Method and device for restraining engine knock and vehicle

Abstract

The invention provides a method and device for inhibiting engine knock and a vehicle, and the method comprises the steps: collecting corresponding engine operation features when the vehicle runs in an actual scene, and obtaining a to-be-detected feature data set; the engine operation characteristics comprise the working condition characteristics of the engine and an initial knock characteristic signal; predicting the to-be-tested feature data set by using a pre-trained limit gradient boosting tree model to obtain the probability that the engine is knocked in a future time period; adjusting control parameters of the engine on the basis of the probability that the engine knocks in the future time period; therefore, the operation characteristics of the engine are predicted through the pre-trained limit gradient boosting tree model, so that the knocking risk of the engine can be accurately pre-judged, control parameters are intervened in advance, the knocking occurrence frequency is restrained, and the operation performance of the engine is improved.

Description

Technical Field

[0001] The present invention relates to the field of engine electronic control technology, and in particular to a method, device and vehicle for suppressing engine knock. Background Art

[0002] The combustion process within an internal combustion engine (ICE) is the core energy conversion process, and its dynamic evolution has a decisive impact on the overall system performance. This process is influenced by a variety of dynamic factors (such as fuel quality, intake air conditions, and ambient temperature). During the combustion process, various control parameters must be adjusted to optimize engine operation and improve performance.

[0003] In related technologies, engine control primarily relies on sensor parameters, lookup through fixed mapping tables (MAPs) (such as ignition MAPs and injection MAPs), or adjustment of control parameters through empirical models. However, these methods are unable to predict combustion conditions in advance, have limitations and lags, and are unable to effectively suppress engine knock.

[0004] Based on this, there is an urgent need for a method to suppress engine knock in order to solve the above technical problems and improve the operating performance of the engine. Summary of the Invention

[0005] In response to the problems existing in the prior art, embodiments of the present invention provide a method, device and vehicle for suppressing engine knock, so as to solve or partially solve the technical problem in the prior art that the engine combustion state cannot be predicted in advance and the engine knock cannot be effectively suppressed, thereby affecting the engine operating performance.

[0006] A first aspect of the present invention provides a method for suppressing engine knock, the method comprising:

[0007] Collecting engine operating characteristics corresponding to a vehicle driving in an actual scenario to obtain a feature data set to be measured; the engine operating characteristics include: the engine operating condition characteristics and the initial knock characteristic signal;

[0008] Using a pre-trained extreme gradient boosting tree model to predict the feature data set to be tested, to obtain a probability of knock occurring in the engine in a future time period;

[0009] Control parameters of the engine are adjusted based on the probability that knock will occur in the engine within a future period.

[0010] In the above solution, the acquisition of engine operating characteristics corresponding to the vehicle driving in the actual scene to obtain the feature data set to be tested includes:

[0011] collecting the engine's operating condition characteristics and initial knock characteristic signals at a predetermined frequency; the operating condition characteristics include: speed, load, ignition advance angle, air-fuel ratio, intake air temperature, coolant temperature, and atmospheric pressure;

[0012] Filtering the initial knock characteristic signal using a bandpass filter to obtain a preprocessed knock characteristic signal; determining time domain characteristics of the preprocessed knock characteristic signal, the time domain characteristics including: a peak value, a root mean square value, and a signal rising slope of the knock characteristic signal;

[0013] Performing a fast Fourier transform on the preprocessed knock characteristic signal to obtain a knock characteristic frequency, summing the power of all knock characteristic frequency points to obtain a knock frequency band energy; and determining a knock energy proportion based on the knock frequency band and the total energy;

[0014] The time domain features, the knock frequency band energy, the knock energy ratio and the operating condition features within a preset time window are vector-joined in a predetermined order to obtain the feature data set to be measured.

[0015] In the above solution, the extreme gradient boosting tree model includes multiple decision trees; the use of the pre-trained extreme gradient boosting tree model to predict the feature data set to be tested to obtain the probability of the engine knocking in the future period includes:

[0016] For each decision tree, the feature value of each feature in the test data set is matched in turn using the splitting condition corresponding to the decision tree, and the sub-probability predicted by the decision tree is output;

[0017] The probability of knock occurring in the engine in a future period is determined based on the sub-probability predicted by each decision tree and the total number of decision trees; wherein,

[0018] The multiple decision trees are processed in series, and the subsequent decision tree needs to be fitted with the residual of the previous decision tree.

[0019] In the above solution, adjusting the control parameters of the engine based on the probability of knock occurring in the engine in a future time period includes:

[0020] determining a knock risk level based on a probability of the engine knocking within a future time period;

[0021] determining a corresponding control strategy according to the knock risk level;

[0022] The control parameters of the engine are adjusted based on the control strategy.

[0023] In the above solution, the step of determining the corresponding control strategy according to the knock risk level includes:

[0024] If the knock risk level is a safe level, the control strategy is: keeping the original control parameters of the engine unchanged;

[0025] If the knock risk level is a warning level, the control strategy is: delaying the ignition advance angle by a first angle;

[0026] If the knock risk level is an emergency level, the control strategy is: delaying the ignition advance angle by a second angle and reducing the boost pressure of the engine according to a preset ratio; the second angle is greater than the first angle.

[0027] In the above solution, after adjusting the control parameters of the engine based on the probability of knock occurring in the engine in a future time period, the method further includes:

[0028] If it is determined that the prediction result of the extreme gradient boosting tree model has a missed warning or a false positive warning; or

[0029] If it is determined that the engine operation characteristic data contains new characteristic data, the self-learning mechanism of the extreme gradient boosting tree model is triggered.

[0030] In the above scheme, the self-learning mechanism of the extreme gradient boosting tree model includes:

[0031] If it is determined that there is a missed warning in the prediction result of the extreme gradient boosting tree model, determine the missed warning feature corresponding to the missed warning of the extreme gradient boosting tree model, and increase the split gain weight corresponding to the missed warning feature;

[0032] If it is determined that there is a false alarm warning in the prediction result of the extreme gradient boosting tree model, the false alarm feature corresponding to the false alarm of the extreme gradient boosting tree model is determined, and the split gain weight corresponding to the false alarm feature is reduced; wherein, the split gain weight is used to adjust the leaf node type into which the false alarm feature and the missed alarm feature fall, thereby updating the sub-probability predicted by the decision tree.

[0033] In the above scheme, the self-learning mechanism of the extreme gradient boosting tree model includes:

[0034] If a new engine operating feature is identified, a corresponding decision tree is added to the extreme gradient boosting tree model.

[0035] A second aspect of the present invention provides a device for suppressing engine knock, the device comprising:

[0036] The acquisition unit is used to collect the engine operation characteristics corresponding to the vehicle when it is driving in the actual scene to obtain a feature data set to be tested;

[0037] A prediction unit, configured to predict the feature data set to be tested using a pre-trained extreme gradient boosting tree model to obtain a probability of knock occurring in the engine within a future period;

[0038] An adjustment unit is configured to adjust a control parameter of the engine based on a probability of knock occurring in the engine within a future period.

[0039] According to a third aspect of the present invention, a vehicle is provided, comprising the device for suppressing engine knock as described in the second aspect.

[0040] The present invention provides a method, device and vehicle for suppressing engine knock, the method comprising: collecting engine operating characteristics corresponding to a vehicle traveling in an actual scenario to obtain a feature data set to be measured; the engine operating characteristics comprising: operating condition characteristics of the engine and an initial knock characteristic signal; using a pre-trained extreme gradient boosting tree model to predict the feature data set to be measured to obtain a probability of knock occurring in the engine in a future time period; and adjusting control parameters of the engine based on the probability of knock occurring in the engine in the future time period; in this way, the engine operating characteristics are predicted by the pre-trained extreme gradient boosting tree model to accurately predict the knock risk of the engine, thereby intervening in control parameters in advance to suppress the occurrence of engine knock, thereby reducing the frequency of engine knock and improving engine operating performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0042] Figure 1 A schematic flow chart of a method for suppressing engine knock according to an embodiment of the present invention is shown;

[0043] Figure 2 A schematic structural diagram of a device for suppressing engine knock according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0044] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0045] In order to better understand the technical solution of the present invention, here we first introduce the current status of engine combustion control:

[0046] 1. Stringent emission regulations: Engines are required to achieve ultra-low emissions under all operating conditions (e.g., NOx ≤ 35mg / km, and exhaust particulate matter number (PN) detection requires particle size detection of 10nm). However, traditional combustion control relies on fixed maps (such as ignition maps and injection maps) for feedback control, which has a lag of one to two combustion cycles, making it impossible to perform active feedforward control and correct transient emissions in real time.

[0047] 2. Breakthrough of thermal efficiency bottleneck: The thermal efficiency of gasoline engines has reached a bottleneck. Therefore, it is necessary to dynamically adjust the control parameters to improve the thermal efficiency of gasoline engines.

[0048] 3. Complex powertrain: Hybrid vehicles frequently start and stop. Traditional methods of adjusting control parameters based on map calibration can cause engine cold-start emissions to exceed standards, and mode switching can cause sudden torque changes, leading to worsening of noise, vibration, and harshness (NVH). Dynamic adjustment of control parameters is required.

[0049] 4. Multi-fuel compatibility challenge: The combustion characteristics of carbon-neutral fuels such as hydrogen fuel, ammonia fuel, and ethanol vary greatly, and control parameters need to be dynamically adjusted to match multi-fuel applications.

[0050] Based on this, when there are changes in fuel quality, carbon deposits, etc. that affect engine combustion, it is necessary to adaptively adjust the engine control parameters to reduce the frequency of knock and optimize engine operating performance.

[0051] The present invention provides a method for suppressing engine knock, such as Figure 1 As shown, the method mainly includes the following steps:

[0052] S110 , collecting engine operating characteristics corresponding to a vehicle traveling in an actual scenario to obtain a feature data set to be measured; the engine operating characteristics include: operating condition characteristics of the engine and an initial knock characteristic signal.

[0053] The main engine operating characteristics associated with engine knock include operating condition characteristics and initial knock characteristic signals. These operating condition characteristics include speed, load, ignition advance angle, air-fuel ratio, intake air temperature, coolant temperature, and atmospheric pressure. Furthermore, the signal collected by the knock sensor changes before and after engine knock occurs. Therefore, it is necessary to determine the time and frequency domain characteristics of the knock characteristic signal and use them as engine operating characteristics.

[0054] In one embodiment, collecting engine operating characteristics corresponding to a vehicle driving in an actual scenario to obtain a feature dataset to be measured includes:

[0055] Collecting engine operating condition characteristics and initial knock characteristic signals for a preset period of time at a predetermined collection frequency; the operating condition characteristics include: speed, load, ignition advance angle, air-fuel ratio, intake air temperature, coolant temperature, and atmospheric pressure;

[0056] Filtering the initial knock characteristic signal using a bandpass filter to obtain a preprocessed knock characteristic signal; determining time domain characteristics of the preprocessed knock characteristic signal, the time domain characteristics including: a peak value, a root mean square value, and a rising slope of the knock characteristic signal;

[0057] Perform fast Fourier transform on the pre-processed knock characteristic signal to obtain the knock characteristic frequency, sum the power of all knock characteristic frequency points to obtain the knock frequency band energy; determine the knock energy ratio based on the knock frequency band and total energy;

[0058] The time domain features, knock frequency band energy, knock energy ratio and working condition features of the preset time window are vector-joined in a predetermined order to obtain the feature data set to be tested.

[0059] Specifically, since the present invention uses the engine operating characteristics of the first N combustion cycles to predict knock for the time period corresponding to the next 1 to 3 combustion cycles, the value of N can be 3 or other values, and the present invention does not impose any limitation.

[0060] The speed, load, ignition advance angle, air-fuel ratio, intake air temperature, coolant temperature, atmospheric pressure can be collected according to the preset collection frequency, and the initial knock signal can be collected using a knock sensor.

[0061] The initial knock signal is an analog electrical signal, and the initial knock signal of the engine's high-frequency vibration needs to be converted into an analog electrical signal, and then converted into a digital electrical signal through an analog-to-digital converter ADC.

[0062] Generally speaking, during normal engine combustion, the energy in the 2kHz-10kHz frequency band accounts for approximately 30%, while during detonation, the energy in the 2kHz-10kHz frequency band accounts for approximately 70%. Therefore, the 2kHz-10kHz frequency band can be considered the detonation frequency band. Therefore, a bandpass filter in the 2kHz-10kHz frequency band can be used to filter the digital electrical signal, removing low-frequency noise (engine mechanical vibration) and high-frequency interference signals (such as electromagnetic noise) in the digital electrical signal to retain the characteristic detonation frequency signal.

[0063] The knock characteristic frequency signal is Fourier transformed by fast Fourier transform to obtain the energy of each frequency point. Based on the energy of each frequency point, the energies of each frequency point contained in the 2kHz to 10kHz frequency band are summed to obtain the energy of the knock frequency band. The knock energy ratio is determined based on the energy of the knock frequency band and the total energy.

[0064] The time-domain features of the preprocessed knock signal are then extracted. These features include the peak value, RMS value, and signal slope. The RMS value is the root mean square of the knock signal's amplitude, and the signal slope is determined using the formula (maximum amplitude - minimum amplitude) / (t2 - t1), where t2 is the time corresponding to the maximum amplitude, and t1 is the time corresponding to the minimum amplitude. A larger signal slope indicates a steeper rising edge, meaning the signal energy is released more quickly. Generally speaking, the slope of a knock signal is much greater than that of a normal mechanical vibration signal.

[0065] Since the engine is in the critical combustion period within the crankshaft angle range of 10° to 60° after ignition, it is necessary to determine the time window based on any crankshaft angle within the range of 10° to 60° after ignition (which can be selected based on actual conditions), extract the engine operation characteristics within the time window, and then determine the extracted engine operation characteristics as the feature data set to be tested.

[0066] The preset time window t can be determined according to the crankshaft angle as shown in formula (1):

[0067]

[0068] In formula (1), θ is the crankshaft angle, and r is the engine speed (in revolutions per second).

[0069] After collecting the above features, they are concatenated in a fixed order to form a vector. For example, concatenation is performed in the order of [peak value, root mean square (RMS), signal rise slope, knock frequency band energy, knock frequency band energy percentage, speed, load, ignition advance angle, air-fuel ratio, intake air temperature, coolant temperature, and atmospheric pressure] to form a concatenated vector. Generally, engine operating characteristics corresponding to three combustion cycles are collected. By extracting the above 11 features for each combustion cycle, a 33-dimensional feature matrix is ​​ultimately determined, which serves as the feature dataset to be tested.

[0070] For example, assuming that when detonation occurs, the time domain characteristic signal is as follows: peak value>500mV, RMS>150mV, signal rising slope>50mV / ms, detonation frequency band energy>500, detonation frequency band energy proportion>70%, speed>4500r / min, load>5bar, ignition advance angle<15°, air-fuel ratio≤14.6, intake temperature>25℃, coolant temperature>80℃, atmospheric pressure<105Kpa

[0071] Then the vector formed can be: [peak value = 520mV, root mean square = 160mV, signal rising slope = 55mV / ms, knock band energy = 750, knock band energy proportion = 80%, speed = 4800rpm, load = 20bar, ignition advance angle = 3°, air-fuel ratio = 14, intake temperature = 45℃, coolant temperature = 100℃, atmospheric pressure = 95Kpa].

[0072] S111, using a pre-trained extreme gradient boosting tree model to predict the feature data set to be tested, to obtain the probability of knock occurring in the engine in a future time period.

[0073] The feature dataset to be tested is fed into a pre-trained extreme gradient boosting tree model to determine the probability of engine knock occurring within a future timeframe. The duration of this future timeframe can be determined based on actual conditions. It should be neither too short nor too long, as this will prevent sufficient time to adjust control parameters. However, if it is too long, control lag will occur. Generally, the future timeframe can be 1 to 3 combustion cycles after engine ignition.

[0074] In one embodiment, the extreme gradient boosting tree model is an XGBoost model, which includes multiple decision trees. The feature data set to be tested is input into the pre-trained extreme gradient boosting tree model for prediction to obtain the probability of engine knock in the future time period, including:

[0075] For each decision tree, the splitting condition corresponding to the decision tree is used to match the feature value of each feature in the test data set in turn, and the sub-probability predicted by the decision tree is output;

[0076] The probability of engine knock occurring in the future period is determined based on the sub-probability predicted by each decision tree and the total number of decision trees;

[0077] Multiple decision trees are processed serially, and the next decision tree needs to fit the residual of the previous decision tree and output the corresponding sub-probability.

[0078] Each decision tree contains a root node (initial splitting node), internal nodes (splitting conditions), and leaf nodes. In one embodiment, the splitting conditions corresponding to the decision tree are used to match the feature values ​​of each feature in the test data set in turn, and the sub-probabilities predicted by the decision tree are output, including:

[0079] For each decision tree, starting from the root node of the decision tree, each feature value is matched with the splitting condition layer by layer until a leaf node is reached, and the predicted value corresponding to the leaf node is output.

[0080] In one embodiment, determining the probability of engine knock occurring in a future time period based on the sub-probabilities predicted by each decision tree and the total number of decision trees includes:

[0081] According to formula (2), the probability P of engine knock in the future period is determined as follows:

[0082]

[0083] In formula (2), n is the total number of decision trees, P i is the sub-probability predicted by the i-th decision tree.

[0084] Specifically, the extreme gradient boosting tree model can be obtained by training the XGboost model.

[0085] When training the XGboost model, sample data must first be acquired. This data can be continuously collected for both knock and normal combustion scenarios under various conditions. In practice, actual engine knock can be induced by varying the ignition angle, using low-octane fuel, performing high-temperature rapid acceleration, and climbing a hill with a heavy load, among other operating conditions. The corresponding knock samples can then be collected.

[0086] The training data set is determined based on the knock samples and normal combustion samples, and the machine learning model is trained using the training data set until the preset number of training times is reached or the prediction accuracy is determined to meet the requirements. The training is then terminated and the trained XGboost model (extreme gradient boosting tree model) is output.

[0087] A trained XGboost model consists of multiple decision trees (e.g., 100), each with a different splitting condition (also known as a decision rule). Each decision tree sequentially predicts the feature dataset under test, generating a corresponding sub-probability. When each subsequent decision tree predicts a corresponding sub-probability, it fits the residual (the error between the actual and predicted values) of the previous decision tree's prediction, thereby improving the prediction accuracy of the extreme gradient boosting tree model.

[0088] Each decision tree contains a root node (initial splitting node), internal nodes (splitting conditions) and leaf nodes. When making decisions on the feature data set to be tested, each decision tree needs to start from the root node and match the characteristic value of each feature layer by layer according to the splitting condition corresponding to the decision tree. Finally, the characteristic value will fall into the leaf node of the decision tree. Each leaf node will correspond to a predicted value, which is the sub-probability of engine knock in the future time period predicted by the decision tree.

[0089] For example, assuming the root node is: knock frequency band energy, if the knock frequency band energy is > 500, then enter the left subtree; otherwise, enter the right subtree;

[0090] Left subtree internal node: If the speed is > 4500 rpm and the peak value is > 400 mV, it falls into the left leaf node (labeled "knock"); otherwise, it falls into the right leaf node (labeled "normal"). The left and right leaf nodes each correspond to a prediction value. Since the present invention predicts the probability of engine knock, the prediction value corresponding to the left leaf node is determined as the sub-probability predicted by this decision tree.

[0091] In this way, each decision tree will eventually output a sub-probability after the above judgment, and finally the probability of engine knock in the future period can be determined based on the sub-probability predicted by each decision tree.

[0092] S112: Adjust control parameters of the engine based on the probability of knock occurring in the engine in a future period.

[0093] To reduce engine knock, the engine control parameters are adjusted based on the probability of engine knock in the future, including:

[0094] Determining a knock risk level based on the probability of engine knock occurring in a future period;

[0095] Determine the corresponding control strategy according to the knock risk level;

[0096] The control parameters of the engine are adjusted based on the control strategy.

[0097] In one embodiment, if the knock risk level is a safe level, the control strategy is: keep the original control parameters of the engine unchanged;

[0098] If the knock risk level is the warning level, the control strategy is: delay the ignition advance angle by a first angle;

[0099] If the knock risk level is an emergency level, the control strategy is: delay the ignition advance angle by a second angle and reduce the engine's boost pressure according to a preset ratio; the second angle is greater than the first angle.

[0100] Specifically, please refer to Table 1:

[0101] Table 1

[0102]

[0103] That is, when the probability of engine knock in the future time period is less than 0.3, the knock risk level is a safety level; when the probability of engine knock in the future time period is greater than or equal to 0.3 and less than 0.8, the knock risk level is a warning level; when the probability of engine knock in the future time period is greater than or equal to 0.8, the knock risk level is an emergency level.

[0104] In terms of the warning level, the corresponding control strategy is: delay the ignition advance angle by 1° (assuming the ignition advance angle was originally 30°, it should now be 29°) to ensure that the piston is ignited when it is further away from the top dead center, making it less likely to cause detonation.

[0105] This allows for early control intervention of engine combustion, adaptive adjustment of control parameters, and suppression of knock, thereby reducing the frequency of engine knock and improving engine performance.

[0106] It should be noted that if the knock risk level is at the warning or emergency level, and the knock risk level remains at the safe level for three consecutive combustion cycles after the control parameter adjustment, the adjusted control parameters will be gradually restored to their original state. During this gradual restoration, the ignition advance angle can be restored by 0.5° and the boost pressure can be restored by 2% per combustion cycle.

[0107] Additionally, in one embodiment, after adjusting the control parameters of the engine based on the probability of engine knock occurring in a future time period, the method further includes:

[0108] If it is determined that the prediction results of the extreme gradient boosting tree model have missed warnings or false positive warnings; or,

[0109] If it is determined that the engine operation characteristic data contains new characteristic data, the self-learning mechanism of the extreme gradient boosting tree model is triggered.

[0110] In one embodiment, the self-learning mechanism of the extreme gradient boosting tree model includes:

[0111] If it is determined that there is a missed warning in the prediction result of the extreme gradient boosting tree model, the missed features corresponding to the missed warning when the extreme gradient boosting tree model misses the warning are determined, and the split gain weight corresponding to the missed warning feature is increased;

[0112] If it is determined that there is a false alarm warning in the prediction result of the extreme gradient boosting tree model, the false alarm feature corresponding to the false alarm of the extreme gradient boosting tree model is determined, and the split gain weight corresponding to the false alarm feature is reduced; wherein the split gain weight is used to adjust the leaf node type into which the false alarm feature and the missed alarm feature fall, thereby updating the sub-probability predicted by the decision tree.

[0113] In one embodiment, the self-learning mechanism of the extreme gradient boosting tree model includes:

[0114] If new engine operating characteristics are identified, a corresponding decision tree is added to the extreme gradient boosting tree model.

[0115] Specifically, if there is a missed warning in the prediction result, it means that this key operating feature that affects the knock has not been given sufficient attention. Therefore, it is necessary to increase the split gain weight corresponding to the missed feature so that the missed feature can be more easily selected when the decision tree splits.

[0116] On the contrary, if it is determined that there are false positive warnings in the prediction results of the extreme gradient boosting tree model, it means that the operating features are over-reliant. Therefore, it is necessary to reduce the split gain weight corresponding to the false positive feature so that the false positive feature is less likely to be selected when the decision tree splits.

[0117] In this way, adjusting the split gain weight will directly change the importance ranking of each operating feature, thereby improving the model prediction accuracy.

[0118] Taking the knock frequency band energy, speed, and intake air temperature as examples, through the autonomous learning mechanism, the split gain weights of these three features are adjusted as shown in Table 2:

[0119] Table 2

[0120]

[0121] In addition, if new engine operating characteristics are identified (for example, the use of a new fuel in the vehicle causes the knock frequency band to shift), a corresponding decision tree can be added to the extreme gradient boosting tree model. The new decision tree makes decisions based on "fuel type + high-frequency harmonics".

[0122] In this way, the present invention predicts the engine operating characteristics through the pre-trained extreme gradient boosting tree model, so as to accurately predict the engine knock risk, and then intervene in the control parameters in advance to suppress the occurrence of engine knock and improve the engine operating performance.

[0123] The present invention establishes feedforward prediction and adaptive control adjustment of knock tendency, achieving a double breakthrough in ensuring engine power performance and reliability: control intervention is performed 1-3 combustion cycles in advance based on the knock sensor signal. Compared with the traditional torque reduction feedback mechanism after knock, it can ensure uninterrupted power output and improve engine operating performance; knock occurrence is suppressed through knock prediction, the frequency of knock occurrence is reduced, the wear of key moving parts of the engine is reduced, and the reliability and durability of the engine are improved.

[0124] Based on the same inventive concept as in the above embodiment, this embodiment also provides a device for suppressing engine knock, such as Figure 2 As shown, the device includes:

[0125] The acquisition unit 21 is used to acquire engine operating characteristics corresponding to the vehicle driving in an actual scenario to obtain a feature data set to be measured; the engine operating characteristics include: the engine operating condition characteristics and the initial knock characteristic signal;

[0126] A prediction unit 22 is configured to predict the feature data set to be tested using a pre-trained extreme gradient boosting tree model to obtain a probability of knock occurring in the engine in a future period;

[0127] The adjusting unit 23 is configured to adjust the control parameters of the engine based on the probability of knock occurring in the engine within a future period.

[0128] The collection unit 21 is specifically used for:

[0129] collecting the engine's operating condition characteristics and initial knock characteristic signals at a predetermined frequency; the operating condition characteristics include: speed, load, ignition advance angle, air-fuel ratio, intake air temperature, coolant temperature, and atmospheric pressure;

[0130] Filtering the initial knock characteristic signal using a bandpass filter to obtain a preprocessed knock characteristic signal; determining time domain characteristics of the preprocessed knock characteristic signal, the time domain characteristics including: a peak value, a root mean square value, and a signal rising slope of the knock characteristic signal;

[0131] Performing a fast Fourier transform on the preprocessed knock characteristic signal to obtain a knock characteristic frequency, summing the power of all knock characteristic frequency points to obtain a knock frequency band energy; and determining a knock energy proportion based on the knock frequency band and the total energy;

[0132] The time domain features, the knock frequency band energy, the knock energy ratio and the operating condition features within a preset time window are vector-joined in a predetermined order to obtain the feature data set to be measured.

[0133] The prediction unit 22 is configured to:

[0134] For each decision tree, the feature value of each feature in the test data set is matched in turn using the splitting condition corresponding to the decision tree, and the sub-probability predicted by the decision tree is output;

[0135] The probability of knock occurring in the engine in a future period is determined based on the sub-probability predicted by each decision tree and the total number of decision trees; wherein,

[0136] The multiple decision trees are processed in series, and the subsequent decision tree needs to be fitted with the residual of the previous decision tree.

[0137] The adjustment unit 23 is configured to:

[0138] determining a knock risk level based on a probability of the engine knocking within a future time period;

[0139] determining a corresponding control strategy according to the knock risk level;

[0140] The control parameters of the engine are adjusted based on the control strategy.

[0141] The adjustment unit 23 is configured to:

[0142] If the knock risk level is a safe level, the control strategy is: keeping the original control parameters of the engine unchanged;

[0143] If the knock risk level is a warning level, the control strategy is: delaying the ignition advance angle by a first angle;

[0144] If the knock risk level is an emergency level, the control strategy is: delaying the ignition advance angle by a second angle and reducing the boost pressure of the engine according to a preset ratio; the second angle is greater than the first angle.

[0145] The device further comprises a self-learning unit 24 for:

[0146] If it is determined that the prediction result of the extreme gradient boosting tree model has a missed warning or a false positive warning; or

[0147] If it is determined that the engine operation characteristic data contains new characteristic data, the self-learning mechanism of the extreme gradient boosting tree model is triggered.

[0148] The self-learning mechanism of the extreme gradient boosting tree model includes:

[0149] If it is determined that there is a missed warning in the prediction result of the extreme gradient boosting tree model, determine the missed warning feature corresponding to the missed warning of the extreme gradient boosting tree model, and increase the split gain weight corresponding to the missed warning feature;

[0150] If it is determined that there is a false alarm warning in the prediction result of the extreme gradient boosting tree model, the false alarm feature corresponding to the false alarm of the extreme gradient boosting tree model is determined, and the split gain weight corresponding to the false alarm feature is reduced; wherein, the split gain weight is used to adjust the leaf node type into which the false alarm feature and the missed alarm feature fall, thereby updating the sub-probability predicted by the decision tree.

[0151] If a new engine operating feature is identified, a corresponding decision tree is added to the extreme gradient boosting tree model.

[0152] Since the device described in the embodiments of the present invention is used to implement the method for suppressing engine knock in the embodiments of the present invention, the specific structure and variations of the device are readily apparent to those skilled in the art based on the method described in the embodiments of the present invention, and therefore, no further description is given here. All devices used in the methods of the embodiments of the present invention fall within the scope of protection of the present invention.

[0153] Based on the same inventive concept as the aforementioned embodiment, the present invention further provides a vehicle comprising the engine knock suppression device described in the aforementioned embodiment. The specific structure and implementation of the device can be referred to the corresponding description in the aforementioned embodiment and will not be repeated here.

[0154] Through one or more embodiments of the present invention, the present invention has the following beneficial effects or advantages:

[0155] The present invention provides a method, device and vehicle for suppressing engine knock, the method comprising: collecting engine operating characteristics corresponding to a vehicle traveling in an actual scenario to obtain a feature data set to be measured; the engine operating characteristics comprising: operating condition characteristics of the engine and an initial knock characteristic signal; using a pre-trained extreme gradient boosting tree model to predict the feature data set to be measured to obtain a probability of knock occurring in the engine in a future time period; and adjusting control parameters of the engine based on the probability of knock occurring in the engine in the future time period; in this way, the engine operating characteristics are predicted by the pre-trained extreme gradient boosting tree model to accurately predict the knock risk of the engine, thereby intervening in control parameters in advance to suppress engine knock, thereby reducing the frequency of knock and improving engine operating performance.

[0156] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0157] The above description is only a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for suppressing engine knock, characterized in that: The method comprises: Collecting engine operating characteristics corresponding to a vehicle driving in an actual scenario to obtain a feature data set to be measured; the engine operating characteristics include: the engine operating condition characteristics and the initial knock characteristic signal; Using a pre-trained extreme gradient boosting tree model to predict the feature data set to be tested, to obtain a probability of knock occurring in the engine in a future time period; Control parameters of the engine are adjusted based on the probability that knock will occur in the engine within a future period.

2. The method according to claim 1, wherein The collecting of engine operation characteristics corresponding to the vehicle driving in the actual scene to obtain a feature data set to be tested includes: collecting the engine's operating condition characteristics and initial knock characteristic signals at a predetermined frequency; the operating condition characteristics include: speed, load, ignition advance angle, air-fuel ratio, intake air temperature, coolant temperature, and atmospheric pressure; Filtering the initial knock characteristic signal using a bandpass filter to obtain a preprocessed knock characteristic signal; determining time domain characteristics of the preprocessed knock characteristic signal, the time domain characteristics including: a peak value, a root mean square value, and a signal rising slope of the knock characteristic signal; Performing a fast Fourier transform on the preprocessed knock characteristic signal to obtain a knock characteristic frequency, summing the power of all knock characteristic frequency points to obtain a knock frequency band energy; and determining a knock energy proportion based on the knock frequency band and the total energy; The time domain features, the knock frequency band energy, the knock energy ratio and the operating condition features within a preset time window are vector-joined in a predetermined order to obtain the feature data set to be measured.

3. The method according to claim 1, wherein The extreme gradient boosting tree model includes a plurality of decision trees; the method of using the pre-trained extreme gradient boosting tree model to predict the feature data set to be tested to obtain the probability of the engine knocking in a future time period includes: For each decision tree, the feature value of each feature in the test data set is matched in turn using the splitting condition corresponding to the decision tree, and the sub-probability predicted by the decision tree is output; The probability of knock occurring in the engine in a future period is determined based on the sub-probability predicted by each decision tree and the total number of decision trees; wherein, The multiple decision trees are processed serially, and the subsequent decision tree needs to fit the residual of the previous decision tree and output the corresponding sub-probability.

4. The method according to claim 1, wherein The adjusting the control parameters of the engine based on the probability of knock occurring in the engine within a future period includes: determining a knock risk level based on a probability of the engine knocking within a future time period; determining a corresponding control strategy according to the knock risk level; The control parameters of the engine are adjusted based on the control strategy.

5. The method according to claim 4, wherein The determining of a corresponding control strategy according to the knock risk level includes: If the knock risk level is a safe level, the control strategy is: keeping the original control parameters of the engine unchanged; If the knock risk level is a warning level, the control strategy is: delaying the ignition advance angle by a first angle; If the knock risk level is an emergency level, the control strategy is: delaying the ignition advance angle by a second angle and reducing the boost pressure of the engine according to a preset ratio; the second angle is greater than the first angle.

6. The method according to claim 1, wherein After adjusting the control parameters of the engine based on the probability of knock occurring in the engine within a future time period, the method further includes: If it is determined that the prediction result of the extreme gradient boosting tree model has a missed warning or a false positive warning; or If it is determined that the engine operation characteristic data contains new characteristic data, the self-learning mechanism of the extreme gradient boosting tree model is triggered.

7. The method according to claim 6, wherein The self-learning mechanism of the extreme gradient boosting tree model includes: If it is determined that there is a missed warning in the prediction result of the extreme gradient boosting tree model, determine the missed warning feature corresponding to the missed warning of the extreme gradient boosting tree model, and increase the split gain weight corresponding to the missed warning feature; If it is determined that there is a false alarm warning in the prediction result of the extreme gradient boosting tree model, the false alarm feature corresponding to the false alarm of the extreme gradient boosting tree model is determined, and the split gain weight corresponding to the false alarm feature is reduced; wherein, the split gain weight is used to adjust the leaf node type into which the false alarm feature and the missed alarm feature fall, thereby updating the sub-probability predicted by the decision tree.

8. The method according to claim 6, wherein The self-learning mechanism of the extreme gradient boosting tree model includes: If a new engine operating feature is identified, a corresponding decision tree is added to the extreme gradient boosting tree model.

9. A device for suppressing engine knock, characterized in that: The device comprises: The acquisition unit is used to collect the engine operation characteristics corresponding to the vehicle when it is driving in the actual scene to obtain a feature data set to be tested; A prediction unit, configured to predict the feature data set to be tested using a pre-trained extreme gradient boosting tree model to obtain a probability of knock occurring in the engine within a future period; An adjustment unit is configured to adjust a control parameter of the engine based on a probability of knock occurring in the engine within a future period.

10. A vehicle, characterized in that: The vehicle includes the device for suppressing engine knock according to claim 9.

Images