Single-rotor engine stratified fuel injection fuel supply system and control method
By calculating the angle advance compensation in a single-rotor engine and predicting the in-cylinder flow field using a long short-term memory network, combined with dual knock sensors to locate the knock source, and implementing a differentiated suppression strategy, the problems of injection initiation angle positioning accuracy and knock risk in single-rotor engines under high-speed rapid acceleration conditions are solved, thereby improving combustion efficiency and safety.
Patent Information
- Application Number
- CN202611421673.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-14
- Publication Date
- 2026-11-17
AI Technical Summary
Under high-speed, rapid acceleration transient conditions, the stratified fuel injection system of a single-rotor engine suffers from problems such as insufficient accuracy in positioning the injection initiation angle, inadequate ability to predict and suppress knock risks, and imperfect safety constraints on injection parameters.
By acquiring the eccentric shaft angle and pedal opening parameters, the angle advance compensation is calculated, the in-cylinder flow field is predicted using a long short-term memory network, the knock source is located by combining dual knock sensors, and a differentiated suppression strategy is implemented to dynamically adjust the injection parameters to improve the injection control accuracy and knock suppression capability.
It achieves precise control of stratified injection under high-speed and rapid acceleration conditions, reduces the continuous development of knock, and improves combustion efficiency and safety.
Smart Images

Figure CN123109897A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of engine fuel injection control technology, and more specifically, to a stratified fuel injection supply system and control method for a single-rotor engine. Background Technology
[0002] Due to its unique structure, single-rotor engines require precise control of the initiation angles and injection quantity distribution of each stage of pilot injection and main injection in a stratified injection fuel supply system to achieve efficient combustion. Under high-speed, rapid acceleration transient conditions, existing control methods typically employ a fixed-delay compensation strategy based on the current engine speed to control the initiation angle of each injection stage, use a post-detection response method to suppress knock, rely on the spark plug position ion current signal for knock detection, and use a fixed energy threshold to determine the knock source.
[0003] However, existing control methods have the following main technical problems: Under high-speed rapid acceleration, the eccentric shaft angle offset corresponding to the system delay increases significantly, causing the actual injection initiation angle to lag far behind the calculated target value, and deteriorating the spatial positioning accuracy of stratified injection; knock suppression adopts a post-detection response method, which lags in response during the rapid rise of cylinder temperature during rapid acceleration, and knock continues to develop for several cycles after being detected; single-point ion current signals based on spark plug position lack spatial resolution and cannot distinguish whether the knock source is located in the pilot layer or at the end of the main injection layer, and a uniform suppression strategy may exacerbate knock in specific layer regions; during rapid acceleration, the knock source dynamically migrates between the pilot layer and the end of the main layer, and the fixed energy judgment threshold causes knock source misjudgment during the migration transition period. Summary of the Invention
[0004] This invention provides a stratified injection fuel supply system and control method for a single-rotor engine, which solves the technical problems in related technologies such as insufficient accuracy of stratified injection timing control, lack of knock risk prediction and suppression capability, and imperfect safety constraints of injection parameters for single-rotor engines under transient operating conditions.
[0005] This invention discloses a stratified fuel injection control method for a single-rotor engine, comprising the following steps: obtaining an eccentric shaft angle sequence and calculating the current angular velocity and angular acceleration values through fitting; obtaining the pedal opening value and the pedal opening change rate. When the angular acceleration value exceeds the preset acceleration threshold and the rate of change of the pedal opening is positive, the advance prediction control mode is activated, and the angular advance compensation amount is calculated based on the current angular velocity value, the angular acceleration value and the total system delay time. Based on the angular acceleration value and the rate of change of intake pressure, the end gas temperature of the future preset number of cycles is extrapolated and predicted. The predicted end gas temperature is compared with the fuel auto-ignition critical temperature to calculate the knock risk level value. When the knock risk level value exceeds the preset prevention threshold and no actual knock occurs, the pilot injection volume is reduced proportionally and the main injection start angle is delayed to generate preventive correction parameters. The angular velocity value, the angular acceleration value, and the pedal opening value are input into the long short-term memory network turbulent kinetic energy prediction model, which outputs a turbulent kinetic energy prediction sequence and a vortex direction angle prediction sequence. The predicted value that matches the actual injection execution time is selected, and the starting angle of each segment is determined by combining the empirical formula for spray penetration distance. The angle advance compensation amount is superimposed to generate the advance compensation injection starting angle of each segment. A safety constraint check is performed on the injection start angle of each segment after advance compensation. When the interval between the injection angles of adjacent segments does not meet the angle margin corresponding to the minimum atomization mixing time, the injection pulse width of the subsequent segment is reduced and the injection pressure is increased. Bandpass filtering is performed on the signals collected by the near-end detonation sensor and the far-end detonation sensor respectively, and the detonation energy value is calculated. The detonation source location is determined based on the detonation energy ratio and the time delay value corresponding to the peak value of the cross-correlation function. The judgment threshold is dynamically adjusted based on the changing trend of the detonation risk level value. Differential suppression is performed based on the detonation source location identifier. The suppression correction is superimposed on the stratified injection timing parameters adjusted by safety constraints. After a safety range check, the final stratified injection control command is generated.
[0006] Furthermore, the angle advance compensation amount is calculated as follows: the product of the current angular velocity value and the total system delay time is taken as the uniform rotation component, and half of the product of the angular acceleration value and the square of the total system delay time is taken as the accelerated rotation component. The uniform rotation component and the accelerated rotation component are summed to obtain the angle advance compensation amount. The total system delay is the sum of the sensor acquisition delay, the model inference calculation delay, and the injector needle valve mechanical response delay.
[0007] Furthermore, the knock risk level value is the ratio of the predicted end gas temperature to the fuel auto-ignition critical temperature; when the angular acceleration value continues to increase, a positive correction term is added to the predicted end gas temperature value, the positive correction term being the product of the rate of change of the angular acceleration value and the positive correction proportional coefficient, and the corrected predicted end gas temperature value is substituted into the calculation of the knock risk level value. The rate of change of the angular acceleration value is obtained by calculating the first-order time difference of a series of consecutive cycles of angular acceleration values and dividing it by the corresponding sampling time interval. When all consecutive difference values in the angular acceleration value series are positive, it is determined that the angular acceleration value is continuously increasing.
[0008] Furthermore, the preventive correction parameter is generated as follows: the preventive correction pilot injection quantity is equal to the pilot injection quantity calculated by the conventional control logic multiplied by the reduction factor, and the reduction factor is equal to the product of the pilot injection quantity reduction ratio coefficient and the knock risk level value. The preventative correction of the main injection initiation angle is equal to the main injection initiation angle calculated by the conventional control logic plus the product of the main injection initiation angle delay ratio coefficient and the knock risk level value; The product of the pilot injection reduction ratio coefficient and the knock risk level value is greater than or equal to zero and less than one. When the product approaches or exceeds one, it is truncated to an upper limit value less than one.
[0009] Furthermore, before inputting the angular velocity value, the angular acceleration value, and the pedal opening value into the long short-term memory network turbulent kinetic energy prediction model, Z-score normalization is performed on the above three input features respectively; The input to the Long Short-Term Memory Network Turbulent Kinetic Energy Prediction Model is a multivariate time series consisting of the angular velocity, angular acceleration, and pedal opening values at the current moment and several historical moments. The output layer is a fully connected layer, and the output is the turbulent kinetic energy prediction value and eddy current direction angle prediction value corresponding to each of the future preset number of cycles. During the training phase, samples of rapidly accelerating transient scenarios are copied and perturbed and expanded. If the predicted eddy current direction angle exceeds the range of zero to two times pi after denormalization, a modulo two pi operation is performed to reduce it to the effective value range.
[0010] Furthermore, the method for selecting the predicted value that matches the actual injection execution time is as follows: divide the total system delay time by the single cycle duration and round up to obtain the number of cycles to be skipped, and take the predicted value at the index position from the turbulent kinetic energy prediction sequence and the vortex direction angle prediction sequence as the effective turbulent kinetic energy estimate and the effective vortex direction angle estimate. The duration of a single cycle is calculated from the current angular velocity value; During the safety constraint check, the angular margin corresponding to the minimum atomization mixing time is the product of the minimum atomization mixing time and the current angular velocity value. When the interval between any adjacent spray angles does not meet the angular margin, the pulse width of the subsequent spray segment is compressed according to the priority rule of the pilot spray segment having the highest priority, the main spray segment having the second highest priority, and the supplementary spray segment having the lowest priority. When the required spray pressure exceeds the rated maximum pressure of the injector, it is executed according to the rated maximum pressure and the uncompensated spray volume difference is allocated to the next adjacent spray segment.
[0011] Furthermore, the method for determining the location of the detonation source is as follows: when the detonation energy ratio is greater than a preset near-end ratio threshold and the time delay value indicates that the vibration propagates from the near end to the far end, the location of the detonation source is identified as a pilot layer detonation. When the detonation energy ratio is less than a preset far-end ratio threshold and the time delay value indicates that the vibration is propagating from the far end to the near end, the location of the detonation source is identified as the detonation at the end of the main layer. When the knock energy ratio is between the preset near-end ratio threshold and the preset far-end ratio threshold, the knock source location is identified as dual-source knock; wherein the near-end knock sensor is installed on the cylinder block at a circumferential position close to the spark plug, and the far-end knock sensor is installed on the cylinder block at a circumferential position away from the spark plug.
[0012] Furthermore, the method of dynamically adjusting the judgment threshold is as follows: calculate the first-order difference for a series of consecutive cycles of detonation risk level values; when the difference value is continuously positive and the absolute value exceeds the preset rise rate threshold, it is determined to be in the rapid rise phase; and multiply the near-end judgment threshold by a reduction coefficient less than one to improve the detection sensitivity of the pilot layer detonation. When the absolute value of the difference is lower than the preset stable rate threshold, it is determined to be in a high-level stable stage. The far-end judgment threshold is multiplied by a reduction coefficient of less than one to improve the detection sensitivity of the detonation at the end of the main layer. Both the near-end determination threshold and the far-end determination threshold are adjusted based on the baseline threshold.
[0013] Furthermore, the differential suppression method is as follows: when the detonation source location is identified as pilot layer detonation, the pilot injection amount is reduced and the pilot injection initiation angle is delayed; When the detonation source location is identified as detonation at the end of the main layer, the main injection start angle is advanced and the main injection volume is increased; when the detonation source location is identified as dual-source detonation, both of the above suppression strategies are executed simultaneously and their respective adjustment ranges are halved. The suppression magnitude in each strategy is determined by multiplying the base suppression amount by a scaling factor. The scaling factor is equal to the product of the scaling sensitivity coefficient and the difference between the detonation risk level value and the risk level benchmark value. When the detonation risk level value is less than the risk level benchmark value, the scaling factor is set to one. The risk level benchmark value is the detonation risk level value recorded synchronously when an actual detonation is confirmed and updated each time a detonation is confirmed. After performing the differential suppression, the spark plug ion current signal is collected and wavelet multi-resolution decomposition is performed to extract the stratified combustion time sequence features. The deviation between the combustion quality index and the target value is used as the cycle-by-cycle feedback correction amount and superimposed on the injection quantity allocation ratio of the next cycle.
[0014] This invention provides a stratified fuel injection control system for a single-rotor engine, comprising: The parameter acquisition module is used to acquire the eccentric shaft angle sequence through the eccentric shaft angle sensor and fit and calculate the current angular velocity value and angular acceleration value, and to acquire the pedal opening value and pedal opening change rate through the accelerator pedal position sensor; The transient judgment and compensation module is used to activate the advance prediction control mode when the angular acceleration value exceeds the preset acceleration threshold and the change rate of the pedal opening is positive, and to calculate the angular advance compensation amount based on the current angular velocity value, the angular acceleration value and the total system delay time; The knock risk prediction module is used to extrapolate and predict the end gas temperature of a preset number of cycles based on the angular acceleration value and the rate of change of the intake pressure, and to calculate the knock risk level value by comparing the predicted end gas temperature with the fuel auto-ignition critical temperature. The preventive correction module is used to generate preventive correction parameters by proportionally reducing the pilot injection volume and delaying the main injection start angle when the knock risk level value exceeds the preset prevention threshold and no actual knock occurs. The flow field prediction and injection angle generation module is used to input the angular velocity value, the angular acceleration value and the pedal opening value into the long short-term memory network turbulent kinetic energy prediction model to output the turbulent kinetic energy prediction sequence and the vortex direction angle prediction sequence, select the prediction value that matches the actual injection execution time, combine the empirical formula of spray penetration distance to determine the target starting angle of each segment, and superimpose the angle advance compensation amount to generate the advance compensation injection starting angle of each segment. The safety constraint check module is used to perform safety constraint checks on the starting angle of each injection segment after advance compensation. When the interval between the injection angles of adjacent segments does not meet the angle margin corresponding to the minimum atomization mixing time, the injection pulse width of the subsequent segment is compressed and the injection pressure is increased. The detonation source location and threshold adjustment module is used to perform bandpass filtering on the signals collected by the near-end detonation sensor and the far-end detonation sensor respectively and calculate the detonation energy value. Based on the detonation energy ratio and the time delay value corresponding to the peak value of the cross-correlation function, the detonation source location identifier is determined, and the judgment threshold is dynamically adjusted based on the changing trend of the detonation risk level value. The differential suppression and command generation module is used to perform differential suppression based on the detonation source location identifier, superimpose the suppression correction onto the layered injection timing parameters adjusted by safety constraints, and generate the final layered injection control command after a safety range check.
[0015] This invention solves the technical problem of deteriorated spatial positioning accuracy of stratified injection caused by system delay under high-speed rapid acceleration conditions by calculating the angular advance compensation amount containing the acceleration component based on the angular acceleration value and the current angular velocity value, and by using a long short-term memory network turbulent kinetic energy prediction model to select the in-cylinder flow field prediction value that matches the actual injection execution time. This allows the calculation of the injection start angle of each segment to be moved forward to the state corresponding to the actual execution time. This invention also solves the technical problem of the delayed response of the after-the-fact detection method during rapid acceleration and rapid temperature rise, leading to the continuous development of knock, by extrapolating and predicting the end gas temperature of several future cycles based on the empirical correlation of the variable compression process temperature and calculating the knock risk level value. When the knock risk level value exceeds the preset prevention threshold, preventive injection correction is performed in advance, solving the technical problem of the delayed response of the after-the-fact detection method during rapid acceleration and rapid temperature rise, leading to the continuous development of knock. Finally, this invention solves the technical problem that a single-point signal cannot distinguish the knock source, and a uniform suppression strategy may aggravate knock in specific layer areas. This is achieved by arranging near-end and far-end knock sensors at different circumferential distances from the spark plug on the cylinder block, and using the knock energy ratio and cross-correlation time delay value as dual criteria to determine the spatial location of the knock source, and then implementing a differentiated suppression strategy accordingly. This invention dynamically adjusts the near-end and far-end judgment thresholds based on the changing trend of the detonation risk level value, so that the threshold sensitivity distribution matches the physical law of detonation source migration, thus solving the technical problem of misjudgment caused by fixed thresholds during the dynamic migration transition period of the detonation source. Attached Figure Description
[0016] Figure 1 This is a flowchart of the single-rotor engine stratified fuel injection supply control method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the acquisition results of the eccentric shaft angular velocity and angular acceleration provided in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the system delay time composition analysis provided in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the prediction of the risk level of detonation in the next three cycles provided by an embodiment of the present invention; Figure 5 This is a schematic diagram comparing the preventive injection correction parameters provided in the embodiments of the present invention; Figure 6 This is a schematic diagram of the predicted output of turbulent kinetic energy and eddy direction angle of the LSTM model provided in the embodiment of the present invention; Figure 7 This is a schematic diagram of the evolution process of the injection initiation angle (pilot injection) provided in an embodiment of the present invention; Figure 8 This is a schematic diagram illustrating the upward trend of the continuous cyclic detonation risk level value provided in the embodiments of the present invention; Figure 9 This is a schematic diagram summarizing the final injection control command parameters provided in an embodiment of the present invention. Detailed Implementation
[0017] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be described in detail below with reference to the accompanying drawings. The description in this part is only exemplary and explanatory, and should not be used to limit the scope of protection of the present invention in any way.
[0018] Due to its unique structure, single-rotor engines require precise control of the initiation angles and injection quantity distribution of each stage of pilot injection and main injection in a stratified injection fuel supply system to achieve efficient combustion. However, under high-speed, rapid acceleration transient conditions, existing control methods face multiple technical challenges. First, the system delay between sensor acquisition, model inference, and the actual opening of the injector needle valve at high speeds corresponds to a significant increase in the eccentric shaft angle offset, causing the actual injection initiation angle to lag severely behind the calculated target value, and drastically deteriorating the spatial positioning accuracy of each stage of stratified injection. Second, existing knock suppression strategies employ a post-detection response approach, while during rapid acceleration, the peak in-cylinder compression temperature and end-gas temperature rise sharply over several consecutive cycles. The inherent lag in the post-response means that knock continues to develop for several cycles after it is detected. Third, single-point measurement based on spark plug position ion current signals lacks spatial resolution, making it impossible to distinguish whether the knock source is located in the overly rich combustion region of the pilot layer or the lean mixture region at the end of the main injection layer. Knock from both sources requires injection adjustment strategies with opposite directions to uniformly suppress knocking that may exacerbate specific layer regions. Fourth, during rapid acceleration, the knock source dynamically migrates between the pilot layer and the end of the main injection layer. A fixed energy judgment threshold can lead to misjudgment of the knock source during the migration transition period.
[0019] It should be understood that the single-rotor engine stratified injection fuel supply system involved in this embodiment includes the following hardware environment: an eccentric shaft angle sensor, a throttle pedal position sensor, an intake manifold pressure sensor, a near-end knock sensor and a far-end knock sensor mounted on the cylinder block at different circumferential distances from the spark plug, a spark plug ion current detection circuit, an injector drive circuit, and an electronic control unit that executes the control algorithm. The near-end knock sensor is mounted on the cylinder block at a circumferential position close to the spark plug, and the far-end knock sensor is mounted on the cylinder block at a circumferential position away from the spark plug. Together, they provide differentiated vibration propagation path information for the spatial positioning of the knock source.
[0020] At least one embodiment of the present invention discloses a stratified fuel injection control method for a single-rotor engine, such as... Figure 1 As shown, it includes the following steps: Step 1: Obtain the eccentric shaft motion parameters and pedal opening parameters; An eccentric shaft angle sequence is acquired at high resolution using an eccentric shaft angle sensor. A second-order polynomial least-squares fitting is performed on a preset number of consecutive sampling points, with time as the independent variable. Twice the coefficients of the second-order term yields the angular acceleration value, and the coefficients of the first-order term yield the current angular velocity value. The preset number of sampling points should balance filtering effect and response speed; for example, it could be eight to sixteen recently acquired angle sampling points. The accelerator pedal position sensor acquires the pedal opening value, and the first-order time derivative of the pedal opening value is calculated to obtain the rate of change of pedal opening.
[0021] Step 2: Determine the transient operating condition and calculate the angle advance compensation; The system determines whether the current condition is a rapid acceleration transient based on angular acceleration values and the rate of change of pedal opening. When the angular acceleration value exceeds a preset acceleration threshold and the rate of change of pedal opening is positive, the advanced prediction control mode is activated. The preset acceleration threshold is an angular acceleration boundary value determined during calibration tests based on engine characteristics, distinguishing between steady-state and rapid acceleration transient conditions. When the angular acceleration value does not exceed the preset acceleration threshold or the rate of change of pedal opening is non-positive, the advanced prediction control mode is not activated, and the injection initiation angle for each segment is calculated according to conventional control logic without adding angle advance compensation.
[0022] In the anticipatory control mode, the angle advance compensation amount corresponding to the total system delay time is calculated based on the current angular velocity and angular acceleration values. The total system delay time is the sum of the sensor acquisition delay, the model inference calculation delay, and the injector needle valve mechanical response delay. The angle advance compensation amount is calculated according to the following formula: in, This represents the angle advance compensation amount, expressed in radians. This is the current angular velocity value, in radians per second. This is the angular acceleration value, expressed in radians squared per second. The total system delay time is expressed in seconds. This formula, under the assumption of uniform acceleration, incorporates the angle through which the eccentric shaft rotates during the system delay period into the calculation. The first term corresponds to the uniform rotation component, and the second term corresponds to the accelerated rotation component.
[0023] Furthermore, the total system latency The delay is obtained by summing the sensor acquisition delay, the model inference calculation delay, and the injector needle valve mechanical response delay. The sensor acquisition delay and the model inference calculation delay can be measured during actual operation using the internal timer of the electronic control unit. The injector needle valve mechanical response delay is determined by the injector's factory calibration data or bench test measurements. All three are directly added together in seconds. .
[0024] Step 3: Predict the knock risk level for future cycles; The current intake pressure and intake pressure change rate are obtained through an intake manifold pressure sensor. Based on the angular acceleration value and the intake pressure change rate, the peak in-cylinder compression temperature and final gas temperature for each cycle in the future preset number of cycles are extrapolated and predicted using the polytropic compression process temperature empirical correlation. The polytropic compression process temperature empirical correlation refers to extrapolating the compression termination temperature to the future cycle intake conditions based on the measured intake temperature and pressure of the current cycle, combined with the polytropic index and compression ratio parameters, according to the temperature and pressure relationship of the polytropic compression process. The intake conditions of the future cycle are obtained by linear extrapolation of the current intake pressure value and the intake pressure change rate. The preset number of cycles can be, for example, two to four consecutive combustion cycles.
[0025] The predicted end-gas temperatures for each future cycle are compared with the auto-ignition critical temperature of the currently used fuel to calculate the knock risk level for each future cycle. The closer the predicted end-gas temperature is to or exceeds the auto-ignition critical temperature, the higher the corresponding knock risk level. Specifically, the knock risk level... Calculate using the following formula: in, This is the predicted end-gas temperature, in Kelvin. The auto-ignition critical temperature of the currently used fuel, in Kelvin; both are absolute temperatures, and the ratio is... It is a dimensionless number. When The closer to or beyond The higher the value of the knockout risk level, the greater the risk level.
[0026] To adapt to the rapid changes in intake conditions during rapid acceleration, a correction logic is introduced into the extrapolation prediction: as the angular acceleration value continues to increase, a positive correction term is added to the predicted end-gas temperature. This positive correction term is proportional to the rate of change of the angular acceleration value to compensate for the nonlinear upward trend of the intake temperature caused by the intensified compression effect. Specifically, the corrected predicted end-gas temperature... Calculate using the following formula: in, The final gas temperature is the predicted value before correction, in Kelvin. ω represents the rate of change of angular acceleration, expressed in radians cubic per second. This is the positive correction proportionality coefficient, measured in Kelvin-second cubed per radian, determined by calibration tests to ensure that the positive correction term is consistent with... The dimensions are consistent.
[0027] Furthermore, The method for obtaining the angular acceleration value sequence is as follows: Based on the angular acceleration value sequence obtained in step 1, calculate the first-order time difference of the angular acceleration value sequence for several consecutive cycles, and then divide it by the corresponding sampling time interval to obtain the rate of change of the angular acceleration value. When all consecutive difference values in the angular acceleration value sequence are positive, it is determined that the angular acceleration value continues to increase, triggering the correction logic. The value is determined by minimizing the deviation between the predicted and measured values of the end gas temperature before and after correction under different rapid acceleration conditions through bench calibration tests.
[0028] Furthermore, the knock risk level value calculated in step 3 The corrected predicted end gas temperature will be used directly in subsequent steps 4, 7, and 8. If the correction logic is triggered in step 3, the corrected predicted end gas temperature will be used. Substitute the knock risk level value The calculation formula is as follows: replace Detonation risk level value in subsequent steps The calculations ensure that the knock risk level values used in each step are consistent.
[0029] Step 4: Generate preventative injection correction parameters; When the knock risk level exceeds the preset prevention threshold, but the near-end knock energy and far-end knock energy do not exceed the knock judgment threshold, the system is determined to be in a knock risk warning state but no actual knock has occurred. The preset prevention threshold, being lower than the knock judgment threshold, serves as an early intervention boundary set for the knock risk level. When the knock risk level does not reach the preset prevention threshold, preventative injection correction is not performed, and the pilot injection quantity and main injection starting angle remain at the output values of the normal control logic.
[0030] Under detonation risk warning conditions, the pilot injection quantity is reduced proportionally based on the detonation risk level value, and the main injection initiation angle is adjusted to a delayed direction, generating a preventatively corrected pilot injection quantity and a preventatively corrected main injection initiation angle. Both the reduction ratio and the delayed angle increase with the increase of the detonation risk level value. Specifically, the preventatively corrected pilot injection quantity... and preventative correction of the main injection initiation angle Calculate according to the following formulas respectively: in, The pilot injection quantity is calculated using conventional control logic, and the unit is milligrams. The starting angle of the main injection is calculated using conventional control logic, in radians. The value represents the risk level of the explosion, and is a dimensionless number. is the pilot injection reduction ratio coefficient, and is a dimensionless number; The main injection initiation angle retardation ratio factor, in radians, is determined by calibration tests to ensure... and The dimensions are consistent. Both proportionality coefficients were determined by calibration experiments.
[0031] Furthermore, The range of values should satisfy To ensure the pilot injection volume for preventative corrections Maintain a positive value, meaning the pilot injection quantity is not zero or negative after preventative correction; when and The product approaches or exceeds the value during calibration or operation. At that time, should be Cut off to less than The upper limit of this value is determined by calibration tests based on the engine's minimum pilot injection quantity requirements.
[0032] Step 5: Predict turbulent kinetic energy and vortex direction angle and generate the injection initiation angle after advance compensation; Before inputting the angular velocity, angular acceleration, and pedal opening values into the long short-term memory network turbulent kinetic energy prediction model, Z-score standardization is performed on the three input features to eliminate the dimensional and numerical scale differences between different physical quantities, ensuring the numerical stability of the training and inference process of the long short-term memory network turbulent kinetic energy prediction model.
[0033] The standardized angular velocity, angular acceleration, and pedal opening values are input into the long short-term memory network turbulent kinetic energy prediction model trained with data augmentation under rapid acceleration conditions. The model outputs a turbulent kinetic energy prediction sequence and an eddy current direction angle prediction sequence for a preset number of future cycles.
[0034] The input to the Long Short-Term Memory (LSTM) network turbulent kinetic energy prediction model is a multivariate time series consisting of angular velocity, angular acceleration, and pedal opening values from the current moment and several historical moments. Each time step corresponds to a vector containing these three features. The output layer is a fully connected layer, outputting the predicted turbulent kinetic energy and vortex direction angle values for each of the predetermined number of future iterations. Each time step outputs a prediction vector containing both turbulent kinetic energy and vortex direction angle components. The LSM network turbulent kinetic energy prediction model uses measured turbulent kinetic energy and vortex direction angle data collected from engine bench tests as training labels. The loss function is the mean square error between the predicted and labeled values, and the Adam optimization algorithm is used for parameter updates. During the training phase, samples from rapidly accelerating transient scenarios are copied and perturbed to improve the prediction accuracy of the LSM network turbulent kinetic energy prediction model for rapidly changing angular velocity and angular acceleration inputs. The turbulent kinetic energy prediction value and eddy current direction angle prediction value output by the Long Short-Term Memory Network Turbulent Kinetic Energy Prediction Model are values in the true scale of the physical quantities. By performing an inverse normalization transformation on the output layer results, they are restored to the original physical units. The unit of the turbulent kinetic energy prediction value is joules per kilogram, and the unit of the eddy current direction angle prediction value is radians, which can be directly used in subsequent steps.
[0035] From the turbulent kinetic energy prediction sequence and the vortex direction angle prediction sequence, the predicted value that matches the actual injection execution time of the current cycle is selected to obtain the effective turbulent kinetic energy estimate and the effective vortex direction angle estimate. Since the long short-term memory network turbulent kinetic energy prediction model outputs predicted values for multiple future cycles, selecting the value that matches the actual injection execution time means skipping the time period corresponding to the system delay and directly using the in-cylinder flow field prediction result corresponding to the actual injection action time after the delay ends.
[0036] Furthermore, the prediction sequence index corresponding to the actual injection execution time is determined as follows: based on the total system delay time. Divide by the single cycle duration and round up to get the number of cycles to be skipped. Take the predicted value at the index position from the turbulent kinetic energy prediction sequence and the eddy current direction angle prediction sequence as the effective turbulent kinetic energy estimate and the effective eddy current direction angle estimate. The single cycle duration is calculated from the current angular velocity value.
[0037] Furthermore, the effective range of predicted eddy current direction angle values is as follows: The unit is radians, corresponding to the complete directional space of the in-cylinder vortex in the circumferential plane; if the predicted vortex direction angle output by the long short-term memory network turbulent kinetic energy prediction model exceeds this range after denormalization, the predicted vortex direction angle should be subjected to a modeling test. The operation reduces it to Within the interval, it is then used as an effective vortex direction angle estimate for subsequent spray penetration distance calculation.
[0038] Based on the effective turbulent kinetic energy estimate and the preventive correction pilot injection quantity, a stratified injection mode is selected and the injection quantity allocation ratio of each segment is determined. When it is determined in step 4 that the knock risk warning state is in effect, the preventive correction pilot injection quantity is used as the input; when it is determined in step 4 that the knock risk warning state is not in effect, the pilot injection quantity calculated by the conventional control logic is used as the input.
[0039] The target starting angle for each injection segment is calculated using the estimated effective vortex direction angle and the empirical formula for spray penetration distance. The empirical formula for spray penetration distance estimates the time required for the spray front to reach the target spatial position based on parameters such as injection pressure, nozzle diameter, and in-cylinder gas density. This is then combined with the estimated effective vortex direction angle to determine the actual landing point of the spray after deflection under the influence of the vortex, thereby calculating the eccentric axis angle at which each injection segment should begin, i.e., the target starting angle. Angle advance compensation is then applied. The angles of each segment are superimposed on the starting angles of the target segments to generate the advanced compensated starting angles of each segment's injection.
[0040] Step 6: Perform a safety constraint check on the injection initiation angle after advance compensation; After advance compensation, a safety constraint check is performed on the starting angle of each injection segment, the angular interval between each injection segment is calculated, and it is determined whether each angular interval meets the angular margin corresponding to the minimum atomization mixing time. The angular margin corresponding to the minimum atomization mixing time refers to the shortest time required to ensure that the fuel injected in the previous segment completes atomization and initial mixing at the current angular velocity value, converted into the angle value through which the eccentric shaft rotates.
[0041] Furthermore, the conversion method for the angle margin corresponding to the minimum atomization mixing time is as follows: multiply the minimum atomization mixing time (the shortest atomization completion time determined by calibration tests for different injection pressures and in-cylinder gas density conditions, in seconds) by the current angular velocity value (in radians per second) to obtain the corresponding minimum angle margin (in radians); when the calculated interval between adjacent injection angle segments is less than the minimum angle margin, the subsequent compression adjustment logic is triggered.
[0042] When the angular interval between any adjacent injection segments does not meet the minimum angular margin, the pulse width of the subsequent injection segment is compressed according to priority, and the injection pressure of that segment is increased accordingly to maintain the injection volume of that segment unchanged, generating stratified injection timing parameters adjusted by safety constraints. The priority rule is: the starting angle of the pilot injection segment has the highest priority, followed by the starting angle of the main injection segment, and the starting angle of the subsequent supplementary injection segments has the lowest priority.
[0043] To prevent the injection pressure from exceeding the injector's rated operating range, an upper limit constraint is set for the increased injection pressure. When the injection pressure required by the compression pulse width exceeds the injector's rated maximum pressure, the rated maximum pressure is applied, and the uncompensated injection volume difference is allocated to the next adjacent injection segment.
[0044] Step 7: Spatial location of the detonation source based on dual detonation sensors and dynamic adjustment of the judgment threshold; Bandpass filtering is performed on the near-end knock signal collected by the near-end knock sensor and the far-end knock signal collected by the far-end knock sensor within the knock characteristic frequency range. The energy values of the filtered signals are calculated to obtain the near-end knock energy value and the far-end knock energy value. The knock characteristic frequency range is the frequency interval of the knock vibration response set determined based on the inherent vibration mode analysis of the single-rotor engine cylinder block structure. When either the near-end knock energy value or the far-end knock energy value exceeds the corresponding knock detection threshold, actual knocking is confirmed to have occurred in the current cycle.
[0045] After confirming the occurrence of an actual detonation, the ratio of the near-end detonation energy value to the far-end detonation energy value is calculated to obtain the detonation energy ratio. Simultaneously, the cross-correlation function between the near-end and far-end detonation signals is calculated, and the time delay value corresponding to the peak value of the cross-correlation function is extracted. The sign and amplitude of the time delay value together indicate the time difference and propagation direction of the detonation pressure wave from the origin point to the near-end and far-end detonation sensors, respectively.
[0046] The location of the detonation source is determined based on the detonation energy ratio and time delay value: when the detonation energy ratio is greater than a preset near-end ratio threshold and the time delay value indicates that the vibration propagates from the near end to the far end, the location of the detonation source is identified as a pilot layer detonation; when the detonation energy ratio is less than a preset far-end ratio threshold and the time delay value indicates that the vibration propagates from the far end to the near end, the location of the detonation source is identified as a main layer end detonation; when the detonation energy ratio is between the preset near-end ratio threshold and the preset far-end ratio threshold, the location of the detonation source is identified as a dual-source detonation.
[0047] To adapt to the dynamic migration of the knock source between the pilot layer and the end of the main injection layer during rapid acceleration, the near-end and far-end judgment thresholds are dynamically adjusted based on the changing trend of the knock risk level value: when the knock risk level value is in a rapidly rising phase, the near-end judgment threshold is lowered to improve the detection sensitivity of knock in the pilot layer; when the knock risk level value has reached a high level and tends to stabilize, the far-end judgment threshold is lowered to improve the detection sensitivity of knock at the end of the main injection layer. This dynamic adjustment is based on the following physical understanding: during the rapid rise of in-cylinder temperature in the initial stage of rapid acceleration, the local high temperature in the rich region of the pilot layer is more likely to trigger auto-ignition first; after the temperature rises to a high level and stabilizes, the long residence time in the lean mixture region at the end of the main injection layer makes it easier to reach the auto-ignition condition. Through the above dynamic adjustments, the knock source location identifier after dynamic threshold correction is obtained.
[0048] Furthermore, the criterion for determining whether the detonation risk level value in step 7 is in a rapidly rising phase or has reached a high level and is trending towards stability is based on the detonation risk level value over several consecutive cycles. The sequence is calculated using the first-order difference. When the difference value is consistently positive and its absolute value exceeds the preset rate of increase threshold, it is determined to be in the rapid rise phase. When the absolute value of the difference value is lower than the preset stable rate threshold, it is determined to be in the high-level stable phase. Both the preset rate of increase threshold and the preset stable rate threshold are determined by calibration experiments.
[0049] Furthermore, the dynamic adjustment method for the near-end and far-end judgment thresholds is as follows: starting from the baseline threshold, during the rapid rise phase, the near-end judgment threshold is multiplied by a factor less than [a certain value]. The reduction coefficient, during the high-level stable phase, multiplies the far-end judgment threshold by a factor less than [a certain value]. The reduction coefficient; the above reduction coefficient and the benchmark threshold are determined by the comprehensive optimal principle of comparing the detection rate and the false judgment rate under different working conditions in the calibration test.
[0050] Step 8: Perform differentiated suppression based on the detonation source location identifier and generate the final injection control command; Differential suppression is implemented based on the location of the detonation source: when the detonation source is identified as a pilot layer detonation, the pilot injection volume is reduced and the pilot injection initiation angle is delayed; when the detonation source is identified as a main layer terminal detonation, the main injection initiation angle is advanced and the main injection volume is increased to enrich the terminal region and move it away from spontaneous combustion conditions; when the detonation source is identified as a dual-source detonation, both of the above suppression strategies are implemented simultaneously, but the adjustment range of each is halved. The suppression range in each strategy is determined by multiplying the base suppression amount by a scaling factor, where the scaling factor is determined by the detonation risk level value. The mapping process generates a scaling factor that increases with the higher the knock risk level, resulting in greater suppression. Specifically, the scaling factor... Calculate using the following formula: in, is the scaling factor, and is a dimensionless number; The value represents the risk level of the explosion, and is a dimensionless number. The threshold value for determining the risk level of a knockout event is a dimensionless number. Here, is the scaling sensitivity coefficient, and is a dimensionless coefficient determined by calibration experiments; and when Time to take .
[0051] Furthermore, the risk level benchmark value The value is the knock risk level value corresponding to the actual knock event confirmed in step 7. That is, the knock risk level value recorded synchronously when the knock detection threshold is triggered. This serves as the benchmark reference point for scaling factor calculation, reflecting the thermodynamic boundary conditions at which detonation occurs under the current operating conditions; risk level benchmark value. Update each time detonation is confirmed to adapt to the differences in detonation triggering conditions under different rapid acceleration conditions.
[0052] Differential suppression corrections are superimposed onto the safety-constrained stratified injection timing parameters to obtain the superimposed injection parameters. The superimposed injection parameters undergo total injection quantity safety range checks and equivalence ratio safety range checks. These safety ranges refer to the allowable variation ranges of injection quantity and equivalence ratio under current speed and load conditions to ensure engine structural safety and emission performance. After confirming that both the total injection quantity and equivalence ratio are within the safety boundaries, the final stratified injection control command is generated and output to the injector drive circuit. If the superimposed injection parameters cause the total injection quantity or equivalence ratio to exceed the safety range, the excess portion is truncated, and the injection quantity is redistributed after truncation to maintain the stratification ratio as close as possible to the target value.
[0053] After differential suppression in step 8, the following feedback correction step is also included: acquiring the spark plug ion current signal of the current cycle, performing wavelet multi-resolution decomposition on the spark plug ion current signal to extract stratified combustion timing features; evaluating the combustion quality index of each layer in the current cycle based on the stratified combustion timing features, and using the deviation between the combustion quality index and the target value as the cycle-by-cycle feedback correction amount, which is superimposed on the injection quantity allocation ratio of the next cycle to form closed-loop control. This feedback correction step further eliminates the cumulative effects of modeling errors and execution deviations on the basis of advance compensation and differential suppression.
[0054] Furthermore, the extraction method of stratified combustion timing features is as follows: after performing wavelet multi-resolution decomposition on the spark plug ion current signal, the energy concentration time windows corresponding to the pilot layer combustion and the main injection layer combustion are identified in the detail coefficients of each decomposition scale, and the peak time and energy integral value in each time window are extracted as stratified combustion timing features; wherein the time window corresponding to the pilot layer is earlier than the time window corresponding to the main injection layer, and the time boundary between the two is determined by the spark plug ion current signal response features under different injection sequences in the calibration test.
[0055] This implementation addresses the control challenges of a single-rotor engine's stratified fuel injection system under high-speed, rapid acceleration transient conditions by establishing a control architecture that combines predictive compensation, preventative adjustment, and differentiated suppression.
[0056] Because this implementation calculates the angle advance compensation amount including the acceleration component based on the angular acceleration value and the current angular velocity value in step 2, and selects the predicted value matching the actual injection execution time after outputting the prediction sequence of future multi-cycles using the long short-term memory network turbulent kinetic energy prediction model in step 5, the calculation of the target starting angle of each injection segment and the estimation of the in-cylinder flow field are both moved forward to the state corresponding to the actual execution time after delay compensation. This overcomes the angle offset caused by system delay at high speeds and solves the problem of the actual injection starting angle lagging far behind the calculated target value, resulting in the deterioration of the spatial positioning accuracy of layered injection. At the same time, the angle margin constraint corresponding to the minimum atomization mixing time in step 6 ensures that a sufficient mixing time window is still maintained between each injection segment under high speed conditions.
[0057] Because in step 3 of this implementation, the peak compression temperature and end gas temperature in the cylinder for several future cycles are extrapolated and predicted based on the empirical correlation of the variable compression process temperature, and the knock risk level value is calculated, in step 4, when the knock risk level value exceeds the preset prevention threshold, preventive pilot injection quantity reduction and preventive correction of the main injection start angle are executed. Therefore, the control system intervenes to adjust the in-cylinder thermodynamic boundary conditions several cycles before knock actually occurs, overcoming the inherent response lag of the post-detection response method in the process of rapid acceleration and rapid temperature rise, and solving the problem that knock continues to develop for several cycles after it is detected.
[0058] Because this embodiment simultaneously collects vibration signals in step 7 by placing near-end knock sensors and far-end knock sensors at different circumferential distances from the spark plug, and uses the knock energy ratio of the two signals and the time delay value corresponding to the peak value of the cross-correlation function as dual criteria to determine the spatial location of the knock source, it overcomes the limitation of single-point ion current signals lacking spatial resolution. This allows the differentiated suppression strategy in step 8 to perform the correct injection adjustment based on whether the knock source is located in the pilot layer or the end of the main layer, solving the problem that the unified suppression strategy may aggravate knocking in specific layer areas because it cannot distinguish the knock source.
[0059] Because this implementation method dynamically adjusts the near-end and far-end judgment thresholds based on the changing trend of the detonation risk level value in step 7, the sensitivity of the near-end judgment threshold is increased during the rapid temperature rise phase in the initial stage of rapid acceleration, and the sensitivity of the far-end judgment threshold is increased during the stage when the temperature reaches a high and stable level. Therefore, the sensitivity distribution of the near-end and far-end judgment thresholds matches the physical law of detonation source migration, overcoming the problem of misjudgment caused by fixed thresholds during the transition period of detonation source migration from the pilot layer to the end of the main layer, and solving the problem of inaccurate positioning caused by the dynamic migration of detonation source during rapid acceleration.
[0060] The following is an example of an application of the present invention, such as... Figure 2-9As shown, the implementation process is as follows: A certain type of single-rotor engine is installed in a sports passenger car (model code: RX-T01). This engine is equipped with a complete stratified fuel injection system hardware environment, including an eccentric shaft angle sensor, accelerator pedal position sensor, intake manifold pressure sensor, near-end knock sensor (installed in the cylinder block circumferential position near the spark plug), far-end knock sensor (installed in the cylinder block circumferential position away from the spark plug), spark plug ion current detection circuit, injector drive circuit, and electronic control unit (model code: ECU-M2).
[0061] The test condition was as follows: During a bench calibration test in 20XX, starting from a stable cruising state (eccentric shaft speed approximately 4800 rpm), the driver rapidly and deeply pressed the accelerator pedal, triggering a high-speed, rapid acceleration transient condition. The electronic control unit continuously collected data from various sensors at 1-millisecond sampling intervals. The current control cycle corresponds to the Nth cycle. The system entered the advanced predictive control mode and initiated the coordinated control process for each step. The fuel used was 93-octane gasoline, with a critical auto-ignition temperature... K.
[0062] The electronic control unit continuously reads the angle values of the most recent 12 sampling points from the eccentric shaft angle sensor. It performs a second-order polynomial least-squares fitting on these 12 angle sampling points with time as the independent variable. From the fitting result, it extracts the coefficients of the first-order term as the current angular velocity value, and twice the coefficients of the second-order term as the current angular acceleration value. Simultaneously, it reads the current pedal opening value from the accelerator pedal position sensor and calculates the first-order time derivative of the pedal opening values at two adjacent sampling times to obtain the rate of change of pedal opening.
[0063] Table 1. Results of Eccentric Shaft Motion Parameters and Pedal Opening Parameters The current angular velocity of 502.65 rad / s corresponds to an eccentric shaft speed of approximately 4800 rpm and an angular acceleration of 38.40 rad / s². 2 The value is positive, and the rate of change of pedal opening is positive at 142.6% / s. Both judgment conditions are met, providing input for the rapid acceleration judgment in step 2.
[0064] The angular acceleration value obtained in step 1 is 38.40 rad / s. 2 The preset acceleration threshold of 25.00 rad / s was determined by the calibration test. 2 Comparing the values, 38.40 > 25.00, and the pedal opening change rate of 142.6% / s is positive. Since both conditions are met simultaneously, the advanced prediction control mode is activated.
[0065] The total system delay consists of three parts: the sensor acquisition delay, measured by the internal timer of the electronic control unit, is 0.0008s; the model inference calculation delay, measured, is 0.0012s; and the injector needle valve mechanical response delay, determined by factory calibration data, is 0.0005s. The sum of these three is: The angular velocity value is 502.65 rad / s and the angular acceleration value is 38.40 rad / s. 2 Substituting the total system delay of 0.0025s into the angle advance compensation formula: The first item, 1.2566 rad, is the uniform rotation component, and the second item, 0.0001 rad, is the accelerated rotation component. The total angular advance compensation is 1.2567 rad, which is used in step 5 to superimpose the values onto the starting angles of each target segment.
[0066] The intake manifold pressure sensor collected a current intake pressure value of 107.5 kPa, with an intake pressure change rate of 18.3 kPa / s. Based on the current cycle's measured intake temperature of 336 K and intake pressure of 107.5 kPa, combined with a polytropic index of 1.35 and a compression ratio of 8.5, and using the empirical correlation formula for polytropic compression process temperature, the intake conditions for the next three cycles are linearly extrapolated, and the predicted peak compression temperature and terminal gas temperature for each cycle are calculated respectively.
[0067] The consecutive difference values of the angular acceleration value sequence were checked, and the three consecutive difference values were +2.1, +3.4, and +2.8 rad / s. 3 All values are positive, indicating a continuous increase in angular acceleration, triggering positive correction logic. The current rate of change of angular acceleration is 2.8 rad / s². 3 The positive correction scaling factor determined during calibration is 4.5 K·s. 3 / rad, the correction amount is: Taking the second future cycle (the cycle that best matches the actual injection execution time) as an example, the predicted end gas temperature before correction is 598K. After correction: The knock risk level is calculated using the corrected predicted end gas temperature: Table 2. Knock Risk Level Prediction for the Next 3 Cycles The knock risk level for the N+2 cycle is 0.9801. This value will be used consistently in steps 4, 7, and 8 to ensure data consistency across steps.
[0068] The current near-end detonation energy is 0.031V. 2 •s, the distal detonation energy value is 0.024V 2 The values of ·s did not exceed the knock detection threshold (calibrated value is 0.085V). 2 The value ·s indicates that no actual detonation has occurred in the current cycle. The detonation risk level value of 0.9801 is compared with the preset prevention threshold of 0.92. Since 0.9801 > 0.92, the current state is determined to be under detonation risk warning, and preventive injection correction is activated.
[0069] The pilot injection reduction ratio determined by calibration is 0.08 (dimensionless), and the main injection initiation angle delay ratio is 0.12 rad (determined by calibration tests). The pilot injection quantity calculated by conventional control logic is 3.85 mg, and the main injection initiation angle is 4.712 rad.
[0070] First, verify the truncation condition: It satisfies the positive value constraint and does not require truncation.
[0071] Preventive correction pilot injection quantity: Preventative correction of main injection starting angle: The pilot injection quantity was reduced from 3.85 mg to 3.548 mg, a decrease of approximately 7.8%; the main injection initiation angle was delayed from 4.712 rad to 4.830 rad, a delay of 0.118 rad. These two preventative correction parameters will be passed to step 5 for subsequent calculations.
[0072] The angular velocity value of 502.65 rad / s and the angular acceleration value of 38.40 rad / s obtained in step 1 are... 2 The pedal opening of 78.3% was subjected to Z-score standardization. Based on the training set statistics (mean and standard deviation determined by bench test data), the three standardized input features constitute the input vector of the current time step. Together with the corresponding vectors of the previous 8 time steps, they form the multivariate time series input of the long short-term memory network turbulent kinetic energy prediction model. The output is the turbulent kinetic energy prediction sequence and eddy current direction angle prediction sequence for the next 3 cycles. After destandardization, the physical quantities are restored to their true scale.
[0073] Determine the prediction sequence index corresponding to the actual injection execution time: single cycle duration s, the number of loops to skip Therefore, the predicted value at the first index position (i.e., the N+1 cycle) in the prediction sequence is taken as the effective estimate. The predicted eddy current direction angle, after inverse standardization, is 6.41 rad, which exceeds... scope( rad), execution mode Operations: The effective vortex direction angle estimate after reduction is 0.127 rad.
[0074] Table 3 Output and Effective Estimates of the Long Short-Term Memory Network Turbulent Kinetic Energy Prediction Model The estimated effective turbulent kinetic energy of 312.7 J / kg and the estimated effective vortex direction angle of 0.127 rad were used as inputs for subsequent flow field calculations. Since step 4 determined that the system was in a knock risk warning state, the stratified injection mode and the injection quantity distribution ratio of each section were determined using a preventive correction of 3.548 mg as input.
[0075] Using the estimated effective vortex direction angle of 0.127 rad, combined with injection pressure, nozzle diameter, and in-cylinder gas density parameters, the initial angle of each target segment is calculated using an empirical formula for spray penetration distance: the initial angle of the pilot injection target is 1.047 rad, and the initial angle of the main injection target uses the preventive correction value of 4.830 rad from step 4. An angle advance compensation of 1.2567 rad is then added to the initial angle of each target segment. Pilot injection advance compensation initial angle: rad Main injection advance compensation starting angle: rad Calculate the angular interval between the pilot injection and the main injection: rad.
[0076] At the current angular velocity of 502.65 rad / s, the calibrated minimum atomization mixing time is 0.0062 s, corresponding to the minimum angular margin: The actual angle interval of 3.783 rad > the minimum angle margin of 3.116 rad, which meets the constraint conditions, and there is no need to trigger the compression adjustment logic. The injection start angle, pulse width, and injection pressure of each segment remain at their original values, generating stratified injection timing parameters confirmed by safety constraints.
[0077] Table 4 Safety Constraint Inspection Results Actual detonation was confirmed in N+2 cycles: the near-end detonation energy value after bandpass filtering by the near-end detonation sensor was 0.127V. 2·s, exceeding the current dynamically adjusted near-end determination threshold of 0.098V 2 •s; the distal detonation energy value is 0.043V. 2 •s, not exceeding the remote judgment threshold. Confirm that an actual detonation has occurred in the current cycle.
[0078] Calculate the detonation energy ratio: The preset near-end ratio threshold is 2.0, and the preset far-end ratio threshold is 0.6. 2.953 > 2.0, meaning the detonation energy ratio is greater than the near-end ratio threshold.
[0079] The cross-correlation function between the near-end and far-end detonation signals was calculated, and the peak value corresponding to a time delay of +0.00018s was extracted. The positive sign indicates that the vibration propagates from the near end to the far end. Both criteria point to the leader layer, and the detonation source location is identified as a leader layer detonation.
[0080] The first-order difference was calculated for the sequence of four consecutive cycles of knock risk level values (0.9412, 0.9578, 0.9714, 0.9801). The difference values were +0.0166, +0.0136, and +0.0087, respectively, all positive. The maximum difference value of 0.0166 exceeded the preset rate-of-rise threshold of 0.010, indicating that the current stage was rapidly rising. Dynamic threshold adjustment was then performed: the near-end judgment threshold was multiplied by a reduction factor of 0.88, based on the baseline value of 0.111V. 2 ·s adjusted to 0.098V 2 •s (already used above); the remote determination threshold remains unchanged from the baseline value. The detonation source location identifier after dynamic threshold correction is still the pilot layer detonation.
[0081] Simultaneously record the current baseline value for the knock risk level: This is used in step 8 to calculate the scaling factor.
[0082] The location of the detonation source is identified as a pilot layer detonation, and the corresponding suppression strategy is implemented: reduce the pilot injection volume and delay the pilot injection initiation angle.
[0083] Calculate the scaling factor; the calibrated scaling sensitivity coefficient is 8.5 (dimensionless). This represents the current knock risk level. benchmark value : This detonation occurred precisely on the first trigger of the current cycle. The scaling factor is 1.0, and the suppression amplitude is taken as the original value of the baseline suppression amount. The baseline suppression amount for the pilot injection quantity determined by calibration is 0.28 mg, and the baseline retardation amount for the pilot injection initiation angle is 0.087 rad. The scaled suppression amounts are as follows: mg and rad.
[0084] The differential suppression correction is superimposed on the adjusted safety constraint parameters output from step 6: Pilot injection quantity: mg Pilot injection initiation angle: rad Main injection parameters remain unchanged from the output value of step 6 (main injection starting angle 6.087 rad).
[0085] Perform a total fuel injection quantity safety range check: Under current speed and load conditions, the permissible total fuel injection quantity range is 14.5–19.8 mg. The total fuel injection quantity after aggregation is... mg (main injection volume 14.2mg from conventional control logic), within the safe range. Equivalence ratio check: calculated equivalence ratio is 0.94, within the allowable range of 0.85 to 1.05. Both checks pass, generating the final stratified injection control command and outputting it to the injector drive circuit.
[0086] Table 5 Summary of Final Injection Control Command Parameters The feedback correction step is then executed: the spark plug ion current signal of the current cycle is acquired, wavelet multi-resolution decomposition is performed, and the energy concentration time window corresponding to the pilot layer (earlier than the main injection layer time window) is identified in the detail coefficients. The peak time and energy integral value of the pilot layer are extracted. The combustion quality index of the pilot layer is evaluated, and the deviation from the target value is -0.032 (normalized). This deviation is used as the cycle-by-cycle feedback correction amount and added to the pilot injection quantity allocation ratio of the next cycle (N+3 cycles) to form a closed-loop correction.
[0087] The data flow throughout the entire implementation process demonstrates a tight logical chain: Step 1 collects the angular velocity value of 502.65 rad / s and the angular acceleration value of 38.40 rad / s. 2 This is the source of all subsequent calculations. Step 2 uses these two parameters to calculate the total system delay time of 0.0025s and the angle advance compensation of 1.2567rad. Step 3 triggers the correction logic based on the intake pressure data and the rate of change of angular acceleration to obtain the corrected knock risk level value of 0.9801. This value is used throughout the preventive correction in Step 4 (pilot injection quantity reduced from 3.85mg to 3.548mg, main injection angle delayed to 4.830rad) and the dynamic threshold adjustment in Step 7 (proximal detection threshold reduced to 0.098V). 2Step 5 calculates the scaling factor (s) and scaling factor 1.0 in step 8; Step 5 superimposes the preventive correction amount from step 4 with the advance compensation amount from step 2, and outputs the starting angle after advance compensation for each segment; Step 6 checks the safety constraints (actual interval 3.783rad > minimum margin 3.116rad) to ensure that the output of step 5 can be directly transmitted to steps 7 and 8; Step 7 uses dual-sensor positioning (energy ratio 2.953 combined with time delay +0.00018s) to determine the knock indicator of the pilot layer, drives step 8 to execute the correct direction of differential suppression, and finally generates a safety control command with a total injection amount of 17.468mg. The combustion quality deviation of this cycle is transmitted to the next cycle through the feedback correction step, forming a complete data closed loop.
Claims
1. A method for controlling the stratified fuel injection supply of a single-rotor engine, characterized in that, Includes the following steps: Obtain the eccentric shaft angle sequence and calculate the current angular velocity and angular acceleration values through fitting; obtain the pedal opening value and the rate of change of pedal opening. When the angular acceleration value exceeds the preset acceleration threshold and the rate of change of the pedal opening is positive, the advance prediction control mode is activated, and the angular advance compensation amount is calculated based on the current angular velocity value, the angular acceleration value and the total system delay time. Based on the angular acceleration value and the rate of change of intake pressure, the end gas temperature of the future preset number of cycles is extrapolated and predicted. The predicted end gas temperature is compared with the fuel auto-ignition critical temperature to calculate the knock risk level value. When the knock risk level value exceeds the preset prevention threshold and no actual knock occurs, the pilot injection volume is reduced proportionally and the main injection start angle is delayed to generate preventive correction parameters. The angular velocity value, the angular acceleration value, and the pedal opening value are input into the long short-term memory network turbulent kinetic energy prediction model, which outputs a turbulent kinetic energy prediction sequence and a vortex direction angle prediction sequence. The predicted value that matches the actual injection execution time is selected, and the starting angle of each segment is determined by combining the empirical formula for spray penetration distance. The angle advance compensation amount is superimposed to generate the advance compensation injection starting angle of each segment. A safety constraint check is performed on the injection start angle of each segment after advance compensation. When the interval between the injection angles of adjacent segments does not meet the angle margin corresponding to the minimum atomization mixing time, the injection pulse width of the subsequent segment is reduced and the injection pressure is increased. Bandpass filtering is performed on the signals collected by the near-end detonation sensor and the far-end detonation sensor respectively, and the detonation energy value is calculated. The detonation source location is determined based on the detonation energy ratio and the time delay value corresponding to the peak value of the cross-correlation function. The judgment threshold is dynamically adjusted based on the changing trend of the detonation risk level value. Differential suppression is performed based on the detonation source location identifier. The suppression correction is superimposed on the stratified injection timing parameters adjusted by safety constraints. After a safety range check, the final stratified injection control command is generated.
2. The single-rotor engine stratified fuel injection supply control method according to claim 1, characterized in that, The angle advance compensation amount is calculated as follows: the product of the current angular velocity value and the total system delay time is taken as the uniform rotation component, and half of the product of the angular acceleration value and the square of the total system delay time is taken as the accelerated rotation component. The uniform rotation component and the accelerated rotation component are summed to obtain the angle advance compensation amount. The total system delay is the sum of the sensor acquisition delay, the model inference calculation delay, and the injector needle valve mechanical response delay.
3. The single-rotor engine stratified fuel injection supply control method according to claim 1, characterized in that, The knock risk level value is the ratio of the predicted end gas temperature to the fuel auto-ignition critical temperature; when the angular acceleration value continues to increase, a positive correction term is added to the predicted end gas temperature value, which is the product of the rate of change of the angular acceleration value and the positive correction ratio coefficient, and the corrected predicted end gas temperature value is substituted into the calculation of the knock risk level value. The rate of change of the angular acceleration value is obtained by calculating the first-order time difference of a series of consecutive cycles of angular acceleration values and dividing it by the corresponding sampling time interval. When all consecutive difference values in the angular acceleration value series are positive, it is determined that the angular acceleration value is continuously increasing.
4. The single-rotor engine stratified fuel injection supply control method according to claim 1, characterized in that, The preventive correction parameters are generated as follows: the pilot injection quantity for preventive correction is equal to the pilot injection quantity calculated by the conventional control logic multiplied by a reduction factor, and the reduction factor is equal to the product of a pilot injection quantity reduction ratio coefficient and the knock risk level value. The preventative correction of the main injection initiation angle is equal to the main injection initiation angle calculated by the conventional control logic plus the product of the main injection initiation angle delay ratio coefficient and the knock risk level value; The product of the pilot injection reduction ratio coefficient and the knock risk level value is greater than or equal to zero and less than one. When the product approaches or exceeds one, it is truncated to an upper limit value less than one.
5. The stratified fuel injection control method for a single-rotor engine according to claim 1, characterized in that, Before inputting the angular velocity value, the angular acceleration value, and the pedal opening value into the long short-term memory network turbulent kinetic energy prediction model, Z-score normalization is performed on the above three input features respectively; The input to the Long Short-Term Memory Network Turbulent Kinetic Energy Prediction Model is a multivariate time series consisting of the angular velocity, angular acceleration, and pedal opening values at the current moment and several historical moments. The output layer is a fully connected layer, and the output is the turbulent kinetic energy prediction value and eddy current direction angle prediction value corresponding to each of the future preset number of cycles. During the training phase, samples of rapidly accelerating transient scenarios are copied and perturbed and expanded. If the predicted eddy current direction angle exceeds the range of zero to two times pi after denormalization, a modulo two pi operation is performed to reduce it to the effective value range.
6. The stratified fuel injection control method for a single-rotor engine according to claim 1, characterized in that, The method for selecting the predicted value that matches the actual injection execution time is as follows: divide the total system delay time by the single cycle duration and round up to obtain the number of cycles to be skipped, and take the predicted value at the index position from the turbulent kinetic energy prediction sequence and the vortex direction angle prediction sequence as the effective turbulent kinetic energy estimate and the effective vortex direction angle estimate. The duration of a single cycle is calculated from the current angular velocity value; During the safety constraint check, the angular margin corresponding to the minimum atomization mixing time is the product of the minimum atomization mixing time and the current angular velocity value. When the interval between any adjacent spray angles does not meet the angular margin, the pulse width of the subsequent spray segment is compressed according to the priority rule of the pilot spray segment having the highest priority, the main spray segment having the second highest priority, and the supplementary spray segment having the lowest priority. When the required spray pressure exceeds the rated maximum pressure of the injector, it is executed according to the rated maximum pressure and the uncompensated spray volume difference is allocated to the next adjacent spray segment.
7. The stratified fuel injection control method for a single-rotor engine according to claim 1, characterized in that, The method for determining the location of the detonation source is as follows: when the detonation energy ratio is greater than a preset near-end ratio threshold and the time delay value indicates that the vibration propagates from the near end to the far end, the location of the detonation source is identified as a pilot layer detonation. When the detonation energy ratio is less than a preset far-end ratio threshold and the time delay value indicates that the vibration is propagating from the far end to the near end, the location of the detonation source is identified as the detonation at the end of the main layer. When the knock energy ratio is between the preset near-end ratio threshold and the preset far-end ratio threshold, the knock source location is identified as dual-source knock; wherein the near-end knock sensor is installed on the cylinder block at a circumferential position close to the spark plug, and the far-end knock sensor is installed on the cylinder block at a circumferential position away from the spark plug.
8. The stratified fuel injection control method for a single-rotor engine according to claim 7, characterized in that, The method of dynamically adjusting the judgment threshold is as follows: calculate the first-order difference for a series of consecutive cycles of detonation risk level values. When the difference value is continuously positive and the absolute value exceeds the preset rise rate threshold, it is determined to be in the rapid rise stage. Multiply the near-end judgment threshold by a reduction coefficient less than one to improve the detection sensitivity of the pilot layer detonation. When the absolute value of the difference is lower than the preset stable rate threshold, it is determined to be in the high-level stable stage. The far-end judgment threshold is multiplied by a reduction coefficient of less than one to improve the detection sensitivity of the detonation at the end of the main layer. Both the near-end determination threshold and the far-end determination threshold are adjusted based on the baseline threshold.
9. The stratified fuel injection control method for a single-rotor engine according to claim 1, characterized in that, The differential suppression method is as follows: when the detonation source location is identified as pilot layer detonation, the pilot injection amount is reduced and the pilot injection initiation angle is delayed; When the detonation source location is identified as detonation at the end of the main layer, the main injection start angle is advanced and the main injection volume is increased; when the detonation source location is identified as dual-source detonation, both of the above suppression strategies are executed simultaneously and their respective adjustment ranges are halved. The suppression magnitude in each strategy is determined by multiplying the base suppression amount by a scaling factor. The scaling factor is equal to the product of the scaling sensitivity coefficient and the difference between the detonation risk level value and the risk level benchmark value. When the detonation risk level value is less than the risk level benchmark value, the scaling factor is set to one. The risk level benchmark value is the detonation risk level value recorded synchronously when an actual detonation is confirmed and updated each time a detonation is confirmed. After performing the differential suppression, the spark plug ion current signal is collected and wavelet multi-resolution decomposition is performed to extract the stratified combustion time sequence features. The deviation between the combustion quality index and the target value is used as the cycle-by-cycle feedback correction amount and superimposed on the injection quantity allocation ratio of the next cycle.
10. A stratified fuel injection fuel supply control system for a single-rotor engine, used to execute the stratified fuel injection fuel supply control method for a single-rotor engine according to any one of claims 1 to 9, characterized in that, include: The parameter acquisition module is used to acquire the eccentric shaft angle sequence through the eccentric shaft angle sensor and fit and calculate the current angular velocity value and angular acceleration value, and to acquire the pedal opening value and pedal opening change rate through the accelerator pedal position sensor; The transient judgment and compensation module is used to activate the advance prediction control mode when the angular acceleration value exceeds the preset acceleration threshold and the change rate of the pedal opening is positive, and to calculate the angular advance compensation amount based on the current angular velocity value, the angular acceleration value and the total system delay time; The knock risk prediction module is used to extrapolate and predict the end gas temperature of a preset number of cycles based on the angular acceleration value and the rate of change of the intake pressure, and to calculate the knock risk level value by comparing the predicted end gas temperature with the fuel auto-ignition critical temperature. The preventive correction module is used to generate preventive correction parameters by proportionally reducing the pilot injection volume and delaying the main injection start angle when the knock risk level value exceeds the preset prevention threshold and no actual knock occurs. The flow field prediction and injection angle generation module is used to input the angular velocity value, the angular acceleration value and the pedal opening value into the long short-term memory network turbulent kinetic energy prediction model to output the turbulent kinetic energy prediction sequence and the vortex direction angle prediction sequence, select the prediction value that matches the actual injection execution time, combine the empirical formula of spray penetration distance to determine the target starting angle of each segment, and superimpose the angle advance compensation amount to generate the advance compensation injection starting angle of each segment. The safety constraint check module is used to perform safety constraint checks on the starting angle of each injection segment after advance compensation. When the interval between the injection angles of adjacent segments does not meet the angle margin corresponding to the minimum atomization mixing time, the injection pulse width of the subsequent segment is compressed and the injection pressure is increased. The detonation source location and threshold adjustment module is used to perform bandpass filtering on the signals collected by the near-end detonation sensor and the far-end detonation sensor respectively and calculate the detonation energy value. Based on the detonation energy ratio and the time delay value corresponding to the peak value of the cross-correlation function, the detonation source location identifier is determined, and the judgment threshold is dynamically adjusted based on the changing trend of the detonation risk level value. The differential suppression and command generation module is used to perform differential suppression based on the detonation source location identifier, superimpose the suppression correction onto the layered injection timing parameters adjusted by safety constraints, and generate the final layered injection control command after a safety range check.
