Engine combustion prediction control method and device based on cylinder pressure signal and vehicle

Abstract

The invention provides an engine combustion prediction control method and device based on a cylinder pressure signal and a vehicle, and the method comprises the steps: determining engine operation characteristics corresponding to N combustion cycles through the cylinder pressure, and determining a to-be-detected engine characteristic data set according to the engine operation characteristics; predicting the to-be-detected engine feature data set by using the two-channel convolutional neural network model to obtain the probability of knocking of the engine in a future time period; adaptively adjusting control parameters of the engine on the basis of the probability that the engine knocks in the future time period; therefore, due to the fact that the short-term fluctuation intensity of the cylinder pressure and the number of the abnormal pulses of the cylinder pressure can reflect the abrupt change characteristics of the engine in the combustion process, and the long-term trend slope and the peak-valley ratio can reflect the development direction of the combustion stability; the knock risk of the engine can be accurately pre-judged in combination with the short-term sudden change and long-term change trend of the engine, engine control parameters are intervened in advance, and the operation performance of the engine is improved.

Description

Technical Field

[0001] This invention relates to the field of engine electronic control technology, and in particular to an engine combustion prediction control method, device, and vehicle based on cylinder pressure signals. Background Technology

[0002] The in-cylinder combustion process in an internal combustion engine is a core link in energy conversion, and its dynamic evolution characteristics have a decisive impact on the overall system performance. The combustion process in an internal combustion engine is affected by a variety of dynamic factors (such as fuel quality, intake conditions, ambient temperature, etc.), and various control parameters need to be adjusted during combustion to optimize engine operation and improve engine performance.

[0003] In related technologies, engine control mainly relies on sensor parameters, which are looked up using fixed mapping tables (such as ignition MAP, fuel injection MAP, etc.) or adjusted using empirical models. However, these control methods cannot predict the combustion state in advance, and have certain limitations and lag.

[0004] Therefore, there is an urgent need for an engine combustion predictive control method to solve the above-mentioned technical problems and improve engine performance. Summary of the Invention

[0005] To address the problems existing in the prior art, embodiments of the present invention provide an engine combustion prediction control method, device, and vehicle based on cylinder pressure signals, in order to solve or partially solve the technical problem that the inability to predict the engine combustion state in advance leads to certain limitations and lags in adjusting engine control parameters, thereby affecting engine operating performance.

[0006] A first aspect of the present invention provides an engine combustion predictive control method based on cylinder pressure signals, characterized in that the method comprises:

[0007] The cylinder pressure of a vehicle engine is collected, and the engine operating characteristics corresponding to N combustion cycles are determined using the cylinder pressure. Based on the engine operating characteristics corresponding to the N combustion cycles, a characteristic dataset of the engine to be tested is determined. The engine operating characteristics include: the short-term fluctuation intensity of the cylinder pressure, the long-term trend slope of the cylinder pressure, the peak-to-valley ratio of the cylinder pressure, and the number of abnormal pulses in the cylinder pressure.

[0008] The probability of the engine detonating in the future is obtained by using a pre-trained dual-channel convolutional neural network model to predict the feature dataset of the engine under test.

[0009] The risk level of knocking is determined based on the probability of the engine knocking in the future, and the control parameters of the engine are adaptively adjusted according to the risk level.

[0010] In the above scheme, the step of using a pre-trained dual-channel convolutional neural network model to predict the feature dataset of the engine under test to obtain the probability of the engine detonating in the future includes:

[0011] The first channel of the dual-channel convolutional neural network model is used to process the short-term fluctuation intensity and the number of abnormal pulses in the feature dataset of the engine under test, respectively, to obtain the first feature value used to characterize cylinder pressure mutation.

[0012] The second channel of the dual-channel convolutional neural network model is used to process the long-term trend slope and the number of abnormal pulses in the feature dataset of the engine under test, respectively, to obtain a second feature value that characterizes the cylinder pressure trend.

[0013] The probability of engine knocking in the future is determined using the first feature value and the second feature value.

[0014] In the above scheme, the step of using the first channel of the dual-channel convolutional neural network model to process the short-term fluctuation intensity and the number of abnormal pulses in the feature dataset of the engine under test to obtain a first feature value for characterizing cylinder pressure mutation includes:

[0015] In the first channel, the short-term fluctuation intensity and the number of abnormal pulses in the feature dataset of the engine under test are processed by the first convolution kernel respectively to obtain the corresponding short-term fluctuation intensity feature value sequence and the abnormal pulse number feature value sequence.

[0016] The short-term fluctuation intensity feature value sequence and the abnormal pulse number feature value sequence are respectively pooled using a pooling layer to obtain the corresponding target short-term fluctuation intensity feature value and target abnormal pulse number feature value.

[0017] The first feature value is obtained by processing the target short-term fluctuation intensity feature value and the target abnormal pulse quantity feature value using the first fully connected layer.

[0018] In the above scheme, the step of using the second channel of the dual-channel convolutional neural network model to process the long-term trend slope and peak-to-valley ratio in the feature dataset of the engine under test to obtain a second feature value characterizing the cylinder pressure trend includes:

[0019] In the second channel, the long-term trend slope and peak-to-valley ratio in the feature dataset of the engine under test are processed by the second convolution kernel to obtain the corresponding long-term trend slope feature value sequence and peak-to-valley ratio feature value sequence.

[0020] The long-term trend slope feature value sequence and the peak-to-valley ratio feature value sequence are respectively pooled using a pooling layer to obtain the corresponding target long-term trend slope feature value and target peak-to-valley ratio feature value.

[0021] The second feature value is obtained by processing the target long-term trend slope feature value and the target peak-to-valley ratio feature value using the second fully connected layer.

[0022] In the above scheme, determining the probability of engine knocking in the future using the first feature value and the second feature value includes:

[0023] Obtain the first weight corresponding to the first feature value, the second weight corresponding to the second feature value, and the bias parameter;

[0024] The fused feature value is determined using the first weight, the first feature value, the second weight, the second feature value, and the bias parameter;

[0025] According to the formula Determine the probability P of the engine detonating in the future;

[0026] Wherein, e is an exponential function, and z is the fusion feature value.

[0027] In the above scheme, determining the fused feature value using the first weight, the first feature value, the second weight, the second feature value, and the bias parameter includes:

[0028] The fusion feature value z is determined using the formula z = w1 × η1 + w2 × η2 + b; where,

[0029] w1 is the first weight, η1 is the first feature value, w2 is the second weight, η2 is the second feature value, and b is the bias parameter.

[0030] In the above scheme, before adaptively adjusting the control parameters of the engine based on the probability of engine knocking in the future, the method further includes:

[0031] Obtain the engine speed of the combustion cycle with the highest sequence number among the N combustion cycles and the engine's average effective braking pressure;

[0032] The ignition advance angle threshold is determined based on the engine speed and the average effective braking pressure, and the control parameters corresponding to different knock probabilities are determined based on the ignition advance angle threshold.

[0033] In the above scheme, determining the risk level of engine knocking based on the probability of knocking occurring in the future, and adaptively adjusting the engine control parameters based on the risk level, includes:

[0034] If it is determined that the probability of the engine knocking in the future period is less than the first threshold, then the risk level is determined to be a safe level, and the current control parameters are maintained unchanged.

[0035] If it is determined that the probability of the engine knocking in the future period is greater than or equal to the first threshold and less than the second threshold, then the risk level is determined to be the warning level, and the ignition advance angle of the engine is controlled to be delayed by the first angle.

[0036] If it is determined that the probability of the engine knocking in the future period is greater than or equal to the second threshold and less than the third threshold, then the risk level is determined to be a high risk level, and the ignition advance angle of the engine is delayed by a second angle and the fuel injection quantity of the engine is increased by a first proportion; the second angle is greater than the first angle.

[0037] If it is determined that the probability of engine knocking in the future period is greater than the third threshold, then the risk level is determined to be an emergency level, the ignition advance angle is delayed by the third angle, and the engine torque is reduced according to the second ratio; the third angle is greater than the second angle.

[0038] A second aspect of the present invention provides an engine combustion prediction control device based on cylinder pressure signals, the device comprising:

[0039] The determination unit is used to collect the cylinder pressure of the vehicle engine, use the cylinder pressure to determine the engine operating characteristics corresponding to N combustion cycles, and determine the engine feature dataset to be tested based on the engine operating characteristics corresponding to the N combustion cycles; the engine operating characteristics include: the short-term fluctuation intensity of the cylinder pressure, the long-term trend slope of the cylinder pressure, the peak-to-valley ratio of the cylinder pressure, and the number of abnormal pulses of the cylinder pressure.

[0040] The prediction unit is used to predict the probability of the engine detonating in the future using a pre-trained dual-channel convolutional neural network model on the feature dataset of the engine under test.

[0041] The adjustment unit is used to determine the risk level of detonation based on the probability of detonation occurring in the engine in the future period, and to adaptively adjust the control parameters of the engine according to the risk level.

[0042] A third aspect of the present invention provides a vehicle comprising the engine combustion prediction control device based on cylinder pressure signals as described in the second aspect.

[0043] This invention provides an engine combustion prediction control method, device, and vehicle based on cylinder pressure signals, comprising: acquiring cylinder pressure of a vehicle engine; using the cylinder pressure to determine engine operating characteristics corresponding to N combustion cycles; and determining a test engine feature dataset based on the engine operating characteristics corresponding to the N combustion cycles. The engine operating characteristics include: the short-term fluctuation intensity of the cylinder pressure, the long-term trend slope of the cylinder pressure, the peak-to-valley ratio of the cylinder pressure, and the number of abnormal pulses in the cylinder pressure. A pre-trained dual-channel convolutional neural network model is used to predict the test engine feature dataset to obtain the probability of engine knocking in the future. The probability of engine knocking in the future is used to determine the risk level of knocking, and the engine control parameters are adaptively adjusted according to the risk level. In this way, since the short-term fluctuation intensity of cylinder pressure and the number of abnormal cylinder pressure pulses can reflect the abrupt changes in the engine combustion process, and the long-term trend slope of cylinder pressure and the peak-to-valley ratio of cylinder pressure can reflect the development direction of combustion stability, the pre-trained dual-channel convolutional neural network model can accurately predict the engine knocking risk by combining the engine's short-term abrupt changes and long-term trends when predicting the above features. This allows for early intervention in engine control parameters, reducing the probability of knocking and improving engine performance. Attached Figure Description

[0044] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0045] Figure 1 A schematic flowchart of an engine combustion predictive control method based on cylinder pressure signal according to an embodiment of the present invention is shown.

[0046] Figure 2 A schematic diagram of an engine combustion prediction control device based on cylinder pressure signals according to an embodiment of the present invention is shown. Detailed Implementation

[0047] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0048] To better understand the technical solution of this invention, let's first introduce the current status of engine combustion control:

[0049] 1. Stringent emission regulations: Engines are required to achieve ultra-low emissions under all operating conditions (e.g., NOx ≤ 35 mg / km, and particulate matter number (PN) in exhaust gas must be measured for particles with a diameter of 10 nm). However, traditional combustion control relies on fixed mapping tables (such as ignition map, fuel injection map, etc.) for feedback control, which has a lag of 1 to 2 combustion cycles, making it impossible to perform active feedforward control and correct transient emissions in real time.

[0050] 2. Breakthrough in thermal efficiency bottleneck: The thermal efficiency of gasoline engines has reached a bottleneck, so it is necessary to dynamically adjust the control parameters to improve the thermal efficiency of gasoline engines.

[0051] 3. Complex powertrain system: Hybrid vehicles experience frequent start-stop cycles. Traditional methods of adjusting control parameters based on map calibration can lead to excessive emissions during engine cold starts and sudden torque changes during mode switching, resulting in deterioration of noise, vibration, and harshness (NVH). Dynamic adjustment of control parameters is required.

[0052] 4. Multi-fuel compatibility challenge: The combustion characteristics of fuels such as hydrogen, ammonia, and ethanol vary greatly, requiring dynamic adjustment of control parameters to match multi-fuel applications.

[0053] Therefore, there is an urgent need to adjust engine control parameters in advance through prediction to reduce engine emissions. Simultaneously, in situations where changes in fuel quality or carbon buildup affect engine combustion, adaptive adjustments to engine control parameters should be made to optimize engine performance.

[0054] This invention provides an engine combustion predictive control method based on cylinder pressure signals, such as... Figure 1 As shown, the method mainly includes the following steps:

[0055] S110, Collect the cylinder pressure of the vehicle engine, use the cylinder pressure to determine the engine operating characteristics corresponding to N combustion cycles, and determine the engine feature dataset to be tested based on the engine operating characteristics corresponding to the N combustion cycles; the engine operating characteristics include: the short-term fluctuation intensity of the cylinder pressure, the long-term trend slope of the cylinder pressure, the peak-to-valley ratio of the cylinder pressure, and the number of abnormal pulses of the cylinder pressure.

[0056] During normal vehicle operation, the cylinder pressure of the engine in each combustion cycle can be collected at a preset frequency. The cylinder pressure is used to determine the engine operating characteristics corresponding to N combustion cycles. Based on these N operating characteristics, a feature dataset of the engine under test is determined. The engine operating characteristics include: the short-term fluctuation intensity of cylinder pressure, the long-term trend slope of cylinder pressure, the peak-to-valley ratio of cylinder pressure, and the number of abnormal pulses in cylinder pressure. The value of N can be set based on actual needs, for example, it can be 30, without limitation.

[0057] For example, if N is 30, and the calculation starts from the 50th combustion cycle, then N combustion cycles are from the 50th combustion cycle to the 21st combustion cycle.

[0058] In one implementation, determining the engine operating characteristics corresponding to N combustion cycles using cylinder pressure includes:

[0059] When the engine operating characteristic is the short-term fluctuation intensity of cylinder pressure (SFI), the cylinder pressure of the first target combustion cycle is obtained by taking any one of the N combustion cycles as the reference combustion cycle in descending order of combustion cycle number; the first target combustion cycle includes: the reference combustion cycle and the first number of combustion cycles before the reference combustion cycle.

[0060] The first standard deviation of the rate of change of cylinder pressure with crankshaft angle for the first target combustion cycle is determined, and the first standard deviation is determined as the short-term fluctuation intensity corresponding to the benchmark combustion cycle.

[0061] Specifically, the first target combustion cycle consists of 5 cycles. This invention uses the engine operating characteristics of the first N combustion cycles to predict the probability of engine knocking in the (N+1) to (N+3)th combustion cycles. The time period corresponding to the (N+1) to (N+3)th combustion cycles is referred to as the future time period in this invention. Therefore, when calculating the short-term fluctuation intensity of cylinder pressure, it is necessary to take any one of the N combustion cycles as the reference combustion cycle, and then, starting from the reference cycle, find the rate of change of cylinder pressure with crankshaft angle for the previous 4 (first number) adjacent combustion cycles (a total of 5 combustion cycles). Based on the cylinder pressure of these 5 cycles, the short-term fluctuation intensity corresponding to the reference combustion cycle is determined, until the short-term fluctuation intensity corresponding to all N combustion cycles is determined.

[0062] For example, assuming the reference combustion cycle number is the 50th combustion cycle, it is necessary to obtain the rate of change of cylinder pressure with crankshaft angle for the 50th, 49th, 48th, 47th and 46th combustion cycles, and then determine the short-term fluctuation intensity corresponding to the 50th combustion cycle.

[0063] Assuming the baseline combustion cycle number is the 49th combustion cycle, it is necessary to obtain the rate of change of cylinder pressure with crankshaft angle for the 49th, 48th, 47th, 46th, and 45th combustion cycles, and then determine the short-term fluctuation intensity corresponding to the 49th combustion cycle. This process is repeated until the short-term fluctuation intensity corresponding to 30 combustion cycles is determined.

[0064] When determining the intensity of short-term fluctuations, the 50th combustion cycle will be used as an example for explanation:

[0065] Obtain the rate of change of cylinder pressure with crankshaft angle for each combustion cycle in the first target combustion cycle. The cylinder pressure can be the average rate of change or the peak rate of change. Then, average the rate of change of the five cylinder pressures with crankshaft angle to obtain the first average value.

[0066] According to the formula The first standard deviation is determined and defined as the short-term fluctuation intensity of the 50th combustion cycle. Here, i represents the i-th combustion cycle in the first target combustion cycle, and x... i Let be the rate of change of cylinder pressure with crankshaft angle in the i-th combustion cycle. This is the first average value.

[0067] In one implementation, determining the engine operating characteristics corresponding to N combustion cycles using cylinder pressure includes:

[0068] When the engine operating characteristic is the long-term trend slope of cylinder pressure (TTF), the cylinder pressure of the second target combustion cycle is obtained by taking any one of the N combustion cycles as the reference combustion cycle in descending order of combustion cycle number; the second target combustion cycle includes: the reference combustion cycle and a second number of combustion cycles preceding the reference combustion cycle; the second number is greater than the first number;

[0069] Linear fitting is performed on the cylinder pressure of the second target combustion cycle to obtain a straight line. The slope of the straight line is determined as the long-term trend slope of the cylinder pressure. The long-term trend slope is used to characterize the change trend of cylinder pressure with the number of combustion cycles.

[0070] The number of the second target combustion cycles can be 50, and the second number is 49. Since the number of the second target combustion cycles is much larger than the number of the first target combustion cycles, in order to determine the short-term fluctuation intensity and long-term trend slope of the N combustion cycles at the same time, the sequence number of the last combustion cycle of the N combustion cycles is generally at least 79. That is to say, the short-term fluctuation intensity and long-term trend slope of the N (30) combustion cycles are determined starting from the 79th combustion cycle.

[0071] For example, assuming the reference combustion cycle number is the 79th combustion cycle, then it is necessary to obtain the cylinder pressure of the 79th to 30th combustion cycles in order to determine the long-term trend slope corresponding to the 79th combustion cycle.

[0072] Assuming the baseline combustion cycle number is the 78th combustion cycle, then the cylinder pressures for combustion cycles 78 to 29 need to be obtained to determine the long-term trend slope corresponding to the 78th combustion cycle. This process is repeated until the long-term fluctuation intensity corresponds to 30 combustion cycles.

[0073] When determining the slope of the long-term trend, the 79th combustion cycle will be used as an example for illustration:

[0074] Obtain the target cylinder pressure for each combustion cycle in the second target combustion cycle. The target cylinder pressure can be the average cylinder pressure for each combustion cycle or the peak cylinder pressure for each combustion cycle; thus, 30 target cylinder pressures are obtained.

[0075] Then, the least squares method was used to perform linear fitting on the 30 target cylinder pressures to obtain the straight line y = kx + a between the cylinder pressure and the combustion cycle mark. The slope k of this straight line was determined as the long-term fluctuation intensity corresponding to the 79th combustion cycle.

[0076] It should be noted that when fitting the straight line, the sequence numbers of combustion cycles 79 to 30 need to be replaced with combustion cycle markers 30 to 1. Similarly, in each straight line fitting, the sequence numbers corresponding to the 30 combustion cycles need to be replaced with 30 to 1.

[0077] In one implementation, determining the engine operating characteristics corresponding to N combustion cycles using cylinder pressure includes:

[0078] When the engine operating characteristic is the peak-to-valley ratio of cylinder pressure, the maximum value, minimum value, and average value of the rate of change of cylinder pressure with crankshaft angle are determined sequentially for N combustion cycles in descending order of combustion cycle number.

[0079] The peak-to-valley ratio of the corresponding combustion cycle is determined based on the maximum, minimum, and average rates of change of cylinder pressure with crankshaft angle for each combustion cycle.

[0080] Among them, it can be based on the formula Determine the peak-to-valley ratio (PVR) for the corresponding combustion cycle; This represents the maximum rate of change of cylinder pressure with respect to crankshaft angle. denoted as the minimum rate of change of cylinder pressure with crankshaft angle, denoted as mean, and denoted as average rate of change of cylinder pressure with crankshaft angle. Here, p is the cylinder pressure and θ is the crankshaft angle.

[0081] In one implementation, determining the engine operating characteristics corresponding to N combustion cycles using cylinder pressure includes:

[0082] When the engine operating characteristic is the abnormal pulse count (APC) of cylinder pressure, the cylinder pressure of the third target combustion cycle is obtained by taking any one of the N combustion cycles as the reference combustion cycle in descending order of combustion cycle number; the third target combustion cycle includes: the reference combustion cycle and the third number of combustion cycles before the reference combustion cycle; the third number is greater than the first number and less than the second number;

[0083] The cylinder pressure variance and second standard deviation are determined based on the cylinder pressure of the third target combustion cycle; the cylinder pressure threshold is determined based on the cylinder pressure variance and second standard deviation.

[0084] In the third target combustion cycle, cylinder pressure exceeding the cylinder pressure threshold is identified as an abnormal pulse, and the number of abnormal pulses in the third target combustion cycle is determined.

[0085] Specifically, if the number of the third target combustion cycles can be 10, then the third number is 9. For example, assuming the baseline combustion cycle number is the 79th combustion cycle, then it is necessary to obtain the cylinder pressure of the 79th to 70th combustion cycles, and then determine the long-term trend slope corresponding to the 79th combustion cycle.

[0086] The 3σ threshold is determined based on the cylinder pressure of the 79th to 70th combustion cycles. This 3σ threshold is then used as the cylinder pressure threshold. The number of cylinder pressures exceeding the threshold in the third target combustion cycle is counted and recorded as the number of abnormal pulses. This process can be repeated to determine the number of abnormal pulses for N combustion cycles.

[0087] It is understandable that each combustion cycle in the N combustion cycles contains 4 of the above-mentioned operational features, so the final feature dataset of the engine under test is a 30*4 matrix, as shown in Table 1:

[0088] Table 1

[0089] Combustion cycle number SFI TTS PVR APC N 2.3 0.04 0.5 3 N-1 1.8 0.03 0.4 2 …… …… …… …… …… N-29 1.1 0.01 0.2 0

[0090] S111, using a pre-trained dual-channel convolutional neural network model to predict the characteristic dataset of the engine under test, the probability of the engine experiencing knocking in the future time period is obtained.

[0091] After determining the aforementioned characteristic dataset of the engine to be tested, a pre-trained dual-channel convolutional neural network model is used to predict the probability of engine knocking in the future.

[0092] The training process of the dual-channel convolutional neural network model is as follows:

[0093] First, training data samples are obtained. Knock can be artificially induced through bench testing, allowing for the collection of positive samples of cylinder pressure under multiple combustion cycles in a knock-prone state. These positive samples include: short-term fluctuation intensity, long-term trend slope of cylinder pressure, peak-to-valley ratio of cylinder pressure, and the number of abnormal cylinder pressure pulses. Induction methods include: excessively retarding the ignition timing, using low-octane fuel, and increasing intake air temperature.

[0094] Collect negative sample data under multiple normal combustion cycles: negative sample data includes the short-term fluctuation intensity of cylinder pressure, the long-term trend slope of cylinder pressure, the peak-to-valley ratio of cylinder pressure, and the number of abnormal pulses in cylinder pressure.

[0095] The positive and negative samples are combined in a 1:3 ratio to obtain the training data sample set.

[0096] Then, the loss function is determined. The loss function of this invention can be: Loss = -α × (1 - P) γ ×y×log(P); where α is the weight adjustment coefficient, used to reduce the weight of negative samples; α=0.25; γ is the focusing coefficient, mainly used to focus on samples that are difficult to classify; γ=2; y is the sample label (positive sample label=1, negative sample label=0), and P is the prediction probability.

[0097] A dual-channel convolutional neural network model was trained using training data samples. During the training process, the initial learning rate was set to 0.001, and it was decayed to 1 / 10 of the original rate every 50 epochs.

[0098] The training process is divided into stages: Stage 1: Training is performed on only one channel. After training one channel is completed, the second stage of training begins. Stage 2: Training is performed on both channels simultaneously until the training termination condition is reached (accuracy reaches the required level or the number of iterations is reached), at which point the trained dual-channel convolutional neural network model is output.

[0099] In one implementation, a pre-trained dual-channel convolutional neural network model is used to predict the probability of engine knocking in the future using a feature dataset of the engine under test, including:

[0100] The first channel of the dual-channel convolutional neural network model is used to process the short-term fluctuation intensity and the number of abnormal pulses in the characteristic dataset of the engine under test, respectively, to obtain the first feature value used to characterize cylinder pressure mutation.

[0101] The second channel of the dual-channel convolutional neural network model is used to process the long-term trend slope and the number of abnormal pulses in the feature dataset of the engine under test, respectively, to obtain a second feature value that characterizes the cylinder pressure trend.

[0102] The probability of engine knocking in the future is determined using the first and second eigenvalues.

[0103] The dual-channel convolutional neural network model includes a first channel and a second channel. The first channel is mainly used to detect abrupt changes in cylinder pressure, while the second channel is mainly used to detect long-term trends in cylinder pressure.

[0104] In one implementation, the first channel of a dual-channel convolutional neural network model is used to process the short-term fluctuation intensity and the number of abnormal pulses in the characteristic dataset of the engine under test, respectively, to obtain a first feature value for characterizing cylinder pressure mutations, including:

[0105] In the first channel, the short-term fluctuation intensity and the number of abnormal pulses in the feature dataset of the engine under test are processed by the first convolution kernel to obtain the corresponding short-term fluctuation intensity feature value sequence and the abnormal pulse number feature value sequence.

[0106] Pooling layers are used to process the short-term volatility intensity feature value sequence and the abnormal pulse quantity feature value sequence respectively to obtain the corresponding target short-term volatility intensity feature value and target abnormal pulse quantity feature value.

[0107] The first feature value is obtained by processing the short-term fluctuation intensity feature value and the abnormal pulse quantity feature value of the target using the first fully connected layer.

[0108] In one embodiment, the second channel of the dual-channel convolutional neural network model is used to process the long-term trend slope and peak-to-valley ratio in the feature dataset of the engine under test, respectively, to obtain a second feature value characterizing the cylinder pressure trend, including:

[0109] In the second channel, the long-term trend slope and peak-to-valley ratio in the feature dataset of the engine under test are processed by the second convolution kernel to obtain the corresponding long-term trend slope feature value sequence and peak-to-valley ratio feature value sequence.

[0110] Pooling layers are used to process the long-term trend slope feature value sequence and the peak-to-valley ratio feature value sequence respectively to obtain the corresponding target long-term trend slope feature value and target peak-to-valley ratio feature value.

[0111] The second fully connected layer is used to process the target long-term trend slope feature value and the target peak-to-valley ratio feature value to obtain the second feature value.

[0112] In this configuration, the width of the first convolutional kernel is smaller than the width of the second convolutional kernel. The width of the first convolutional kernel can be 3, and the width of the second convolutional kernel can be 5. In other words, the first channel uses three-loop window convolutional kernels to capture the characteristics of sudden abnormal fluctuations in cylinder pressure, while the second channel uses five-loop window convolutional kernels to observe the evolution trend of cylinder pressure data within five loops.

[0113] Taking the detection of the second channel as an example, the feature dataset of the engine to be tested contains 30 long-term trend slopes and 30 peak-to-valley ratios. Assume the long-term trend slope sequence is as follows:

[0114] [0.01,0.015,0.02,0.025,0.03,0.03,0.025,0.02,0.015,0.01,0.005,0,-0.005,-0.01,-0.01,-0.01,-0.015,-0.02,-0.025,-0.03,-0.03,-0.025,-0.02,-0.015,-0.01,-0.005,0,0.005,0.01,0.015].

[0115] Each time the second convolutional kernel slides once, it can calculate a weighted sum of 5 long-term trend slopes based on the weights corresponding to each long-term trend slope. When the second convolutional kernel slides from the beginning of the sequence to the end of the sequence, it can finally obtain a weighted sum of 26 (30-5+1) long-term trend slopes. The target long-term trend slope feature value is obtained by pooling and averaging the 26 weighted sums.

[0116] The target peak-to-valley ratio characteristic value can be obtained in the same way.

[0117] Then, a fully connected layer is used to process the target long-term trend slope feature value and the target peak-to-valley ratio feature value to obtain the second feature value, including:

[0118] The second eigenvalue η2 is determined according to the formula η2=w1′×λ1+w2′×λ2+b′;

[0119] w1′ is the weight corresponding to the target long-term trend slope feature value, w2′ is the weight corresponding to the target peak-to-valley ratio feature value, λ1 is the target long-term trend slope feature value, λ2 is the target peak-to-valley ratio feature value, and b′ is the first bias parameter; w1′, w2′, and b′ are all known values.

[0120] The first eigenvalue η1 can be determined using the same method described above.

[0121] After the first and second characteristic values ​​are determined, in one embodiment, the probability of engine knocking in a future time period is determined using the first and second characteristic values, including:

[0122] Obtain the first weight corresponding to the first feature value, the second weight corresponding to the second feature value, and the bias parameter;

[0123] The fusion feature value is determined using the first weight, the first feature value, the second weight, the second feature value, and the bias parameter;

[0124] According to the formula Determine the probability P of engine knocking in the future;

[0125] Where e is an exponential function and z is a fusion feature value.

[0126] In one implementation, determining the fused feature value using a first weight, a first eigenvalue, a second weight, a second eigenvalue, and a bias parameter includes:

[0127] The fusion feature value z is determined using the formula z = w1 × η1 + w2 × η2 + b; where,

[0128] w1 is the first weight, η1 is the first eigenvalue, w2 is the second weight, η2 is the second eigenvalue, and b is the bias parameter. w1, w2, and b are also known values.

[0129] In this way, the probability of engine knocking in the future can be determined through the above processing.

[0130] S112 determines the risk level of engine knocking based on the probability of engine knocking in the future, and adaptively adjusts the engine control parameters based on the risk level.

[0131] To prevent engine knocking in the future, this invention requires determining the risk level of knocking based on the probability of engine knocking in the future, and adaptively adjusting the engine control parameters based on the risk level, including:

[0132] If it is determined that the probability of engine knocking in the future is less than the first threshold, then the risk level is determined to be the safe level, and the current control parameters are maintained unchanged.

[0133] If it is determined that the probability of engine knocking in the future period is greater than or equal to the first threshold and less than the second threshold, then the risk level is determined to be the warning level, and the engine's ignition advance angle is controlled to be delayed by the first angle.

[0134] If it is determined that the probability of engine knocking in the future is greater than or equal to the second threshold and less than the third threshold, then the risk level is determined to be high risk level, and the engine ignition advance angle is delayed by the second angle and the engine fuel injection quantity is increased by the first proportion; the second angle is greater than the first angle.

[0135] If it is determined that the probability of engine knocking in the future period is greater than the third threshold, then the risk level is determined to be an emergency level, the ignition advance angle is delayed by the third angle, and the engine torque is reduced according to the second ratio; the third angle is greater than the second angle.

[0136] The first and second proportions can be determined according to the actual situation. For example, the first proportion can be 3% and the second proportion can be 10%.

[0137] Since the first ignition advance angle, the second ignition advance angle, and the third ignition advance angle need to be determined based on the engine speed and the mean effective braking pressure of the last combustion cycle in N combustion cycles, in one embodiment, before adaptively adjusting the engine control parameters based on the probability of engine knocking in the future, the method further includes:

[0138] Obtain the engine speed and engine braking mean effective pressure for the combustion cycle with the highest sequence number among N combustion cycles;

[0139] The ignition advance angle threshold is determined based on engine speed and average effective braking pressure, and the control parameters corresponding to different knock probabilities are determined based on the ignition advance angle threshold.

[0140] Specifically, the ignition advance angle threshold Threshold can be determined using the formula Threshold=0.5+0.1×(n / 6000)-0.05×(BEMP / 20), where n is the engine speed and BEMP is the engine's average effective braking pressure.

[0141] In the above formula, n ranges from 0 to 6000 rpm; the average effective braking pressure ranges from 0 to 20 bar; 0.5 is the basic threshold under normal operating conditions in the calibration experiment; 0.1 is the compensation coefficient for the first operating condition and 0.05 is the compensation coefficient for the second operating condition, which is used to balance the influence of speed and load on knock sensitivity.

[0142] The first angle is 0.5 + Threshold, the second angle is 3 × (0.5 + Threshold), and the third angle is 5 × (0.5 + Threshold). The units for the first, second, and third angles are °CA.

[0143] For example, when the engine speed n = 3000 rpm and the mean effective pressure BMEP = 10 bar, Threshold = 0.5 + 0.05 - 0.025 = 0.525.

[0144] The first, second, and third thresholds can be determined based on actual conditions. For example, the first threshold could be 0.3, the second threshold could be 0.6, and the third threshold could be 0.8. The adjustment strategy for the engine control parameters is shown in Table 2.

[0145] Table 2

[0146] Detonation probability Level determination Control parameter adjustment strategy P<0.3 Safety Maintain current control parameters and execute fuel injection quantity according to the calibrated fuel injection quantity Map. 0.3≤P<0.6 Warning Controlling the first angle of ignition advance 0.6≤P<0.8 High risk Control the ignition advance angle by the second angle and increase the fuel injection quantity by 3%. P≥0.8 urgent Control the ignition advance angle by the third angle and reduce torque by 10%.

[0147] In Table 2, the characteristics of a safety level are generally SFI < 1.5 and APC = 0. The characteristics of a warning level are generally TTS > 0.02 and sustained for 5 combustion cycles. The characteristics of a high-risk level are generally APC ≥ 3 and PVR > 0.4. The characteristics of an emergency level are TTS > 0.1 or a predicted detonation probability P > 0.7 for 3 consecutive times.

[0148] In practical applications, the prediction method of this invention for a certain 1.5T engine running at full load at 4500rpm is as follows:

[0149] First, the operating characteristics of the engine are extracted. When the above operating characteristics are predicted using a dual-channel convolutional neural network model, the first weight w1 is 0.7 and the second weight w2 is 0.9. The final probability of knocking is 0.78.

[0150] Triggering the "high-risk" level involves delaying the ignition advance angle by a second angle, increasing the fuel injection quantity by 3% to reduce combustion temperature. For example, assuming the current ignition advance angle is 30°CA and the second angle is 3°CA, delaying the ignition advance angle by 3°CA results in ignition at 27°CA, ensuring the piston is further away from top dead center and thus reducing the likelihood of knocking.

[0151] After adjusting the control parameters, the P value dropped to 0.45 in the subsequent 3 combustion cycles, switching to the "warning" level, and the ignition advance angle was changed from the second delay angle to the first delay angle.

[0152] This approach combines short-term engine mutations with long-term trends to accurately predict engine knock risk, thereby allowing for early intervention in engine control parameters, reducing the probability of knocking, and improving engine performance.

[0153] Based on the same inventive concept as the foregoing embodiments, this embodiment also provides an engine combustion prediction and control device based on cylinder pressure signals, such as... Figure 2 As shown, the device includes:

[0154] The determination unit 21 is used to collect the cylinder pressure of the vehicle engine, use the cylinder pressure to determine the engine operating characteristics corresponding to N combustion cycles, and determine the engine feature dataset to be tested based on the engine operating characteristics corresponding to the N combustion cycles; the engine operating characteristics include: the short-term fluctuation intensity of the cylinder pressure, the long-term trend slope of the cylinder pressure, the peak-to-valley ratio of the cylinder pressure, and the number of abnormal pulses of the cylinder pressure.

[0155] The prediction unit 22 is used to predict the probability of the engine detonating in the future by using a pre-trained dual-channel convolutional neural network model to predict the feature dataset of the engine under test.

[0156] The adjustment unit 23 is used to determine the risk level of knocking based on the probability of knocking in the engine in the future period, and to adaptively adjust the control parameters of the engine according to the risk level.

[0157] In one implementation, the prediction unit 22 is specifically used for:

[0158] The first channel of the dual-channel convolutional neural network model is used to process the short-term fluctuation intensity and the number of abnormal pulses in the feature dataset of the engine under test, respectively, to obtain the first feature value used to characterize cylinder pressure mutation.

[0159] The second channel of the dual-channel convolutional neural network model is used to process the long-term trend slope and the number of abnormal pulses in the feature dataset of the engine under test, respectively, to obtain a second feature value that characterizes the cylinder pressure trend.

[0160] The probability of engine knocking in the future is determined using the first feature value and the second feature value.

[0161] Prediction unit 22 is specifically used for:

[0162] In the first channel, the short-term fluctuation intensity and the number of abnormal pulses in the feature dataset of the engine under test are processed by the first convolution kernel respectively to obtain the corresponding short-term fluctuation intensity feature value sequence and the abnormal pulse number feature value sequence.

[0163] The short-term fluctuation intensity feature value sequence and the abnormal pulse number feature value sequence are respectively pooled using a pooling layer to obtain the corresponding target short-term fluctuation intensity feature value and target abnormal pulse number feature value.

[0164] The first feature value is obtained by processing the target short-term fluctuation intensity feature value and the target abnormal pulse quantity feature value using the first fully connected layer.

[0165] Prediction unit 22 is specifically used for:

[0166] In the second channel, the long-term trend slope and peak-to-valley ratio in the feature dataset of the engine under test are processed by the second convolution kernel to obtain the corresponding long-term trend slope feature value sequence and peak-to-valley ratio feature value sequence.

[0167] The long-term trend slope feature value sequence and the peak-to-valley ratio feature value sequence are respectively pooled using a pooling layer to obtain the corresponding target long-term trend slope feature value and target peak-to-valley ratio feature value.

[0168] The second feature value is obtained by processing the target long-term trend slope feature value and the target peak-to-valley ratio feature value using the second fully connected layer.

[0169] Prediction unit 22 is specifically used for:

[0170] Obtain the first weight corresponding to the first feature value, the second weight corresponding to the second feature value, and the bias parameter;

[0171] The fused feature value is determined using the first weight, the first feature value, the second weight, the second feature value, and the bias parameter;

[0172] According to the formula Determine the probability P of the engine detonating in the future;

[0173] Wherein, e is an exponential function, and z is the fusion feature value.

[0174] Prediction unit 22 is specifically used for:

[0175] The fusion feature value z is determined using the formula z = w1 × η1 + w2 × η2 + b; where,

[0176] w1 is the first weight, η1 is the first feature value, w2 is the second weight, η2 is the second feature value, and b is the bias parameter.

[0177] The determining unit 21 is also used for:

[0178] Obtain the engine speed of the combustion cycle with the highest sequence number among the N combustion cycles and the engine's average effective braking pressure;

[0179] The ignition advance angle threshold is determined based on the engine speed and the average effective braking pressure, and the control parameters corresponding to different knock probabilities are determined based on the ignition advance angle threshold.

[0180] Adjustment unit 23 is used for:

[0181] If it is determined that the probability of the engine knocking in the future period is less than the first threshold, then the risk level is determined to be a safe level, and the current control parameters are maintained unchanged.

[0182] If it is determined that the probability of the engine knocking in the future period is greater than or equal to the first threshold and less than the second threshold, then the risk level is determined to be the warning level, and the ignition advance angle of the engine is controlled to be delayed by the first angle.

[0183] If it is determined that the probability of the engine knocking in the future period is greater than or equal to the second threshold and less than the third threshold, then the risk level is determined to be a high risk level, and the ignition advance angle of the engine is delayed by a second angle and the fuel injection quantity of the engine is increased by a first proportion; the second angle is greater than the first angle.

[0184] If it is determined that the probability of engine knocking in the future period is greater than the third threshold, then the risk level is determined to be an emergency level, the ignition advance angle is delayed by the third angle, and the engine torque is reduced according to the second ratio; the third angle is greater than the second angle.

[0185] Since the apparatus described in the embodiments of this invention is used to implement the engine combustion predictive control method based on cylinder pressure signals according to the embodiments of this invention, those skilled in the art can understand the specific structure and variations of the apparatus based on the method described in the embodiments of this invention, and therefore will not be described in detail here. All apparatuses used in the methods of the embodiments of this invention fall within the scope of protection of this invention.

[0186] Based on the same inventive concept as the foregoing embodiments, the present invention also provides a vehicle, the vehicle including the aforementioned engine combustion prediction control device based on cylinder pressure signals.

[0187] Through one or more embodiments of the present invention, the present invention has the following beneficial effects or advantages:

[0188] This invention provides an engine combustion prediction control method, device, and vehicle based on cylinder pressure signals, comprising: acquiring cylinder pressure of a vehicle engine; using the cylinder pressure to determine engine operating characteristics corresponding to N combustion cycles; and determining a test engine feature dataset based on the engine operating characteristics corresponding to the N combustion cycles. The engine operating characteristics include: the short-term fluctuation intensity of the cylinder pressure, the long-term trend slope of the cylinder pressure, the peak-to-valley ratio of the cylinder pressure, and the number of abnormal pulses in the cylinder pressure. A pre-trained dual-channel convolutional neural network model is used to predict the test engine feature dataset to obtain the probability of engine knocking in the future. The probability of engine knocking in the future determines the risk level of knocking, and the engine control parameters are adaptively adjusted according to the risk level. Since the short-term fluctuation intensity and abnormal pulse number of cylinder pressure reflect the abrupt changes in the engine's combustion process, and the long-term trend slope and peak-to-valley ratio of cylinder pressure reflect the development direction of combustion stability, the pre-trained dual-channel convolutional neural network model can accurately predict the engine knocking risk by combining short-term abrupt changes and long-term trends when predicting these features. This allows for early intervention in engine control parameters, reducing the probability of knocking and improving engine performance.

[0189] Furthermore, by using data from N combustion cycles, the coupling effect between short-term fluctuations and long-term trends can be identified, enabling a more accurate prediction of the probability of knocking. Based on the predicted knocking probability, a graded ignition delay control and fuel-coordinated control algorithm is used to perform graded adaptive control of knocking. Compared with the traditional torque reduction feedback mechanism after knocking, this ensures uninterrupted engine power output. By predicting knocking, the frequency of knocking is reduced, wear on key moving parts of the engine is decreased, and engine reliability and durability are improved.

[0190] 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.

[0191] The above description is merely 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 within the scope of protection of the present invention.

Claims

1. An engine combustion predictive control method based on cylinder pressure signals, characterized in that, The method includes: The cylinder pressure of a vehicle engine is collected, and the engine operating characteristics corresponding to N combustion cycles are determined using the cylinder pressure. Based on the engine operating characteristics corresponding to the N combustion cycles, a characteristic dataset of the engine to be tested is determined. The engine operating characteristics include: the short-term fluctuation intensity of the cylinder pressure, the long-term trend slope of the cylinder pressure, the peak-to-valley ratio of the cylinder pressure, and the number of abnormal pulses in the cylinder pressure. The probability of the engine detonating in the future is obtained by using a pre-trained dual-channel convolutional neural network model to predict the feature dataset of the engine under test. The risk level of knocking is determined based on the probability of the engine knocking in the future, and the control parameters of the engine are adaptively adjusted according to the risk level.

2. The method as described in claim 1, characterized in that, The method of using a pre-trained dual-channel convolutional neural network model to predict the probability of engine detonation in the future includes: The first channel of the dual-channel convolutional neural network model is used to process the short-term fluctuation intensity and the number of abnormal pulses in the feature dataset of the engine under test, respectively, to obtain the first feature value used to characterize cylinder pressure mutation. The second channel of the dual-channel convolutional neural network model is used to process the long-term trend slope and the number of abnormal pulses in the feature dataset of the engine under test, respectively, to obtain a second feature value that characterizes the cylinder pressure trend. The probability of engine knocking in the future is determined using the first feature value and the second feature value.

3. The method as described in claim 2, characterized in that, The first channel of the dual-channel convolutional neural network model is used to process the short-term fluctuation intensity and the number of abnormal pulses in the feature dataset of the engine under test, respectively, to obtain a first feature value for characterizing cylinder pressure mutations, including: In the first channel, the short-term fluctuation intensity and the number of abnormal pulses in the feature dataset of the engine under test are processed by the first convolution kernel respectively to obtain the corresponding short-term fluctuation intensity feature value sequence and the abnormal pulse number feature value sequence. The short-term fluctuation intensity feature value sequence and the abnormal pulse number feature value sequence are respectively pooled using a pooling layer to obtain the corresponding target short-term fluctuation intensity feature value and target abnormal pulse number feature value. The first feature value is obtained by processing the target short-term fluctuation intensity feature value and the target abnormal pulse quantity feature value using the first fully connected layer.

4. The method as described in claim 2, characterized in that, The second channel of the dual-channel convolutional neural network model is used to process the long-term trend slope and peak-to-valley ratio in the feature dataset of the engine under test, respectively, to obtain a second feature value characterizing the cylinder pressure trend, including: In the second channel, the long-term trend slope and peak-to-valley ratio in the feature dataset of the engine under test are processed by the second convolution kernel to obtain the corresponding long-term trend slope feature value sequence and peak-to-valley ratio feature value sequence. The long-term trend slope feature value sequence and the peak-to-valley ratio feature value sequence are respectively pooled using a pooling layer to obtain the corresponding target long-term trend slope feature value and target peak-to-valley ratio feature value. The second feature value is obtained by processing the target long-term trend slope feature value and the target peak-to-valley ratio feature value using the second fully connected layer.

5. The method as described in claim 2, characterized in that, The step of determining the probability of engine knocking in the future using the first feature value and the second feature value includes: Obtain the first weight corresponding to the first feature value, the second weight corresponding to the second feature value, and the bias parameter; The fused feature value is determined using the first weight, the first feature value, the second weight, the second feature value, and the bias parameter; According to the formula Determine the probability P of the engine detonating in the future; Wherein, e is an exponential function, and z is the fusion feature value.

6. The method as described in claim 5, characterized in that, The step of determining the fused feature value using the first weight, the first feature value, the second weight, the second feature value, and the bias parameter includes: The fusion feature value z is determined using the formula z = w1 × η1 + w2 × η2 + b; where, w1 is the first weight, η1 is the first feature value, w2 is the second weight, η2 is the second feature value, and b is the bias parameter.

7. The method as described in claim 1, characterized in that, Before adaptively adjusting the engine control parameters based on the probability of engine knocking in the future, the method further includes: Obtain the engine speed of the combustion cycle with the highest sequence number among the N combustion cycles and the engine's average effective braking pressure; The ignition advance angle threshold is determined based on the engine speed and the average effective braking pressure, and the control parameters corresponding to different knock probabilities are determined based on the ignition advance angle threshold.

8. The method as described in claim 1, characterized in that, The process of determining the risk level of engine knocking based on the probability of knocking occurring in the future, and adaptively adjusting the engine control parameters based on the risk level, includes: If it is determined that the probability of the engine knocking in the future period is less than the first threshold, then the risk level is determined to be a safe level, and the current control parameters are maintained unchanged. If it is determined that the probability of the engine knocking in the future period is greater than or equal to the first threshold and less than the second threshold, then the risk level is determined to be the warning level, and the ignition advance angle of the engine is controlled to be delayed by the first angle. If it is determined that the probability of the engine knocking in the future period is greater than or equal to the second threshold and less than the third threshold, then the risk level is determined to be a high risk level, and the ignition advance angle of the engine is delayed by a second angle and the fuel injection quantity of the engine is increased by a first proportion; the second angle is greater than the first angle. If it is determined that the probability of engine knocking in the future period is greater than the third threshold, then the risk level is determined to be an emergency level, the ignition advance angle is delayed by the third angle, and the engine torque is reduced according to the second ratio; the third angle is greater than the second angle.

9. An engine combustion prediction control device based on cylinder pressure signals, characterized in that, The device includes: The determination unit is used to collect the cylinder pressure of the vehicle engine, use the cylinder pressure to determine the engine operating characteristics corresponding to N combustion cycles, and determine the engine feature dataset to be tested based on the engine operating characteristics corresponding to the N combustion cycles; the engine operating characteristics include: the short-term fluctuation intensity of the cylinder pressure, the long-term trend slope of the cylinder pressure, the peak-to-valley ratio of the cylinder pressure, and the number of abnormal pulses of the cylinder pressure. The prediction unit is used to predict the probability of the engine detonating in the future using a pre-trained dual-channel convolutional neural network model on the feature dataset of the engine under test. The adjustment unit is used to determine the risk level of detonation based on the probability of detonation occurring in the engine in the future period, and to adaptively adjust the control parameters of the engine according to the risk level.

10. A vehicle, characterized in that, The vehicle includes the engine combustion prediction control device based on cylinder pressure signal as described in claim 9.

Images