and regeneration indicator light activation status; The current DPF carbon load refers to the amount of carbon accumulated in the diesel engine particulate filter DPF system up to now; the aftertreatment temperature refers to the temperature of the aftertreatment device in the exhaust system; the intake flow rate refers to the air flow rate entering the engine; the regeneration indicator light activation state refers to the indicator light state of the DPF regeneration process, including activation and inactivation; The preprocessing includes identifying missing data, detecting outliers, deleting duplicate data, etc.; Specifically, statistical tools or data analysis libraries are used to detect whether there are missing data in the data set. If missing data is detected, the data records containing the missing data are deleted or the missing data are filled by interpolation. If no missing data is detected, no processing is performed. Use statistical methods or model-based detection methods to identify whether there are outliers in the data set, and the model-based detection methods include box plot method, Z-score, etc. If an outlier is detected, replace the outlier with a value within a reasonable range or delete the data record containing the outlier; if no outlier is detected, do not process it; Use the pandas library to identify whether there is duplicate data in the dataset. If there is missing data, use the delete function to remove the duplicate data. If there is no missing data, no processing is done. S12: Calculate the future carbon load increment per kilometer as a data label; Collect the carbon load of the vehicle when regeneration is triggered and the vehicle mileage when regeneration is triggered; The future carbon load increment per kilometer is calculated by the following formula: ; ; in, To trigger the regeneration of carbon load, To trigger regeneration, the vehicle mileage, is the current DPF carbon load, The total mileage of the current vehicle; The future carbon load increase per kilometer will be As data labels; S13: Segment the collected vehicle history data to obtain a sample set; The collected vehicle historical data is divided according to the driving distance per kilometer, with one kilometer as one measurement, and only one piece of data is retained within one kilometer; one piece of data includes eight values ​​of the current DPF carbon load, the current total mileage of the vehicle, the vehicle speed, the engine speed, the engine torque, the post-processing temperature, the intake air flow, and the current fuel consumption of the same vehicle; if there are multiple pieces of data within one kilometer, the multiple pieces of data are averaged and the average value is used as the only piece of data within this kilometer; Divide into sample sets with N kilometers as the dimension; N is a positive integer, that is, a sample contains N data; The data labels obtained in step S12 are saved in the sample set; S14: normalize the data in the sample; The data in the sample is normalized using the following formula; ; Wherein, i is a positive integer in the interval [1, 8], representing the current DPF carbon load, the current vehicle total mileage, vehicle speed, engine speed, engine torque, aftertreatment temperature, intake air flow, and current fuel consumption respectively; is the normalized value of the corresponding parameter, is the original value of the corresponding parameter, is the minimum value of this parameter in the sample, is the maximum value of the parameter in the sample; Get the normalized sample set; S15: calculating vehicle driving style; The vehicle driving styles include normal, energy-saving, and energy-consuming; The following sub-steps are included: S151: Calculate the average value of current fuel consumption, vehicle speed, and engine speed in the sample set; Calculate the average value of current fuel consumption, vehicle speed, and engine speed in the sample set respectively, and record the current fuel consumption as o, vehicle speed as s, and engine speed as r; then the average value of current fuel consumption is , average speed , average engine speed as follows: ; Where N is the number of samples contained in the sample set; Respectively represent the current fuel consumption, vehicle speed, and engine speed of the i-th data in the sample; S152: Randomly select 3 data points as cluster centers and evaluate the cluster centers; Randomly select 3 data points as the initial cluster centers, assign each sample in the sample set to the cluster center closest to it, and obtain three categories; The mean of all samples in each cluster is calculated respectively, and the mean is used as the new cluster center; the compactness and distribution of the cluster are evaluated by the objective function WCSS of the K-means algorithm, and the objective function WCSS is as follows: ; Among them, E is the evaluation result of the objective function WCSS on each cluster; K is the number of classes, that is, K=3. There are 3 categories, for The mean vector in ; ; Where M is the number of samples in each category, and x is the sample vector in each category; ; S153: Mark the clusters into three driving styles: normal, energy-saving, and energy-consuming; According to the obtained E value, step S512 is repeated until the cluster center does not change, that is, convergence is reached; and according to the current fuel consumption, vehicle speed, and engine speed values ​​in the final three cluster centers, the three clusters are manually marked as normal, energy-saving, and energy-consuming driving styles respectively; S16: Divide the samples into training set and training set; In step S153, the samples in the sample set are divided into three categories: normal, energy-saving, and energy-consuming. In each category, the samples are divided into a training set and a test set according to a proportion.

[0008] By preprocessing the vehicle historical data, invalid data and repeated redundant data can be effectively filtered out. Using the processed vehicle historical data as a training set to train the model can effectively improve the accuracy of model prediction.

[0009] S2: Input the processed vehicle history data into the EDO-GRU model for model training; The following sub-steps are included: S21: Build EDO-GRU model; Each individual unit module of the GRU network consists of an update gate and a reset gate. The update gate memorizes historical data through a gate. The update gate is calculated by the following formula: ; Among them, σ is the sigmoid function, and each element of the output of this function is a real number between 0 and 1, which represents the weight of allowing the corresponding information to pass through. represents the output of the previous sequence, represents the input of the current sequence, and is the weight of the update gate to be trained; The reset gate determines the forgetting process of the data. The reset gate is calculated by the following formula: ;

[0010] in, and is the weight of the reset gate to be trained.

[0011] Use the exponential optimization algorithm EDO to find the best combination of learning rate, number of hidden nodes and maximum number of iterations; the EDO algorithm includes four parts: population initialization, population development, population exploration and population update; Population initialization refers to randomly generating some initial solutions. The initialization formula is as follows: ; Among them, i=1, 2, ...N; j=1, 2, ...D; N is the number of initialized individuals, D is the dimension of the solution. In this scenario, the solution dimension is the learning rate, the number of hidden nodes and the maximum number of iterations. is the minimum value of the j-th dimension variable. is the maximum value; The population development phase uses the memorylessness, exponential rate, variance, and mean of the exponential distribution, and adopts a guidance solution to guide the population to the global optimal direction; the guidance solution is the average of the three best solutions of the population, namely: ; Where t is the current iteration number, They are the top three best individuals in the current iterative population; Define a memoryless position matrix, the first generation position matrix is ​​equal to the population matrix. The update formula of the individual is as follows: ;

[0012] in, is the i-th individual in the position matrix, is a random number in [0,1]; the calculation formulas for other parameters are as follows: ; In the population exploration phase, two random individuals in the population are used for updating to avoid local convergence of the population; the update formula of the individual is: ; in, is the average value of the population in generation t; other parameters are calculated as: ; In the population update of the EDO algorithm, population development and population exploration are randomly selected, that is, each individual is randomly updated with a probability of 50% by development or exploration. For the population update, two matrices need to be updated, the population matrix X and the memoryless position matrix Y. The specific operations are as follows: ; in, is the ith individual in the position matrix of generation t+1, It is the i-th individual of the t+1 generation in the population matrix, which is updated according to the fitness of individuals in the position matrix and the population matrix, that is, individuals with better fitness in the position matrix and the population matrix are selected; fit() is the fitness function of the individual, which is calculated as follows: ; in, is the actual carbon load per kilometer, G is the predicted carbon load per kilometer, and N is the number of training samples.

[0013] S22: According to the training set and the optimization function, iteratively update the weights of the GRU network; The data in the training set is input into the GRU network for training. The GRU network outputs the predicted future carbon load increment per kilometer. The predicted future carbon load increment per kilometer is compared with the future carbon load increment per kilometer calculated in step S12 in the training set. For comparison, based on the optimization function, the weights of the GRU network are continuously updated iteratively until the predicted future carbon load increment per kilometer is infinitely close to the future carbon load increment per kilometer included in the training set. ; S23: Calculate the loss function based on the test set. When the loss function converges, the EDO-GRU model training is completed. Set a threshold in advance; During the EDO-GRU model training process, the test set data is input into the EDO-GRU model, and the loss function is monitored regularly. When the change of the loss function is less than the threshold, that is, the loss function converges, the EDO-GRU model training is completed; the EDO-GRU model training for the three driving styles is completed.

[0014] S3: Predict the future carbon load increment per kilometer based on the trained EDO-GRU model, and calculate the remaining regeneration mileage based on the predicted value; Collect vehicle data within the last L kilometers of the current vehicle, and normalize the collected vehicle data to obtain a data set; the data set includes the current DPF carbon load, the current vehicle total mileage, vehicle speed, engine speed, engine torque, post-processing temperature, intake air flow, current fuel consumption, regeneration indicator activation status, vehicle carbon load when regeneration is triggered, etc.; Determine the driving style of the current vehicle through the method in step S152 based on the current vehicle speed, engine speed, and current fuel consumption; Find the corresponding EDO-GRU model according to the driving style, input the data in the data set into the EDO-GRU model in sequence, and predict the future carbon load increment per kilometer of the current vehicle, recorded as ; The remaining regeneration mileage threshold is set in advance, and the remaining regeneration mileage is calculated based on the following formula and the future carbon load increment per kilometer; ; in, To trigger the regeneration of carbon load, is the current DPF carbon load, The future carbon load increment per kilometer of current vehicles; If the remaining regeneration mileage calculated by the data in three consecutive data sets is less than 100 kilometers, a vehicle operation planning reminder is issued to the driver.

[0015] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides a DPF regeneration warning method based on an EDO-GRU model. The method determines the driving style of a current vehicle by collecting historical vehicle data, prepares training sets and test sets of different driving styles, and uses the training sets to train the EDO-GRU model to obtain a prediction model for the carbon load increment per kilometer. After the model is trained, the current new vehicle data can be processed and input into the corresponding model to predict the future carbon load increment per kilometer, and the regeneration mileage is calculated according to the predicted result. Through timely reminders, it is convenient to control the vehicle to reasonably arrange the DPF regeneration time. The DPF regeneration of the vehicle can be effectively avoided from being frequently interrupted, the operation efficiency of the vehicle is improved, and the energy consumption of the vehicle is reduced. The problems of long-term dependence and slow convergence speed caused by the inability to predict the future DPF regeneration mileage and the failure to consider the difference in driving styles in the prior art are avoided. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The present invention is a flowchart of the steps of a DPF regeneration warning method based on the EDO-GRU model. DETAILED DESCRIPTION

[0017] In order to provide a further understanding of the purpose, structure, features, and functions of the present invention, the following detailed description is given in conjunction with the embodiments.

[0018] like Figure 1 As shown, a DPF regeneration warning method based on the EDO-GRU model includes the following steps: S1: Collect vehicle historical data to obtain training sets and test sets; The following sub-steps are included: S11: Collect vehicle historical data and perform preprocessing; Engineers collect multiple vehicle historical data from vehicle sensors and control units and save them into a dataset; The vehicle history data includes the current DPF carbon load, the current total vehicle mileage, vehicle speed, engine speed, engine torque, post-treatment temperature, intake air flow, current fuel consumption, and regeneration indicator light activation status; The current DPF carbon load refers to the amount of carbon accumulated in the diesel engine particulate filter DPF system up to now; the aftertreatment temperature refers to the temperature of the aftertreatment device in the exhaust system; the intake flow rate refers to the air flow rate entering the engine; the regeneration indicator light activation state refers to the indicator light state of the DPF regeneration process, including activation and inactivation; The preprocessing includes identifying missing data, detecting outliers, deleting duplicate data, etc.; Specifically, statistical tools or data analysis libraries are used to detect whether there are missing data in the data set. If missing data is detected, the data records containing the missing data are deleted or the missing data are filled by interpolation. If no missing data is detected, no processing is performed. Use statistical methods or model-based detection methods to identify whether there are outliers in the data set, and the model-based detection methods include box plot method, Z-score, etc. If an outlier is detected, replace the outlier with a value within a reasonable range or delete the data record containing the outlier; if no outlier is detected, do not process it; Use the pandas library to identify whether there is duplicate data in the dataset. If there is missing data, use the delete function to remove the duplicate data. If there is no missing data, no processing is done. S12: Calculate the future carbon load increment per kilometer as a data label; Collect the carbon load of the vehicle when regeneration is triggered and the vehicle mileage when regeneration is triggered; The future carbon load increment per kilometer is calculated by the following formula: ; ; in, To trigger the regeneration of carbon load, To trigger regeneration, the vehicle mileage, is the current DPF carbon load, The total mileage of the current vehicle; The future carbon load increase per kilometer will be As data labels; S13: Segment the collected vehicle history data to obtain a sample set; The collected vehicle historical data is divided according to the driving distance per kilometer, with one kilometer as one measurement, and only one piece of data is retained within one kilometer; one piece of data includes eight values ​​of the current DPF carbon load, the current total mileage of the vehicle, the vehicle speed, the engine speed, the engine torque, the post-processing temperature, the intake air flow, and the current fuel consumption of the same vehicle; if there are multiple pieces of data within one kilometer, the multiple pieces of data are averaged and the average value is used as the only piece of data within this kilometer; Divide into sample sets with N kilometers as the dimension; N is a positive integer, that is, a sample contains N data; The data labels obtained in step S12 are saved in the sample set; S14: normalize the data in the sample; The data in the sample is normalized using the following formula; ; Wherein, i is a positive integer in the interval [1, 8], representing the current DPF carbon load, the current vehicle total mileage, vehicle speed, engine speed, engine torque, aftertreatment temperature, intake air flow, and current fuel consumption respectively; is the normalized value of the corresponding parameter, is the original value of the corresponding parameter, is the minimum value of this parameter in the sample, is the maximum value of the parameter in the sample; Get the normalized sample set; S15: calculating vehicle driving style; The vehicle driving styles include normal, energy-saving, and energy-consuming; The following sub-steps are included: S151: Calculate the average value of current fuel consumption, vehicle speed, and engine speed in the sample set; Calculate the average value of current fuel consumption, vehicle speed, and engine speed in the sample set respectively, and record the current fuel consumption as o, vehicle speed as s, and engine speed as r; then the average value of current fuel consumption is , average vehicle speed , average engine speed as follows: ; Where N is the number of samples contained in the sample set; Respectively represent the current fuel consumption, vehicle speed, and engine speed of the i-th data in the sample; S152: Randomly select 3 data points as cluster centers and evaluate the cluster centers; Randomly select 3 data points as the initial cluster centers, assign each sample in the sample set to the cluster center closest to it, and obtain three categories; The mean of all samples in each cluster is calculated respectively, and the mean is used as the new cluster center; the compactness and distribution of the cluster are evaluated by the objective function WCSS of the K-means algorithm, and the objective function WCSS is as follows: ; Among them, E is the evaluation result of the objective function WCSS on each cluster; K is the number of classes, that is, K=3. There are 3 categories, for The mean vector in ; ; Where M is the number of samples in each category, and x is the sample vector in each category; ; S153: Mark the clusters into three driving styles: normal, energy-saving, and energy-consuming; According to the obtained E value, step S512 is repeated until the cluster center does not change, that is, convergence is reached; and according to the current fuel consumption, vehicle speed, and engine speed values ​​in the final three cluster centers, the three clusters are manually marked as normal, energy-saving, and energy-consuming driving styles respectively; S16: Divide the samples into training set and training set; In step S153, the samples in the sample set are divided into three categories: normal, energy-saving, and energy-consuming. In each category, the samples are divided into a training set and a test set according to a proportion.

[0019] By preprocessing the vehicle historical data, invalid data and repeated redundant data can be effectively filtered out. Using the processed vehicle historical data as a training set to train the model can effectively improve the accuracy of model prediction.

[0020] S2: Input the processed vehicle history data into the EDO-GRU model for model training; The following sub-steps are included: S21: Build EDO-GRU model; Each individual unit module of the GRU network consists of an update gate and a reset gate. The update gate memorizes historical data through a gate. The update gate is calculated by the following formula: ; Among them, σ is the sigmoid function, and each element of the output of this function is a real number between 0 and 1, which represents the weight of allowing the corresponding information to pass through. represents the output of the previous sequence, represents the input of the current sequence, and is the weight of the update gate to be trained; The reset gate determines the forgetting process of the data. The reset gate is calculated by the following formula: ;

[0021] in, and is the weight of the reset gate to be trained.

[0022] Use the exponential optimization algorithm EDO to find the best combination of learning rate, number of hidden nodes and maximum number of iterations; the EDO algorithm includes four parts: population initialization, population development, population exploration and population update; Population initialization refers to randomly generating some initial solutions. The initialization formula is as follows: ; Among them, i=1, 2, ...N; j=1, 2, ...D; N is the number of initialized individuals, D is the dimension of the solution. In this scenario, the solution dimension is the learning rate, the number of hidden nodes and the maximum number of iterations. is the minimum value of the j-th dimension variable. is the maximum value; The population development phase uses the memorylessness, exponential rate, variance, and mean of the exponential distribution, and adopts a guidance solution to guide the population to the global optimal direction; the guidance solution is the average of the three best solutions of the population, namely: ; Where t is the current iteration number, They are the top three best individuals in the current iterative population; Define a memoryless position matrix, the first generation position matrix is ​​equal to the population matrix. The update formula of the individual is as follows: ;

[0023] in, is the i-th individual in the position matrix, is a random number in [0,1]; the calculation formulas for other parameters are as follows: ; In the population exploration phase, two random individuals in the population are used for updating to avoid local convergence of the population; the update formula of the individual is: ; in, is the average value of the population in generation t; other parameters are calculated as: ; In the population update of the EDO algorithm, population development and population exploration are randomly selected, that is, each individual is randomly updated with a probability of 50% by development or exploration. For the population update, two matrices need to be updated, the population matrix X and the memoryless position matrix Y. The specific operations are as follows: ; in, is the ith individual in the position matrix of generation t+1, It is the i-th individual of the t+1 generation in the population matrix, which is updated according to the fitness of individuals in the position matrix and the population matrix, that is, individuals with better fitness in the position matrix and the population matrix are selected; fit() is the fitness function of the individual, which is calculated as follows: ; in, is the actual carbon load per kilometer, G is the predicted carbon load per kilometer, and N is the number of training samples.

[0024] S22: According to the training set and the optimization function, iteratively update the weights of the GRU network; The data in the training set is input into the GRU network for training. The GRU network outputs the predicted future carbon load increment per kilometer. The predicted future carbon load increment per kilometer is compared with the future carbon load increment per kilometer calculated in step S12 in the training set. For comparison, based on the optimization function, the weights of the GRU network are continuously updated iteratively until the predicted future carbon load increment per kilometer is infinitely close to the future carbon load increment per kilometer included in the training set. ; S23: Calculate the loss function based on the test set. When the loss function converges, the EDO-GRU model training is completed. Set a threshold in advance; During the EDO-GRU model training process, the test set data is input into the EDO-GRU model, and the loss function is monitored regularly. When the change of the loss function is less than the threshold, that is, the loss function converges, the EDO-GRU model training is completed; the EDO-GRU model training for the three driving styles is completed.

[0025] S3: Predict the future carbon load increment per kilometer based on the trained EDO-GRU model, and calculate the remaining regeneration mileage based on the predicted value; Collect vehicle data within the last L kilometers of the current vehicle, and normalize the collected vehicle data to obtain a data set; the data set includes the current DPF carbon load, the current vehicle total mileage, vehicle speed, engine speed, engine torque, post-processing temperature, intake air flow, current fuel consumption, regeneration indicator activation status, vehicle carbon load when regeneration is triggered, etc.; Determine the driving style of the current vehicle through the method in step S152 based on the current vehicle speed, engine speed, and current fuel consumption; Find the corresponding EDO-GRU model according to the driving style, input the data in the data set into the EDO-GRU model in sequence, and predict the future carbon load increment per kilometer of the current vehicle, recorded as ; The remaining regeneration mileage threshold is set in advance, and the remaining regeneration mileage is calculated based on the following formula and the future carbon load increment per kilometer; ; in, To trigger the regeneration of carbon load, is the current DPF carbon load, The future carbon load increment per kilometer of current vehicles; If the remaining regeneration mileage calculated by the data in three consecutive data sets is less than 100 kilometers, a vehicle operation planning reminder is issued to the driver.

[0026] The present invention has been described by the above-mentioned relevant embodiments, however, the above-mentioned embodiments are only examples for implementing the present invention. It must be pointed out that the disclosed embodiments do not limit the scope of the present invention. On the contrary, changes and modifications made without departing from the spirit and scope of the present invention are all within the scope of patent protection of the present invention.

Claims

1. A DPF regeneration early warning method based on the EDO-GRU model, characterized in that: The following steps are involved: S1: Collect vehicle historical data to obtain training sets and test sets; The following sub-steps are included: S11: Collect vehicle historical data and perform preprocessing; Engineers collect multiple vehicle historical data from vehicle sensors and control units and save them into a dataset; The vehicle history data includes the current DPF carbon load, the current total vehicle mileage, vehicle speed, engine speed, engine torque, post-treatment temperature, intake air flow, current fuel consumption, and regeneration indicator light activation status; S12: Calculate the future carbon load increment per kilometer as a data label; Collect the carbon load of the vehicle when regeneration is triggered and the vehicle mileage when regeneration is triggered; The future carbon load increment per kilometer is calculated by the following formula: ; ； in, To trigger the regeneration of carbon load, To trigger regeneration, the vehicle mileage, is the current DPF carbon load, The total mileage of the current vehicle; The future carbon load increase per kilometer will be As data labels; S13: Segment the collected vehicle history data to obtain a sample set; S14: normalize the data in the sample; S15: calculating vehicle driving style; The vehicle driving styles include normal, energy-saving, and energy-consuming; S16: Divide the samples into training set and training set; In step S153, the samples in the sample set are divided into three categories: normal, energy-saving, and energy-consuming. In each category, the samples are divided into a training set and a test set according to a proportion. S2: Input the processed vehicle history data into the EDO-GRU model for model training; The following sub-steps are included: S21: Build EDO-GRU model; S22: According to the training set and the optimization function, iteratively update the weights of the GRU network; S23: Calculate the loss function based on the test set. When the loss function converges, the EDO-GRU model training is completed. S3: Predict the future carbon load increment per kilometer based on the trained EDO-GRU model, and calculate the remaining regeneration mileage based on the predicted value.

2. The DPF regeneration early warning method based on the EDO-GRU model according to claim 1, characterized in that: In step S11, the preprocessing includes identifying missing data, detecting outliers, and deleting duplicate data; Specifically, statistical tools or data analysis libraries are used to detect whether there are missing data in the data set. If missing data is detected, the data records containing the missing data are deleted or the missing data are filled by interpolation. If no missing data is detected, no processing is performed. Using statistical methods or model-based detection methods to identify whether there are outliers in the data set, the model-based detection methods include box plot method, Z-score; If an outlier is detected, the outlier is replaced with a value within a reasonable range or the data record containing the outlier is deleted; if no outlier is detected, no processing is performed; Use the pandas library to identify whether there is duplicate data in the dataset. If there is missing data, delete the identified duplicate data through the delete function; if there is no missing data, no processing is performed.

3. The DPF regeneration early warning method based on the EDO-GRU model as claimed in claim 1, characterized in that: In step S13, the collected vehicle historical data is divided according to the driving distance per kilometer, with one kilometer as one measurement, and only one piece of data is retained within one kilometer; wherein one piece of data includes eight values ​​of the current DPF carbon load, the current total mileage of the vehicle, the vehicle speed, the engine speed, the engine torque, the post-processing temperature, the intake air flow, and the current fuel consumption of the same vehicle; if there are multiple pieces of data within one kilometer, the multiple pieces of data are averaged, and the obtained average value is used as the only piece of data within this kilometer; Divide into sample sets with N kilometers as the dimension; N is a positive integer, that is, a sample contains N data; The data labels obtained in step S12 are saved in the sample set.

4. The DPF regeneration early warning method based on the EDO-GRU model as claimed in claim 1, characterized in that: In step S14, the data in the sample is normalized using the following formula; ； Wherein, i is a positive integer in the interval [1, 8], representing the current DPF carbon load, the current vehicle total mileage, vehicle speed, engine speed, engine torque, aftertreatment temperature, intake air flow, and current fuel consumption respectively; is the normalized value of the corresponding parameter, is the original value of the corresponding parameter, is the minimum value of this parameter in the sample, is the maximum value of the parameter in the sample; Get the normalized sample set.

5. The DPF regeneration early warning method based on the EDO-GRU model as claimed in claim 1, characterized in that: Step S15 includes the following sub-steps: S151: Calculate the average value of current fuel consumption, vehicle speed, and engine speed in the sample set; Calculate the average value of current fuel consumption, vehicle speed, and engine speed in the sample set respectively, and record the current fuel consumption as o, vehicle speed as s, and engine speed as r; then the average value of current fuel consumption is , average vehicle speed , average engine speed as follows: ； Where N is the number of samples contained in the sample set; Respectively represent the current fuel consumption, vehicle speed, and engine speed of the i-th data in the sample; S152: Randomly select 3 data points as cluster centers and evaluate the cluster centers; Randomly select 3 data points as the initial cluster centers, assign each sample in the sample set to the cluster center closest to it, and obtain three categories; The mean of all samples in each cluster is calculated respectively, and the mean is used as the new cluster center; the compactness and distribution of the cluster are evaluated by the objective function WCSS of the K-means algorithm, and the objective function WCSS is as follows: ； Among them, E is the evaluation result of the objective function WCSS on each cluster; K is the number of classes, that is, K=3. There are 3 categories, for The mean vector in ; ; Where M is the number of samples in each category, and x is the sample vector in each category; ； S153: Mark the clusters into three driving styles: normal, energy-saving, and energy-consuming; According to the obtained E value, step S512 is repeated until the cluster center no longer changes, that is, convergence is reached; and according to the current fuel consumption, vehicle speed, and engine speed values ​​in the final three cluster centers, the three clusters are manually marked as normal, energy-saving, and energy-consuming driving styles.

6. The DPF regeneration early warning method based on the EDO-GRU model as claimed in claim 1, characterized in that: S22: According to the training set and the optimization function, iteratively update the weights of the GRU network; The data in the training set is input into the GRU network for training. The GRU network outputs the predicted future carbon load increment per kilometer. The predicted future carbon load increment per kilometer is compared with the future carbon load increment per kilometer calculated in step S12 in the training set. For comparison, based on the optimization function, the weights of the GRU network are continuously updated iteratively until the predicted future carbon load increment per kilometer is infinitely close to the future carbon load increment per kilometer included in the training set. ; S23: Calculate the loss function based on the test set. When the loss function converges, the EDO-GRU model training is completed. Set a threshold in advance; During the EDO-GRU model training process, the test set data is input into the EDO-GRU model, and the loss function is monitored regularly. When the change of the loss function is less than the threshold, that is, the loss function converges, the EDO-GRU model training is completed; the EDO-GRU model training for the three driving styles is completed.

7. The DPF regeneration early warning method based on the EDO-GRU model as claimed in claim 1, characterized in that: The specific contents of step S3 are as follows: Collect vehicle data within the last L kilometers of the current vehicle, and normalize the collected vehicle data to obtain a data set; the data set includes the current DPF carbon load, the current vehicle total mileage, vehicle speed, engine speed, engine torque, post-processing temperature, intake air flow, current fuel consumption, regeneration indicator activation status, and vehicle carbon load when regeneration is triggered; Determine the driving style of the current vehicle through the method in step S152 based on the current vehicle speed, engine speed, and current fuel consumption; Find the corresponding EDO-GRU model according to the driving style, input the data in the data set into the EDO-GRU model in sequence, and predict the future carbon load increment per kilometer of the current vehicle, recorded as ; The remaining regeneration mileage threshold is set in advance, and the remaining regeneration mileage is calculated based on the following formula and the future carbon load increment per kilometer; ； in, To trigger the regeneration of carbon load, is the current DPF carbon load, The future carbon load increment per kilometer of current vehicles; If the remaining regeneration mileage calculated by the data in three consecutive data sets is less than 100 kilometers, a vehicle operation planning reminder is issued to the driver.

Images