Emission prediction and optimization method for digital twin-driven marine diesel engine

Abstract

The invention discloses a digital twin-driven marine diesel engine emission prediction and optimization method, which comprises the following steps: (1) constructing a comprehensive performance mechanism model, and obtaining twin data through representation and verification of the comprehensive performance mechanism model; (2) designing a calibration module based on a deep learning algorithm, taking real data and twin data as input, and outputting calibrated data through model training; (3) constructing a combustion stage mechanism module, and optimizing a gas inlet pipeline structure and improving mixing uniformity through simulation of a combustion mechanism model on oil-gas mixing characteristics and combustion efficiency; (4) constructing a data driving model, and fitting a relationship between the turbulent energy index and emission; and (5) taking the constructed comprehensive performance mechanism model, the calibration module, the combustion stage mechanism model and the data driving model fusion technology system as a fitness function, searching an optimal oil injection control parameter based on a particle swarm optimization algorithm, and formulating an optimization control strategy adapting to different working conditions.

Description

Technical Field

[0001] The present invention relates to digital twin technology, and in particular to a digital twin-driven marine diesel engine emission prediction and optimization method. Background Art

[0002] Under the objective factors of strict international emission standards and rising fuel costs, energy conservation and emission reduction have become important guidelines for the engine industry. With the continuous improvement of my country's energy conservation and emission reduction indicators and the increasingly stringent emission regulations, the demand for new low-emission and high-efficiency internal combustion engines has increased dramatically. In the context of the new energy revolution, new energy technology is in a period of rapid development. There are still some problems in its energy storage, use, endurance, power, stability and reliability that need to be solved urgently. In the next few decades, diesel engines will continue to shine in the fields of engineering machinery and military equipment until the arrival of mature new energy power units. During this period, marine diesel engines need to continuously update their technology to meet the requirements of energy conservation, emission reduction, low carbon and environmental protection in the context of the times.

[0003] With the increasing global environmental awareness and increasingly stringent emission standards, the shipping industry is facing the dual pressure of reducing emissions and improving fuel efficiency. As the main power source of ships, the emission problem of marine diesel engines has become a research hotspot. Traditional emission control methods, such as hardware upgrades and emission aftertreatment technologies, are effective but often costly and difficult to monitor and adjust emissions in real time. Currently in the field of computer data regression analysis. Artificial neural networks can make reasonable predictions on different data after good training and show relatively superior performance. However, general deep learning methods usually require a long time of training and a large amount of historical data. To address this problem, digital twin technology, as an emerging digital means, realizes real-time monitoring, simulation and optimization of complex systems by building virtual models of physical entities. Applying digital twin technology to marine diesel engines can not only accurately predict the emissions of the engine, but also find ways to optimize operating parameters through simulation and data analysis, thereby minimizing emissions. Summary of the invention

[0004] In view of the problems existing in the prior art, the present invention provides a method for predicting and optimizing emissions of marine diesel engines driven by digital twins: First, a comprehensive performance mechanism model is constructed based on Simulink to describe the working principle and internal operation mechanism of the diesel engine, and to obtain the operating parameters of the diesel engine, such as in-cylinder pressure, in-cylinder temperature and CO. The model adopts the idea of ​​decomposition and combination, decomposing the diesel engine system into multiple independent and interconnected subsystems to be more explanatory. Secondly, a calibration module is constructed, and a fusion technology system for calibrating the comprehensive performance mechanism model based on deep learning is proposed. The module has the characteristics of non-intrusive data monitoring. The comprehensive performance mechanism model is calibrated by a deep learning algorithm, which improves the generalization ability of the model and solves a series of problems of traditional sensor monitoring methods. Then, a combustion stage mechanism model is constructed to solve the problem that the comprehensive mechanism model cannot reflect the heat conduction problem of the diesel engine combustion process. The model deeply explores the influence of in-cylinder temperature distribution and oil-gas mixing characteristics on combustion efficiency, and improves the uniformity and flow characteristics of oil-gas mixing by optimizing the intake pipe structure. On this basis, a data-driven model was established, and a deep learning algorithm was further used to fit the relationship between turbulent kinetic energy indicators and emissions to understand the intrinsic relationship between high-fidelity deduction of the combustion stage and emissions. Finally, from the perspective of optimizing the control strategy, with the goal of reducing pollutant emissions, a particle swarm optimization algorithm was used to find a control strategy that minimizes emissions. These strategies are based on the constructed comprehensive performance mechanism model, calibration module, combustion stage mechanism model, and data-driven model fusion technology model as fitness functions to optimize the injection control parameters and achieve the goal of more complete combustion and lower pollutant emissions.

[0005] The technical solution of the present invention is as follows:

[0006] (1) Construct a comprehensive performance mechanism model to describe the working principle and internal operation mechanism of the diesel engine, and obtain the operating parameters of the diesel engine, including the in-cylinder pressure, in-cylinder temperature, CO, etc. during the combustion stage. The present invention adopts the idea of ​​decomposition and combination to decompose the diesel engine system into a series of independent and interconnected subsystems, including a compressor submodule, an intercooler submodule, an intake pipe submodule, a cylinder submodule, an exhaust pipe submodule, a turbine submodule, a supercharger rotor submodule, and a fuel injection pump submodule. The present invention considers the interaction and influence between different subsystems and is more explanatory.

[0007] (2) Construct a calibration module and propose a fusion technology system based on deep learning optimization and calibration of the comprehensive performance mechanism model, which also has the characteristics of non-intrusive data monitoring. The mechanism model is highly interpretable, but due to the fixed formula, it has insufficient generalization ability and cannot cope with complex working conditions. Therefore, the present invention proposes a calibration method that uses a deep learning algorithm to calibrate the comprehensive performance mechanism model, improves the generalization ability of the comprehensive performance mechanism model, and solves the problems of high cost, insufficient economy and short service life of traditional sensor monitoring methods.

[0008] (3) Constructing a combustion stage mechanism model. The main factor for high pollutant emissions is incomplete combustion. In order to solve the problem that the comprehensive mechanism model cannot reflect the heat conduction problem of the diesel engine combustion process, the present invention combines the combustion stage mechanism model, including the intake manifold, resonance cavity, intake manifold, injector, etc., to deeply explore the influence of the temperature distribution in the cylinder and the oil-gas mixing characteristics on the combustion efficiency. First, the output of the calibrated mechanism model, including: injection pressure, inlet and outlet pressure, and cylinder temperature, is used as the inlet and outlet conditions, wall conditions, and initial conditions of the combustion stage mechanism model. Secondly, the turbulent kinetic energy is used as an indicator of the uniformity of the oil-gas mixing, and the intake pipe is structurally optimized to improve the uniformity and flow characteristics of the oil-gas mixing.

[0009] (4) A data-driven model was established to further understand the intrinsic relationship between high-fidelity simulation of the combustion stage and emissions. Based on the above combustion process mechanism model, a deep learning algorithm was further used to fit the relationship between turbulent kinetic energy indicators and emissions.

[0010] (5) Finally, from the perspective of optimizing the control strategy, we seek the optimal strategy to make combustion more complete and pollutant emissions lower. We use the above-mentioned comprehensive performance mechanism model, calibration module, combustion stage mechanism model, and data-driven model fusion technology model as the fitness function. With the goal of minimizing emissions, we design an artificial intelligence optimization algorithm to find the optimal values ​​of injection control parameters such as main injection timing, pre-injection timing, and injection amount under different conditions, optimize the control strategy, and find the control strategy that minimizes emissions.

[0011] Furthermore, the steps of preprocessing the historical data in step (2) are as follows:

[0012] (1-1) Clean the historical data, delete abnormal and duplicate data, and use the difference method to fill in the missing data. The interpolation formula is as follows:

[0013] l t =(l t+1 +l t-1 ) / 2

[0014] In the formula, l t represents the missing value at time t; l t+1is the normal eigenvalue at time t+1; l t-1 is the normal eigenvalue at time t-1. That is, the mean of the time before and after the missing value is taken as the interpolation value.

[0015] (1-2) Use variance filtering method to remove engine operating parameters with variance of 0.

[0016] Furthermore, the parameters of the BiLstm in step (3) are set as follows: the number of layers of the BiLstm model is 3, dropout = 0.5, hiden size = 400, input size = 64. The setting variables of the hidden layer node for the CNN model are: channel = 3, stride = 1, padding = 1, and the activation function is the mean square error.

[0017] (2-1) First, for each time step t, the input gate decides how much new information should be added to the cell state. The output i of the input gate t It is calculated by the following formula:

[0018] i t =σ(w i [h t-1 ,x t ])+b i

[0019] At the same time, the tanh activation function is used to add candidate memory units:

[0020] c' t =tanh(w c [h t-1 ,x t ]+b c )

[0021] Among them, σ is the Sigmoid activation function, w i 、w c i t 、c t The weight of i , b c i t 、c t The bias term, x t is the input sequence at time step t, that is, the input, h t-1 is the hidden state at the previous time step t-1.

[0022] (2-2) The forget gate determines which information in the new state should be forgotten or retained. The output of the forget gate is f t It is calculated by the following formula:

[0023] ft =σ(w f [h t-1 ,x t ]+b f )

[0024] (2-3) Next, use the output i of the input gate t And the forget gate output f t To update the cell state c t :

[0025] c t =f t c t-1 +i t c' t

[0026] (2-4) The output gate is based on h t-1 、x t Calculate output information o t , tanh activation function combined with output gate information o t Get the current hidden layer state

[0027] o t =σ(w o [h t-1 ,x t ]+b o )

[0028] h t =o t *tanh(c t )

[0029] In the formula, w o is the output gate weight; b o is the output gate bias term.

[0030] Hidden layer state h t is transferred to the next time step, and the cell state c t It is also passed to the next step to maintain long-term memory. At this point, the neurons inside the LSTM have completed a calculation process.

[0031] Specifically, the calculation formula of the mean square error loss function is:

[0032]

[0033] In the formula, N is the number of samples, y is the true value, and a is the predicted value. The calculation formula for the predicted value a is:

[0034] a=f(z)=f(w*x+b0

[0035] Where x is the input, w is the weight, b is the bias term, and f is the activation function.

[0036] Furthermore, the state value of the Attention hidden layer in step (3) is:

[0037] e i =tanh(Wh i +b)

[0038]

[0039] Where: W and b are weight and threshold respectively; h i is the hidden unit state value output by CNN, and the score e of each hidden unit state value is obtained i , and normalize the score of the attention mechanism, and finally e i and a i Perform weighted summation to obtain the hidden layer state value c t .

[0040] Furthermore, in the step (4), the standard k-ε model is selected when calculating the turbulence model, and the SIMPLE algorithm is selected when solving the velocity-pressure correlation model.

[0041] Furthermore, in step (5), the particle optimization process:

[0042]

[0043] Where: i is 1, 2....N; N is the total number of particles in the population; t is the number of particle iterations; ω is the weight; v is the speed of each particle in the population; r1, r2 are random numbers; x i is the current position of particle i; is the optimal individual position of particle i at the moment; it is also the optimal position of the particle population at the moment; c1 and c2 are learning factors.

[0044] The beneficial effects of the present invention are:

[0045] Compared with the prior art, the present invention has the following significant advantages: the present invention discloses a method for predicting and optimizing emissions of marine diesel engines driven by a digital twin. The digital twin system consists of a one-dimensional physical model, a calibration module, and a three-dimensional physical model, and a comprehensive digital twin system for diesel engines is constructed. This system can not only accurately simulate the emissions of diesel engines under different operating conditions, but also improve the accuracy and reliability of the model through a non-invasive method of the calibration module, solving the challenges of low accuracy of one-dimensional physical models and heat conduction problems. At the same time, the present invention realizes emission prediction and multi-angle emission optimization based on high-fidelity data produced by digital twins, including optimization of the intake pipe structure and injection timing, which improves combustion efficiency and emission control capabilities. In addition, the online model update is carried out using real-time big data of diesel engines, so that the digital twin model can adapt to various complex working conditions in a timely manner, and improve the accuracy and practicality of emission prediction. In summary, the present invention brings a comprehensive and accurate solution to the field of diesel engine emission prediction and optimization, which has important technical and economic value. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a structural diagram of a calibration module of the present invention;

[0047] Figure 2 The NO of each working condition before and after the optimization of the present invention X Emission experiment diagram;

[0048] Figure 3 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0049] In order to more clearly understand the purpose, features and advantages of the present invention, the present invention is further described below in conjunction with the accompanying drawings and embodiments. In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention can also be implemented in other ways different from those described herein, and therefore, the present invention is not limited to the specific embodiments disclosed below.

[0050] The present invention discloses a digital twin-driven marine diesel engine emission prediction and optimization method, comprising:

[0051] (1) Construct a comprehensive performance mechanism model to describe the working principle and internal operation mechanism of the diesel engine, and obtain the operating parameters of the diesel engine, including the in-cylinder pressure, in-cylinder temperature, CO, etc. during the combustion stage. The present invention adopts the idea of ​​decomposition and combination to decompose the diesel engine system into a series of independent and interconnected subsystems, including a compressor submodule, an intercooler submodule, an intake pipe submodule, a cylinder submodule, an exhaust pipe submodule, a turbine submodule, a supercharger rotor submodule, and a fuel injection pump submodule. The present invention considers the interaction and influence between different subsystems and is more explanatory.

[0052] (1-1) Fuel injection pump submodule

[0053] The fuel injection pump sub-model obtains the fuel injection amount of the fuel injection pump per cylinder cycle according to the diesel engine speed and the fuel injection pump rack position output by the speed regulator. The relationship between the fuel injection pump single cylinder cycle fuel supply amount, the fuel injection pump rack position and the diesel engine speed is obtained according to the speed characteristic curve and load characteristic curve of the fuel injection pump:

[0054] g c =f(F r ,n s )

[0055] where g c It is the fuel supply of the injection pump per cylinder in kg / cyl.

[0056] (1-2) Diesel engine cylinder submodule

[0057] The cylinder charging efficiency is only a function of the diesel engine speed, so its change with speed can be approximately expressed by a semi-empirical formula:

[0058]

[0059] Where n s is the diesel engine speed, in r / min; is the diesel engine speed for determining the optimal timing, in r / min; η v is the cylinder charging efficiency; is the cylinder charging efficiency with optimal timing; B is a constant, which is 0.014 for a four-stroke diesel engine.

[0060] For a four-stroke diesel engine, the air flow through the intake valve can be divided into two parts: the intake volume and the scavenging volume. The charge mass flow q1 (kg / s), the scavenging mass flow q2 (kg / s) and the total air mass flow q a The concept of (kg / s):

[0061] q a =q1+q2

[0062] The quality of the inflation flow can be obtained by the following formula:

[0063]

[0064] where p in Air pressure in the intake pipe, Pa; T in is the average temperature of the air in the intake pipe, K; Ri is the gas constant on the intake side, which is 286.846 J / kg·K; V is the cylinder displacement of the diesel engine, m 3 ; S is the stroke coefficient, for a four-stroke diesel engine, it is 2. The scavenging mass flow rate can be calculated by the following formula:

[0065]

[0066] When p out / p in >0.98, When p out / p in When ≤0.98,

[0067] S q is the average value of the flow cross section during scavenging, m 3 ;k i The constant entropy index of the intake side gas is taken as 1.4; the average exhaust temperature of the engine is calculated according to the first law of thermodynamics of the cylinder process:

[0068] q f H u -L w -N e =(q a +q f )C Pe T out -q a C Pi T in

[0069] where q f The amount of fuel injected into the cylinder per unit time, in kg / s; H u is the lower calorific value of the fuel, in J / kg; L w is the heat taken away by cooling water per unit time, in J / s; N e is the effective power of the diesel engine, in W; C Pe is the constant pressure specific heat capacity of the intake working fluid, in J / (kg·K); C Pi is the constant pressure specific heat capacity of the exhaust medium, in J / (kg·K); T out is the average exhaust temperature in K.

[0070] Diesel engine effective power:

[0071] N e =n e q f H u

[0072] The heat removed by cooling water is:

[0073] L w =δ w q f H u

[0074] Where ne is the estimated effective efficiency of diesel engine; δ w is the cooling loss percentage. Arranging the above three equations, we can get:

[0075]

[0076] Where T cool is the outlet temperature of the intercooler, in K. For diesel engines, The value of usually varies between 0.94 and 1.0, so it can be considered to be approximately equal to 1, so the above can be simplified to:

[0077]

[0078] Engine effective thermal efficiency n e Mainly affected by a / q f The influence of cooling loss δ w Mainly affected by a / q f and engine speed n s Assume the engine exhaust temperature is K T =(H u / C Pe )(1-n e -δ w )=f(q a / q f ,n s ), and finally the average exhaust temperature of the engine is:

[0079]

[0080] (1-3) Exhaust pipe module

[0081] Consider the volume of the exhaust manifold and each exhaust branch pipe as one volume, let the volume be V e (m 3 ), the exhaust gas mass in the volume is Q e (kg), the mass flow rate flowing through the diesel engine exhaust valve is q Exh (kg / s), equal to the total air mass flow rate q a (kg / s), fuel mass flow rate q f The sum of (kg / s) is the mass flow rate q flowing through the supercharger turbine tur (kg / s). In the non-steady state, q Exh With q t The difference makes the mass Q in the intake pipe volume e Changes occur, namely:

[0082]

[0083] The average exhaust temperature from the cylinder into the exhaust pipe is T Exh (K), and then the air is evenly mixed in the exhaust pipe, flows out of the exhaust pipe, and enters the turbine. The temperature of the gas flowing into the turbine (i.e. the average temperature of the air in the exhaust pipe) is T tur (K), from the energy equation and the uniform mixing assumption, we can get:

[0084]

[0085] The temperature T of the exhaust pipe can be obtained from the above three equations: Aout (K):

[0086]

[0087] Changes in mass and temperature in the volume cause changes in pressure. According to the thermodynamic state equation, the exhaust pressure is:

[0088]

[0089] Where T Aout is the average temperature of the exhaust pipe, in K; p Exh is the exhaust pressure of the exhaust pipe, in Pa; R e is the gas constant on the exhaust side, which is 286.354 J / kg·K; k e is the constant entropy index of the exhaust gas, which is taken as 1.33.

[0090] NOx in diesel engines is mainly generated by the reaction of nitrogen and oxygen in the air at high temperatures, mainly including thermal NO, fast NO and fuel NO. Thermal NO is the most important generation mechanism, which can be calculated by the Zeldovich mechanism:

[0091]

[0092] CO is mainly generated by partial oxidation of hydrocarbons during incomplete combustion. Its generation is related to fuel type, air-fuel ratio, and combustion temperature. The generation and consumption of CO can be expressed by the following equilibrium equation:

[0093]

[0094] (2) Calibration module construction, a fusion technology system based on deep learning optimization and calibration of comprehensive performance mechanism model is proposed, which also has the characteristics of non-intrusive data monitoring. The mechanism model is highly interpretable, but due to the fixed formula, the generalization ability is insufficient and it cannot cope with complex working conditions. Therefore, the present invention proposes a calibration method, which uses a deep learning algorithm to calibrate the comprehensive performance mechanism model, improves the generalization ability of the comprehensive performance mechanism model, and solves the problems of high cost, insufficient economy and short service life of traditional sensor monitoring methods.

[0095] (2-1) Use the historical data to clean up, delete abnormal and duplicate data, and use the difference method to fill in the missing data. The interpolation formula is as follows:

[0096] l t =(l t+1 +l t-1 ) / 2

[0097] In the formula, l t represents the missing value at time t; l t+1 is the normal eigenvalue at time t+1; l t-1 is the normal eigenvalue at time t-1. That is, the mean of the time before and after the missing value is taken as the interpolation value.

[0098] (2-2) Use variance filtering method to remove engine operating parameters with variance of 0.

[0099] (2-3) Using CNN-Bilstm-Attention as the calibration module, the neurons of the LSTM long short-term memory neural network will extract and filter the features of the data in the model, thereby improving the effect of the model. The internal calculation process of the neuron is as follows: First, for each time step t, the input gate determines how much new information should be added to the cell state, and the output i of the input gate is t It is calculated by the following formula:

[0100] i t =σ(w i [h t-1 ,x t ])+b i

[0101] At the same time, the tanh activation function is used to add candidate memory units:

[0102] c' t =tanh(w c [h t-1 ,x t ]+b c )

[0103] Among them, σ is the Sigmoid activation function, wi 、w c i t 、c' t The weight of i , b c i t 、c' t The bias term, x t is the input sequence at time step t, that is, the input, h t-1 is the hidden state at the previous time step t-1.

[0104] The forget gate determines which information in the new state should be forgotten or retained. The output f of the forget gate t It is calculated by the following formula:

[0105] f t =σ(w f [h t-1 ,x t ]+b f )

[0106] Next, use the output i of the input gate t And the forget gate output f t To update the cell state c t :

[0107] c t =f t c t-1 +i t c' t

[0108] The output gate is based on h t-1 、x t Calculate output information o t , tanh activation function combined with output gate information o t Get the current hidden layer state

[0109] o t =σ(w o [h t-1 ,x t ]+b o )

[0110] h t =o t *tanh(c t )

[0111] In the formula, w o is the output gate weight; b o is the output gate bias term.

[0112] Hidden layer state h tis transferred to the next time step, and the cell state C t It is also passed to the next step to maintain long-term memory. So far, the neurons of the LSTM neural network have completed a calculation process based on the input.

[0113] Specifically, the calculation formula of the mean square error loss function is:

[0114]

[0115] In the formula, N is the number of samples, y is the true value, and a is the predicted value. The calculation formula for the predicted value a is:

[0116] a=f(z)=f(w*x+b)

[0117] Where x is the input, w is the weight, b is the bias term, and f is the activation function.

[0118] Furthermore, the state value of the Attention hidden layer in step (3) is:

[0119] e i =tanh(Wh i +b)

[0120]

[0121] Where: W and b are weight and threshold respectively; h i is the hidden unit state value output by CNN, and the score e of each hidden unit state value is obtained i , and normalize the score of the attention mechanism, and finally e i and a i Perform weighted summation to obtain the hidden layer state value c t .

[0122] (3) Constructing a combustion stage mechanism model. The main factor for high pollutant emissions is incomplete combustion. In order to solve the problem that the comprehensive mechanism model cannot reflect the heat conduction problem of the diesel engine combustion process, the present invention combines the combustion stage mechanism model, including the intake manifold, resonance cavity, intake manifold, injector, etc., to deeply explore the influence of the temperature distribution in the cylinder and the oil-gas mixing characteristics on the combustion efficiency. First, the output of the calibrated mechanism model, including: injection pressure, inlet and outlet pressure, and cylinder temperature, is used as the inlet and outlet conditions, wall conditions, and initial conditions of the combustion stage mechanism model. Secondly, the turbulent kinetic energy is used as an indicator of the uniformity of the oil-gas mixing, and the intake pipe is structurally optimized to improve the uniformity and flow characteristics of the oil-gas mixing.

[0123] (3-1) Establish a CFD numerical simulation model, including the intake manifold, resonant cavity, intake manifold, injector and other structures.

[0124] (3-2) Intake mesh division: extract the fluid domain of the intake pipe 3D model and divide the fluid domain into tetrahedral meshes. Set the front end of the intake manifold as the fluid inlet and the end of the intake manifold as the fluid outlet. Set the end of the injector as the methanol injection inlet, and the remaining surfaces will be set as the wall of the intake pipe by default. The mesh at the end of the injector will be locally encrypted.

[0125] (3-3) The data in step (3) are used as the inlet and outlet conditions, wall conditions and initial conditions of the combustion stage mechanism model, including: cylinder temperature, inlet and outlet pressure, and injector pressure.

[0126] (3-4) Determination of initialization conditions and boundary conditions. When performing computational fluid dynamics solutions, appropriate initial and boundary conditions must be established. The selection of appropriate boundary conditions requires consideration of the physical properties of the fluid, the calculation range, and the control equations. This can make the fluid calculation results more consistent with the actual situation. In fluid simulation calculations, there are a variety of inlet and outlet boundary conditions to choose from. If the pressure, flow rate, and mass of the fluid are known, the inlet (outlet) can be set to the corresponding pressure, velocity, and mass boundary conditions. During the simulation process, the inlet uses mass and pressure boundaries, and the outlet uses pressure boundaries. At the same time, the wall is set to be smooth and frictionless, and the standard wall function method is used for solution.

[0127] (3-5) Structural optimization, analyze the flow characteristics in the intake pipe. When the injector is not spraying fuel, all intake manifolds are pressure outlets, and the initial pressure is atmospheric pressure. Analyze the oil-gas mixing characteristics, and use the maximum turbulent kinetic energy of the intake manifold as an indicator of the uniformity of the oil-gas mixing. Adjust the length of the intake manifold in the vertical direction.

[0128] (3-6) Optimization verification. After the optimization of the above steps (4-5) is completed, the mixing characteristics analysis is performed again to check the maximum turbulent kinetic energy and turbulent kinetic energy cloud map of the intake manifold to verify the effectiveness of the structural optimization.

[0129] (4) Construct a data-driven model to further understand the intrinsic relationship between high-fidelity simulation of the combustion stage and emissions. Based on the above combustion process mechanism model, a deep learning algorithm is further used to fit the relationship between turbulent kinetic energy indicators and emissions.

[0130] A convolutional neural network (CNN) module is established to extract the maximum value of turbulent kinetic energy and the characteristics of operating data. Through the CNN module, the maximum value of turbulent kinetic energy and characteristic information related to the operating conditions can be automatically extracted from the input data. And the relationship between the maximum value of turbulent kinetic energy and operating information and emission data is predicted. Then the model is trained, and the back propagation algorithm and optimizer are used to adjust the parameters of the model to minimize the loss function MAE.

[0131]

[0132] (5) From the perspective of optimizing the control strategy, find the optimal strategy to make the combustion more complete and the pollutant emission lower. Use the above-mentioned comprehensive performance mechanism model, calibration module, combustion stage mechanism model, and data-driven model fusion technology model as the fitness function. With the goal of minimizing emissions, use the particle swarm optimization algorithm to find the optimal values ​​of injection control parameters such as main injection timing, pre-injection timing, and injection amount under different conditions, optimize the control strategy, and find the control strategy that minimizes emissions.

[0133] Particle optimization process:

[0134]

[0135] Where: i is 1, 2....N; N is the total number of particles in the population; t is the number of particle iterations; ω is the weight; v is the speed of each particle in the population; r1, r2 are random numbers; x i is the current position of particle i; is the optimal individual position of particle i at the moment; it is also the optimal position of the particle population at the moment; c1 and c2 are learning factors.

[0136] Matters not covered by the present invention are known technologies.

[0137] The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable people familiar with the technology to understand the content of the present invention and implement it accordingly, and they cannot be used to limit the protection scope of the present invention. Any equivalent changes or modifications made according to the spirit of the present invention should be included in the protection scope of the present invention.

Claims

1. A digital twin-driven marine diesel engine emission prediction and optimization method, characterized in that: The following steps are involved: (1) Construct a comprehensive performance mechanism model to describe the working principle and internal operation mechanism of the diesel engine, obtain the operating parameters of the diesel engine, and use the idea of ​​decomposition and combination to decompose the diesel engine system into a series of independent and interconnected subsystems, considering the interaction and influence between different subsystems; (2) Construct a calibration module and propose a fusion technology system based on deep learning to optimize and calibrate the comprehensive performance mechanism model. Use deep learning algorithms to calibrate the comprehensive performance mechanism model, improve the generalization ability of the comprehensive performance mechanism model, and solve the problems of high cost, insufficient economy, and short service life of traditional sensor monitoring methods. (3) Construct a combustion stage mechanism model, use the output of the calibrated comprehensive performance mechanism model as the inlet and outlet conditions, wall conditions, and initial conditions of the combustion stage mechanism model, use turbulent kinetic energy as an indicator of the uniformity of oil-gas mixing, and optimize the structure of the intake pipe to improve the uniformity and flow characteristics of oil-gas mixing; (4) Establish a data-driven model to further understand the intrinsic relationship between high-fidelity simulation of the combustion stage and emissions. Based on the combustion process mechanism model, a deep learning algorithm is used to fit the relationship between turbulent kinetic energy indicators and emissions. (5) From the perspective of optimizing the control strategy, we seek the optimal strategy to make combustion more complete and pollutant emissions lower. We use the constructed comprehensive performance mechanism model, calibration module, combustion stage mechanism model and data-driven model fusion technology system as the fitness function, and take minimizing emissions as the goal. We design an artificial intelligence optimization algorithm, optimize the injection control parameters, and determine the optimal injection control strategy under different operating conditions.

2. The method for predicting and optimizing emissions of marine diesel engines driven by digital twins according to claim 1, characterized in that: Construction of the comprehensive performance mechanism model in step (1): (1-1) constructing a comprehensive performance mechanism model, including a compressor submodule, an intercooler submodule, an intake pipe submodule, a cylinder submodule, an exhaust pipe submodule, a turbine submodule, a supercharger rotor submodule, and a fuel injection pump submodule; (1-2) Determine the output parameters of the comprehensive performance mechanism model, including the in-cylinder pressure and temperature during the combustion phase, as well as emission parameters such as CO, NO, and NO2.

3. The digital twin-driven marine diesel engine emission prediction and optimization method according to claim 1 is characterized in that: The calibration module construction in step (2) comprises the following steps: (2-1) preprocessing the output of the comprehensive performance mechanism model in step (2) and historical data, including cleaning the data to delete abnormal data and duplicate data, and fill in missing data; using variance filtering method to remove data with variance of 0, and dividing the data into training set, test set and validation set; (2-2) Establish a CNN module, input the divided data set into the convolutional neural network for feature extraction, extract data through the convolution layer and pass it to the pooling layer after processing with the activation function, use the neurons in the pooling layer to perform secondary local feature extraction and simplification on the data extracted for the first time, and perform hierarchical extraction on the input data, integrate the extracted local information, and lay the foundation for using BiLSTM to extract deep temporal features; (2-3) Establish a BiLSTM module to process and analyze data to capture long-term dependencies in time series data, use a bidirectional long short-term memory network model to extract the time series information of the data, and analyze the patterns and regularities therein; (2-4) The Attention mechanism is introduced, and the information extracted by CNN is passed into LSTM as a new input variable. The Attention mechanism is used to calculate the weight of the LSTM hidden layer through the forward and back propagation of deep learning. The gate weights and bias matrices corresponding to the hidden layer states of each time series point are propagated, and the error between the predicted value and the actual value is adjusted and optimized. The gate weights and bias values ​​optimized by the Attention mechanism are updated through the error. The Attention mechanism is added to the hidden layer of BiLSTM to deeply learn the time series change rules, accurately capture key information and prevent data generalization, and finally obtain a suitable and stable deep learning model.

4. The method for predicting and optimizing emissions of marine diesel engines driven by digital twins according to claim 1, characterized in that: The construction of the combustion stage mechanism model in step (3) comprises the following steps: (3-1) Establish a CFD numerical simulation model and define the structures of the intake manifold, resonant cavity, intake manifold and injector; (3-2) Perform intake meshing, extract the fluid domain of the intake pipe three-dimensional model, and perform tetrahedral meshing on the fluid domain; set the front end of the intake manifold as the fluid inlet, and the end of the intake manifold as the fluid outlet; set the end of the injector as the methanol injection inlet, and the remaining surface as the intake pipe wall; perform local encryption on the mesh of the injector end; (3-3) using the data in step (3) as the inlet and outlet conditions, wall conditions and initial conditions of the combustion stage mechanism model, wherein the data includes the cylinder temperature, inlet and outlet pressures and injector pressure; (3-4) Determine the initialization conditions and boundary conditions, and establish the initial and boundary conditions based on the computational fluid dynamics solution; when selecting boundary conditions, consider the physical properties of the fluid, the calculation range, and the control equations; set the inlet boundary to the mass and pressure boundary, and the outlet boundary to the pressure boundary; set the wall surface to be smooth and frictionless, and use the standard wall function method for solution; (3-5) Optimize the structure and analyze the flow characteristics in the intake pipe; when the injector is not spraying fuel, set all intake manifolds as pressure outlets and set the initial pressure to atmospheric pressure; analyze the oil-gas mixing characteristics and use the maximum value of the intake manifold turbulent kinetic energy as an indicator of the uniformity of oil-gas mixing; adjust the length of the intake manifold in the vertical direction to improve the mixing characteristics; (3-6) Carry out optimization verification. After completing the structural optimization, re-analyze the mixing characteristics; observe the maximum turbulent kinetic energy and turbulent kinetic energy cloud diagram of the intake manifold to verify the effectiveness of the structural optimization.

5. The method for predicting and optimizing emissions of marine diesel engines driven by digital twins according to claim 1, characterized in that: The step (3) of establishing a data-driven model includes the following steps: (4-1) Preprocess the maximum turbulent kinetic energy and emission data under different working conditions, including cleaning and normalization, and divide them into training sets and test sets; (4-2) Establishing a convolutional neural network (CNN) module to extract the maximum value of turbulent kinetic energy and the characteristics of operating condition data; automatically extracting the maximum value of turbulent kinetic energy and characteristic information related to the operating condition from the input data through the CNN module, and predicting the relationship between the maximum value of turbulent kinetic energy and the operating condition information and emission data; (4-3) Model training is performed, and the back propagation algorithm and optimizer are used to adjust the model parameters to minimize the loss function.

6. The method for predicting and optimizing emissions of a marine diesel engine driven by a digital twin according to claim 1, characterized in that: The control strategy optimization in step (5) includes the following steps: (5-1) Determine the optimization parameters as injection control parameters, including pilot injection timing, main injection timing and injection amount; (5-2) Based on the particle swarm optimization algorithm, the data in step (5-1) is used as the initialization particle swarm parameters, and the position and speed of each particle are randomly initialized; the fusion system composed of step (1), step (2), step (3), and step (4) is used as the fitness function, and the minimum emission is used as the objective function to perform iterative optimization; (5-3) Through the fusion system composed of step (1), step (2), step (3), and step (4), the optimized pilot injection timing, main injection timing, and injection amount are verified to determine the corresponding emission amount.

Images