Combustion prediction method and device for vehicle engine, medium and product

Abstract

The invention discloses a combustion prediction method and device of a vehicle engine, a medium and a product, and the method comprises the steps: obtaining a target data set of a target vehicle, the target data set comprises a first target data set for fuel injection quantity prediction, a second target data set for gas quantity prediction and a third target data set for ignition prediction; based on the first target data set and a preset fuel injection quantity prediction model, a fuel injection quantity prediction result corresponding to an engine of the target vehicle is determined; based on the second target data set and a preset gas flow prediction model, an exhaust gas recirculation rate prediction result corresponding to the engine is determined; based on the third target data set and a preset ignition prediction model, a knock sensor vibration signal prediction result corresponding to the engine is determined; and based on the fuel injection quantity prediction result, the exhaust gas recirculation rate prediction result and the knock sensor vibration signal prediction result, a combustion control instruction of the target vehicle is determined, so that engine combustion is accurately controlled.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicles, in particular to a combustion prediction method and device for a vehicle engine, a medium and a product. BACKGROUND

[0002] In the research and optimization of engine combustion process, accurate prediction of combustion state is crucial to improve efficiency and reduce emissions. In related technologies, there are various techniques and methods for engine combustion prediction, including computational fluid dynamics (CFD) simulation technology, real-time monitoring of engine operating parameters, etc.

[0003] Computational fluid dynamics simulation technology is a simulation method based on computational fluid dynamics, which can simulate complex fluid dynamics, heat conduction and chemical reaction processes. Through CDF technology, the fine details of fuel injection, ignition and flame propagation in the engine combustion chamber can be revealed, but there is a problem of poor real-time performance, which is difficult to adapt to dynamic conditions.

[0004] Real-time monitoring of engine operating parameters, such as monitoring of in-cylinder pressure, combustion noise, engine body vibration, etc., can evaluate the combustion state of the engine. For example, in-cylinder pressure detection can identify whether the engine has knock phenomenon, and combustion noise analysis can detect whether the engine has knock. However, the monitoring of engine operating parameters has certain limitations, such as poor durability of pressure detection and insufficient accuracy and response sensitivity of combustion noise.

[0005] In summary, engine combustion prediction is a complex and important process, and the existing technical solutions cannot meet the dynamic management needs of hybrid vehicle models for combustion efficiency and energy consumption. SUMMARY

[0006] The present application provides a combustion prediction method, device, medium and product for a vehicle engine.

[0007] In a first aspect, the present application provides a combustion prediction method for a vehicle engine, comprising: obtaining a target data set of a target vehicle, the target data set comprising a first target data set for predicting fuel injection, a second target data set for predicting gas volume, and a third target data set for predicting ignition; determining a fuel injection prediction result corresponding to the engine of the target vehicle based on the first target data set and a preset fuel injection prediction model; determining a waste gas recirculation rate prediction result corresponding to the engine based on the second target data set and a preset gas volume prediction model; Based on the third target dataset and the preset ignition prediction model, the predicted vibration signal of the knock sensor corresponding to the engine is determined. Based on the predicted fuel injection quantity, the predicted exhaust gas recirculation rate, and the predicted vibration signal from the knock sensor, the combustion control command for the target vehicle is determined so that the target vehicle executes the combustion control command. In some implementations, the preset fuel injection quantity prediction model is constructed through the following steps: Obtain the first sample dataset for fuel injection quantity prediction; The initial fuel injection quantity prediction model is trained based on the first sample dataset to obtain the trained fuel injection quantity prediction model, which is used as the preset fuel injection quantity prediction model.

[0008] In some implementations, obtaining the first sample dataset for fuel injection quantity prediction includes: For each cylinder of the engine, a first raw dataset corresponding to that cylinder is determined. The first raw dataset includes data related to the fuel injection quantity under different engine operating conditions, the cylinder knocking angle and cylinder knocking intensity of the target vehicle. For each cylinder of the engine, principal component analysis is performed on the first original dataset of the cylinder, and based on the principal component analysis results, the data corresponding to the first Q principal components whose cumulative contribution value is greater than the first threshold are taken as the first dataset corresponding to the cylinder, where Q is a positive integer; The first sample dataset is constructed based on the first dataset corresponding to each cylinder.

[0009] In some implementations, the preset gas volume prediction model is constructed through the following steps: For each cylinder of the engine, a second raw dataset corresponding to that cylinder is determined. The second raw dataset includes data related to air volume prediction under different engine operating conditions. For each cylinder of the engine, principal component analysis is performed on the second original dataset of the cylinder, and based on the principal component analysis results, the data corresponding to the top M principal components whose cumulative contribution value is greater than the second threshold are taken as the second dataset corresponding to the cylinder, where M is a positive integer; Based on the second dataset corresponding to each cylinder, a second sample dataset is constructed for gas volume prediction. Based on the second sample dataset, the initial gas volume prediction model is trained to obtain a trained gas volume prediction model, which is used as the preset gas volume prediction model.

[0010] In some implementations, the preset ignition prediction model is constructed through the following steps: A third sample dataset for ignition prediction is obtained, the third sample dataset including historical detonation sensor vibration signals; The initial ignition prediction model is trained based on the third sample dataset to obtain a trained ignition prediction model, which is then used as the preset ignition prediction model.

[0011] In some implementations, obtaining the third sample dataset for ignition prediction includes: Obtain a third raw dataset, which includes data related to ignition prediction under different engine operating conditions; Principal component analysis is performed on the third original dataset, and based on the results of the principal component analysis, the data corresponding to the top P principal components whose cumulative contribution value is greater than the third threshold are determined, where P is a positive integer; The historical knock sensor vibration signal of the engine is acquired, and the third sample dataset is constructed based on the data corresponding to the first P principal components and the historical knock sensor vibration signal.

[0012] In some embodiments, the preset fuel injection quantity prediction model, the preset air volume prediction model, and the preset ignition prediction model are deployed both locally on the target vehicle and in the cloud corresponding to the target vehicle. The method further includes: When the target vehicle establishes a connection with the cloud, the target dataset is sent to the cloud through the target vehicle, so as to perform engine combustion prediction through the preset fuel injection quantity prediction model, the preset air volume prediction model and the preset ignition prediction model deployed in the cloud; When the target vehicle disconnects from the cloud, combustion prediction of the engine is performed using the preset fuel injection quantity prediction model, the preset air volume prediction model, and the preset ignition prediction model deployed on the target vehicle itself.

[0013] In some embodiments, the method further includes: When an update is detected in the preset fuel injection quantity prediction model, the preset air volume prediction model, and / or the preset ignition prediction model deployed in the cloud, the updated target model is obtained through the target vehicle to update the local model in the target vehicle corresponding to the target model.

[0014] In some embodiments, the method further includes: Based on a preset update cycle, the fuel injection quantity prediction model, the preset air volume prediction model, and / or the preset ignition prediction model deployed in the cloud are updated.

[0015] In some implementations, both the preset fuel injection quantity prediction model and the preset air volume prediction model are gated recurrent neural network models.

[0016] In some implementations, the ignition prediction model is a wavelet neural network model.

[0017] In a second aspect, embodiments of the present invention provide a combustion prediction device for a vehicle engine, comprising: The acquisition module is used to acquire the target dataset of the target vehicle. The target dataset includes a first target dataset for fuel injection quantity prediction, a second target dataset for gas volume prediction, and a third target dataset for ignition prediction. The first processing module is used to determine the fuel injection quantity prediction result corresponding to the engine of the target vehicle based on the first target dataset and the preset fuel injection quantity prediction model. The second processing module is used to determine the predicted exhaust gas recirculation rate corresponding to the engine based on the second target dataset and the preset gas volume prediction model. The third processing module is used to determine the predicted vibration signal of the knock sensor corresponding to the engine based on the third target dataset and the preset ignition prediction model. The execution module is used to determine the combustion control command of the target vehicle based on the fuel injection quantity prediction result, the exhaust gas recirculation rate prediction result, and the knock sensor vibration signal prediction result, so as to cause the target vehicle to execute the combustion control command.

[0018] Thirdly, embodiments of the present invention provide a combustion prediction device for a vehicle engine, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the above-described combustion prediction method for a vehicle engine.

[0019] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-described combustion prediction method for a vehicle engine.

[0020] Fifthly, embodiments of the present invention provide a computer program product, the computer program product including a computer program, which, when executed by a processor, is used to load and execute the steps in the above-described combustion prediction method for a vehicle engine.

[0021] The above-described one or more technical solutions in the embodiments of this application have at least the following technical effects: The combustion prediction method for vehicle engines provided in this specification involves acquiring a target dataset for the target vehicle. This target dataset includes a first target dataset for fuel injection quantity prediction, a second target dataset for air volume prediction, and a third target dataset for ignition prediction. Based on the first target dataset and a preset fuel injection quantity prediction model, the fuel injection quantity prediction result corresponding to the engine of the target vehicle is determined. Based on the second target dataset and the preset air volume prediction model, the exhaust gas recirculation prediction result corresponding to the engine is determined. Based on the third target dataset and the preset ignition prediction model, the knock sensor signal prediction result corresponding to the engine is determined. Based on the fuel injection quantity prediction result, exhaust gas recirculation prediction result, and knock sensor vibration signal prediction result, the combustion control command for the target vehicle is determined, enabling the target vehicle to execute the combustion control command. This solution, by establishing preset fuel injection quantity prediction models, preset air volume prediction models, and preset ignition prediction models, systematically considers the control strategies of important modules related to engine combustion. It can accurately predict engine combustion, control and adjust engine combustion-related parameters in real time, effectively improve engine reliability and thermal efficiency, and reduce energy consumption. Attached Figure Description

[0022] Figure 1 A schematic flowchart illustrating a combustion prediction method for a vehicle engine provided in an embodiment of this specification; Figure 2 A schematic diagram of a combustion prediction system for a vehicle engine provided as an embodiment of this specification; Figure 3 A schematic diagram of a network structure for predicting engine air-fuel ratio based on a GRU neural network, provided for an embodiment of this specification; Figure 4 This is a schematic diagram of the structure of a WNN neural network provided in an embodiment of this specification; Figure 5 A schematic diagram of the functional modules of a combustion prediction device for a vehicle engine provided in an embodiment of this specification; Figure 6 This is a schematic diagram of the structure of a combustion prediction device for a vehicle engine provided in an embodiment of this specification. Detailed Implementation

[0023] The overall technical solution of this specification embodiment is as follows: A target dataset for the target vehicle is obtained, comprising a first target dataset for fuel injection quantity prediction, a second target dataset for air volume prediction, and a third target dataset for ignition prediction; based on the first target dataset and a preset fuel injection quantity prediction model, a fuel injection quantity prediction result corresponding to the engine of the target vehicle is determined; based on the second target dataset and the preset air volume prediction model, an exhaust gas recirculation rate prediction result corresponding to the engine is determined; based on the third target dataset and the preset ignition prediction model, a knock sensor vibration signal prediction result corresponding to the engine is determined; based on the fuel injection quantity prediction result, the exhaust gas recirculation rate prediction result, and the knock sensor vibration signal prediction result, a combustion control command for the target vehicle is determined, so that the target vehicle executes the combustion control command.

[0024] The solutions in the embodiments of this specification, by building preset fuel injection quantity prediction models, preset air volume prediction models, and preset ignition prediction models, systematically take into account the control strategies of important modules related to engine combustion. This enables accurate prediction of engine combustion, real-time control and adjustment of engine combustion-related parameters, effectively improving engine reliability and thermal efficiency, while reducing energy consumption.

[0025] To better understand the above technical solutions, the technical solutions of the embodiments of this specification will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of this specification and the specific features in the embodiments are detailed descriptions of the technical solutions of the embodiments of this specification, rather than limitations on the technical solutions of this specification. Unless otherwise specified, the embodiments of this specification and the technical features in the embodiments can be combined with each other.

[0026] First, it should be clarified that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0027] like Figure 1 The diagram shown is a flowchart of a combustion prediction method for a vehicle engine provided in an embodiment of this specification. The method includes the following steps: Step S101: Obtain the target dataset of the target vehicle, wherein the target dataset includes a first target dataset for fuel injection quantity prediction, a second target dataset for gas volume prediction, and a third target dataset for ignition prediction. Step S102: Based on the first target dataset and the preset fuel injection quantity prediction model, determine the fuel injection quantity prediction result corresponding to the engine of the target vehicle; Step S103: Based on the second target dataset and the preset gas volume prediction model, determine the predicted exhaust gas recirculation rate corresponding to the engine; Step S104: Based on the third target dataset and the preset ignition prediction model, determine the predicted result of the knock sensor vibration signal corresponding to the engine; Step S105: Based on the fuel injection quantity prediction result, the exhaust gas recirculation rate prediction result, and the knock sensor vibration signal prediction result, determine the combustion control command for the target vehicle so that the target vehicle executes the combustion control command.

[0028] The methods described in this specification can be applied to vehicles, to the cloud that communicates with vehicles, or to a system consisting of vehicles and the cloud; no limitation is made here.

[0029] like Figure 2 The diagram shown is a schematic of a combustion prediction system for a vehicle engine provided in an embodiment of this specification, including a data extraction module, a combustion prediction module, and a combustion execution module. Each module will be described in detail below.

[0030] The data extraction module can collect various data from the engine, battery, motor and vehicle in real time. Based on this data, the engine combustion can be pre-judged. For example, an initial engine combustion state can be generated as a feedforward. In some embodiments, the engine combustion can be pre-judged in a traditional way, such as by using engine parameters.

[0031] The combustion prediction module may include a fuel injection quantity prediction module, an air volume prediction module, and an ignition prediction module. For each module, a machine learning prediction module can be established based on the characteristics of its respective prediction data to accurately predict the control target. In the embodiments of this specification, considering the nonlinearity and temporal nature of fuel injection control and air volume control, principal component analysis (PCA) and gated recurrent unit (GRU) neural network modeling methods can be used to determine the prediction results of fuel injection quantity and EGR (Exhaust Gas Recirculation) rate, respectively. Considering the significant vibration and noise caused by engine ignition angle control during knocking, principal component analysis and wavelet neural network (WNN) modeling methods are selected to accurately predict future changes in the knock sensor vibration signal.

[0032] like Figure 2 As shown, the fuel injection quantity prediction module can predict the air-fuel ratio, thereby predicting the fuel injection quantity. In the air-fuel ratio prediction module, since there is a correlation between the EGR rate and the in-cylinder EGR flow rate, throttle flow rate, intercooler pipe air volume, and mixture flow rate, the corresponding prediction model can capture the correlation between the input data and each flow rate, as well as the correlation between each flow rate and the EGR rate. Based on the in-cylinder EGR flow rate prediction, throttle flow rate prediction, intercooler pipe air volume prediction, and mixture flow rate prediction, the final EGR rate prediction result is output. The ignition prediction module can predict the knock sensor vibration signal, including transient combustion knock prediction and knock retraction angle prediction.

[0033] The combustion execution module can be divided into two execution schemes: cloud-based and vehicle-based. The cloud stores an online combustion prediction model, which is used for predicting fuel injection quantity, air volume, and ignition. The cloud model can iteratively optimize its prediction performance in real time. In some embodiments, the model can fully consider the impact of vehicle aging and degradation. For example, during model training, relevant data from aging vehicles is collected for model training to enable the model to accurately predict the combustion state of aging vehicles. When the vehicle and the cloud can communicate well, the cloud-based combustion prediction model predicts the vehicle's combustion results online, generates corresponding control commands based on the combustion prediction results, and sends the control commands to the vehicle in real time.

[0034] The vehicle-side system can store offline combustion prediction models for predicting fuel injection quantity, air volume, and ignition. In some embodiments, the vehicle-side model can be identical to the cloud-based model. In other embodiments, considering the limitations of vehicle-side memory and computing power, the online combustion prediction model can be simplified; this is not a limitation here. When the vehicle communicates with the cloud terminal, the combustion prediction results can be output based on the vehicle-side combustion prediction model, and corresponding control commands can be generated to achieve engine combustion control.

[0035] like Figure 2As shown, the engine combustion prediction system interacts in real time with the vehicle's intelligent energy management module. In some embodiments, the engine combustion prediction system sends the generated control commands to the engine-related actuators to achieve actuator coordination. The actuators include, but are not limited to, the turbocharger, throttle, mixing valve, EGR valve, and ignition injection mechanism, etc., and feeds the control results back to the energy management controller so that the energy management controller can perform vehicle energy management, such as controlling the engine and / or motor speed, SOC (State of Charge), engine and / or motor start-stop, engine and / or motor torque according to the corresponding scenario algorithm, and feeding the control results back to the combustion prediction control system. This realizes bidirectional interaction between engine combustion control and vehicle energy management control, thereby effectively improving the working efficiency of the engine and motor and improving the overall vehicle energy consumption performance.

[0036] In step S101, the target dataset for the target vehicle is obtained. In this embodiment of the specification, combustion prediction includes fuel injection quantity prediction, air volume prediction, and ignition prediction. The data used for prediction of different types of parameters are also different. Therefore, the target dataset includes a first target dataset for fuel injection quantity prediction, a second target dataset for air volume prediction, and a third target dataset for ignition prediction. The first target dataset, the second target dataset, and the third target dataset may contain the same data or may be completely different.

[0037] In some embodiments, the first target dataset, the second target dataset, and the third target dataset can be selected data that are highly sensitive to the fuel injection quantity prediction results, gas volume prediction results, and ignition prediction results. The specific data types contained in each dataset can be determined during model training, and each dataset will be specifically described in subsequent model training processes.

[0038] In some embodiments, the target dataset can be a preprocessed dataset, for example, by first obtaining the raw data of the target vehicle for combustion prediction, and then performing outlier removal, noise processing, etc. on the raw data.

[0039] In some embodiments, the preprocessing of the collected raw data may include missing value handling. Specifically, the raw data may contain missing values, possibly due to acquisition failures (such as sensor disconnection). When missing values ​​exist, the missing rate of the data can be determined first. If the missing rate is greater than or equal to a preset missing rate, the data is not used and is re-acquired. If the missing rate is less than the preset missing rate, it can be filled using methods such as mean, median, or interpolation. The preprocessing of the collected raw data may also include outlier handling. Outliers are values ​​that deviate significantly from the overall data distribution. When handling outliers, they can be replaced with a threshold. For example, if an outlier is larger than the maximum value that can be collected, it is replaced with the maximum value of the data; if an outlier is smaller than the minimum value that can be collected, it is replaced with the minimum value of the data. Of course, the data preprocessing process may also include other types, which will not be listed here.

[0040] In steps S102 to S104, after obtaining the first target dataset, the second target dataset, and the third target dataset, the combustion state of the engine is predicted in multiple dimensions based on the trained preset fuel injection quantity prediction model, preset gas volume prediction model, and preset ignition prediction model, so as to obtain the fuel injection quantity prediction result, the exhaust gas recirculation rate prediction result, and the knock sensor vibration signal prediction result.

[0041] To better understand the combustion prediction process in the embodiments of this specification, the preset fuel injection quantity prediction model, the preset gas volume prediction model, and the preset ignition prediction model, as well as their corresponding model input data, will be described in detail below.

[0042] I. Prediction of fuel injection quantity The preset fuel injection quantity prediction model is constructed through the following steps: obtaining a first sample dataset for fuel injection quantity prediction; training the initial fuel injection quantity prediction model based on the first sample dataset to obtain a trained fuel injection quantity prediction model, which serves as the preset fuel injection quantity prediction model.

[0043] In the embodiments described in this specification, the fuel injection control of the engine is considered to be a dynamic process, influenced by engine operating conditions and various data, exhibiting complex dynamic characteristics over time. To improve the accuracy of fuel injection quantity prediction, when determining the first sample dataset, data that can affect the fuel injection quantity prediction results can be selected based on experience, or data can be selected using other methods. For example, based on the sensitivity analysis of various data on fuel injection control, data with a sensitivity greater than a preset sensitivity can be selected to form the first sample dataset.

[0044] In some embodiments, the first sample dataset can be determined by the following steps: for each cylinder of the engine, determine the first original dataset corresponding to that cylinder, the first original dataset including data related to fuel injection quantity under different engine operating conditions, cylinder knocking angle and cylinder knocking intensity of the target vehicle; for each cylinder of the engine, perform principal component analysis on the first original dataset of that cylinder, and based on the principal component analysis results, take the data corresponding to the top N principal components whose cumulative contribution value is greater than a first threshold as the first dataset corresponding to that cylinder, where N is a positive integer; construct the first sample dataset based on the first dataset corresponding to each cylinder.

[0045] Specifically, considering the numerous factors influencing engine fuel injection control, directly using all data that could affect it as input to the fuel injection quantity prediction model would result in an enormous amount of data to fit, posing significant challenges to the model training process. Therefore, PCA (Programmatical Conversion Aspect) can be used to map high-dimensional data to a low-dimensional space, improving the model's data processing speed while preserving the original data features. PCA is an unsupervised learning method that uses orthogonal transformations to convert observed data represented by linearly correlated variables into a few linearly independent variables, thus representing most of the feature information of the original data.

[0046] In the embodiments of this specification, when predicting the fuel injection quantity, cylinder prediction and control are performed on the engine. For each cylinder of the engine, a first original dataset corresponding to that cylinder is determined. To consider fuel injection control under different operating conditions, the first original dataset includes data related to the fuel injection quantity under different engine operating conditions. Engine operating conditions can be characterized by parameters such as engine speed and torque; that is, different operating conditions correspond to different speeds and torques. In some embodiments, the first original dataset may include the following feature data: engine speed, torque, coolant temperature, throttle position, intake airflow, wheel-end power demand, hybrid mode (power split, direct drive, etc.), engine coolant temperature, intake manifold temperature, and historical engine fuel injection quantity. It should be noted that because direct injection gasoline engines use a high compression ratio, the in-cylinder heat load and pressure increase significantly, easily causing knocking, which greatly restricts the improvement of engine performance. Therefore, in the embodiments of this specification, cylinder knocking angle and cylinder knocking intensity are added to the first original data to optimize the final fuel injection prediction model.

[0047] The specific process for performing principal component analysis on the first set of raw data can be found in the following steps: (1) The first original data contains m types of feature data, and the sequence length of each type of feature data is n, where m and n are both positive integers. For each type of feature data, the following formula is used to standardize each type of feature data to eliminate the influence of units and orders of magnitude on the correlation of the data:

[0048] in, x For each feature data, the original data sequence, x min The minimum value in the original data sequence. x max The maximum value in the original data sequence. This is the standardized feature data sequence.

[0049] (2) Calculate the correlation coefficient matrix R.

[0050] First, the standardized feature data are combined into a feature data matrix, represented as follows:

[0051] Secondly, based on the correlation between the various feature data, the correlation coefficient matrix R is calculated using the following formula:

[0052] in, r ij The calculation formula is:

[0053] in, , They are column vectors , The average value.

[0054] (3) Calculate the eigenvalues ​​and eigenvectors, and solve for the principal components.

[0055] According to the characteristic equation We obtain m eigenvalues ​​λ1≥λ2≥……≥λ m eigenvectors corresponding to ≥0 α 1, α 2,……, α m , where E is the identity matrix. The magnitude of the eigenvalues ​​can be used to measure the role of each principal component in describing the evaluated object.

[0056] m new feature data are composed of the feature values:

[0057] Furthermore, the cumulative contribution rate of the first Q principal components is determined as follows:

[0058] when When the value is close to 1, the top Q principal components are selected to replace the m feature data to achieve dimensionality reduction. In some embodiments, when When the value is ≥0.95, the first Q principal components are considered to form the first dataset, that is, the first threshold is 0.95.

[0059] Using the above method, the first dataset corresponding to each cylinder is determined, and the first sample dataset includes the first dataset corresponding to each cylinder.

[0060] After determining the first sample dataset, the initial fuel injection quantity prediction model is trained based on this dataset. In this embodiment, the fuel injection quantity prediction model uses a GRU neural network to describe the nonlinearity and temporal dependence of the fuel injection quantity. The GRU neural network not only possesses the self-learning and nonlinear data processing capabilities of neural networks, enabling it to better adapt to engine fuel injection control under different scenarios, but its hidden layer neuron loop network structure also has short-term memory capabilities, allowing the output to consider both the current input and inputs from a past period, thus better handling temporal issues. Furthermore, the unique gating structure of the GRU neural network's hidden layer neurons can control the rate of information accumulation, selectively adding new information and previously accumulated information, improving long-term dependency issues. Therefore, the GRU neural network can better learn the changing patterns of engine fuel injection quantity and accurately predict the engine fuel injection quantity at future moments.

[0061] like Figure 3 The diagram shown is a schematic representation of a network structure for predicting the air-fuel ratio of a single-cylinder engine based on a GRU neural network, as provided in an embodiment of this specification. Please refer to... Figure 3 A GRU neural network can include an input layer, a hidden layer, and an output layer.

[0062] The training process of the initial fuel injection quantity prediction model is explained below. When the engine is a multi-cylinder engine, the first sample dataset contains the first dataset corresponding to each cylinder. For ease of explanation, taking a four-cylinder engine as an example, the first sample dataset contains four first datasets, each corresponding to one of the four cylinders of the engine. During training, the first dataset corresponding to each cylinder is selected as the input data for the model. Since the engine fuel injection quantity is controlled by a four-cylinder integrated system, future fuel injection quantities can be predicted. l oil The air-fuel ratio of each cylinder in the step engine y 1, y 2,……, y loil The amount of fuel injected into the engine can be calculated based on the air-fuel ratio statistics.

[0063] Specifically, refer to Figure 3 In the network structure, the input layer reads the first dataset and passes it to the hidden layer. The computation results of each neuron in the hidden layer can be represented as follows:

[0064] in, The activation function for the hidden layer. and The outputs of each neuron in the input layer and hidden layer. That is, the Q principal components contained in the first dataset input to the GRU neural network; and For the network weights of the hidden layer, and n represents the number of neurons in the input layer.

[0065] In particular, the hidden layer neurons of the GRU neural network have special gating structures, including update gates and reset gates. Maintaining a balance between input and forgetting:

[0066] in, , and Update the network weights of the gates in the hidden layer neurons; The output of each neuron in the hidden layer is used to characterize the state of the hidden layer neurons and store nonlinear output information; when z=0, y Hid ( t )and y Hid ( t -1) represents a nonlinear functional relationship; when z=1, y Hid ( t )and y Hid ( t -1) represents a linear function relationship.

[0067] GRU neural network uses a reset gate To control candidate states Does it depend on the state of the previous moment?

[0068]

[0069] in, , , and , , These are the network weights for the reset gate and candidate state update in the hidden layer neurons, respectively. When r=0, the state of the hidden layer neuron at the current time depends only on the input at the current time and is independent of the historical state; when r=1, the current state depends on the input at the previous time step and the state at the previous time step.

[0070] Therefore, the state update method for the hidden layer neurons of GRU is as follows:

[0071] Through the above process, the GRU neural network ultimately outputs the air-fuel ratio for each cylinder of the future engine. Furthermore, based on the predicted air-fuel ratio... Ratio Air-fuel and the intake air volume collected in real time during driving. Air intake The predicted fuel injection quantity is calculated using the following formula. Fuel This allows for precise fuel injection control by compensating for the amount of fuel injected into each cylinder.

[0072] In the embodiments of this specification, an initial fuel injection quantity prediction model is trained based on a first sample dataset. The training process for each round is described above. In each round of training, a predicted air-fuel ratio is output, and the predicted fuel injection quantity is calculated based on this output. In some embodiments, the predicted fuel injection quantity can be compared with the actual fuel injection quantity, and the various model parameters of the fuel injection quantity prediction model can be adjusted according to the comparison results until the error between the two is less than a preset error, or the number of training iterations reaches a preset number, indicating that the model training is complete. The finally obtained trained fuel injection quantity prediction model is used as the preset fuel injection quantity prediction model.

[0073] In the embodiments described in this specification, the trained preset fuel injection quantity prediction model can be put into use to predict the fuel injection quantity in real time based on the collected first target dataset. It should be noted that during the training process of the fuel injection quantity prediction model, since principal component analysis was performed on the original data to select the data ultimately used for model training, in some embodiments, the data type of the first target dataset is consistent with the data type obtained after principal component analysis.

[0074] II. Gas Volume Prediction The preset air volume prediction model is constructed through the following steps: For each cylinder of the engine, a second original dataset corresponding to that cylinder is determined. The second original dataset includes data related to air volume prediction under different engine operating conditions. For each cylinder of the engine, principal component analysis is performed on the second original dataset of that cylinder. Based on the principal component analysis results, the data corresponding to the top M principal components with a cumulative contribution value greater than a second threshold are used as the second dataset corresponding to that cylinder, where M is a positive integer. Based on the second dataset corresponding to each cylinder, a second sample dataset for air volume prediction is constructed. Based on the second sample dataset, the initial air volume prediction model is trained to obtain a trained air volume prediction model, which is used as the preset air volume prediction model.

[0075] Specifically, the EGR system is a crucial factor in improving the thermal efficiency of hybrid engines; however, inaccurate EGR rate prediction and the engine mixture lag effect pose significant challenges to active engine control. Accurately reproducing combustion lag and revealing its mechanism through numerical methods is one of the most challenging problems in EGR system combustion research. This specification's embodiments propose a combustion model system with the EGR rate model at its core.

[0076] Similar to the air-fuel ratio prediction mentioned above, the EGR rate, as an important parameter characterizing the ratio of recirculated exhaust gas to the total exhaust gas volume discharged from the engine cylinders, is also affected by the complex actions of multiple feature variables, exhibiting significant nonlinearity and time-series dependence. Therefore, in the training process of the air volume prediction model, the same modeling algorithm as that used for fuel injection quantity prediction is employed, namely PCA fused with a GRU neural network, to accurately predict the EGR rate at future moments.

[0077] In the embodiments of this specification, cylinder-specific prediction and control can be used when predicting air volume. That is, for each cylinder of each engine, a second original dataset corresponding to that cylinder is determined. In order to consider air volume prediction under different operating conditions, the first original dataset includes data related to air volume prediction under different engine operating conditions. The engine operating conditions can be characterized by parameters such as engine speed and torque, that is, the speed and torque corresponding to different operating conditions are different.

[0078] In some embodiments, the second raw data may include the following feature data: engine speed, torque, engine coolant temperature, intake manifold temperature, intake manifold pressure, throttle inlet pressure, air filter inlet pressure, air filter inlet temperature, altitude, ambient temperature, humidity, and historical engine fuel injection quantity. Further, principal component analysis is performed on the second raw dataset. The specific process can be referred to the explanation of principal component analysis above, and will not be repeated here. Based on the principal component classification results, the data corresponding to the top M principal components with a cumulative contribution value greater than the second threshold are taken as the second dataset corresponding to that cylinder. The second threshold can be set according to actual needs; for example, the second threshold can be 0.95, 0.9, etc., and the second threshold can be the same as or different from the first threshold.

[0079] For each cylinder, there is a corresponding second dataset, which may include the second dataset corresponding to each cylinder. In the embodiments of this specification, after determining the second sample dataset, the initial air volume prediction model is trained based on the second sample dataset. In some embodiments, the air volume prediction model still uses a GRU neural network. The structure and principle of the GRU neural network can be referred to the description above, and will not be repeated here.

[0080] Based on the second sample dataset, the initial gas volume prediction model is trained. In some embodiments, during each training round, the predicted results of the future engine throttle flow, mixing chamber flow, and intercooler pipe gas volume are output, and the EGR rate prediction result, i.e., the predicted exhaust gas recirculation rate, is calculated based on the prediction results. In some embodiments, the EGR rate prediction result can also be directly output during each training round, which is not limited here. For each training round, the EGR rate prediction result is compared with the actual EGR rate, and the various model parameters of the gas volume prediction model are adjusted according to the comparison results until the error between the two is less than the preset error, or the number of training iterations reaches the preset number, indicating that the model training has been completed. The finally obtained trained gas volume prediction model is used as the preset gas volume prediction model.

[0081] In the embodiments described in this specification, the trained preset gas volume prediction model can be put into use to predict gas volume in real time on the collected second target dataset. It should be noted that during the training process of the gas volume prediction model, since principal component analysis was performed on the original data to select the data that would ultimately be used for model training, in some embodiments, the data type of the second target dataset is consistent with the data type obtained after principal component analysis.

[0082] In some embodiments, after obtaining the exhaust gas recirculation prediction result based on the preset gas volume prediction model, the turbocharger, throttle, mixing valve and EGR valve can be coordinated and precisely controlled based on the exhaust gas recirculation prediction result.

[0083] III. Ignition Prediction The preset ignition prediction model is constructed through the following steps: obtaining a third sample dataset for ignition prediction, the third sample dataset including historical knock sensor vibration signals; training the initial ignition prediction model based on the third sample dataset to obtain a trained ignition prediction model, which is used as the preset ignition prediction model.

[0084] Specifically, accurate prediction of transient combustion knock is crucial for ignition monitoring and control. Especially with the development of high-compression engines in recent years, the embodiments in this specification employ a neural network model to accurately predict knock changes during combustion oscillations while maintaining prediction accuracy.

[0085] Unlike fuel injection prediction and air volume prediction, the correlation between the ignition angle during knock and historical ignition angles is weaker, and it is more related to factors such as engine intake air volume and intake pressure change rate. Therefore, when predicting the vibration signal from the knock sensor, the aforementioned time-series-based GRU neural network is not used; instead, a wavelet neural network for signal analysis is employed. Wavelet neural networks can be used for boundary processing and filtering, video analysis, signal-to-noise separation, etc. Integrating wavelet transform into the neural network allows it to combine the advantages of wavelet transform in time-frequency local analysis with the learning capabilities of a neural network. Therefore, considering the characteristics of the knock signal, the embodiments in this specification use a wavelet neural network to predict the vibration signal from the knock sensor.

[0086] Before using wavelet neural networks to predict the vibration signals from the knock sensor, a third sample dataset for ignition prediction can be obtained through the following steps: First, obtain a third original dataset, which includes data related to ignition prediction under different engine operating conditions; second, perform principal component analysis on the third original dataset, and based on the principal component analysis results, determine the data corresponding to the top P principal components whose cumulative contribution value is greater than a third threshold, where P is a positive integer; third, obtain the historical knock sensor vibration signals of the engine, and construct the third sample dataset based on the data corresponding to the top P principal components and the historical knock sensor vibration signals.

[0087] In the embodiments of this specification, the prediction of the knock sensor vibration signal can be performed by cylinder or not. When performing cylinder-by-cylinder prediction and control of the knock sensor vibration signal, a third raw dataset corresponding to each cylinder of the engine is determined. To consider fuel injection control under different operating conditions, the third raw dataset includes data related to fuel injection quantity under different engine operating conditions. Engine operating conditions can be characterized by parameters such as engine speed and torque; that is, different operating conditions correspond to different speeds and torques. However, when only one knock sensor is installed in the engine (e.g., installed in the middle of the cylinder block or a shared location with the cylinder head), the knock sensor collects the superimposed vibration signal generated by combustion in all cylinders, rather than the independent vibration of a single cylinder. In this case, the third raw dataset can be the dataset corresponding to the entire cylinder.

[0088] Since the third raw dataset can include multiple data related to ignition prediction, in some embodiments, the third raw dataset may include the following feature data: intake air volume, intake pressure change rate, water temperature, intake air temperature, engine speed, load, and pedal information. Further, principal component analysis is performed on the third raw dataset; the specific process can be referred to the explanation of principal component analysis above, and will not be repeated here. Based on the principal component classification results, the data corresponding to the top P principal components with a cumulative contribution value greater than the third threshold are used as the third dataset. The third threshold can be set according to actual needs; for example, the third threshold can be 0.98, 0.95, 0.9, etc., and the third threshold can be the same as or different from the first and second thresholds.

[0089] After obtaining the data corresponding to the first P principal components, the historical knock sensor vibration signal of the engine is acquired. The data corresponding to the first P principal components and the historical knock sensor vibration signal are used as the third sample dataset, that is, the third sample dataset includes P+1 types of data.

[0090] After determining the third sample dataset, the initial ignition prediction model is trained based on the third sample dataset. In the embodiments of this specification, the ignition prediction model is selected as a WNN neural network model, such as... Figure 4 The diagram shown is a schematic representation of a WNN neural network provided in an embodiment of this specification, including an input layer, a hidden layer, and an output layer. The principle of using WNN for predicting vibration signals from a detonation sensor will be explained below.

[0091] In some embodiments, the overall prediction process for the vibration signal of the knock sensor is divided into four parts: initialization, forward propagation of the signal, backward propagation of the error, and model prediction.

[0092] During the initialization process, the weight parameters of the WNN neural network are randomly initialized.

[0093] During the forward propagation of the signal, the third sample dataset is used as the input to the WNN neural network. For ease of explanation, let's take an example where the third sample dataset contains K sets of data, i.e., P+1=K as mentioned above. Then, the third sample dataset can be represented as: The weights between the input layer and the hidden layer The output of the hidden layer is calculated, and then the weights between the hidden layer and the output layer are applied. The output layer value is calculated, which is the predicted value of the knock sensor vibration signal at the current iteration number.

[0094] The formula for calculating the output of the hidden layer can be:

[0095] in, The output value of the hidden layer. This is the input data for the i-th node in the input layer. The wavelet basis functions used may, in some embodiments, be the Morlet mother wavelet basis functions. and These are the scaling factor and translation factor corresponding to the wavelet basis function, respectively. The Morlet mother wavelet basis function formula can be: .

[0096] Output layer The calculation formula can be:

[0097] in, This represents the number of nodes in the output layer.

[0098] During the backpropagation of errors, when one forward propagation of the signal is completed, the vibration signal value of the detonation sensor predicted by the training can be obtained. Based on this, the error between the actual detonation sensor vibration signal and the predicted value is calculated. The overall prediction is completed by updating and iterating through backpropagation of errors.

[0099] In some embodiments, error backpropagation specifically involves: First, RMSE is used as the objective function for error calculation to reflect the difference between the predicted knock sensor vibration signal and the actual knock sensor vibration signal:

[0100] in, This is the actual vibration signal from the knock sensor. To predict the vibration signal of the knock sensor.

[0101] Secondly, update the weights of the WNN neural network according to the aforementioned error. and :

[0102]

[0103] in, It is the network's learning rate.

[0104] During the model prediction process, after iterative cycles, if the error between the final predicted knock sensor signal and the actual knock sensor signal is within a certain range, or if the number of iterations reaches the preset number, the model training process is completed, and the trained model is used as the final preset ignition prediction model.

[0105] In the embodiments described in this specification, the trained preset ignition prediction model can be put into use to perform ignition prediction on the collected third target dataset in real time. It should be noted that during the training process of the ignition prediction model, since principal component analysis was performed on the original data to select the data that would ultimately be used for model training, in some embodiments, the data type of the third target dataset is consistent with the data type obtained after principal component analysis.

[0106] In some embodiments, after obtaining the knock sensor vibration signal prediction result based on the preset ignition prediction model, the coordinated and precise control of the turbocharger, VVT (Variable Valve Timing), and ignition angle can be completed based on the knock sensor vibration signal prediction result.

[0107] It should be noted that, in order to ensure that the combustion prediction method in the embodiments of this specification can be adapted to various vehicle types, during the training process of the above-mentioned models, data corresponding to different vehicle types can be collected as training data. For example, data corresponding to PHEV (Plug-in Hybrid Electric Vehicle), HEV (Hybrid Electric Vehicle), and REV (Range Extended Electric Vehicle) can be collected as training data for each prediction model, thereby enabling the combustion prediction model to be adapted to multiple vehicle types.

[0108] In step S105, after obtaining the prediction results from the preset fuel injection quantity prediction model, the preset ignition prediction model, and the preset ignition prediction model, combustion control commands for the target vehicle can be determined based on these prediction results. It should be noted that the combustion control commands can involve control commands for multiple actuators, such as control commands for the throttle valve, engine speed, turbocharger, and mixing valve. Each command is sent to its corresponding actuator so that each actuator executes the corresponding control command, thereby precisely controlling the combustion process of the target vehicle to achieve optimal performance and emissions.

[0109] In the embodiments of this specification, since preset fuel injection quantity prediction models, preset air volume prediction models, and preset ignition prediction models are deployed in both the cloud and the vehicle, it is possible to choose whether to perform engine combustion prediction in the cloud or in the vehicle.

[0110] In some embodiments, when the target vehicle establishes a connection with the cloud, the target dataset is sent to the cloud through the target vehicle to perform engine combustion prediction using the preset fuel injection quantity prediction model, the preset air volume prediction model, and the preset ignition prediction model deployed in the cloud; when the target vehicle disconnects from the cloud, engine combustion prediction is performed using the preset fuel injection quantity prediction model, the preset air volume prediction model, and the preset ignition prediction model deployed on the target vehicle itself.

[0111] In the embodiments of this specification, considering the limitations of vehicle-side memory and computing power, the combustion prediction model stored on the vehicle-side can be a simplified version of the cloud-based combustion prediction model. In this case, to ensure more accurate prediction results, when performing combustion prediction, it can be determined whether the vehicle-side has established a connection with the cloud. If a connection has been established, the cloud-based model is selected for prediction. That is, the target vehicle uploads the currently collected target datasets (first target dataset, second target dataset, and third target dataset) to the cloud, performs combustion prediction through the preset fuel injection quantity prediction model, preset air volume prediction model, and preset ignition prediction model deployed on the cloud, and generates combustion control commands based on the combustion prediction results and sends them to the target vehicle so that the target vehicle can execute the combustion control commands.

[0112] When the cloud and the vehicle are not connected, the preset fuel injection quantity prediction model, preset air volume prediction model and preset ignition prediction model deployed on the vehicle are used to predict combustion. Based on the combustion prediction results (fuel injection quantity prediction results, exhaust gas recirculation rate prediction results and knock sensor vibration signal prediction results), corresponding combustion control commands are generated so that the target vehicle can execute the combustion control commands.

[0113] In addition, when establishing a connection between the cloud and the vehicle, the network status of both the cloud and the vehicle can be checked first. If the network status is relatively stable, the target dataset is uploaded to the cloud for engine combustion prediction. If the network status is unstable, there is a risk of missing, duplicate, or incorrect transmission of the target dataset. In this case, engine combustion prediction is performed through the vehicle.

[0114] In some embodiments, if a network interruption occurs during the process of the target vehicle uploading the target dataset to the cloud, the upload of the target dataset can be stopped, and a model deployed locally on the vehicle can be selected for combustion prediction.

[0115] In this embodiment of the specification, in order to ensure that the preset fuel injection quantity prediction model, the preset air volume prediction model, and the preset ignition prediction model are all up-to-date models, the following steps may also be performed: when it is detected that the preset fuel injection quantity prediction model, the preset air volume prediction model, and / or the preset ignition prediction model deployed in the cloud are updated, the updated target model is obtained through the target vehicle to update the local model in the target vehicle corresponding to the target model.

[0116] In some embodiments, model updates can be performed first in the cloud. For example, the cloud can update preset fuel injection quantity prediction models, preset air volume prediction models, and / or preset ignition prediction models according to a preset update cycle, or update the corresponding models in the cloud when it detects that the prediction accuracy of a model is lower than a preset accuracy. Furthermore, when communication is established between the cloud and the target vehicle, and the current state of the target vehicle can meet the requirements for model updates, for example, when the vehicle enters the download update state, the cloud can send the updated target model to the target vehicle. In some embodiments, the cloud can send an update data packet of the target model to the target vehicle, and the target vehicle updates its corresponding local model after receiving the update data packet.

[0117] In summary, the method provided in the embodiments of this specification can achieve the following effects: The system is based on machine learning algorithms to build a precise prediction model to optimize control precision. It can support the improvement of the reliability of high-efficiency engines, and cope with the losses caused by abnormal combustion such as high compression ratio knock and pre-ignition. It can support the selection of engine points in actual scenarios for PHEV, HEV and REV models. It can also achieve the ultimate exploration of engine combustion potential under complex scenarios and working conditions, improve the thermal efficiency in actual scenarios, and reduce the energy consumption of the whole vehicle. Machine learning algorithms can systematically acquire signals from the vehicle, motor, and battery, enabling dynamic updates in the cloud and offline storage. This supports the replacement of some sensors to reduce the cost of hybrid systems. For example, model-based predictions can predict fuel injection, air intake, and ignition. Since the combustion prediction results in the embodiments of this specification are more accurate, it is possible to eliminate or reduce the specifications of certain components in the vehicle, such as reducing water temperature sensors, eliminating EGR differential pressure sensors, VVT, etc., and promoting the deep integration of hybrid electric drive assemblies. It supports the integrated control of multiple domains such as engines and motors, and can implement the algorithms provided in the embodiments of this specification based on the same chip, thereby improving the efficiency of the all-in-one controller algorithm; The PCA+GRU neural network algorithm / WNN neural network algorithm used in the embodiments of this specification is data-driven, which can effectively reduce the repeated iterative optimization on benches or actual vehicles required for rule-based calibration methods, thereby reducing development costs.

[0118] Based on the same inventive concept, embodiments of this specification also provide a combustion prediction device for a vehicle engine, such as... Figure 5 As shown, the device includes: The acquisition module 501 is used to acquire the target dataset of the target vehicle. The target dataset includes a first target dataset for fuel injection quantity prediction, a second target dataset for gas volume prediction, and a third target dataset for ignition prediction. The first processing module 502 is used to determine the fuel injection quantity prediction result corresponding to the engine of the target vehicle based on the first target dataset and the preset fuel injection quantity prediction model. The second processing module 503 is used to determine the predicted exhaust gas recirculation rate corresponding to the engine based on the second target dataset and the preset gas volume prediction model. The third processing module 504 is used to determine the predicted result of the knock sensor vibration signal corresponding to the engine based on the third target dataset and the preset ignition prediction model. The execution module 505 is used to determine the combustion control command of the target vehicle based on the fuel injection quantity prediction result, the exhaust gas recirculation rate prediction result, and the knock sensor vibration signal prediction result, so that the target vehicle executes the combustion control command.

[0119] In some embodiments, the combustion prediction device for the vehicle engine further includes: The fuel injection quantity prediction model construction module is used to obtain a first sample dataset for fuel injection quantity prediction; and to train an initial fuel injection quantity prediction model based on the first sample dataset to obtain a trained fuel injection quantity prediction model, which is used as the preset fuel injection quantity prediction model.

[0120] In some embodiments, the fuel injection quantity prediction model building module is used for: For each cylinder of the engine, a first raw dataset corresponding to that cylinder is determined. The first raw dataset includes data related to the fuel injection quantity under different engine operating conditions, the cylinder knocking angle and cylinder knocking intensity of the target vehicle. For each cylinder of the engine, principal component analysis is performed on the first original dataset of the cylinder, and based on the principal component analysis results, the data corresponding to the first Q principal components whose cumulative contribution value is greater than the first threshold are taken as the first dataset corresponding to the cylinder, where Q is a positive integer; The first sample dataset is constructed based on the first dataset corresponding to each cylinder.

[0121] In some embodiments, the combustion prediction device for the vehicle engine further includes: The air volume prediction model construction module is used to determine the second original dataset corresponding to each cylinder of the engine. The second original dataset includes data related to air volume prediction under different engine operating conditions. For each cylinder of the engine, principal component analysis is performed on the second original dataset of that cylinder. Based on the principal component analysis results, the data corresponding to the top M principal components with a cumulative contribution value greater than a second threshold are used as the second dataset corresponding to that cylinder, where M is a positive integer. Based on the second dataset corresponding to each cylinder, a second sample dataset for air volume prediction is constructed. Based on the second sample dataset, the initial air volume prediction model is trained to obtain a trained air volume prediction model, which is used as the preset air volume prediction model.

[0122] In some embodiments, the combustion prediction device for the vehicle engine further includes: The ignition prediction model construction module is used to acquire a third sample dataset for ignition prediction, the third sample dataset including historical detonation sensor vibration signals; and to train the initial ignition prediction model based on the third sample dataset to obtain a trained ignition prediction model, which is used as the preset ignition prediction model.

[0123] In some implementations, the ignition prediction model building module is used for: A third raw dataset is obtained, which includes data related to ignition prediction under different engine operating conditions; principal component analysis is performed on the third raw dataset, and based on the principal component analysis results, the data corresponding to the top P principal components with a cumulative contribution value greater than a third threshold are determined, where P is a positive integer; historical knock sensor vibration signals of the engine are obtained, and the third sample dataset is constructed based on the data corresponding to the top P principal components and the historical knock sensor vibration signals.

[0124] In some embodiments, the preset fuel injection quantity prediction model, the preset air volume prediction model, and the preset ignition prediction model are deployed both locally on the target vehicle and in the cloud corresponding to the target vehicle. The combustion prediction device for the vehicle engine further includes: The cloud prediction module is used to send the target dataset to the cloud through the target vehicle when the target vehicle establishes a connection with the cloud, so as to perform engine combustion prediction through the preset fuel injection quantity prediction model, the preset air volume prediction model and the preset ignition prediction model deployed in the cloud. The vehicle-side prediction module is used to predict engine combustion by means of the preset fuel injection quantity prediction model, the preset air volume prediction model, and the preset ignition prediction model deployed on the target vehicle itself when the target vehicle is disconnected from the cloud.

[0125] In some embodiments, the combustion prediction device for the vehicle engine further includes: The update module is used to obtain the updated target model through the target vehicle when it is detected that the preset fuel injection quantity prediction model, the preset air volume prediction model and / or the preset ignition prediction model deployed in the cloud have been updated, so as to update the local model in the target vehicle corresponding to the target model.

[0126] In some implementations, the update module is further configured to update the fuel injection quantity prediction model, the preset air volume prediction model, and / or the preset ignition prediction model deployed in the cloud based on a preset update cycle.

[0127] In some implementations, both the preset fuel injection quantity prediction model and the preset air volume prediction model are gated recurrent neural network models.

[0128] In some implementations, the ignition prediction model is a wavelet neural network model.

[0129] Regarding the above-mentioned device, the specific implementation of each step has been described in detail in the embodiment of the combustion prediction method for vehicle engines provided in the specification, and will not be elaborated here.

[0130] The combustion prediction device for vehicle engines provided in the embodiments of this specification can achieve the following effects: The precise prediction model built on machine learning algorithms can optimize control accuracy, support the improvement of high-efficiency engine reliability, and cope with the losses caused by abnormal combustion such as high compression ratio knock and pre-ignition. It can support the selection of engine points in actual scenarios for PHEV, HEV and REV models, and can achieve the ultimate exploration of engine combustion potential under complex scenarios and working conditions, improve the thermal efficiency in actual scenarios, and reduce the energy consumption of the whole vehicle. Machine learning algorithms can systematically acquire signals from the vehicle, motor, and battery, enabling dynamic updates in the cloud and offline storage. This supports the replacement of some sensors to reduce the cost of hybrid systems. For example, model-based predictions can predict fuel injection, air intake, and ignition. Since the combustion prediction results in the embodiments of this specification are more accurate, it is possible to eliminate or reduce the specifications of certain components in the vehicle. For example, the number of water temperature sensors can be reduced, and VVT and EGR differential pressure sensors can be eliminated. This can promote the deep integration of hybrid electric drive assemblies. Supporting the fusion of multi-domain control for engines and motors, in REV models, the integrated GECU (Generator and Engine Control Unit) integrates the generator controller, engine controller, and vehicle controller; or in PCU (Powertrain Control Unit) models, the integrated PCU integrates the generator controller, engine controller, vehicle controller, shift controller, and drive motor controller. The algorithms provided in the embodiments of this specification can be implemented on the same chip, improving the efficiency of the integrated controller algorithm. The PCA+GRU neural network algorithm / WNN neural network algorithm used in the embodiments of this specification is data-driven, which can effectively reduce the repeated iterative optimization on benches or actual vehicles required for rule-based calibration methods, thereby reducing development costs.

[0131] Based on the same inventive concept, embodiments of the present invention also provide a combustion prediction method for vehicle engines, such as... Figure 6 The device includes a memory 604, a processor 602, and a computer program stored in the memory 604 and executable on the processor 602. When the processor 602 executes the program, it implements any of the above-described embodiments of the vehicle engine combustion prediction method. In some embodiments, the electronic device may be a vehicle infotainment system.

[0132] Among them, Figure 6In this document, a bus architecture (represented by bus 600) is used. Bus 600 may include any number of interconnected buses and bridges, linking various circuits including one or more processors represented by processor 602 and memory represented by memory 604. Bus 600 may also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. Bus interface 605 provides an interface between bus 600 and receiver 601 and transmitter 603. Receiver 601 and transmitter 603 may be the same element, i.e., a transceiver, providing a unit for communicating with various other devices over a transmission medium. Processor 602 is responsible for managing bus 600 and general processing, while memory 604 can be used to store data used by processor 602 during operation.

[0133] Based on the same inventive concept, embodiments of this specification provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described combustion prediction method for a vehicle engine: acquiring a target dataset for a target vehicle, the target dataset including a first target dataset for fuel injection quantity prediction, a second target dataset for air volume prediction, and a third target dataset for ignition prediction; determining a fuel injection quantity prediction result corresponding to the engine of the target vehicle based on the first target dataset and a preset fuel injection quantity prediction model; determining an exhaust gas recirculation rate prediction result corresponding to the engine based on the second target dataset and the preset air volume prediction model; determining a knock sensor vibration signal prediction result corresponding to the engine based on the third target dataset and the preset ignition prediction model; and determining a combustion control command for the target vehicle based on the fuel injection quantity prediction result, the exhaust gas recirculation rate prediction result, and the knock sensor vibration signal prediction result, so that the target vehicle executes the combustion control command.

[0134] Based on the same inventive concept, embodiments of this specification provide a computer program product, which includes a computer program. When executed by a processor, the computer program loads and executes the combustion prediction method steps of the above-described vehicle engine: acquiring a target dataset of a target vehicle, the target dataset including a first target dataset for fuel injection quantity prediction, a second target dataset for air volume prediction, and a third target dataset for ignition prediction; determining a fuel injection quantity prediction result corresponding to the engine of the target vehicle based on the first target dataset and a preset fuel injection quantity prediction model; determining an exhaust gas recirculation rate prediction result corresponding to the engine based on the second target dataset and the preset air volume prediction model; determining a knock sensor vibration signal prediction result corresponding to the engine based on the third target dataset and the preset ignition prediction model; and determining a combustion control command for the target vehicle based on the fuel injection quantity prediction result, the exhaust gas recirculation rate prediction result, and the knock sensor vibration signal prediction result, so that the target vehicle executes the combustion control command.

[0135] The functions described herein can be implemented in hardware, software executed by a processor, firmware, or any combination thereof. If implemented in software executed by a processor, the functions can be stored as one or more instructions or code on or transmitted via a computer-readable medium. Other examples and embodiments are within the scope of this invention and the appended claims. For example, due to the nature of software, the functions described above can be implemented using software executed by a processor, hardware, firmware, hardwired, or any combination thereof. Furthermore, the functional units can be integrated into a single processing unit, or each unit can exist physically separately, or two or more units can be integrated into a single unit.

[0136] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0137] The units described as separate components may or may not be physically separate. Similarly, the components of the control device may or may not be physical units; they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0138] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0139] The above description is merely an embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A combustion prediction method of a vehicle engine, characterized by, The method comprises the following steps: acquiring a target data set of a target vehicle, the target data set comprising a first target data set for fuel injection amount prediction, a second target data set for gas amount prediction, and a third target data set for ignition prediction; determining a fuel injection amount prediction result corresponding to an engine of the target vehicle based on the first target data set and a preset fuel injection amount prediction model; determining an exhaust gas recirculation rate prediction result corresponding to the engine based on the second target data set and a preset gas amount prediction model; determining a knock sensor vibration signal prediction result corresponding to the engine based on the third target data set and a preset ignition prediction model; determining a combustion control instruction of the target vehicle based on the fuel injection amount prediction result, the exhaust gas recirculation rate prediction result, and the knock sensor vibration signal prediction result, so that the target vehicle executes the combustion control instruction.

2. The method of claim 1, wherein, The preset fuel injection amount prediction model is constructed by the following steps: acquiring a first sample data set for fuel injection amount prediction; training an initial fuel injection amount prediction model based on the first sample data set to obtain a trained fuel injection amount prediction model as the preset fuel injection amount prediction model.

3. The method of claim 2, wherein, The acquiring of the first sample data set for fuel injection amount prediction comprises the following steps: for each cylinder of the engine, determining a first raw data set corresponding to the cylinder, the first raw data set comprising data associated with fuel injection amount under different engine operating conditions, cylinder-specific knock back angle, and cylinder-specific knock intensity of the target vehicle; for each cylinder of the engine, performing principal component analysis on the first raw data set of the cylinder, and based on the principal component analysis result, taking data corresponding to the first Q principal components with cumulative contribution value greater than a first threshold value as a first data set corresponding to the cylinder, Q being a positive integer; constructing the first sample data set based on the first data set corresponding to each cylinder.

4. The method of claim 1, wherein, The preset gas amount prediction model is constructed by the following steps: for each cylinder of the engine, determining a second raw data set corresponding to the cylinder, the second raw data set comprising data associated with gas amount prediction under different engine operating conditions; for each cylinder of the engine, performing principal component analysis on the second raw data set of the cylinder, and based on the principal component analysis result, taking data corresponding to the first M principal components with cumulative contribution value greater than a second threshold value as a second data set corresponding to the cylinder, M being a positive integer; constructing a second sample data set for gas amount prediction based on the second data set corresponding to each cylinder; training an initial gas amount prediction model based on the second sample data set to obtain a trained gas amount prediction model as the preset gas amount prediction model.

5. The method of claim 1, wherein, The preset ignition prediction model is constructed by the following steps: acquiring a third sample data set for ignition prediction, the third sample data set comprising historical knock sensor vibration signals; training an initial ignition prediction model based on the third sample data set to obtain a trained ignition prediction model as the preset ignition prediction model.

6. The method of claim 5, wherein, The acquiring of the third sample data set for ignition prediction comprises the following steps: obtain a third original data set, the third original data set comprising data associated with ignition prediction under different engine working conditions; perform principal component analysis on the third original data set, and determine, based on a result of the principal component analysis, data corresponding to a first P principal components with cumulative contribution values greater than a third threshold value, P being a positive integer; obtain a historical knock sensor vibration signal of the engine, and construct the third sample data set based on the data corresponding to the first P principal components and the historical knock sensor vibration signal.

7. The method of claim 1, wherein, The preset fuel injection amount prediction model, the preset air amount prediction model and the preset ignition prediction model are deployed in the target vehicle and a cloud corresponding to the target vehicle, and the method further comprises: When the target vehicle is connected to the cloud, the target data set is sent to the cloud by the target vehicle, so that the preset fuel injection amount prediction model, the preset air amount prediction model and the preset ignition prediction model deployed in the cloud are used to perform combustion prediction of the engine; When the target vehicle is disconnected from the cloud, the preset fuel injection amount prediction model, the preset air amount prediction model and the preset ignition prediction model deployed in the target vehicle are used to perform combustion prediction of the engine.

8. The method of claim 1, wherein, The method further comprises: When it is detected that the preset fuel injection amount prediction model, the preset air amount prediction model and / or the preset ignition prediction model deployed in the cloud exist updates, an updated target model is obtained by the target vehicle, so as to update a local model corresponding to the target model in the target vehicle.

9. The method of claim 8, wherein, The method further comprises: The preset fuel injection amount prediction model, the preset air amount prediction model and / or the preset ignition prediction model deployed in the cloud are updated based on a preset update period.

10. The method of claim 1, wherein, The preset fuel injection amount prediction model and the preset air amount prediction model are both gated recurrent neural network models.

11. The method of claim 1, wherein, The ignition prediction model is a wavelet neural network model.

12. A combustion prediction device of a vehicle engine, characterized by, The method comprises: an obtaining module, configured to obtain a target data set of a target vehicle, the target data set comprising a first target data set for fuel injection amount prediction, a second target data set for air amount prediction and a third target data set for ignition prediction; a first processing module, configured to determine, based on the first target data set and a preset fuel injection amount prediction model, a fuel injection amount prediction result corresponding to an engine of the target vehicle; a second processing module, configured to determine, based on the second target data set and a preset air amount prediction model, an exhaust gas recirculation rate prediction result corresponding to the engine; a third processing module, configured to determine, based on the third target data set and a preset ignition prediction model, a knock sensor vibration signal prediction result corresponding to the engine; an executing module, configured to determine, based on the fuel injection amount prediction result, the exhaust gas recirculation rate prediction result and the knock sensor vibration signal prediction result, a combustion control instruction of the target vehicle, so that the target vehicle executes the combustion control instruction.

13. A combustion prediction device of a vehicle engine, characterized by, A computer program product comprising a computer program stored on a memory and loadable into the internal memory of a digital computer, comprising executable software code portions for realizing the method according to any one of claims 1 to 11 when the program is run on a processor of the digital computer.

14. A computer-readable storage medium, characterized in that, A computer program product comprising a computer program stored on a memory and loadable into the internal memory of a digital computer, comprising executable software code portions for realizing the method according to any one of claims 1 to 11 when the program is run on a processor of the digital computer.

15. A computer program product, characterised in that, The computer program product comprises a computer program which, when executed by a processor, is for loading and executing the method according to any one of claims 1 to 11.

Images