Self-correction method for carbon deposition of nitrogen-oxygen sensor

Abstract

The invention discloses a self-correction method for carbon deposition of a nitrogen-oxygen sensor, relates to the technical field of motor vehicle exhaust emission monitoring, and solves the problems that an existing solution for carbon deposition of the nitrogen-oxygen sensor causes irreversible damage to the nitrogen-oxygen sensor and the compensation error limitation is small. The method comprises the following steps: acquiring driving parameters and real-time NOx concentration of a vehicle, inputting the driving parameters into a constructed driving parameter-NOx concentration related neural network model to obtain predicted NOx concentration, obtaining a difference between the real-time NOx concentration and the predicted NOx concentration to obtain a NOx concentration deviation value, and calculating the NOx concentration deviation value according to the calculated NOx concentration deviation value. And a NOx concentration correction strategy is triggered according to the NOx concentration deviation value to obtain a NOx concentration correction value, and when preset engine parameter conditions are met, the fuel injection quantity is controlled according to the NOx concentration correction value. The phenomenon of insufficient combustion of the engine is reduced, the carbon deposition generation rate is reduced, and the requirement for frequent disassembly and cleaning of the sensor due to carbon deposition failure is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of motor vehicle exhaust emission monitoring, and more specifically, it relates to a method for self-correcting carbon deposition of a nitrogen oxide sensor. Background Art

[0002] The nitrogen oxide sensor plays a crucial role in the automotive exhaust emission system. It is mainly used to detect the NOx concentration in the exhaust gas and transmit this information to the electronic control unit (ECU) of the vehicle. The ECU then adjusts the fuel injection volume based on this information to ensure that the engine operates in the best state. However, the flow rate of the exhaust gas flowing into the sensor is often unstable, which directly affects the contact degree between the sensor and the reactants in the exhaust gas and the measurement accuracy. The traditional sensor design lacks an effective flow rate adjustment mechanism, resulting in the measurement results being easily interfered by external factors. In addition, during the use of the nitrogen oxide sensor, especially when the engine combustion is poor, carbon deposition is likely to occur, which has a significant impact on the performance and emissions of the vehicle.

[0003] Carbon deposition is mainly caused by incomplete combustion of the engine mixture. These unburned residues are discharged through the exhaust pipe, but a part of them adheres to the nitrogen oxide sensor. Over time, more and more carbon deposits accumulate, resulting in a decrease in the sensitivity of the sensor or even its failure. The presence of carbon deposits will hinder or block the external air from entering the inside of the nitrogen oxide sensor, resulting in inaccurate signals output by the sensor. After receiving the wrong information, the ECU cannot correct the air-fuel ratio in time, resulting in an increase in the fuel consumption of the engine and a significant increase in the emission concentration.

[0004] The impact of carbon deposition on the nitrogen oxide sensor is not limited to fuel consumption and emissions. Since the ECU cannot accurately control the fuel injection volume, the air-fuel ratio of the engine will be incorrect, which may cause the engine to have problems such as unstable idle speed, vibration, misfire, and surge. In addition, carbon deposition may also cause the nitrogen oxide sensor to be poisoned. Especially when the vehicle uses leaded gasoline, lead will invade the inside of the sensor, hinder the diffusion of oxygen ions, and cause the sensor to fail. At this time, even if a new sensor is replaced, if leaded gasoline is still used, the sensor may soon fail.

[0005] Regarding the problem of carbon deposition on the nitrogen oxide sensor, traditional carbon deposition detection relies on periodic physical cleaning. By using oxalic acid solution or other cleaning agents, the carbon deposits on the sensor can be effectively removed and its normal operation can be restored. However, it should be noted that it is easy to cause irreversible damage to the nitrogen oxide sensor when cleaning the carbon deposits, and the measurement errors caused by carbon deposition cannot be compensated in real time. When cleaning the carbon deposits, the sensor has already suffered a significant performance decay (sensitivity decrease > 30%), and the errors during the carbon deposition accumulation process cannot be compensated in real time, resulting in an additional increase in fuel consumption and emissions of 15% - 20%.

[0006] Existing electronic compensation schemes only perform linear correction through the temperature-impedance curve and do not consider the non-linear impact of vehicle dynamic conditions (acceleration / load) on NOx generation, resulting in a compensation error > 15%. In addition, existing neural networks predict NOx emissions, but the input parameters are limited to the engine speed and intake air volume, without integrating the accelerating effect of ambient temperature on the carbon deposition rate of the sensor (the carbon deposition rate increases by 2.3 times at high temperatures). Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide a method for self-correcting carbon deposition of a nitrogen oxide sensor in view of the deficiencies of the prior art, and to solve the problems that the existing solutions for carbon deposition of nitrogen oxide sensors will cause irreversible damage to the nitrogen oxide sensors and have a small compensation error limitation.

[0008] A method for self-correcting carbon deposition of a nitrogen oxide sensor according to the present invention includes obtaining the driving parameters of a vehicle and the real-time NOx concentration, inputting the driving parameters into a pre-constructed neural network model related to driving parameters-NOx concentration to obtain a predicted NOx concentration, taking the difference between the real-time NOx concentration and the predicted NOx concentration to obtain a NOx concentration deviation value, triggering a NOx concentration correction strategy according to the NOx concentration deviation value to obtain a NOx concentration correction value, and controlling the fuel injection amount according to the NOx concentration correction value when preset engine parameter conditions are satisfied.

[0009] For further improvement, the method for constructing the neural network model related to driving parameters-NOx concentration is as follows:

[0010] Utilize big data to obtain the driving parameters and NOx concentration of similar vehicle models, establish a training database according to the driving parameters and NOx concentration of the similar vehicle models, and construct a neural network model related to driving parameters-NOx concentration through a multi-layer perceptron neural network algorithm model and according to the training database.

[0011] Furthermore, the neural network model related to driving parameters-NOx concentration includes

[0012] Input layer: There are four nodes, namely the driving speed, acceleration, load of the driving parameters, and the external ambient temperature;

[0013] First hidden layer: There are 64 nodes, and the weight initialization of the first hidden layer adopts the He normal distribution:

[0014] where x is the number of nodes, and Leaky ReLU(α = 0.01)) is the activation function;

[0015] Dropout layer: To prevent overfitting;

[0016] Second hidden layer: There are 32 nodes, and the expression of the second hidden layer is

[0017] Swish(x) = x * Sigmoid(x); where Swish() and Sigmoid() are activation functions, and x is the number of nodes;

[0018] Output layer: 1 node, used to output the predicted NOx concentration.

[0019] Furthermore, the NOx concentration deviation correction strategy is as follows

[0020] Set the first NOx concentration threshold and the second NOx concentration threshold. When the NOx concentration deviation value is greater than the first NOx concentration threshold, it indicates that the nitrogen oxide sensor is in the severe carbon deposition stage, and the NOx concentration detection abnormal alarm is activated;

[0021] When the NOx concentration deviation value is greater than the second NOx concentration threshold and less than or equal to the first NOx concentration threshold, it indicates that the nitrogen oxide sensor is in the moderate carbon deposition stage, and the real-time NOx concentration is corrected to obtain the NOx concentration correction value;

[0022] When the NOx concentration deviation value is less than or equal to the second NOx concentration threshold, it indicates that the nitrogen oxide sensor is in the mild carbon deposition stage, and there is no need to correct the real-time NOx concentration.

[0023] Furthermore, the method for correcting the real-time NOx concentration is to fine-tune the weights of all hidden layers in the driving parameter - NOx concentration related neural network model through the NOx concentration deviation value to obtain the NOx concentration correction value.

[0024] Furthermore, the expression for obtaining the NOx concentration correction value is

[0025] y corrected = f MLP (V; W new );

[0026] where y corrected is the NOx concentration correction value, f MLP is the multi-layer perceptron neural network algorithm model function, V is the feature vector of the driving parameters, and W new represents the new weight parameters after fine-tuning of all hidden layers in the driving parameter - NOx concentration related neural network model.

[0027] Furthermore, the expression for obtaining the new weight parameters after fine-tuning of all hidden layers in the driving parameter - NOx concentration related neural network model is

[0028]

[0029] Among them, ΔNOx is the NOx concentration deviation value, L is the weighted MAE loss function, η ∈ (0.0001, 0.01), and W new represents the new weight parameters after fine-tuning of all hidden layers in the driving parameter - NOx concentration related neural network model, and W old represents the original weight parameters of all hidden layers in the driving parameter - NOx concentration related neural network model before fine-tuning.

[0030] Furthermore, the expression of the weighted MAE loss function is

[0031]

[0032] Among them, L is the weighted MAE loss function, and w i represents the weight parameter of the i-th sample in the loss function, and y pred represents the predicted NOx concentration, and y true represents the NOx concentration value in the training database, and N represents the total number of samples in the loss function calculation.

[0033] Furthermore, the engine parameter condition is that the engine load rate is greater than a preset first load threshold and the exhaust temperature of the engine is greater than a preset temperature threshold.

[0034] Furthermore, the driving parameters include the speed, acceleration, load, and external environment temperature of the vehicle.

[0035] Beneficial Effects

[0036] The advantages of the present invention are as follows:

[0037] The present invention obtains the driving parameters of the vehicle and the real-time NOx concentration, inputs the driving parameters into the constructed driving parameter - NOx concentration related neural network model to obtain the predicted NOx concentration, subtracts the real-time NOx concentration from the predicted NOx concentration to obtain the NOx concentration deviation value, triggers the NOx concentration correction strategy according to the NOx concentration deviation value to obtain the NOx concentration correction value, and when the preset engine parameter conditions are met, controls the fuel injection volume according to the NOx concentration correction value. Compared with the traditional periodic cleaning, this method can achieve error compensation at the initial stage of carbon deposition, avoid error accumulation, keep the sensor measurement accuracy, and significantly delay the carbon deposition deterioration process of the nitrogen oxide sensor; reduce the phenomenon of incomplete combustion of the engine, reduce the carbon deposition generation rate, and avoid the need for frequent disassembly and cleaning of the sensor due to carbon deposition failure. Description of the Drawings

[0038] Figure 1 is the flow chart of the carbon deposition self-correction method for the nitrogen oxide sensor of the present invention;

[0039] Figure 2Schematic diagram of the driving parameter - NOx concentration related neural network model of the present invention. Detailed implementation manners

[0040] The following describes the present invention in further detail in conjunction with embodiments, but does not constitute any limitation to the present invention. Any limited number of modifications made by anyone within the scope of the claims of the present invention are still within the scope of the claims of the present invention.

[0041] Refer to Figure 1 - Figure 2 , a carbon deposit self - correction method for a nitrogen - oxygen sensor of the present invention. The method includes: acquiring the driving parameters of the vehicle and the real - time NOx concentration, inputting the driving parameters into the constructed driving parameter - NOx concentration related neural network model to obtain the predicted NOx concentration, taking the difference between the real - time NOx concentration and the predicted NOx concentration to obtain the NOx concentration deviation value, triggering a NOx concentration correction strategy according to the NOx concentration deviation value to obtain the NOx concentration correction value, and when the preset engine parameter conditions are met, controlling the fuel injection amount according to the NOx concentration correction value.

[0042] As Figure 1 shown, the present invention integrates an intelligent compensation module in the sensor control system. This module constructs a driving parameter - NOx concentration related neural network model by collecting multi - dimensional driving parameters such as the vehicle speed (v), acceleration (a), load (m), and external environment temperature (T) in real - time, and is used to output the predicted NOx concentration (y_pred). By comparing and analyzing the real - time NOx concentration (y_meas) of the nitrogen - oxygen sensor with the predicted NOx concentration (y_pred) in real - time, the system can dynamically identify the measurement deviation caused by carbon deposits and trigger a NOx concentration correction strategy based on the deviation magnitude.

[0043] Since the generation of the carbon deposit phenomenon in the nitrogen - oxygen sensor is a long - term process, and each time the nitrogen - oxygen sensor is cleaned, the carbon deposit phenomenon is already relatively serious, resulting in relatively large errors. Although the nitrogen - oxygen sensor can return to its original accuracy after cleaning, the fuel consumption and environmental pollution caused during the generation process of the carbon deposit phenomenon are usually ignored. During the generation process of the carbon deposit phenomenon, as the carbon deposit accumulates continuously, the sensitivity of the nitrogen - oxygen sensor will gradually decrease, making the nitrogen - oxygen sensor less sensitive to changes in the concentration of nitrogen oxides in the exhaust gas. Since the nitrogen - oxygen sensor cannot accurately measure the content of nitrogen oxides in the exhaust gas, the ECU cannot accurately control the fuel injection amount, and the emission control system cannot timely and accurately adjust the operating state of the engine, resulting in increased fuel consumption and excessive nitrogen oxide emissions. Before cleaning the nitrogen - oxygen sensor, the fuel loss and emission pollution caused by the carbon deposit phenomenon in the nitrogen - oxygen sensor are huge. This application proposes a self - correction method for the measurement accuracy of the nitrogen - oxygen sensor due to the carbon deposit phenomenon.

[0044] This patent mainly considers several aspects of driving parameters, including the speed, acceleration, load, and external environmental temperature of vehicle driving. Define the characteristic vector V of driving parameters as V=(v, a, m, T)∈R 4 , where:

[0045] Speed v (0 - 120 km, sampling frequency 10 Hz): The emission concentration of nitrogen oxides in the exhaust gas shows a stable increasing trend with the increase of the vehicle driving speed. When the vehicle driving speed is above 60 km / h, the nitrogen oxide emission concentration is 3 times that when the driving speed is below 40 km / h. At the same time, some studies also point out that the nitrogen oxide concentration will increase with the increase of the vehicle driving speed, while the concentrations of carbon monoxide and hydrocarbons will decrease with the increase of speed.

[0046] Acceleration a (-3~+3m / s 2 , resolution 0.1m / s 2 ): Although the specific impact of acceleration on nitrogen oxide emissions may not be as direct as that of speed, the changes in the engine load and combustion state during the acceleration process will also affect the generation of nitrogen oxides. And sudden acceleration (a>2m / s 2 ) causes an instantaneous increase in fuel injection volume, and the unburned HC emission increases by 25%, significantly accelerating the carbon deposition adhesion.

[0047] Load m (0 - 5 tons, calibrated by suspension strain sensor): The load of the vehicle will also affect the nitrogen oxide emissions and the carbon deposition rate. Generally speaking, the greater the load, the greater the power required by the engine, and the more intense the combustion, thus increasing the nitrogen oxide emissions and the carbon deposition rate. The load m is calculated by fusing the suspension strain sensor data and the tire pressure sensor data. The calculation formula is: m = k1·ε + k2·(P - P0), where ε is the suspension microstrain, P is the tire pressure, k1 and k2 are calibration coefficients, and P o is the tire pressure at no load.

[0048] Environmental temperature T (-40~125℃, collected by NTC sensor): The external environmental temperature will affect the engine cooling effect and combustion state, and thus affect the nitrogen oxide emissions. And high temperature exacerbates the oxidation and condensation of unburned HC, thus increasing the carbon deposition rate.

[0049] Then, through analysis of variance, the priority weights of the above driving parameters are determined as: W=(0.35, 0.25, 0.2, 0.2).

[0050] As Figure 2 shown, the construction of the neural network model related to driving parameters - NOx concentration:

[0051] Using big data to statistically analyze the NOx content detected by the nitrogen oxide sensor when the four driving parameters of the same type of vehicle, namely speed, acceleration, load, and external environmental temperature, change, and establish a database. Then, a neural network model related to the four driving parameters - NOx concentration during vehicle driving is constructed through the multi-layer perceptron (MLP) neural network algorithm. The framework of the neural network model related to the driving parameters - NOx concentration is consistent with the framework of the multi-layer perceptron neural network algorithm model.

[0052] The neural network model related to the driving parameters - NOx concentration includes:

[0053] Input layer: It has 4 nodes, namely driving speed v, acceleration a, load m, and external environmental temperature T.

[0054] The first hidden layer: It has 64 nodes, the activation function is Leaky ReLU (α = 0.01)), and the weight initialization adopts the He normal distribution: x is the number of nodes.

[0055] Dropout layer (probability = 0.2): To prevent overfitting.

[0056] The second hidden layer: It has 32 nodes, the activation functions are Swish (β = 1.0) and Sigmoid, and the regularization adopts the L1+L2 mixed penalty term: Swish(x) = x*Sigmoid(x). In information science, due to its properties such as being monotonically increasing and its inverse function being monotonically increasing, the Sigmoid function is often used as the activation function of the neural network to map the variable to between 0 and 1.

[0057] Output layer: 1 node, used to output the predicted NOx concentration, with the unit of ppm.

[0058] Loss function: Custom weighted MAE:

[0059] Among them, L is the weighted MAE loss function, w i represents the weight parameter of the i-th sample in the loss function, y pred represents the predicted NOx concentration, y true represents the NOx concentration value in the training database, and N represents the total number of samples in the loss function calculation. The weight is positively correlated with the engine load and is dynamically adjusted according to the working conditions. When accelerating suddenly, w i = 2.0, strengthening the correction response to high carbon deposit risk working conditions.

[0060] When the vehicle is in motion, the vehicle's central control system receives and processes the signals of four key parameters in real time. These parameters include the vehicle speed, acceleration, load, and external ambient temperature. These parameters are crucial for predicting the NOx emission concentration. Based on these input signals, the central control system uses the established multi-layer perceptron (MLP) neural network model for calculation to predict the quasi-NOx emission concentration of the vehicle under the current operating conditions.

[0061] The multi-layer perceptron (MLP) neural network model has been trained with a large amount of historical data and already has a high prediction accuracy.

[0062] The NOx concentration deviation correction strategy is as follows:

[0063] Set the first NOx concentration threshold and the second NOx concentration threshold. In this embodiment, the first NOx concentration threshold is 50 ppm, and the second NOx concentration threshold is 20 ppm. When the NOx concentration deviation value is greater than the first NOx concentration threshold, it indicates that the nitrogen oxide sensor is in the severe carbon deposition stage, and an abnormal alarm for NOx concentration detection is activated. That is, when the NOx concentration deviation value ΔNOx = |y_pred - y_meas| > 50 ppm, the system determines that the nitrogen oxide sensor has serious carbon deposition, triggers an alarm and recommends cleaning.

[0064] When the NOx concentration deviation value is greater than the second NOx concentration threshold and less than or equal to the first NOx concentration threshold, it indicates that the nitrogen oxide sensor is in the moderate carbon deposition stage, and the real-time NOx concentration is corrected. When 20 ppm ≤ ΔNOx ≤ 50 ppm, parameter fine-tuning is started to fine-tune the hidden layer weights. The NOx concentration correction value is obtained through the NOx concentration deviation value and the engine load, so as to compensate for the sensitivity attenuation in real time.

[0065] The expression for obtaining the NOx concentration correction value is as follows:

[0066] y corrected =f MLP (V;W new );

[0067] where, y corrected is the NOx concentration correction value, f MLP is the multi-layer perceptron neural network algorithm model function, V is the feature vector of the driving parameters, and W new represents the new weight parameters after fine-tuning of all hidden layers in the driving parameter - NOx concentration related neural network model.

[0068] The expression for obtaining the new weight parameters after fine-tuning of all hidden layers in the driving parameter - NOx concentration related neural network model is as follows:

[0069]

[0070] where ΔNOx is the NOx concentration deviation value, L is the weighted MAE loss function, η ∈ (0.0001, 0.01), and W new represents the new weight parameters after fine-tuning all the hidden layers in the driving parameter-NOx concentration related neural network model, and W old represents the original weight parameters of all the hidden layers in the driving parameter-NOx concentration related neural network model before fine-tuning.

[0071] When the engine parameter conditions are met, then the NOx concentration correction value is sent to the ECU to control the fuel injection amount. The engine parameter conditions are that the engine load rate is greater than the first load threshold and the exhaust temperature of the engine is greater than the temperature threshold. The first load threshold is 30% of the engine full load, and the temperature threshold is 200 °C.

[0072] When the NOx concentration deviation value is less than or equal to the second NOx concentration threshold, it indicates that the nitrogen oxide sensor is in the mild carbon deposition stage, and there is no need to correct the real-time NOx concentration. That is, when ΔNOx ≤ 20 ppm, the system determines that the NOx concentration deviation value is within the acceptable range, and directly uses the real-time NOx concentration (y_meas).

[0073] The core of the present invention lies in constructing a self-correcting system with dynamically coupled parameters. By real-time monitoring of vehicle operating conditions and environmental parameters, using the MLP neural network to output and predict the NOx concentration (y_pred), and comparing it with the real-time NOx concentration (y_meas), dynamically correct the measurement deviation of the nitrogen oxide sensor caused by carbon deposition. At the same time, the system optimizes the fuel injection strategy to reduce the emission of unburned hydrocarbons (HC) from the engine, and suppresses the generation of carbon deposition from the source.

[0074] The NOx concentration correction value is regarded as the concentration of the actual emissions of the vehicle, and this numerical signal is sent to the engine control center. The engine control center adjusts the working state and fuel injection strategy of the engine according to the corrected NOx concentration, so as to achieve the following goals:

[0075] 1. Optimize the combustion process, reduce the emission concentration of NOx, and ensure that the vehicle meets the environmental protection standards.

[0076] 2. Improve fuel utilization rate, reduce fuel consumption, and lower operating costs.

[0077] 3. Ensure the power performance and economy of the engine, and enhance the overall performance of the vehicle.

[0078] The above are only the preferred embodiments of the present invention. It should be noted that for those skilled in the art, without departing from the structure of the present invention, several deformations and improvements can be made, and these will not affect the implementation effect of the present invention and the practicability of the patent.

Claims

1. A method for self-correcting carbon deposition of a nitrogen oxide sensor, characterized in that, The method includes obtaining the driving parameters of the vehicle and the real-time NOx concentration, inputting the driving parameters into the constructed neural network model related to driving parameters-NOx concentration to obtain the predicted NOx concentration, taking the difference between the real-time NOx concentration and the predicted NOx concentration to obtain the NOx concentration deviation value, triggering the NOx concentration correction strategy according to the NOx concentration deviation value to obtain the NOx concentration correction value, and controlling the fuel injection quantity according to the NOx concentration correction value when the preset engine parameter conditions are met.

2. The carbon deposit self-correction method of a nitrogen-oxygen sensor according to claim 1, characterized in that The method for constructing the neural network model related to driving parameters-NOx concentration is as follows: Using big data to obtain the driving parameters and NOx concentration of the same type of vehicle, establishing a training database according to the driving parameters and NOx concentration of the same type of vehicle, and constructing a neural network model related to driving parameters-NOx concentration through a multi-layer perceptron neural network algorithm model and according to the training database.

3. A carbon deposition self-correction method for a nitrogen-oxygen sensor according to claim 2, characterized in that, The neural network model related to driving parameters-NOx concentration includes: Input layer: There are four nodes, namely the driving speed, acceleration, load, and external environment temperature of the driving parameters; The first hidden layer: It has 64 nodes, and the weight initialization of the first hidden layer adopts the He normal distribution: where x is the number of nodes, and Leaky ReLU(α = 0.01)) is the activation function; Dropout layer: To prevent overfitting; Second hidden layer: There are 32 nodes, and the expression of the second hidden layer is: Swish(x) = x * Sigmoid(x); where both Swish() and Sigmoid() are activation functions, and x is the number of nodes; Output layer: 1 node, used to output the predicted NOx concentration.

4. A carbon deposit self-correction method for a nitrogen-oxygen sensor according to claim 2, characterized in that, The NOx concentration deviation correction strategy is as follows: Set the first NOx concentration threshold and the second NOx concentration threshold. When the NOx concentration deviation value is greater than the first NOx concentration threshold, it indicates that the nitrogen oxide sensor is in the severe carbon deposition stage, and the NOx concentration detection abnormal alarm is started; When the NOx concentration deviation value is greater than the second NOx concentration threshold and less than or equal to the first NOx concentration threshold, it indicates that the nitrogen oxide sensor is in the moderate carbon deposition stage, and the real-time NOx concentration is corrected to obtain the NOx concentration correction value; When the NOx concentration deviation value is less than or equal to the second NOx concentration threshold, it indicates that the nitrogen oxide sensor is in the mild carbon deposition stage, and there is no need to correct the real-time NOx concentration.

5. The carbon deposit self-correction method of a nitrogen-oxygen sensor according to claim 4, characterized in that The method for correcting the real-time NOx concentration is to fine-tune the weights of all hidden layers in the neural network model related to driving parameters-NOx concentration through the NOx concentration deviation value to obtain the NOx concentration correction value.

6. A method for self-correcting carbon deposition of a nitrogen-oxygen sensor according to claim 5, characterized in that, The expression for obtaining the NOx concentration correction value is: y corrected = f MLP (V; W new ); where y corrected is the NOx concentration correction value, f MLP is the multi-layer perceptron neural network algorithm model function, V is the feature vector of driving parameters, W new represents the new weight parameters after fine-tuning all hidden layers in the driving parameter - NOx concentration related neural network model.

7. A carbon deposition self-correction method for a nitrogen-oxygen sensor according to claim 6, characterized in that, The expression for obtaining the new weight parameters of all hidden layers in the neural network model related to driving parameters-NOx concentration after fine-tuning is: where ΔNOx is the NOx concentration deviation value, L is the weighted MAE loss function, η ∈ (0.0001, 0.01), and W new represents the new weight parameters after fine-tuning of all hidden layers in the driving parameter-NOx concentration related neural network model, and W old represents the original weight parameters of all hidden layers in the driving parameter-NOx concentration related neural network model before fine-tuning.

8. A method for self-correcting carbon deposition of a nitrogen-oxygen sensor according to claim 7, characterized in that, The weighted MAE loss function expression is: Among them, L is the weighted MAE loss function, w i represents the weight parameter of the i-th sample in the loss function, y pred represents the predicted NOx concentration, y true represents the NOx concentration value in the training database, and N represents the total number of samples in the loss function calculation.

9. A method for self-correcting carbon deposition of a nitrogen-oxygen sensor according to claim 1, characterized in that, The engine parameter conditions are that the engine load rate is greater than the preset first load threshold and the exhaust temperature of the engine is greater than the preset temperature threshold.

10. A carbon deposition self-correction method for a nitrogen-oxygen sensor according to claim 1, characterized in that, The driving parameters include the speed, acceleration, load, and external environment temperature of the vehicle.

Images