An Adaptive Urea Injection Control Method for a Diesel Engine SCR System

Abstract

This invention relates to diesel engine exhaust aftertreatment technology, and more particularly to an adaptive urea injection control method for a diesel engine SCR system. The method includes: automatically selecting an ammonia storage state prediction model based on catalyst temperature; calculating a basic value for urea injection quantity according to a three-segment control strategy based on the predicted ammonia storage quantity and the monitored catalyst temperature; modeling historical engine operating condition characteristic sequences using an LSTM network to output a feedforward prediction value for the current urea injection quantity; fusing the feedforward prediction value and the basic value for the current urea injection quantity to obtain the final urea injection quantity and outputting it to the urea nozzle; and periodically executing an adaptive coefficient correction process to adjust the adaptive correction coefficient and update the urea injection quantity. Compared with existing technologies, this invention has advantages such as accurate prediction of ammonia storage state and improved accuracy of injection quantity under transient operating conditions.

Description

Technical Field

[0001] This invention relates to diesel engine exhaust aftertreatment technology, and more particularly to an adaptive urea injection control method for a diesel engine SCR system. Background Technology

[0002] To meet the Euro VI / China VI emission regulations regarding nitrogen oxides (NOx) x Due to stringent emission restrictions, selective catalytic reduction (SCR) aftertreatment systems are widely used in road diesel engines. Existing SCR system control strategies mainly include two approaches: pulse spectrum (MAP)-based open-loop control and model-based closed-loop control.

[0003] The pulse spectrum-based open-loop control strategy determines the urea injection quantity through a pre-calibrated MAP (Magnetic Mapping Analysis) diagram. Its advantages lie in its simple controller structure design and low development difficulty. However, its drawbacks include poor transient control performance, decreased control accuracy with system aging, and difficulty adapting to complex real-world operating conditions. The model-based closed-loop control strategy dynamically adjusts the injection quantity by establishing a mathematical model of the SCR system and combining sensor feedback. Its advantages include high control accuracy and adaptability to a certain range of operating condition changes. However, its drawbacks include fixed model parameters, inability to adapt to system aging and component degradation, and limited diagnostic capabilities, making it difficult to accurately distinguish between NH3 leakage and NO2 leakage. x Emissions exceeded standards.

[0004] Existing SCR control technology faces the following four technical problems in practical applications: (1) The model parameters cannot be updated in real time: The parameters of the existing SCR control model (such as the catalyst ammonia storage capacity, reaction activation energy, etc.) are fixed after being determined during calibration and cannot be adaptively adjusted according to factors such as catalyst aging, changes in urea solution concentration, and sensor drift, resulting in a decrease in control accuracy as the usage time increases.

[0005] (2) The ammonia storage state is difficult to predict accurately: the ammonia storage state of the catalyst is a factor affecting the NO in the SCR system. x Conversion efficiency and NH3 leakage are key factors. Existing technologies use a uniform ammonia storage model in both high-temperature (>350℃) and low-temperature (≤350℃) regions, resulting in limited prediction accuracy; there is also a lack of effective correction methods for errors caused by factors such as incomplete urea decomposition.

[0006] (3) Lack of time-series feedforward prediction: The generation of diesel engine emissions and SCR conversion have obvious time-series correlations. The current optimal urea injection quantity is not only related to the current operating conditions, but also affected by changes in operating conditions over a short period of time in history. Traditional MAP methods and static models fail to utilize this information, resulting in large prediction errors for dynamic operating conditions, which can easily lead to transient NO x Emissions exceeded standards.

[0007] (4) Large calibration workload: In order to cover various speed-load-temperature combinations, a large number of calibration experiments are required. As emission requirements increase, the calibration particle size requirement is finer, the number of calibration experiments increases exponentially, and the calibration cycle is long and the cost is high.

[0008] A search revealed Chinese invention patent application publication number CN106837497A, which discloses a diesel engine catalytic reduction urea injection control method based on real-time ammonia storage management. The method establishes an ammonia storage calculation model based on the ammonia mass conservation law of the SCR system to calculate the ammonia storage at the current moment under actual operating conditions. Based on steady-state experiments, the engine NOx emission pulse spectrum, exhaust mass flow pulse spectrum, and ammonia-to-nitrogen ratio pulse spectrum are calibrated to calculate the basic urea injection quantity. The target ammonia storage region, ammonia adsorption time constant, and ammonia release time constant are calibrated experimentally to calculate the corrected urea injection quantity. Under actual operating conditions, the urea injection is controlled by the sum of the basic and corrected urea injection quantities. In cases of sudden increases in exhaust temperature, urea slow injection and injection stoppage are employed to bring the ammonia storage at the current moment of the SCR closer to the target ammonia storage region. This existing patent application employs a fixed-parameter open-loop control strategy, which cannot update the model in real-time to adapt to long-term system degradation. Furthermore, it does not establish accurate ammonia storage prediction models for high and low temperature operating conditions, resulting in decreased control accuracy with use and large ammonia storage prediction errors.

[0009] How to accurately predict the ammonia storage status of the SCR system and achieve adaptive control of urea injection has become a technical problem that needs to be solved. Summary of the Invention

[0010] The purpose of this invention is to overcome the shortcomings of the prior art and provide an adaptive urea injection control method for a diesel engine SCR system.

[0011] The objective of this invention can be achieved through the following technical solutions: According to one aspect of the present invention, an adaptive urea injection control method for a diesel engine SCR system is provided, comprising the following steps: The ammonia storage state prediction model is automatically selected based on the real-time monitored catalyst temperature of the diesel engine SCR system to predict the ammonia storage amount in real time. When the temperature is higher than the first temperature threshold, the slow time-varying EKF model is used; otherwise, the low temperature EKF model is used. Based on the estimated ammonia storage and the monitored catalyst temperature, the baseline value of urea injection rate is calculated according to the three-section control strategy. The historical engine operating condition feature sequence is modeled based on the LSTM network, and the feedforward prediction of the current urea injection quantity is output. By combining the feedforward estimate of the current urea injection volume with the baseline value of the urea injection volume, the final urea injection volume is obtained and output to the urea nozzle; In the method, an adaptive coefficient correction process is periodically executed to dynamically adjust the adaptive correction coefficient and update the final urea injection quantity.

[0012] As a preferred technical solution, the low-temperature EKF model is specifically as follows: Θmax(T) = S1•exp(-S2•T), Where S1 and S2 are the ammonia storage capacity coefficient and temperature sensitivity coefficient, respectively, T is the catalyst temperature, and Θmax(T) is the maximum ammonia storage capacity corresponding to temperature T. The slow time-varying EKF model uses a constant maximum ammonia storage capacity in the high-temperature region.

[0013] As a preferred technical solution, the calculation of urea injection volume according to the three-segment control strategy includes: When the estimated ammonia storage reaches the peak threshold, urea injection is shut off; When the estimated ammonia storage is below the peak threshold and the temperature is below the second temperature threshold, the urea injection rate is increased to establish ammonia reserves. When the estimated ammonia storage is below the peak threshold and the temperature is above or equal to the second temperature threshold, the original urea injection rate is maintained.

[0014] As a preferred technical solution, the calculation process of the feedforward estimate of the current urea injection volume includes: Maintain a sliding window of length T to store the engine operating condition feature sequence of the most recent T time steps. The engine operating condition feature sequence of each time step includes engine speed, engine load, SCR inlet exhaust temperature, and SCR inlet NO. x Concentration and the amount of urea injected in the previous step; The sliding window is updated each time a new engine operating condition feature sequence is acquired. The sliding window data is input into the trained LSTM network, and the feedforward prediction of the current urea injection amount is obtained by forward calculation.

[0015] As a preferred technical solution, the LSTM network adopts a two-level stacked structure: The first LSTM layer contains 128 hidden units and returns the output of all time steps, which is used to extract low-level temporal features; The second LSTM layer contains 64 hidden units and only returns the output of the last time step, which is used to extract high-level temporal features; After BatchNorm normalization and Dropout regularization, the feedforward estimate of the current urea injection amount is obtained through mapping by two fully connected layers.

[0016] As a preferred technical solution, the final urea injection volume is obtained by integrating the feedforward estimated value of the current urea injection volume with the baseline value of the urea injection volume: u{final}=ω*m{lstm}+(1-ω)*m{base}, where u{final} is the final urea injection amount; m{lstm} is the feedforward prediction value of the LSTM output; m{base} is the base value of the calculated urea injection amount; ω is the fusion coefficient, which is adaptively adjusted according to the catalyst aging degree.

[0017] As a preferred technical solution, the theoretical urea injection quantity m_calc is corrected using a temperature correction coefficient, an aging correction coefficient, and an adaptive correction coefficient to update the urea injection quantity, thus obtaining the base value m{base} of the urea injection quantity, specifically: m{base}=m_calc×α×β×γ, Where m{base} is the corrected base value of urea injection rate, α is the temperature correction coefficient, β is the aging correction coefficient, γ is the adaptive correction coefficient, and the theoretical urea injection rate m_calc is based on the stoichiometric ratio and NO. x The theoretical urea demand is calculated from the original emissions.

[0018] As a preferred technical solution, the adaptive correction coefficient is calculated based on the NO value from the SCR reaction kinetic model. x Conversion efficiency and NO x The actual NO measured by the sensor x The deviation in conversion efficiency is dynamically adjusted.

[0019] As a preferred technical solution, the adaptive coefficient correction process is implemented using a five-state machine, which includes, in sequence, a closed-loop injection control state, a state that meets the triggering conditions, a state that stops injection and empties the catalyst, a state that calculates the adaptive correction coefficient, and a stable period of under-injection with a fixed ammonia-nitrogen ratio. The triggering conditions from the closed-loop injection control state to the state that meets the triggering conditions include Class A normal wear triggering conditions and / or Class B abnormal emission triggering conditions.

[0020] As a preferred technical solution, if any one of the Class A normal wear triggering conditions is met, it is determined that the triggering condition is met, and the adaptive correction coefficient is adjusted to perform adaptive urea injection control. The Class A normal wear triggering conditions include: (A1) The engine's current running time exceeds the set value; (A2) The cumulative operating time of the nitrogen oxide treatment system exceeds the threshold; (A3) The cumulative mass of reducing agent sprayed exceeds the limit; If any one of the Class B abnormal emission triggering conditions is met, the triggering condition is determined to be met, and the adaptive correction coefficient is adjusted for adaptive urea injection control. The Class B abnormal emission triggering conditions include: (B1) The deviation between model efficiency and sensor efficiency exceeds a preset ratio; (B2) After the NH3 leakage monitoring function is completed, the system will automatically request to trigger.

[0021] Compared with the prior art, the present invention has the following beneficial effects: 1) This invention addresses the problem of insufficient prediction accuracy for ammonia storage under high and low temperature conditions by using a catalyst temperature-zoned ammonia storage state prediction model. The LSTM network utilizes historical operating conditions and time-series characteristics to improve the prediction accuracy of injection volume under transient conditions. The periodic adaptive coefficient correction process can compensate in real time for performance degradation caused by catalyst aging and sensor drift, ensuring that NO levels remain stable after catalyst aging. x Control accuracy is improved by approximately 10%, extending the service life of the SCR system. Simultaneously, feedforward prediction compensates for the lag in traditional feedback control, improving dynamic operating condition response speed by approximately 30%. A temperature-zoned ammonia storage state prediction model is selected, combined with a three-zone control strategy, LSTM feedforward prediction, and multi-module collaborative control with periodic adaptive correction, achieving precise control of urea injection in the diesel engine SCR system across all operating conditions.

[0022] 2) This invention solves the problem of insufficient prediction accuracy caused by the use of a uniform ammonia storage model in existing technologies for high and low temperature conditions by automatically selecting an ammonia storage state prediction model. In the low-temperature region, the temperature sensitivity coefficient is used to accurately match the change law of catalyst ammonia storage capacity with temperature, improving the accuracy of ammonia storage state prediction under low-temperature conditions and providing a reliable basis for urea injection quantity calculation. In the high-temperature region, a simplified model with a constant maximum ammonia storage capacity is used, which reduces calculation complexity while ensuring prediction accuracy, improves control real-time performance, and is conducive to optimizing the ammonia storage prediction effect under all operating conditions. This provides core support for improving transient control accuracy and catalyst aging compensation, ensuring NO control of the SCR system across the entire temperature range. x The conversion efficiency and NH3 leakage control effectiveness meet stringent emission regulations.

[0023] 3) This invention uses a three-segment control strategy to achieve precise zonal control of urea injection based on the estimated ammonia storage and catalyst temperature, which specifically solves the problem that traditional control strategies cannot adapt to different ammonia storage and temperature conditions.

[0024] 4) This invention constructs a urea injection quantity calculation and correction system. Through a three-level correction formula consisting of temperature correction coefficient, aging correction coefficient, and adaptive correction coefficient, the theoretical urea injection quantity is compensated in multiple dimensions. It accurately corrects temperature fluctuations, long-term catalyst degradation, and real-time control deviations, effectively compensating for performance degradation caused by catalyst aging and obtaining the basic value of urea injection quantity. The LSTM feedforward prediction value and the basic value of urea injection quantity are dynamically weighted and integrated. The weight ratio of feedforward and basic control can be adaptively adjusted according to the degree of catalyst aging. The transient response accuracy and steady-state control stability are balanced throughout the system's life cycle, further improving the control adaptability under complex operating conditions. Attached Figure Description

[0025] Figure 1 This is a schematic flowchart of an adaptive urea injection control method for a diesel engine SCR system according to the present invention. Figure 2 This is a schematic diagram of the dual-mode EKF ammonia storage prediction process in this invention; Figure 3 This is a schematic diagram illustrating the logic of calculating the urea injection volume using the three-segment control strategy in this invention. Figure 4 This is a schematic diagram of the adaptive correction mechanism based on a five-state machine in this invention; Figure 5 This is a schematic diagram of the hierarchical structure of the LSTM network in this invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0027] Example 1 This embodiment relates to an adaptive urea injection control method for a diesel engine SCR system, such as... Figure 1 This includes the following steps: Step S1: Monitor the SCR catalyst temperature in real time using a temperature sensor, and select the ammonia storage state prediction model based on the temperature value: when the temperature is higher than the first temperature threshold (e.g., 350℃), use the slow time-varying EKF model; when the temperature is lower than or equal to the first temperature threshold, use the low temperature EKF model. Step S2: Based on the selected ammonia storage status prediction model, predict the ammonia storage quantity in real time, and combine it with NO x The measured values ​​of the sensor and the NH3 sensor are used to correct the estimated ammonia storage in real time using Kalman filtering to obtain the optimal estimate and solve the problems of unknown initial state, model error and sensor noise. Step S3: Based on the estimated ammonia storage and the real-time monitored catalyst temperature, calculate the base value of urea injection rate according to the three-segment control strategy: when the estimated ammonia storage exceeds the peak threshold, shut off injection; when the estimated ammonia storage is lower than the peak threshold and the temperature is lower than the second temperature threshold (e.g., 250°C), increase injection to build up reserves; when the ammonia storage exceeds the peak threshold and the temperature is higher than or equal to the second temperature threshold, maintain the original urea injection rate. Step S4: Based on the Long Short-Term Memory (LSTM) network, perform time-series modeling of the historical engine operating condition feature sequence, estimate the current optimal urea injection quantity feedforward value, and output the current urea injection quantity feedforward prediction value; Step S5: Combine the feedforward predicted injection value output by the LSTM network with the basic value of urea injection quantity calculated by the three-segment control strategy to obtain the final urea injection quantity and output it to the urea nozzle. Step S6: Periodically execute the adaptive coefficient correction process: By comparing the conversion efficiency calculated by the SCR reaction kinetic model with the conversion efficiency measured by the sensor, calculate the adaptive correction coefficient and update the control parameters (urea injection amount) to obtain the final urea injection amount; Step S7: When the system efficiency deviation exceeds the limit, the NH3 leakage diagnosis process is triggered. The ammonia consumption is calculated by stopping urea injection and integrating the data to distinguish between NH3 leakage and NO leakage. x Types of malfunctions due to excessive emissions.

[0028] In step S1, to achieve accurate prediction of the catalyst's ammonia storage state, this invention employs a dual-mode prediction method based on extended Kalman filtering (EKF), such as... Figure 2 First, the ammonia storage process is divided into two working regions based on catalyst temperature: a high-temperature region and a low-temperature region. Different observer models are used for each region. When the catalyst temperature is above 350℃, the slow time-varying EKF model is used for state prediction because the catalyst's NH3 storage capacity is very low and changes slowly. When the temperature is below or equal to 350℃, the low-temperature EKF model is used for state prediction because the maximum NH3 storage capacity is a strong function of temperature. This temperature-zoned differentiated modeling method can effectively improve the prediction accuracy under different operating conditions.

[0029] To address the prediction error caused by incomplete urea decomposition under low-temperature conditions, this invention introduces a reducing agent energy efficiency ratio parameter to correct the urea injection quantity. This energy efficiency ratio is defined as the ratio of the actual mass of NH3 participating in the reaction to the theoretical mass of injected urea. Through real-time monitoring and calculation, the urea injection quantity in the low-temperature zone is dynamically corrected, thereby improving the accuracy of ammonia storage state prediction.

[0030] To address the problem of prediction when the initial ammonia storage state is unknown, this invention addresses this issue by appropriately setting the initial values ​​of the error covariance matrix and process noise covariance matrix of the EKF model. This allows the observer to achieve accurate predictions even with an unknown initial ammonia storage state, after sufficient computation time. Specifically, a larger initial error covariance and an appropriate process noise covariance are used to ensure the filter has a faster convergence speed in the initial stage. As observation data accumulates, the estimated value gradually converges to the true state.

[0031] The EKF state-space model consists of two parts: the state equation and the observation equation. The state equation describes the dynamic changes in ammonia storage capacity. Θ(k+1)=A·Θ(k)+B·u(k)+w(k), Wherein, Θ(k) is the ammonia storage at time k, i.e., the actual mass of NH3 stored by the catalyst at time k, in grams (g); u(k) is the urea injection rate at time k, i.e., the urea injection mass flow rate calculated and executed by the controller at time k, in grams per second (g / s); w(k) is the process noise, a random variable characterizing model uncertainty and external disturbances, following a normal distribution with a mean of 0 and a covariance of Q, where Q is the observation noise covariance matrix; A is the state transition coefficient, a dimensionless coefficient describing the natural decay characteristics of the ammonia storage over time, which is related to the catalyst temperature and the NH3 desorption rate, and its value range is usually 0.95~0.99; B is the input gain coefficient, a dimensionless coefficient describing the efficiency of injecting urea into ammonia storage, which is related to catalyst activity, temperature, and urea decomposition efficiency.

[0032] Observation equation description NO x Relationship between measured values ​​and ammonia storage: y(k) = C•Θ(k) + v(k), Where y(k) is NO at time k. x sensor NO x The measured value is in ppm (parts per million), and is usually taken from the downstream NO. x Concentration is used as the observed variable; v(k) is the observation noise, which follows a normal distribution with a mean of 0 and a covariance of R, where R is the observation noise covariance matrix; C is the observation coefficient matrix, describing the relationship between ammonia storage and NO. x The coefficient matrix of the relationship between measured values, and NO x Conversion efficiency is related to catalyst characteristics.

[0033] The model for the maximum ammonia storage capacity in the low-temperature zone is: Θmax(T) = S1•exp(-S2•T), Wherein, S1 is the ammonia storage capacity coefficient, a model parameter related to the catalyst coating material, support structure, and total coating amount, in g, reflecting the maximum ammonia storage potential of the catalyst; S2 is the temperature sensitivity coefficient, describing the sensitivity of ammonia storage capacity to decrease with increasing temperature, in 1 / ℃ (per degree Celsius), related to the thermodynamic characteristics of NH3 adsorption, with a typical value range of 0.005~0.015; T is the catalyst temperature, i.e., the real-time temperature of the SCR catalyst bed, in ℃ (degree Celsius); Θmax(T) is the maximum ammonia storage capacity corresponding to temperature T; exp(•) is an exponential function.

[0034] EKF recursive calculation includes the following steps: (1) State prediction: predict the state value at the current time based on the estimated value at the previous time and the current control input; (2) Covariance prediction: calculate the error covariance matrix of the predicted state; (3) Kalman gain calculation: calculate the optimal Kalman gain based on the predicted covariance and observation noise; (4) State update: correct the state estimate using the deviation between the actual observation value and the predicted value; (5) Covariance update: calculate the updated error covariance matrix for the recursive calculation at the next time.

[0035] In step S3, to achieve precise control of the urea injection quantity, this invention employs a three-segment adaptive injection control strategy based on ammonia storage feedback, such as... Figure 3 This strategy divides the control process into three operating zones—a supersaturated zone, a low-temperature ammonia storage zone, and a high-efficiency conversion zone—based on the ratio of real-time ammonia storage to maximum ammonia storage and catalyst temperature. Different injection control logics are employed for each zone, including: When the system reaches the supersaturation zone (actual ammonia storage exceeds the peak threshold), the urea injection is shut off to prevent NH3 leakage. When in the low-temperature ammonia storage zone (ammonia storage is less than the peak value and temperature is below 250℃), the system increases the urea injection rate to establish ammonia reserves, and then gradually reduces the injection rate according to the temperature rise. When the high-efficiency conversion zone (ammonia storage is less than the peak value and the temperature is higher than or equal to 250°C) is reached, the system maintains the original urea injection rate. At this time, the SCR catalyst is in a high-efficiency working state, and the excess NH3 is treated by the downstream ammonia escape catalyst (ASC).

[0036] This partition control strategy can guarantee NO x While improving conversion efficiency, it effectively reduces NH3 leakage and urea consumption.

[0037] To further improve control accuracy, this invention applies multi-factor correction to the theoretical urea injection rate. The correction coefficients include three parts: temperature correction coefficient, aging correction coefficient, and adaptive correction coefficient. The temperature correction coefficient adjusts the injection rate based on the real-time catalyst temperature; the aging correction coefficient compensates for the injection rate based on the catalyst usage time; and the adaptive correction coefficient is dynamically adjusted based on the deviation between the conversion efficiency calculated by the SCR reaction kinetic model and the actual measured efficiency. The final injection rate calculation formula is as follows: m{base}=m_calc×α×β×γ, Wherein, α is the temperature correction coefficient, a dimensionless coefficient, which adjusts the injection quantity according to the real-time catalyst temperature T to compensate for the influence of temperature on reaction kinetics, with a typical range of 0.8~1.2; β is the aging correction factor, a dimensionless coefficient, which is determined based on the cumulative catalyst usage time or the cumulative NO treatment. xMass is used to compensate for the decrease in activity caused by catalyst aging. The initial value is 1.0, which gradually increases with aging. γ is an adaptive correction coefficient, a dimensionless coefficient, which is dynamically adjusted based on the deviation between the conversion efficiency calculated by the SCR reaction kinetic model and the actual sensor measurement efficiency. Its value range is usually limited to 0.5~2.0. m{base} represents the final urea injection rate, which is the actual urea injection mass flow rate after multi-factor correction, in g / s (grams per second). m_calc is the theoretical urea injection rate, based on stoichiometry and NO. x The theoretical urea requirement calculated from the original emissions is determined by engine operating conditions and the original NO. x Emissions decision.

[0038] Compared to traditional open-loop control strategies, the three-segment adaptive injection control strategy of this invention can achieve a 13.3% reduction in urea injection quantity, a 25.6% reduction in ammonia emissions after SDPF (diesel particulate filter coated with SCR catalyst), a 25.4% reduction in tailpipe ammonia emissions, and a reduction in tailpipe NO. x Total emissions decreased by 11.4%, significantly improving the overall performance of the SCR system.

[0039] The adaptive correction factor is calculated according to the following formula: γ = η_sensor / η_model Where γ is the adaptive correction coefficient, a dimensionless coefficient, representing the sensor's measured NO. x Conversion efficiency η_sensor and NO calculated based on SCR reaction kinetics model x The ratio of conversion efficiency η_model is used to compensate for model errors.

[0040] Adaptive correction coefficient γ based on SCR reaction kinetics model and NO x The dynamic adjustment of the deviation of the sensor's measured conversion efficiency combines model calculation with the closed loop of feedback from the vehicle's sensors, realizing online real-time self-calibration of control parameters. This not only improves the control accuracy under all operating conditions but also reduces the reliance on large-scale calibration experiments. At the same time, it works in synergy with the periodic adaptive coefficient correction process to further enhance the system's adaptability to catalyst aging, sensor drift, and changes in operating conditions. This ensures that the SCR system meets China VI and above emission regulations and RDE cycle requirements throughout its entire life cycle, improving system reliability and industrial adaptability.

[0041] To ensure the stability of the correction, the adaptive correction coefficient is limited: γ_min≤γ≤γ_max (usually 0.5~2.0), where γ_min and γ_max are the lower and upper limits of the correction factor, respectively.

[0042] Through periodic adaptive correction, the system can compensate for model errors caused by factors such as catalyst aging and sensor drift, and maintain long-term stable control accuracy.

[0043] Feedforward prediction can adjust the injection quantity in advance when operating conditions change, compensating for the lag of traditional feedback control, and improving the dynamic operating condition response speed by about 30%. Through online adaptive updates, it can track the characteristic changes caused by catalyst aging and automatically adjust the injection quantity prediction. Compared with traditional methods, the NO after catalyst aging is significantly reduced. x The control accuracy is improved by about 10%, extending the service life of the SCR system.

[0044] In step S4, the feedforward prediction of the current urea injection amount is output based on the LSTM network. The LSTM network adopts a network structure of two-stage LSTM stacking + fully connected output, as follows: Figure 5 As shown in Table 1, the engine operating condition feature sequence with time step T and dimension 5 is input into an LSTM network. The first LSTM layer contains 128 hidden units and returns the output of all time steps, used to extract low-level temporal features. The second LSTM layer contains 64 hidden units and returns only the output of the last time step, used to extract high-level temporal features. After the outputs of the two LSTM layers, they are successively normalized by BatchNorm, regularized by Dropout (Dropout probability p=0.2), and activated by ReLU (Dropout probability p=0.2). Finally, the final injection quantity prediction is obtained through mapping by a fully connected layer (FC). The fully connected layer consists of fully connected layer 1 and fully connected layer 2. Fully connected layer 1 maps the 64-dimensional features to 32-dimensional features and connects them to a corrected linear unit (ReLU activation layer) to further introduce nonlinearity. Fully connected layer 2 finally maps the 32-dimensional features to 1-dimensional features and outputs the final injection quantity prediction under the current operating condition.

[0045] Table 1

[0046] At each time step, select the 5-dimensional features from Table 2, maintain a sliding window of length T, and save the 5-dimensional feature sequence of the most recent T time steps as the input of the LSTM network. T is generally 10~30, which can be adjusted according to the computing power of the ECU.

[0047] Table 2

[0048] In the offline training phase, network parameters were trained using multi-condition operating data collected from bench experiments. The training hyperparameters were: optimizer AdamW, weight decay of 1e-4, initial learning rate of 1e-3, cosine annealing decay, batch size of 32, 10,000 training epochs, and features normalized to the [0,1] interval. LSTM was used to learn the mapping relationship from the operating data, capturing the temporal impact of historical operating condition changes, reducing the reliance on large-scale fine-tuning calibration experiments. Compared to traditional static MAP, the prediction error of transient operating condition injection volume was reduced by approximately 15%~20%. x Peak emissions are significantly reduced, meeting the requirements of the RDE cycle. Online inference requires only one LSTM forward propagation, and a single prediction on a typical automotive ECU takes only a few milliseconds, meeting real-time control requirements.

[0049] Online updates via a sliding window. Input the current engine speed (engine_speed), current engine load (engine_load), current exhaust temperature (exhaust_temp), and current NO. x Concentration raw_nox, previous injection amount last_injection, history buffer history_buffer, window size window_size, output updated history buffer history_buffer.

[0050] In step S5, after obtaining the feedforward estimate of the current urea injection volume in step S4, it is fused with the calculated baseline value of the urea injection volume to obtain the final urea injection volume: u{final}=ω*m{lstm}+(1-ω)*m{base}, Wherein, ω is the fusion coefficient, which is adaptively adjusted according to catalyst aging, as shown in Table 3.

[0051] Table 3

[0052] The fusion coefficient ω is adaptively weighted according to the degree of catalyst aging, and increases monotonically with the degree of catalyst aging. Specifically: when the catalyst is fresh, ω is 0.10, and the control is based on the basic value of urea injection amount and supplemented by the LSTM feedforward prediction value to ensure the stability of the basic control; when the catalyst is moderately aged, ω is 0.30, and the weight of the LSTM feedforward prediction value is increased to compensate for the loss of catalytic efficiency; when the catalyst is severely aged, ω is 0.65, and the control is based on the LSTM feedforward prediction value to achieve dynamic correction, realize precise injection compensation at different aging stages, and achieve adaptive and smooth transition of feedforward and basic control weights throughout the entire life cycle of the catalyst.

[0053] In step S6, to achieve dynamic adjustment of the adaptive correction coefficient γ of the SCR reaction kinetic model parameters, this invention designs an adaptive coefficient correction method based on a five-state machine. This method, through state transitions, periodically adjusts the parameter correction process of the adaptive correction coefficient without affecting normal emission control. The entire correction process includes five working states: closed-loop injection control state, trigger condition met state, catalyst purging state, adaptive correction coefficient calculation state, and fixed ammonia-nitrogen ratio (ANR) under-injection stabilization period, as follows: Figure 4 .

[0054] The triggering conditions for state transitions are divided into two types: Category A, normal wear triggering, and Category B, abnormal emission triggering. Category A conditions include engine operating time exceeding a set value and accumulated nitrogen oxides (NOx) emissions from the denitrification system (DeNOx). x Type A conditions include normal wear and tear such as operating time exceeding the threshold and cumulative urea (reducing agent) injection exceeding the limit; Type B conditions include abnormal emissions such as the deviation between model efficiency and sensor efficiency exceeding the ratio and automatic requests after the completion of NH3 leakage monitoring. When any trigger condition is met, the system switches from closed-loop injection control to the state where the trigger condition is met, and then sequentially executes steps such as stopping injection and clearing, adaptive correction coefficient calculation, and under-injection stabilization.

[0055] By periodically calculating the adaptive correction coefficient and updating the final urea injection quantity, the system can compensate for model errors caused by factors such as catalyst aging and sensor drift, and maintain long-term stable control accuracy.

[0056] In step S7, to achieve accurate diagnosis of NH3 leakage faults, this invention employs a diagnostic method based on urea injection stoppage and ammonia storage consumption integral. This method first determines the activation timing of the diagnostic function through a multi-dimensional activation condition determination mechanism. Activation occurs when the enable switch is turned on, the time interval exceeds the calibration value, the injection history exceeds limits, the supply rate exceeds limits, or NO... x The NH3 leak diagnosis process is triggered when seven conditions are met simultaneously, including sufficient level, stable temperature, and efficiency deviation exceeding the limit.

[0057] Once the diagnostic process begins, the system immediately stops urea injection and calculates the cumulative ammonia consumption during the injection stoppage period through integration. Simultaneously, the system continuously monitors NO. x Changes in conversion efficiency. Based on cumulative ammonia storage consumption and NO... x The system can accurately distinguish different fault types based on the combined characteristics of conversion efficiency: if the cumulative ammonia storage consumption exceeds the limit and NO... x If the conversion efficiency does not drop below the set threshold, it is determined to be an NH3 leakage fault (flag bit = 2); if NO x If the conversion efficiency drops below a set threshold and the cumulative ammonia storage consumption is less than the limit, it is judged as NO. xEmissions exceeding standards fault (flag bit = 1); if the diagnosis cannot be completed within the calibrated time, it is determined that the diagnosis is incomplete (flag bit = 0).

[0058] To improve the reliability of diagnostic results, this invention introduces the concept of diagnostic confidence: Confidence=min(ΔΘ / ΔΘ_limit,1.0)×100%, Where Confidence is the diagnostic confidence level; ΔΘ is the cumulative ammonia consumption; ΔΘ_limit is the diagnostic threshold ammonia consumption, which is the minimum ammonia consumption threshold required to determine NH3 leakage fault, in g (grams), and is determined based on catalyst characteristics and operating conditions; min(•) is the minimum value function.

[0059] A higher confidence level indicates a higher degree of reliability in the diagnostic results. This method effectively distinguishes between NH3 leakage and NO leakage by integrating ammonia storage consumption. x It can diagnose two types of faults, namely, exceeding emission standards, with a diagnostic accuracy rate of over 95%, a response time of less than 300 seconds, and a false alarm rate of less than 1%.

[0060] Example 2 This embodiment also relates to an adaptive urea injection control system for a diesel engine SCR system. The system adopts a layered architecture design, including three core layers: the vehicle control layer, the network communication layer, and the cloud platform layer. Through data interaction and collaborative work between the layers, the adaptive control, state prediction, and fault diagnosis functions of the SCR system are realized.

[0061] The vehicle control layer is responsible for collecting real-time SCR system operating data and performing core control functions such as ammonia storage status prediction, adaptive injection control, fault diagnosis, and model correction. It also integrates an LSTM feedforward estimation module to output the feedforward prediction of the current urea injection quantity. The network communication layer realizes 5G / 4G data transmission, V2X vehicle-road collaboration, and GPS / BeiDou positioning functions through the T-Box communication module, and supports OTA upgrades, remote monitoring, and data reporting. The cloud platform layer is responsible for regularly updating and optimizing model parameters and pushing them to the vehicle terminal via OTA to achieve long-term adaptive catalyst aging.

[0062] The vehicle control layer includes: SCR control unit (DCU), whose hardware requirements are main frequency ≥ 200MHz and RAM ≥ 512KB; NO x One sensor each upstream and downstream, measuring range 0~5000ppm; one NH3 sensor downstream, measuring range 0~100ppm; three temperature sensors in total: upstream, downstream, and inside the catalyst; and a T-Box communication module supporting 5G / 4G and GPS / BeiDou positioning.

[0063] The hardware of the cloud platform layer includes: a big data storage cluster with petabyte-level storage capacity; an AI training server equipped with GPU acceleration and supporting deep learning frameworks such as TensorFlow / PyTorch; and a real-time computing engine that supports streaming data processing.

[0064] This system optimizes the SCR system in the vehicle controller by using networked data and considering factors such as the actual environment and road conditions. This reduces urea consumption and ammonia leakage while meeting emission targets. The system can be seamlessly integrated with traditional EKF ammonia storage prediction and PID feedback control frameworks without requiring large-scale modifications to existing control systems, making it easy for industrial applications.

[0065] Bench tests have verified that this invention achieves significant improvements compared to traditional control strategies, as shown in Table 4. Table 4

[0066] Urea consumption decreased from 1.62 g / kWh to 1.43 g / kWh, an improvement of 13.3%; NH3 emissions decreased from 5.15 ppm to 4.1 ppm, and ammonia emissions after SDPF improved by 25.6%; ammonia emissions from the cold-start tailpipe decreased from 13.5 g to 10.8 g, an improvement of 25.4%; tailpipe NO... x Emissions have been reduced from the traditionally controlled level of 0.35 g / kWh to 0.31 g / kWh, and NO2 in the tailpipe has also decreased. x Emissions were reduced by 11.4%. Test results show that, while ensuring emissions meet standards, this invention effectively reduces urea consumption and NH3 leakage, significantly improving the overall performance of the SCR system.

[0067] Example 3 The electronic device of this invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in read-only memory (ROM) or loaded from a storage unit into random access memory (RAM). The RAM may also store various programs and data required for device operation. The CPU, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0068] Multiple components in the device are connected to the I / O interface, including: input units such as keyboards and mice; output units such as various types of displays and speakers; storage units such as disks and optical discs; and communication units such as network interface cards (NICs), modems, and wireless transceivers. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0069] The processing unit performs the various methods and processes described above. For example, in some embodiments, the methods may be implemented as computer software programs tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via ROM and / or a communication unit. When the computer program is loaded into RAM and executed by the CPU, one or more steps of the methods described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute the methods by any other suitable means (e.g., by means of firmware).

[0070] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0071] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0072] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory, optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0073] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An adaptive urea injection control method for a diesel engine SCR system, characterized in that, Includes the following steps: The ammonia storage state prediction model is automatically selected based on the real-time monitored catalyst temperature of the diesel engine SCR system to predict the ammonia storage amount in real time. When the temperature is higher than the first temperature threshold, the slow time-varying EKF model is used; otherwise, the low temperature EKF model is used. Based on the estimated ammonia storage and the monitored catalyst temperature, the baseline value of urea injection rate is calculated according to the three-section control strategy. The historical engine operating condition feature sequence is modeled based on the LSTM network, and the feedforward prediction of the current urea injection quantity is output. By combining the feedforward estimate of the current urea injection volume with the baseline value of the urea injection volume, the final urea injection volume is obtained and output to the urea nozzle; In the method, an adaptive coefficient correction process is periodically executed to adjust the adaptive correction coefficient and update the final urea injection quantity.

2. The adaptive urea injection control method for a diesel engine SCR system according to claim 1, characterized in that, The specific low-temperature EKF model is as follows: Θmax(T) = S1•exp(-S2•T), Where S1 and S2 are the ammonia storage capacity coefficient and temperature sensitivity coefficient, respectively, T is the catalyst temperature, and Θmax(T) is the maximum ammonia storage capacity corresponding to temperature T. The slow time-varying EKF model uses a constant maximum ammonia storage capacity in the high-temperature region.

3. The adaptive urea injection control method for a diesel engine SCR system according to claim 1, characterized in that, The calculation of urea injection volume according to the three-segment control strategy includes: When the estimated ammonia storage reaches the peak threshold, urea injection is shut off; When the estimated ammonia storage is below the peak threshold and the temperature is below the second temperature threshold, the urea injection rate is increased to establish ammonia reserves. When the estimated ammonia storage is below the peak threshold and the temperature is above or equal to the second temperature threshold, the original urea injection rate is maintained.

4. The adaptive urea injection control method for a diesel engine SCR system according to claim 1, characterized in that, The calculation process for the feedforward estimate of the current urea injection volume includes: Maintain a sliding window of length T to store the engine operating condition feature sequence of the most recent T time steps. The engine operating condition feature sequence of each time step includes engine speed, engine load, SCR inlet exhaust temperature, and SCR inlet NO. x Concentration and the amount of urea injected in the previous step; The sliding window is updated each time a new engine operating condition feature sequence is acquired. The sliding window data is input into the trained LSTM network, and the forward calculation yields the feedforward estimate of the current urea injection volume.

5. An adaptive urea injection control method for a diesel engine SCR system according to claim 1 or 4, characterized in that, The LSTM network adopts a two-level stacked structure: The first LSTM layer contains 128 hidden units and returns the output of all time steps, which is used to extract low-level temporal features; The second LSTM layer contains 64 hidden units and only returns the output of the last time step, which is used to extract high-level temporal features; After normalization by BatchNorm and regularization by Dropout, the feedforward estimate of the current urea injection amount is obtained by mapping through two fully connected layers.

6. The adaptive urea injection control method for a diesel engine SCR system according to claim 1, characterized in that, By combining the feedforward estimate of the current urea injection volume with the baseline value of the urea injection volume, the final urea injection volume is obtained as follows: u{final}=ω*m{lstm}+(1-ω)*m{base}, where u{final} is the final urea injection amount; m{lstm} is the feedforward prediction value of the LSTM output; m{base} is the base value of the calculated urea injection amount; ω is the fusion coefficient, which is adaptively adjusted according to the catalyst aging degree.

7. The adaptive urea injection control method for a diesel engine SCR system according to claim 6, characterized in that, The theoretical urea injection rate m_calc is corrected using temperature correction coefficient, aging correction coefficient, and adaptive correction coefficient to obtain the base value m{base} of the urea injection rate, specifically: m{base}=m_calc×α×β×γ, Where m{base} is the corrected base value of urea injection rate, α is the temperature correction coefficient, β is the aging correction coefficient, γ is the adaptive correction coefficient, and the theoretical urea injection rate m_calc is based on the stoichiometric ratio and NO. x The theoretical urea demand is calculated from the original emissions.

8. The adaptive urea injection control method for a diesel engine SCR system according to claim 7, characterized in that, The adaptive correction coefficient is calculated based on the NO value from the SCR reaction kinetics model. x Conversion efficiency and NO x The actual NO measured by the sensor x The deviation in conversion efficiency is dynamically adjusted.

9. The adaptive urea injection control method for a diesel engine SCR system according to claim 1, characterized in that, The adaptive coefficient correction process is implemented using a five-state machine, which includes, in sequence, a closed-loop injection control state, a state that meets the trigger conditions, a state that stops injection and empties the catalyst, a state that calculates the adaptive correction coefficient, and a stable period of under-injection with a fixed ammonia-nitrogen ratio. The trigger conditions from the closed-loop injection control state to the state that meets the trigger conditions include Class A normal wear trigger conditions and / or Class B abnormal emission trigger conditions.

10. The adaptive urea injection control method for a diesel engine SCR system according to claim 9, characterized in that, If any one of the Class A normal wear triggering conditions is met, the triggering condition is determined to be met, and the adaptive correction coefficient is adjusted for adaptive urea injection control. The Class A normal wear triggering conditions include: (A1) The engine's current running time exceeds the set value; (A2) The cumulative operating time of the nitrogen oxide treatment system exceeds the threshold; (A3) The cumulative mass of reducing agent sprayed exceeds the limit; If any one of the Class B abnormal emission triggering conditions is met, the triggering condition is determined to be met, and the adaptive correction coefficient is adjusted for adaptive urea injection control. The Class B abnormal emission triggering conditions include: (B1) The deviation between model efficiency and sensor efficiency exceeds a preset ratio; (B2) After the NH3 leakage monitoring function is completed, the system will automatically request to trigger.

Images