Mpc combustion control method for marine high-pressure common rail diesel engine based on physical informationization neural network

Abstract

The MPC combustion control method for marine high-pressure common rail diesel engines based on physical information neural networks belongs to the field of marine high-pressure common rail diesel engine control. It solves the problems of neglecting the physical consistency between the model and the actual engine and poor model generalization ability caused by simply relying on data-driven methods. The method includes: keeping the combustion midpoint constant, acquiring in-cylinder pressure, maximum in-cylinder pressure rise rate, indicated thermal efficiency, and emission data for different pre-injection fuel ratios and pre-main injection timing intervals; calculating the cyclic variation of the diesel engine's average indicated pressure, filtering data, and calculating the cumulative heat release and combustion midpoint; obtaining the influence law of relevant in-cylinder combustion characterization parameters of the diesel engine, determining the crankshaft angle, pre-injection fuel ratio, pre-main injection interval boundary, and maximum cylinder pressure rise rate upper limit corresponding to the pre-injection timing; constructing a diesel engine in-cylinder combustion prediction output model and MPC controller; updating control parameters and combustion state data to achieve closed-loop control. It has applications in energy conservation and emission reduction.

Description

Technical Field

[0001] This invention belongs to the field of marine high-pressure common rail diesel engine control, and in particular relates to a combustion control method for marine high-pressure common rail diesel engines based on physical information neural networks. Background Technology

[0002] For a long time, diesel engines have been the primary power source in the marine industry due to their advantages of high fuel economy, high torque output, and high durability. Although the national strategy of energy conservation and emission reduction has led to a shift towards cleaner energy sources, a large number of diesel engines still exist in the marine engine market and are not easily replaced.

[0003] To achieve energy conservation and emission reduction in diesel engines, there are currently two main technical routes: in-cylinder combustion control technology and emission after-treatment technology. In-cylinder combustion control technology focuses on methods and strategies to minimize or eliminate pollution during combustion. It enables low-temperature combustion patterns distinct from traditional diffusion combustion. Currently known novel combustion modes include HCCI, PCCI or PPCI, and RCCI. The development of combustion control largely sustains the future use and development of diesel engines. Under the premise of ensuring high diesel engine efficiency or equivalent to traditional combustion modes, optimizing the combustion process through in-cylinder combustion control methods to achieve emission reduction goals has become a research hotspot in this field.

[0004] Currently, combustion control methods in this field mainly rely on cylinder pressure feedback information to extract combustion characteristic data as feedback values ​​input into the control system. This includes combustion start point (CA10), combustion midpoint (CA50), combustion end point (CA90), and mean indicated pressure (IMEP), which are input as feedback variables to the controller. The controller then adjusts control variables such as fuel injection and intake air volume to control the combustion process. Combustion controllers can be designed based on model predictive control (MPC). There are two methods to obtain the model required by the controller: one is to obtain an abstract data-driven model through calculations using measured data, and the other is to model based on empirical formulas related to the physical processes of the engine. The former does not rely on empirical formulas, but the data-driven modeling method often fails to consider physical consistency, has poor generalization ability, and struggles to ensure consistency between the model and the actual engine, thus affecting the controller's performance. The latter relies on traditional experience or physical models, failing to fully utilize the physical information during engine operation and ignoring changes in engine component parameters over long-term use, resulting in a mismatch between the model and the actual engine throughout its lifespan. Meanwhile, when designing diesel engine combustion control methods, the output of the control variables also needs to consider the impact of the in-cylinder working process on the engine's operating state and performance. For example, when using single injection to achieve premixed combustion, the long ignition delay period can trigger multi-point auto-ignition in the cylinder, resulting in an excessively high in-cylinder pressure rise rate. This damages the insulating gas boundary layer in the cylinder, leading to increased heat transfer loss and reduced efficiency. This phenomenon is more pronounced under high-load conditions. Existing in-cylinder combustion research, such as "Improvement of premixed charge compression ignition-based combustion by two-stage injection," has demonstrated that multiple injections can reduce in-cylinder combustion temperature, and that increasing pre-injection control can improve the problem of excessively high pressure rise rate caused by intense combustion in single-stage fuel injection premixed combustion. However, it does not address specific control methods.

[0005] Chinese patent CN118838177A discloses a data-driven multi-objective control method for diesel engine combustion. This patent employs a neural network algorithm combined with model-free adaptive control. It uses NOx and fuel consumption control error data obtained during diesel engine operation as inputs to the neural network and Jacobi estimation algorithm to adjust control parameters. Under the constraints of NOx and fuel consumption objective functions, the model-free adaptive algorithm outputs control quantities to achieve adaptive adjustment of injection timing and rail pressure control, thereby achieving the control objective of the combustion process and realizing the control targets for fuel consumption and emissions under all operating conditions. However, it cannot simultaneously limit the rate of increase in cylinder pressure within an allowable range while controlling multiple injections to achieve combustion control.

[0006] Chinese patent CN115419511A discloses a closed-loop combustion control method and system for marine natural gas engines. This patent aims to address the instability of in-cylinder combustion in natural gas engines, which can lead to large fluctuations in cylinder pressure, misfires, and knocking. It utilizes a data-driven modeling method, combining the control signal under current operating conditions with collected in-cylinder combustion characteristic signals to construct the compact linearized data-driven model described in the patent. A data-driven model parameter update algorithm is designed to update the solved model parameters. Finally, a designed control input criterion function is used to update the solved parameters, resulting in the final control input applied to the natural gas engine's operation, thereby achieving combustion control. However, it still cannot simultaneously limit the in-cylinder pressure rise rate within an allowable range while controlling multiple fuel injections to achieve combustion control.

[0007] Chinese patent CN107269408A discloses an optimized combustion controller for diesel engines and a simulation model control method. To achieve optimized combustion control in diesel engines, this invention designs a data-driven MPC optimized combustion controller. The control variables are injection quantity, injection timing, EGR valve opening, and VGT cross-sectional area. It acquires NOx, output torque, and fuel consumption rate data through open-loop control of the diesel engine to construct the system's input-output Hankel matrix. Based on the Hankel matrix, it constructs an incremental form prediction equation to predict the system's future output. Finally, it obtains the optimal control law by solving the objective function and applies it to the system. While this patent involves the design of an MPC controller for optimized diesel engine combustion, it employs a purely data-driven modeling method. This model lacks physical information and constraints regarding the in-cylinder working process and cannot perform closed-loop control of the in-cylinder pressure rise rate during injection control. Summary of the Invention

[0008] In view of this, the present invention aims to propose a combustion control method for marine high-pressure common rail diesel engines based on physical information neural networks, in order to solve the problems of neglecting the physical consistency between the model and the actual machine and the poor generalization ability of the model due to simply relying on data-driven methods.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: a combustion control method for marine high-pressure common rail diesel engines based on physical information neural networks, the method comprising:

[0010] Step S1: Determine the cyclic injection pulse width and air-fuel ratio according to the diesel engine operating conditions. Keep the combustion midpoint unchanged by changing the crankshaft angle corresponding to the main injection timing. Obtain the in-cylinder pressure, maximum in-cylinder pressure rise rate, indicated thermal efficiency and emission data for different pre-injection fuel ratios and pre-main injection timing intervals.

[0011] Step S2: Calculate the cyclic variation of the diesel engine's average indicated pressure, and use a data classification algorithm to filter data with a cyclic variation of less than 3% in the average indicated pressure;

[0012] Step S3: Calculate the cumulative heat release and combustion midpoint based on the screened data;

[0013] Step S4: Obtain the influence law of relevant characterization parameters of diesel engine in-cylinder combustion based on the pre-injection fuel ratio, pre-main injection interval and crankshaft angle corresponding to the main injection timing, and determine the crankshaft angle boundary, pre-injection fuel ratio boundary, pre-main injection interval boundary and cylinder pressure maximum pressure rise limit corresponding to the pre-injection timing based on the influence law;

[0014] Step S5: Construct a diesel engine in-cylinder combustion prediction output model based on a physical information neural network. The inputs of the model include: crankshaft angle of the current cycle, total injection duration, pre-injection ratio, crankshaft angle corresponding to the main injection timing, pre-main injection interval, cylinder pressure, and combustion midpoint. The outputs are the predicted values ​​of the maximum in-cylinder pressure rise and combustion midpoint for the next cycle. The model uses a fully connected residual network, and the loss function includes regression loss terms and physical constraint terms.

[0015] Step S6: Construct an MPC controller for a diesel engine in-cylinder combustion prediction output model based on a physical information neural network. With the target value of the combustion midpoint and the predicted value of the pressure rise rate of the next cycle as input, under the constraints of the maximum cylinder pressure rise rate, the pre-main injection interval, and the pre-injection fuel ratio, solve the objective function to obtain the crankshaft angle and the pre-injection fuel ratio corresponding to the main injection timing of the next cycle.

[0016] Step S7: Store and update control parameters and combustion state data in real time. Recursively update the parameters of the PINN-based diesel engine in-cylinder combustion prediction output model every ≥1000 time steps to achieve closed-loop control of in-cylinder combustion.

[0017] Furthermore, a preferred embodiment is proposed, wherein the pre-main spray timing interval in step S1 is:

[0018]

[0019] in, It is the first The crankshaft rotation angle corresponding to the main injection timing of each cycle. It is the first The crankshaft angle corresponding to the pre-injection timing of each cycle. The delay in injector activation corresponds to the angle through which the crankshaft has rotated. It is the crankshaft rotation angle corresponding to the duration of pre-injected fuel.

[0020] Furthermore, a preferred method is proposed, in which a support vector machine is used for data filtering in step S2, and the kernel function of the support vector machine is a Gaussian kernel:

[0021]

[0022] in, For the first The feature vector of each sample For the first k The feature vector of each sample Kernel function parameters.

[0023] Furthermore, a preferred method is proposed, wherein step S3 accumulates the heat released. Through cylinder pressure data calculate:

[0024]

[0025] in, For the first The cumulative heat released in each cycle, The crankshaft rotation angle. For instantaneous cylinder volume, Let be the rate of change of cylinder pressure in the i-th cycle. The convective heat transfer rate of the gas inside the cylinder to the cylinder wall. This refers to the crankshaft angle corresponding to the intake valve closing. The instantaneous rate of change of the cylinder's internal volume.

[0026] Furthermore, a preferred embodiment is proposed, wherein the instantaneous cylinder volume is:

[0027]

[0028] in, Where is the combustion chamber volume, D is the cylinder diameter, and S is the piston stroke.

[0029] Furthermore, a preferred embodiment is proposed, wherein the regression loss term of the loss function in step S5 includes:

[0030]

[0031] in, For the first The crankshaft angle corresponding to 50% of the cumulative heat release in the first cycle, i.e., the first... The midpoint of combustion in each cycle, The first output of the physical information neural network The crankshaft angle corresponding to 50% of the cumulative heat release in the first cycle, i.e., the first... Predicted combustion midpoint for each cycle, No. The first cycle Cylinder pressure data at each sampling point For the first The first cycle Cylinder pressure data at each sampling point The sampling interval is based on the crankshaft angle. The first output of the physical information neural network The first cycle Cylinder pressure data at each sampling point For the first The first cycle The cumulative heat release corresponding to each sampling point For the first The first cycle The cumulative heat release corresponding to each sampling point The first output of the physical information neural network The first cycle The cumulative heat release corresponding to each sampling point The first output of the physical information neural network The first cycle The cumulative heat release corresponding to each sampling point For the first The first cycle The heat exchange between the gas inside the cylinder and the cylinder wall corresponding to each sampling point. For the first The first cycle The wall heat exchange corresponding to each sampling point The first output of the physical information neural network The first cycle The heat exchange between the gas inside the cylinder and the cylinder wall corresponding to each sampling point. The first output of the physical information neural network The first cycle The heat exchange between the gas inside the cylinder and the cylinder wall at each sampling point.

[0032] Furthermore, a preferred embodiment is proposed, wherein the physical constraint term of the loss function in step S5 includes:

[0033]

[0034] in, Cylinder pressure p The predicted physical residual term, The first output of the physical information neural network prediction Cylinder pressure change rate per cycle For specific heat ratio, For the first The instantaneous cylinder volume in each cycle, For the first The instantaneous rate of change of the cylinder's internal volume in each cycle. For the first The predicted cylinder pressure value is generated by a cyclic physical information neural network. For the first The predicted heat release rate output by a cyclic physical information neural network. For the first The convective heat transfer rate of the in-cylinder gas to the cylinder wall is predicted and output by a cyclic physical information neural network. The convective heat transfer rate of the gas inside the cylinder to the cylinder wall The predicted physical residual term, For the first Cylinder wall temperature per cycle For the first The total amount of working fluid in the cylinder during each cycle. For heat release rate The predicted physical residual term, For the first The amount of fuel injected per cycle. For the first The crankshaft angle corresponding to the duration of each combustion cycle Midpoint of combustion The predicted physical residual term, The first output of the physical information neural network prediction 50% of the cumulative heat released in each cycle. The first output of the physical information neural network prediction The crankshaft angle at which 50% of the cumulative heat release occurs in each cycle.

[0035] Furthermore, a preferred method is proposed, wherein the objective function in step S6 is:

[0036]

[0037] in, For the first The target crankshaft angle at which 50% of the cumulative heat release occurs in the first cycle, i.e., the [number]th cycle... The target reference value for the combustion midpoint of each cycle. For the first The crankshaft angle at which 50% of the cumulative heat release occurs in the first cycle, i.e., the first... The midpoint of combustion in each cycle, For the first The crankshaft rotation angle corresponding to the main injection timing of each cycle. For the first The crankshaft angle control increment corresponding to the main injection timing of each cycle. For the first The pre-injection fuel ratio control increment for each cycle For the first The diesel engine state variables are input into a cyclical neural network. For the first The diesel engine control variables are input into the physical information neural network in a loop. This refers to the crankshaft angle of the diesel engine. For physical information neural network prediction of diesel engine first stage The status output of each loop, To control the length of the time domain, To predict the length of the time domain, The current cycle number of the diesel engine ( ), These are the initial state parameters of the diesel engine. For the first One cycle of pre-injection fuel ratio This represents the lower limit of the pre-injected fuel ratio. This represents the upper limit of the pre-injected fuel ratio. For the first Maximum rate of increase in cylinder pressure per cycle This represents the upper limit of the rate of increase in cylinder pressure. This is the lower limit of the pre-main spray interval. For the first The pre-main injection interval of each cycle, This is the upper limit of the pre-main spray interval.

[0038] Based on the same inventive concept, the present invention also proposes a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the MPC combustion control method for marine high-pressure common rail diesel engines based on physical information neural networks as described in any one of the preceding claims.

[0039] Based on the same inventive concept, the present invention also proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the MPC combustion control method for marine high-pressure common rail diesel engines based on a physical information neural network as described above.

[0040] Compared with the prior art, the beneficial effects of the present invention are:

[0041] The MPC combustion control method for marine high-pressure common rail diesel engines proposed in this invention does not solely rely on data-driven modeling. Instead, it combines physical models with diesel engine operating data, avoiding the problems of neglecting the physical consistency between the model and the actual engine and poor model generalization ability caused by relying solely on data-driven methods, as well as the problems of relying on experience-based adjustments in pure physical modeling and ignoring the changes in engine component parameters during long-term use. This invention replaces traditional dynamic numerical integration with a deep fusion of Physical Information Neural Network (PINN) and Model Predictive Control (MPC), achieving coordinated and precise control of the multiple injection processes of marine diesel engines. Compared with traditional data-driven models, the PINN model introduces physical constraint terms such as cylinder pressure differential equation residuals, heat release rate function residuals, and wall heat transfer formula residuals, enabling the predictive model to strictly adhere to the laws of thermodynamic conservation while maintaining data adaptability. Experiments show that: the combustion midpoint control deviation is stable within ±0.5°CA; the maximum in-cylinder pressure rise rate is successfully limited to the safe threshold of 8 bar / °CA; and the timing coordination error between pre-injection and main injection is reduced by 62%. This breakthrough in control precision fundamentally solves the problem of excessive pressure rise rate caused by a single injection under high-load conditions.

[0042] Traditional physical models struggle to track parameter drift caused by engine component aging (such as injector carbon buildup and cylinder wear), while purely data-driven models lack physical consistency, resulting in insufficient generalization ability. This invention proposes an innovative dual-recursive update mechanism for the MPC combustion control method of marine high-pressure common rail diesel engines. It updates the PINN parameters using real-time operating data every ≥1000 working cycles; a physics-data hybrid loss function ensures that model parameter updates do not violate fundamental physical laws; and model training is independently executed by an AI chip (RK3588), avoiding the computing power bottleneck of the main control chip (MPC5744). Actual shipboard testing shows that after 5000 hours of continuous operation, the cylinder pressure curve predicted by the model still maintains a 92.3% overlap with the measured value, solving the industry problem of model-engine lifecycle mismatch in existing technologies.

[0043] The method proposed in this invention combines a Physical Information Neural Network (PINN) with an MPC control method and considers the influence of pressure rise rate during the control process to automatically adjust the fuel injection control signal of the marine diesel engine, thereby achieving the purpose of controlling in-cylinder combustion. This avoids the problem of excessively high maximum pressure rise rate in the cylinder, which can damage the insulating gas boundary layer in the cylinder, leading to increased heat transfer loss and reduced efficiency. The in-cylinder combustion control method of this invention adopts a low-temperature combustion organization form, which can effectively control the in-cylinder combustion characteristic parameters near the desired value and achieve the purpose of reducing emissions. Attached Figure Description

[0044] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0045] Figure 1 This is a flowchart of the diesel engine combustion control method described in this invention;

[0046] Figure 2 This is a schematic diagram illustrating the definition of the control parameters described in this invention;

[0047] Figure 3 This is a block diagram of the combustion control method described in this invention;

[0048] Figure 4 This is a schematic diagram of residual connections in the network structure described in this invention;

[0049] Figure 5 This is a schematic diagram of the PINN network training framework described in this invention;

[0050] Figure 6 This is a block diagram of the adaptive online update method for the prediction model described in this invention;

[0051] Figure 7 This is a schematic diagram of the MPC combustion controller structure for marine diesel engines based on PINN as described in this invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of the present invention can be combined with each other, and the described embodiments are only some embodiments of the present invention, not all embodiments.

[0053] Implementation Method 1: The MPC combustion control method for marine high-pressure common rail diesel engines based on physical information neural networks described in this implementation method includes:

[0054] Step S1: Determine the cyclic injection pulse width and air-fuel ratio according to the diesel engine operating conditions. Keep the combustion midpoint unchanged by changing the crankshaft angle corresponding to the main injection timing. Obtain the in-cylinder pressure, maximum in-cylinder pressure rise rate, indicated thermal efficiency and emission data for different pre-injection fuel ratios and pre-main injection timing intervals.

[0055] Step S2: Calculate the cyclic variation of the diesel engine's average indicated pressure, and use a data classification algorithm to filter data with a cyclic variation of less than 3% in the average indicated pressure;

[0056] Step S3: Calculate the cumulative heat release and combustion midpoint based on the screened data;

[0057] Step S4: Obtain the influence law of relevant characterization parameters of diesel engine in-cylinder combustion based on the pre-injection fuel ratio, pre-main injection interval and crankshaft angle corresponding to the main injection timing, and determine the crankshaft angle boundary, pre-injection fuel ratio boundary, pre-main injection interval boundary and cylinder pressure maximum pressure rise limit corresponding to the pre-injection timing based on the influence law;

[0058] Step S5: Construct a diesel engine in-cylinder combustion prediction output model based on a physical information neural network. The inputs of the model include: crankshaft angle of the current cycle, total injection duration, pre-injection ratio, crankshaft angle corresponding to the main injection timing, pre-main injection interval, cylinder pressure, and combustion midpoint. The outputs are the predicted values ​​of the maximum in-cylinder pressure rise and combustion midpoint for the next cycle. The model uses a fully connected residual network, and the loss function includes regression loss terms and physical constraint terms.

[0059] Step S6: Construct an MPC controller for a diesel engine in-cylinder combustion prediction output model based on a physical information neural network. With the target value of the combustion midpoint and the predicted value of the pressure rise rate of the next cycle as input, under the constraints of the maximum cylinder pressure rise rate, the pre-main injection interval, and the pre-injection fuel ratio, solve the objective function to obtain the crankshaft angle and the pre-injection fuel ratio corresponding to the main injection timing of the next cycle.

[0060] Step S7: Store and update control parameters and combustion state data in real time. Recursively update the parameters of the PINN-based diesel engine in-cylinder combustion prediction output model every ≥1000 time steps to achieve closed-loop control of in-cylinder combustion.

[0061] The MPC combustion control method for marine high-pressure common rail diesel engines proposed in this embodiment does not solely rely on data-driven modeling. Instead, it combines the physical model with diesel engine operating data, avoiding the problems of neglecting the physical consistency between the model and the actual engine, poor model generalization ability, and reliance on empirical adjustments in pure physical modeling, which ignores changes in engine component parameters over long-term use. This embodiment replaces traditional dynamic numerical integration with a deep fusion of Physical Information Neural Network (PINN) and Model Predictive Control (MPC), achieving coordinated and precise control of the multiple injection processes of marine diesel engines. Compared to traditional data-driven models, the PINN model introduces physical constraint terms such as cylinder pressure differential equation residuals, heat release rate function residuals, and wall heat transfer formula residuals, ensuring that the predictive model strictly adheres to thermodynamic conservation laws while maintaining data adaptability. Experiments show that the combustion midpoint control deviation is stable within ±0.5°CA; the maximum in-cylinder pressure rise rate is successfully limited to the safe threshold of 8 bar / °CA; and the timing coordination error between pre-injection and main injection is reduced by 62%. This breakthrough in control precision fundamentally solves the problem of excessive pressure rise rate caused by a single injection under high-load conditions.

[0062] Traditional physical models struggle to track parameter drift caused by engine component aging (such as injector carbon buildup and cylinder wear), while purely data-driven models lack physical consistency, resulting in insufficient generalization ability. This implementation proposes an innovative dual-recursive update mechanism for the MPC combustion control method of marine high-pressure common rail diesel engines. It updates the PINN parameters using real-time operating data every ≥1000 working cycles; a physics-data hybrid loss function ensures that model parameter updates do not violate fundamental physical laws; and model training is independently performed by an AI chip (RK3588), avoiding the computing power bottleneck of the main control chip (MPC5744). Actual shipboard testing shows that after 5000 hours of continuous operation, the cylinder pressure curve predicted by the model still maintains a 92.3% overlap with the measured value, solving the industry problem of model-engine lifecycle mismatch in existing technologies.

[0063] This embodiment combines a Physical Information Neural Network (PINN) with the MPC control method and considers the influence of pressure rise rate during the control process to automatically adjust the fuel injection control signal of the marine diesel engine, thereby achieving the purpose of controlling in-cylinder combustion. This avoids excessively high maximum pressure rise rate in the cylinder, which can damage the insulating gas boundary layer in the cylinder, leading to increased heat transfer loss and reduced efficiency. The in-cylinder combustion control method in this embodiment adopts a low-temperature combustion organization form, which can effectively control the in-cylinder combustion characteristic parameters to near the desired value and achieve the purpose of reducing emissions.

[0064] Implementation Method Two: This implementation method further defines the MPC combustion control method for marine high-pressure common rail diesel engines based on physical information neural networks described in Implementation Method One. In step S1, the pre-main injection timing interval is:

[0065]

[0066] in, It is the first The crankshaft angle corresponding to the main injection timing of the first cycle, the first cycle A cycle It is the crankshaft rotation angle corresponding to the pre-injection timing. The delay in injector activation corresponds to the angle through which the crankshaft has rotated. It is the crankshaft rotation angle corresponding to the duration of pre-injected fuel.

[0067] Implementation Method 3: This implementation method further defines the MPC combustion control method for marine high-pressure common rail diesel engines based on physical information neural networks described in Implementation Method 1. In step S2, a support vector machine is used for data filtering, and the kernel function of the support vector machine is a Gaussian kernel.

[0068]

[0069] in, For the first The feature vector of each sample For the first k The feature vector of each sample Kernel function parameters.

[0070] Implementation Method Four: This implementation method further defines the MPC combustion control method for marine high-pressure common rail diesel engines based on physical information neural networks described in Implementation Method One. The heat released in step S3 is accumulated. Through cylinder pressure data calculate:

[0071]

[0072] in, For the first The cumulative heat released in each cycle, The crankshaft rotation angle. For instantaneous cylinder volume, Let be the rate of change of cylinder pressure in the i-th cycle. The convective heat transfer rate of the gas inside the cylinder to the cylinder wall. This refers to the crankshaft angle corresponding to the intake valve closing. The instantaneous rate of change of the cylinder's internal volume.

[0073] Implementation Method 5: This implementation method further defines the MPC combustion control method for marine high-pressure common rail diesel engines based on physical information neural networks described in Implementation Method 4. The instantaneous cylinder volume is:

[0074]

[0075] in, Where is the combustion chamber volume, D is the cylinder diameter, and S is the piston stroke.

[0076] Implementation Method Six: This implementation method further defines the MPC combustion control method for marine high-pressure common rail diesel engines based on physical information neural networks described in Implementation Method One. The regression loss term of the loss function in step S5 includes:

[0077]

[0078] in, For the first The crankshaft angle corresponding to 50% of the cumulative heat release in the first cycle, i.e., the first... The midpoint of combustion in each cycle, The first output of the physical information neural network The crankshaft angle corresponding to 50% of the cumulative heat release in the first cycle, i.e., the first... Predicted combustion midpoint for each cycle, For the first The first cycle Cylinder pressure data at each sampling point For the first The first cycle Cylinder pressure data at each sampling point The sampling interval is based on the crankshaft angle. The first output of the physical information neural network The first cycle Cylinder pressure data at each sampling point For the first The first cycle The cumulative heat release corresponding to each sampling point For the first The first cycle The cumulative heat release corresponding to each sampling point The first output of the physical information neural network The first cycle The cumulative heat release corresponding to each sampling point The first output of the physical information neural network The first cycle The cumulative heat release corresponding to each sampling point For the first The first cycle The heat exchange between the gas inside the cylinder and the cylinder wall corresponding to each sampling point. For the first The first cycle The wall heat exchange corresponding to each sampling point The first output of the physical information neural network The first cycle The heat exchange between the gas inside the cylinder and the cylinder wall corresponding to each sampling point. The first output of the physical information neural network The first cycle The heat exchange between the gas inside the cylinder and the cylinder wall at each sampling point.

[0079] Implementation Method Seven: This implementation method further defines the MPC combustion control method for marine high-pressure common rail diesel engines based on physical information neural networks described in Implementation Method One. The physical constraint terms of the loss function in step S5 include:

[0080]

[0081] in, Cylinder pressure p The predicted physical residual term, The first output of the physical information neural network prediction Cylinder pressure change rate per cycle For specific heat ratio, For the first The instantaneous cylinder volume in each cycle, For the first The instantaneous rate of change of the cylinder's internal volume in each cycle. For the first The predicted cylinder pressure value is generated by a cyclic physical information neural network. For the first The predicted heat release rate output by a cyclic physical information neural network. For the first The convective heat transfer rate of the in-cylinder gas to the cylinder wall is predicted and output by a cyclic physical information neural network. The convective heat transfer rate of the gas inside the cylinder to the cylinder wall The predicted physical residual term, For the first Cylinder wall temperature per cycle For the first The total amount of working fluid in the cylinder during each cycle. For heat release rate The predicted physical residual term, For the first The amount of fuel injected per cycle. For the first The crankshaft angle corresponding to the duration of each combustion cycle Midpoint of combustion The predicted physical residual term, The first output of the physical information neural network prediction 50% of the cumulative heat released in each cycle. The first output of the physical information neural network prediction The crankshaft angle at which 50% of the cumulative heat release occurs in each cycle.

[0082] Implementation Method Eight: This implementation method further defines the MPC combustion control method for marine high-pressure common rail diesel engines based on physical information neural networks described in Implementation Method One. The objective function in step S6 is:

[0083]

[0084] in, For the first The target crankshaft angle at which 50% of the cumulative heat release occurs in the first cycle, i.e., the [number]th cycle... The target reference value for the combustion midpoint of each cycle. For the first The crankshaft angle at which 50% of the cumulative heat release occurs in the first cycle, i.e., the first... The midpoint of combustion in each cycle, For the first The crankshaft rotation angle corresponding to the main injection timing of each cycle. For the first The control increment of crankshaft angle corresponding to the main injection timing of each cycle. For the first The pre-injection fuel ratio control increment for each cycle For the first The diesel engine state variables are input into a cyclical neural network. For the first The diesel engine control variables are input into the physical information neural network in a loop. This provides crankshaft angle information for the diesel engine. For physical information neural network prediction of diesel engine first stage The status output of each loop, To control the length of the time domain, To predict the length of the time domain, The current cycle number of the diesel engine ( ), These are the initial state parameters of the diesel engine. For the first The pre-injection fuel ratio for each cycle, This represents the lower limit of the pre-injected fuel ratio. This represents the upper limit of the pre-injected fuel ratio. For the first Maximum rate of increase in cylinder pressure per cycle This represents the upper limit of the rate of increase in cylinder pressure. This is the lower limit of the pre-main spray interval. For the first The pre-main injection interval of each cycle, This is the upper limit of the pre-main spray interval.

[0085] Implementation Method Nine: A computer device according to this implementation method includes a memory and a processor. The memory stores a computer program. When the processor runs the computer program stored in the memory, the processor executes the MPC combustion control method for marine high-pressure common rail diesel engines based on physical information neural networks as described in any one of Implementation Methods One to Eight.

[0086] Implementation Method 10: A computer-readable storage medium according to this implementation method stores a computer program, which, when executed by a processor, performs the steps of the MPC combustion control method for marine high-pressure common rail diesel engines based on a physical information neural network as described in any one of Implementation Methods 1 to 8.

[0087] Implementation Method 11, see below Figures 1 to 7 This embodiment describes a specific example of the MPC combustion control method for marine high-pressure common rail diesel engines based on physical information neural networks described in Embodiment 1. It also serves to explain Embodiments 2 through 8. Specifically:

[0088] Step S100: Determine the cyclic injection pulse width based on the diesel engine operating conditions. air-fuel ratio (where i represents the loop number, In experiments or calibrated model simulations, the crankshaft angle corresponding to the main injection timing is changed. Maintain the midpoint of combustion With the same conditions, different pre-spray ratios and pre-main injection timing interval The cylinder pressure data is obtained by using a cylinder pressure sensor. (where j represents the interval based on crankshaft rotation angle) The j-th cylinder pressure data point obtained by triggering sampling in the i-th cycle, ), maximum cylinder pressure rise rate Indicating thermal efficiency NOx generation Unburned hydrocarbon generation Carbon soot particulate matter generation data;

[0089] Step S101: Cylinder pressure data obtained in step S100 Calculate the cyclic variation of the diesel engine's mean indicated pressure. The data was classified and labeled using Support Vector Machine (SVM), and the cyclic variation of diesel engine IMEP was filtered and retained. Data;

[0090] Step S102: Using the remaining diesel engine working cycle cylinder pressure data retained in step S101 Calculate the cumulative heat release And based on the cumulative heat release Data calculation of combustion midpoint ;

[0091] Step S103: Determine the pre-injection fuel ratio Pre-main injection timing interval Crankshaft rotation angle corresponding to main injection timing Based on the data retained and filtered in step S101, as well as the influence of cumulative heat release and combustion midpoint on in-cylinder combustion-related characterization parameters, the crankshaft angle corresponding to the pre-injection timing is used to determine the appropriate parameters. Indicative thermal efficiency The influence of emission data determines the crankshaft rotation angle corresponding to the pre-injection timing. Controllable upper and lower boundary range ,according to , right , Determining the impact patterns of emission data and Upper and lower boundaries of the controllable range , and upper boundary ;

[0092] Step S104: Using the diesel engine crankshaft angle The current number Total injection duration of each cycle Pre-injection fuel ratio Crankshaft rotation angle corresponding to main injection timing Pre-main injection timing interval Cylinder pressure data points Midpoint of combustion As model input, the first Predicted cylinder pressure value for each cycle Predicted value of combustion midpoint Predicted cumulative heat release value Predicted heat transfer value of wall The PINN model, used as the model output, is used to build the diesel engine prediction output model. This includes the network framework structure and the total loss function. The data retained after step S102 is divided and packaged to obtain the PINN model training dataset. and verification set ,use and The model is trained and tested, and the training framework is as follows: Figure 2 As shown;

[0093] Step S105: Differentiate the output of the physical information prediction model. Cylinder pressure change rate per cycle Rear-embedded maximum cylinder pressure rise rate The identification module performs post-processing to output the maximum cylinder pressure rise rate for the next cycle. The output of the PINN combustion prediction model and For input, the crankshaft angle corresponding to the main injection timing of the next working cycle. and pre-spray ratio To construct a PINN-based diesel engine MPC combustion controller, its structure is as follows: Figure 3 As shown. Used to control the combustion midpoint , Mainly used to control the maximum pressure rise rate The crankshaft angle corresponding to the pre-injection timing. The value is set within the boundary range determined in step S103, and fine-tuned to a fixed value based on the control effect to meet the control requirements. The cyclic oil supply pulse width is obtained by querying the MAP based on the torque and speed requirements. Inputting into the PINN prediction model, the PINN combustion prediction model predicts the next cycle based on the feedback parameters from the previous cycle. , Provided to the combustion controller, the controller solves online with... , , The objective function of the constraint conditions is used to obtain the crankshaft angle corresponding to the main injection timing in the next cycle. Pre-injection fuel ratio The control quantity enables control over the pressure rise rate and the combustion midpoint.

[0094] Furthermore, such as Figure 4 As shown, in step S100, the total injection pulse width corresponds to the injection quantity required for the current operating condition. ,in, It is the pulse width of pre-injected fuel. It refers to the main injection fuel pulse width; pre-injection ratio; and pre-main injection timing interval. , It is the crankshaft rotation angle corresponding to the main injection timing. It is the crankshaft rotation angle corresponding to the pre-injection timing. The delay in injector activation corresponds to the angle through which the crankshaft has rotated. It can be calibrated by testing; indicating thermal efficiency. The controller can be based on Discrete computation For the current number A cyclic cylinder pressure dataset, For the current number The first cycle Cylinder pressure values ​​at each sampling point For the current number The first cycle Cylinder pressure values ​​at each sampling point For the current number The first cycle The volume of the cylinder corresponding to each sampling point For the current number The first cycle The corresponding cylinder volume for each sampling point This refers to the crankshaft angle corresponding to when the exhaust valve opens. This is the crankshaft angle corresponding to when the intake valve is closed. For the current number The amount of fuel injected per cycle. The lower calorific value of diesel fuel; maximum cylinder pressure rise rate. To ensure calculation accuracy, the resolution of crankshaft angle trigger sampling should be as high as possible. ;

[0095] Furthermore, in step S101, can be calculate, The standard deviation of IMEP , This is the average value of IMEP.

[0096] Furthermore, in step S101, the Support Vector Machine (SVM) is constructed as a non-linear classification problem, mapping the data in the input space X to the feature space F: , , For feature mapping function, For the first in the input space One element, For weight parameters, The input is the prediction fitting function; the classifier is defined as follows: The Lagrange dual objective function is constructed as follows: , , For Lagrange multipliers, , The label value of the sample. Original sample The feature vector is mapped to a high-dimensional space by the feature mapping function. Original sample The feature vector, after being mapped to a high-dimensional space by the feature mapping function, is constrained as follows: The solution obtained by solving the optimization problem using the quadratic programming method is: ,in , and These are the Lagrange multiplier, bias, and class label, respectively. It is the upper boundary of the Lagrange multipliers. The characteristic space transformation and inner product calculation use the Gaussian kernel function. ;

[0097] Furthermore, in step S102, the accumulated heat release The cylinder pressure data can be obtained through step S100 calculate:

[0098]

[0099] Among them, specific heat ratio ;

[0100] Instantaneous cylinder volume

[0101] instantaneous rate of change of cylinder volume

[0102] Where D is the combustion chamber volume, D is the cylinder diameter, and S is the piston stroke; the convective heat transfer rate of the gas inside the cylinder to the cylinder wall. , For diesel engine speed, heat transfer coefficient Determined using the Woschni formula B is the cylinder diameter, T wi The cylinder wall temperature and the average gas flow rate inside the cylinder are given. S p It is the average piston speed. The cylinder pressure during the reverse drag cycle of the diesel engine is C1, C2 are constants, and the cylinder temperature is... , It is the average relative molecular mass of the air and combustion products in the cylinder, and R is the gas constant. It is the total amount of the working fluid in the cylinder;

[0103] Furthermore, in step S102, the combustion initiation point It is the crankshaft angle corresponding to 10% of the cumulative heat release, the combustion midpoint. It is the crankshaft angle corresponding to 50% of the cumulative heat release, the combustion initiation point. Midpoint of combustion It can be determined by the cumulative heat release Integral solution calculation:

[0104]

[0105] This refers to the cumulative heat released from the start of combustion until the exhaust valve opens. From the occurrence of combustion to the crankshaft angle The cumulative heat released so far, This is the theoretical value of the total heat release. The crankshaft angle for sustained combustion. The combustion initiation angle, , These are adjustable parameters;

[0106] Preferably, in step S103, a multi-objective optimization algorithm can be used to construct a multi-objective function based on the diesel engine performance indicators, including emissions, thermal efficiency, maximum pressure rise rate, etc., and solve for the design variables that meet the constraints. , , The Pareto front can be used to determine the upper and lower boundaries of the controllable range of design variables based on the non-dominated solution set.

[0107] Preferably, in step S103, the determined boundary can be used to further define the input space of the PINN model;

[0108] Further, in step S104, the training dataset and validation set All results were obtained through experiments or calibration model simulations.

[0109]

[0110] Each sample data point consists of an input vector. and output vector The set consists of two parts, where a and b are the number of samples in the training set and validation set, respectively.

[0111] Furthermore, in step S104, the PINN model employs a fully connected residual neural network, such as... Figure 5 As shown, this avoids gradient vanishing or gradient explosion due to different magnitudes of loss functions;

[0112] Furthermore, in step S104, the total loss function Includes: regression loss term and physical constraints Total loss function , and These are the weights of the regression loss term and the physical constraint term, respectively.

[0113] Regression loss term:

[0114]

[0115] In the formula, the first term inside the square brackets is the combustion midpoint. The regression loss, the second term is the cylinder pressure change rate. The regression loss, the third term is the heat release rate. The regression loss, the fourth term is the wall heat transfer rate. The regression loss.

[0116] Physical residuals:

[0117]

[0118] In the formula, the first term inside the square brackets is the cylinder pressure. p The predicted physical residual term, The first output of the physical information neural network prediction Cylinder pressure change rate per cycle For specific heat ratio, For the first The instantaneous cylinder volume in each cycle, For the first The instantaneous rate of change of the cylinder's internal volume in each cycle. For the first The predicted cylinder pressure value is generated by a cyclic physical information neural network. For the first The predicted heat release rate output by a cyclic physical information neural network. For the first The convective heat transfer rate of the in-cylinder gas to the cylinder wall is predicted and output by a cyclic physical information neural network. The second term in square brackets is the convective heat transfer rate of the in-cylinder gas to the cylinder wall. The predicted physical residual term, For the first Cylinder wall temperature per cycle For the first The total amount of working fluid in the cylinder during each cycle. The third item in square brackets is the heat release rate. The predicted physical residual term, For the first The amount of fuel injected per cycle. For the first The crankshaft angle corresponding to the duration of each combustion cycle is shown in square brackets; the fourth item is the combustion midpoint. The predicted physical residual term, The first output of the physical information neural network prediction 50% of the cumulative heat released in each cycle. The first output of the physical information neural network prediction The crankshaft angle at which 50% of the cumulative heat release occurs in each cycle.

[0119] Furthermore, in step S105, the objective function is:

[0120]

[0121] In the formula, It controls the length of the time domain. It predicts the length of the time domain. , , , , These are the weights of each term in the objective function. It is the PINN prediction model. It is an increment of injection timing control. It is an increase in the pre-injection fuel ratio control;

[0122] Furthermore, in step S106, the controller hardware employing this control method includes a main control chip MPC5744 and an AI edge computing chip RK3588. The main control chip MPC5744 is primarily responsible for acquiring and processing signals such as engine speed, temperature, and cylinder pressure, as well as calculating and outputting control quantities such as injection timing, injection quantity, and intake valve opening. The AI ​​edge computing chip RK3588 is primarily responsible for training and updating the PINN prediction model network structure parameters and for combustion prediction, thus avoiding the problem of insufficient computing power of a single chip. Its structural diagram is shown below. Figure 7 As shown.

[0123] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0124] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0125] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0126] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this disclosure and not to limit its protection scope. Although this disclosure has been described in detail with reference to the above embodiments, those skilled in the art should understand that after reading this disclosure, they can still make various changes, modifications or equivalent substitutions to the specific implementation of the invention, but these changes, modifications or equivalent substitutions are all within the protection scope of the published pending claims.

Claims

1. A combustion control method for marine high-pressure common rail diesel engines based on physical information neural networks, characterized in that, The method includes: Step S1: Determine the cyclic injection pulse width and air-fuel ratio according to the diesel engine operating conditions. Keep the combustion midpoint unchanged by changing the crankshaft angle corresponding to the main injection timing. Obtain the in-cylinder pressure, maximum in-cylinder pressure rise rate, indicated thermal efficiency and emission data for different pre-injection fuel ratios and pre-main injection timing intervals. Step S2: Calculate the cyclic variation of the diesel engine's average indicated pressure, and use a data classification algorithm to filter data with a cyclic variation of less than 3% in the average indicated pressure; Step S3: Calculate the cumulative heat release and combustion midpoint based on the screened data; Step S4: Obtain the influence law of relevant characterization parameters of diesel engine in-cylinder combustion based on the pre-injection fuel ratio, pre-main injection interval and crankshaft angle corresponding to the main injection timing, and determine the crankshaft angle boundary, pre-injection fuel ratio boundary, pre-main injection interval boundary and cylinder pressure maximum pressure rise limit corresponding to the pre-injection timing based on the influence law; Step S5: Construct a diesel engine in-cylinder combustion prediction output model based on a physical information neural network. The inputs of the model include: crankshaft angle of the current cycle, total injection duration, pre-injection ratio, crankshaft angle corresponding to the main injection timing, pre-main injection interval, cylinder pressure, and combustion midpoint. The outputs are the predicted values ​​of the maximum in-cylinder pressure rise and combustion midpoint for the next cycle. The model uses a fully connected residual network, and the loss function includes regression loss terms and physical constraint terms. Step S6: Construct an MPC controller for a diesel engine in-cylinder combustion prediction output model based on a physical information neural network. With the target value of the combustion midpoint and the predicted value of the pressure rise rate of the next cycle as input, under the constraints of the maximum cylinder pressure rise rate, the pre-main injection interval, and the pre-injection fuel ratio, solve the objective function to obtain the crankshaft angle and the pre-injection fuel ratio corresponding to the main injection timing of the next cycle. Step S7: Store and update control parameters and combustion state data in real time. Recursively update the parameters of the PINN-based diesel engine in-cylinder combustion prediction output model every ≥1000 time steps to achieve closed-loop control of in-cylinder combustion.

2. The MPC combustion control method for marine high-pressure common rail diesel engines based on physical information neural networks according to claim 1, characterized in that, The pre-main injection timing interval in step S1 is: in, It is the first The crankshaft rotation angle corresponding to the main injection timing of each cycle. It is the first The crankshaft angle corresponding to the pre-injection timing of each cycle. The delay in injector activation corresponds to the angle through which the crankshaft has rotated. It is the crankshaft rotation angle corresponding to the duration of pre-injected fuel.

3. The MPC combustion control method for marine high-pressure common rail diesel engines based on physical information neural networks according to claim 1, characterized in that, In step S2, a support vector machine is used for data filtering. The kernel function of the support vector machine is a Gaussian kernel. in, For the first The feature vector of each sample For the first k The feature vector of each sample Kernel function parameters.

4. The MPC combustion control method for marine high-pressure common rail diesel engines based on physical information neural networks according to claim 1, characterized in that, The cumulative heat release rate in step S3 Through cylinder pressure data calculate: in, For the first The cumulative heat released in each cycle, This refers to the crankshaft rotation angle. For instantaneous cylinder volume, Let be the rate of change of cylinder pressure in the i-th cycle. The convective heat transfer rate of the gas inside the cylinder to the cylinder wall. This refers to the crankshaft angle corresponding to the intake valve closing. The instantaneous rate of change of the cylinder's internal volume.

5. The MPC combustion control method for marine high-pressure common rail diesel engines based on physical information neural networks according to claim 4, characterized in that, The instantaneous cylinder volume is: in, Where is the combustion chamber volume, D is the cylinder diameter, and S is the piston stroke.

6. The MPC combustion control method for marine high-pressure common rail diesel engines based on physical information neural networks according to claim 1, characterized in that, The regression loss term of the loss function in step S5 includes: in, For the first The crankshaft angle corresponding to 50% of the cumulative heat release in one cycle. The first output of the physical information neural network The crankshaft angle corresponding to 50% of the cumulative heat release in one cycle. No. The first cycle Cylinder pressure data at each sampling point For the first The first cycle Cylinder pressure data at each sampling point The sampling interval is based on the crankshaft angle. The first output of the physical information neural network The first cycle Cylinder pressure data at each sampling point For the first The first cycle The cumulative heat release corresponding to each sampling point For the first The first cycle The cumulative heat release corresponding to each sampling point The first output of the physical information neural network The first cycle The cumulative heat release corresponding to each sampling point The first output of the physical information neural network The first cycle The cumulative heat release corresponding to each sampling point For the first The first cycle The heat exchange between the gas inside the cylinder and the cylinder wall corresponding to each sampling point. For the first The first cycle The wall heat exchange corresponding to each sampling point The first output of the physical information neural network The first cycle The heat exchange between the gas inside the cylinder and the cylinder wall corresponding to each sampling point. The first output of the physical information neural network The first cycle The heat exchange between the gas inside the cylinder and the cylinder wall at each sampling point.

7. The MPC combustion control method for marine high-pressure common rail diesel engines based on physical information neural networks according to claim 1, characterized in that, The physical constraints of the loss function in step S5 include: in, Cylinder pressure p The predicted physical residual term, The first output of the physical information neural network prediction Cylinder pressure change rate per cycle For specific heat ratio, For the first The instantaneous cylinder volume in each cycle, For the first The instantaneous rate of change of the cylinder's internal volume in each cycle. For the first The predicted cylinder pressure value is generated by a cyclic physical information neural network. For the first The predicted heat release rate output by a cyclic physical information neural network. For the first The convective heat transfer rate of the in-cylinder gas to the cylinder wall is predicted and output by a cyclic physical information neural network. The convective heat transfer rate of the gas inside the cylinder to the cylinder wall The predicted physical residual term, For the first Cylinder wall temperature per cycle For the first The total amount of working fluid in the cylinder during each cycle. For heat release rate The predicted physical residual term, For the first The amount of fuel injected per cycle. For the first The crankshaft angle corresponding to the duration of each combustion cycle Midpoint of combustion The predicted physical residual term, The first output of the physical information neural network prediction 50% of the cumulative heat released in each cycle. The first output of the physical information neural network prediction The crankshaft angle at which 50% of the cumulative heat release occurs in each cycle.

8. The MPC combustion control method for marine high-pressure common rail diesel engines based on physical information neural networks according to claim 1, characterized in that, The objective function in step S6 is: in, For the first The target crankshaft angle at which 50% of the cumulative heat release occurs in each cycle. For the first The crankshaft angle at which 50% of the cumulative heat release occurs in each cycle. For the first The crankshaft rotation angle corresponding to the main injection timing of each cycle. For the first The crankshaft angle control increment corresponding to the main injection timing of each cycle. For the first The pre-injection fuel ratio control increment for each cycle For the first The diesel engine state variables are input into a cyclical neural network. For the first The diesel engine control variables are input into the physical information neural network in a loop. This refers to the crankshaft angle of the diesel engine. For physical information neural network prediction of diesel engine first stage The status output of each loop, To control the length of the time domain, To predict the length of the time domain, This is the current cycle number for the diesel engine. These are the initial state parameters of the diesel engine. For the first One cycle of pre-injection fuel ratio This represents the lower limit of the pre-injected fuel ratio. This represents the upper limit of the pre-injected fuel ratio. For the first Maximum rate of increase in cylinder pressure per cycle This represents the upper limit of the rate of increase in cylinder pressure. This is the lower limit of the pre-main spray interval. For the first The pre-main injection interval of each cycle, This is the upper limit of the pre-main spray interval.

9. A computer device, characterized in that: It includes a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the MPC combustion control method for marine high-pressure common rail diesel engines based on a physical information neural network as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the MPC combustion control method for marine high-pressure common rail diesel engines based on a physical information neural network as described in any one of claims 1-8.

Images