On-line prediction method for two-time micro-injection fuel quantity of dual-fuel engine considering heat effect

Abstract

The invention aims to provide a heat effect-considered online prediction method for two-time micro-injection fuel quantity of a dual-fuel engine, and belongs to the field of engines. The method comprises the steps that firstly, on a high-pressure common-rail fuel injection system experiment table, under the conditions of different rail pressure, first-time micro-injection pulse width, two-time injection interval and second-time micro-injection pulse width, inlet pressure fluctuation, driving current and fuel injection rate of a fuel injector are obtained, and a data set is established; then, a DeepAR model with the oil injector inlet pressure fluctuation and the driving current curve as input and the oil injection rate as output is constructed; thirdly, a fuel temperature prediction model of a fuel injector orifice outlet in the engine cylinder cover is established; and finally, the fuel injection quantity of two times of micro-injection considering fuel pressure drop heat generation and the in-cylinder thermal environment coupling heat effect is corrected and obtained. According to the method, oil supply-oil injection disturbance under the transient working condition can be effectively dealt with, the consistency of the oil injection quantity of each cylinder is improved, and the combustion efficiency and the operation stability of the engine are improved.

Description

Technical Field

[0001] The present invention relates to an engine control method, in particular to a fuel injection control method. Background Art

[0002] With the global energy shortage and the increasing requirements for environmental protection, engine combustion technology has gradually developed towards high efficiency and low emissions, and dual-fuel engines have become a hot topic in recent years. Dual-fuel engines combine the advantages of diesel and low-carbon fuels, which can significantly reduce carbon emissions and fuel costs. However, the combustion control of dual-fuel engines is more complicated, especially in the process of multiple pilot fuel injections, where precise control of the injection amount becomes particularly important.

[0003] During the pilot fuel injection process of a dual-fuel engine, the amount of fuel injected in the two micro-injections is not only affected by the pressure fluctuations of the fuel supply system, but also by the combined effects of the heat generation effect of the fuel itself and the complex thermal environment in the cylinder. Specifically, the injector is in a high-temperature cylinder head, and the high-temperature environment in the cylinder affects the temperature of the entire cylinder head through the heat transfer effect, thereby affecting the temperature of the fuel to be injected in the injector nozzle. At the same time, during the high-pressure fuel injection process, due to the Joule-Thomson effect, the fuel will accelerate and expand in the spray hole, causing the fuel temperature at the spray hole outlet to change significantly, ultimately affecting the density of the injected fuel, directly resulting in a deviation between the actual injection amount of each injection and the target injection amount, and the injection amount information cannot be accurately obtained online. Summary of the invention

[0004] The purpose of the present invention is to provide an online prediction method for the amount of micro-injection oil of a dual-fuel engine taking thermal effects into account, which can solve the problem that the amount of micro-injection oil cannot be accurately predicted online in real time under a thermal environment of the engine.

[0005] The object of the present invention is achieved in that:

[0006] The present invention provides an online prediction method for the amount of two micro-injection ignition oil in a dual-fuel engine considering thermal effects, which is characterized by comprising the following steps:

[0007] (1) On a high-pressure common rail fuel injection system test bench, the injector inlet pressure fluctuation, drive current, and injection rate were obtained under different rail pressures, first micro-injection pulse widths, two injection intervals, and second micro-injection pulse widths to establish a data set;

[0008] (2) A DeepAR model with injector inlet pressure fluctuation and driving current curve as input and injection rate as output was constructed. The DeepAR model was trained based on the data set and its prediction accuracy was optimized so that the model could online predict the injection rate of the first micro-injection and the injection rate of the second micro-injection affected by the pressure fluctuation caused by the previous injection, and output the injection amount of the two micro-injections by time integration at the same time.

[0009] (3) Establish a prediction model for the fuel temperature at the outlet of the injector nozzle in the engine cylinder head, so that the model uses the cylinder head temperature as the initial temperature of the fuel at the nozzle inlet and calculates the fuel temperature at the nozzle outlet;

[0010] (4) During the operation of the engine, the injector inlet pressure, driving current and cylinder head temperature of each cycle are measured in real time. The constructed DeepAR model is used to predict the injection amount of the two micro-injections online. The fuel temperature at the nozzle outlet is calculated using the real-time measured cylinder head temperature. The fuel density at the nozzle outlet is calculated based on this temperature, so as to correct and obtain the injection amount of the two micro-injections taking into account the heat generated by the fuel pressure drop and the coupled thermal effects of the thermal environment in the cylinder.

[0011] The present invention may also include:

[0012] 1. The construction method of DeepAR model in step (2) is:

[0013] (21) Data normalization within the dataset:

[0014]

[0015] Where P max , P min ,I max ,I min They are respectively the injector inlet pressure P in and drive current I drive The maximum and minimum values ​​of ;

[0016] Set the time window length T and divide the data into sequence inputs:

[0017]

[0018] (22) Construct a model structure framework and use a recurrent neural network to model the time dependency of the input sequence:

[0019] h t =RNN(X(t),h t-1 )

[0020] In the formula, h t is the hidden state at the current moment; according to the hidden state, the probability distribution of the output injection rate

[0021]

[0022] In the formula, μ t , σ t are the mean and standard deviation of the injection rate, respectively, which are calculated from the hidden state through the fully connected layer:

[0023]

[0024] Where W μ , W σ is the model weight, b μ , b σ is the model bias; the negative log-likelihood loss function is used to quantify the prediction accuracy of the model:

[0025]

[0026] (23) Construct a 4th-order Runge-Kutta algorithm optimization framework based on multi-point estimation:

[0027]

[0028] Where k1, k2, k3, and k4 are the initial estimate, the second estimate, the third estimate, and the fourth estimate, respectively;

[0029] With the above estimated values, the final predicted value is updated as:

[0030]

[0031] (24) The DeepAR model is trained based on the dataset, and optimized using the fourth-order Runge–Kutta algorithm at each time step to obtain the trained and optimized DeepAR model.

[0032] 2. The process of establishing the prediction model of the fuel temperature at the nozzle outlet in step (3) is as follows:

[0033] According to the nozzle structure parameters, calculate the adiabatic number Ad:

[0034]

[0035] Where D and L are the lengths of the nozzle diameter, Pr is the Prandtl number, Re is the Reynolds number, and Nu is the Nuschel number;

[0036] Based on the Joule-Thomson fuel heat generation theory, calculate the fuel temperature T at the nozzle outlet section out :

[0037]

[0038] In the formula, C p is the specific heat capacity at constant pressure, β is the thermal expansion coefficient, is the average temperature of the nozzle inlet and outlet, ρ is the fuel density, p in 、p out are the inlet and outlet pressures, T in is the nozzle inlet temperature, v out is the fuel flow rate at the nozzle outlet, and δ represents the temperature lost by heat transfer.

[0039] 3. The correction method of the injection amount in step (4) is:

[0040]

[0041] Where M inj,1 、M inj,2 are the corrected injection amount of the first micro-injection and the second micro-injection, ρ(T out ) is the fuel density at the nozzle outlet, t0 and t1 are the start and end times of the first micro-spray, and t1 and t2 are the start and end times of the second micro-spray.

[0042] The advantages of the present invention are:

[0043] The prediction result of the present invention covers the influence of the first micro-injection on the interference of the second micro-injection, takes into account the influence of the coupled thermal effect of the fuel's own heat generation and the thermal environment in the cylinder on the fuel density, and can accurately predict the injection amount of the two micro-injections in real time online. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0045] The present invention is described in more detail below with reference to the accompanying drawings:

[0046] Combination Figure 1 The present invention provides an online prediction method for the amount of two micro-injection ignition oil in a dual-fuel engine considering thermal effect, comprising the following steps:

[0047] Step S1, on a high-pressure common rail fuel injection system test bench, under different rail pressures, first micro-injection pulse widths, two injection intervals, and second micro-injection pulse widths, injector inlet pressure fluctuations, drive currents, and injection rates are obtained to establish a data set;

[0048] Step S2, constructing a DeepAR model with the injector inlet pressure fluctuation and the driving current curve as input and the injection rate as output, training based on the data set, and using the fourth-order Runge-Kutta algorithm to optimize the prediction accuracy of the DeepAR model, so that the model can online predict the injection rate of the first micro-injection and the injection rate of the second micro-injection affected by the pressure fluctuation caused by the previous injection, and simultaneously output the injection amount of the two micro-injections by time integration;

[0049] Step S3, based on the Joule-Thomson fuel heat generation theory, a prediction model for the fuel temperature at the nozzle outlet of the fuel injector in the engine cylinder head is established, so that the model uses the cylinder head temperature as the initial temperature of the fuel at the nozzle inlet, and calculates the fuel temperature at the nozzle outlet;

[0050] Step S4: During the actual engine operation, the injector inlet pressure, drive current and cylinder head temperature of each cycle are measured in real time. The constructed DeepAR model is used to predict the injection amount of the two micro-injections online, and the cylinder head temperature measured in real time is used to calculate the fuel temperature at the nozzle outlet. The fuel density at the nozzle outlet is calculated based on the temperature, so as to correct and obtain the injection amount of the two micro-injections that takes into account the heat generated by the fuel pressure drop and the coupled thermal effects of the thermal environment in the cylinder.

[0051] The construction and optimization of the DeepAR model in step S2 are as follows:

[0052] Step (a), normalize and window the data in the dataset. Normalize the data to reduce the impact of different dimensions on the model:

[0053]

[0054] Where P max , P min ,I max ,I min They are respectively the injector inlet pressure P in and drive current I drive The maximum and minimum values ​​of .

[0055] Set the time window length T and divide the data into sequence inputs:

[0056]

[0057] Step (b) builds the model structure framework. A recurrent neural network is used to model the time dependency of the input sequence:

[0058] h t =RNN(X(t),h t-1 ) (4)

[0059] In the formula, h t is the hidden state at the current moment. According to the hidden state, the probability distribution of the output injection rate is

[0060]

[0061] In the formula, μ t , σ t are the mean and standard deviation of the injection rate, respectively, which are calculated from the hidden state through the fully connected layer:

[0062]

[0063] Where W μ , W σ is the model weight, b μ , bσ is the model bias. The negative log-likelihood loss function is used to quantify the prediction accuracy of the model:

[0064]

[0065] Step (c), construct a 4th-order Runge-Kutta algorithm optimization framework based on multi-point estimation:

[0066]

[0067] In the formula, k1, k2, k3, and k4 are the initial estimate, the second estimate, the third estimate, and the fourth estimate, respectively. Based on the above estimates, the final prediction value update formula is:

[0068]

[0069] This can correct the prediction value of the DeepAR model, reduce the cumulative error, and improve the prediction accuracy.

[0070] Step (d), training the DeepAR model based on the data set, and optimizing it using the 4th-order Runge-Kutta algorithm at each time step to obtain the trained and optimized DeepAR model.

[0071] The fuel temperature prediction model for the nozzle outlet in step S3 is established as follows:

[0072] According to the nozzle structure parameters, calculate the adiabatic number Ad:

[0073]

[0074] Where D and L are the lengths of the nozzle diameter, Pr is the Prandtl number, Re is the Reynolds number, and Nu is the Nuschel number.

[0075] Based on the Joule-Thomson fuel heat generation theory, calculate the fuel temperature T at the nozzle outlet section out :

[0076]

[0077] In the formula, C p is the specific heat capacity at constant pressure, β is the thermal expansion coefficient, is the average temperature of the nozzle inlet and outlet, ρ is the fuel density, p in 、p out are the inlet and outlet pressures, T in is the nozzle inlet temperature, which is replaced by the cylinder head real-time temperature during engine operation. outis the fuel flow rate at the nozzle outlet, which is solved by the Bernoulli equation. δ represents the temperature lost by heat transfer. When Ad>6, the adiabatic assumption is reasonable, then δ=0; when Ad≤6, the calculated T out Significantly affected by heat transfer; the value of δ is calibrated by the experimental data of temperature rise at the injector nozzle.

[0078] The fuel injection amount correction in step S4 is specifically:

[0079]

[0080] Where M inj,1 、M inj,2 are the corrected injection amount of the first micro-injection and the second micro-injection, ρ(T out ) is the fuel density at the nozzle outlet, t0 and t1 are the start and end times of the first micro-spray, and t1 and t2 are the start and end times of the second micro-spray.

Claims

1. An online prediction method for the amount of pilot oil in two micro-injections of a dual-fuel engine considering thermal effects, characterized by: The following steps are involved: (1) On a high-pressure common rail fuel injection system test bench, the injector inlet pressure fluctuation, drive current, and injection rate were obtained under different rail pressures, first micro-injection pulse widths, two injection intervals, and second micro-injection pulse widths to establish a data set; (2) A DeepAR model with injector inlet pressure fluctuation and driving current curve as input and injection rate as output was constructed. The DeepAR model was trained based on the data set and its prediction accuracy was optimized so that the model could online predict the injection rate of the first micro-injection and the injection rate of the second micro-injection affected by the pressure fluctuation caused by the previous injection, and output the injection amount of the two micro-injections by time integration at the same time. (3) Establish a prediction model for the fuel temperature at the outlet of the injector nozzle in the engine cylinder head, so that the model uses the cylinder head temperature as the initial temperature of the fuel at the nozzle inlet and calculates the fuel temperature at the nozzle outlet; (4) During the operation of the engine, the injector inlet pressure, driving current and cylinder head temperature of each cycle are measured in real time. The constructed DeepAR model is used to predict the injection amount of the two micro-injections online. The fuel temperature at the nozzle outlet is calculated using the real-time measured cylinder head temperature. The fuel density at the nozzle outlet is calculated based on this temperature, so as to correct and obtain the injection amount of the two micro-injections taking into account the heat generated by the fuel pressure drop and the coupled thermal effects of the thermal environment in the cylinder.

2. The method for online prediction of the amount of pilot oil for two micro-injections of a dual-fuel engine considering thermal effects according to claim 1 is characterized by: The construction method of the DeepAR model in step (2) is: (21) Data normalization within the dataset: Where P max , P min ,I max ,I min They are respectively the injector inlet pressure P in and drive current I drive The maximum and minimum values ​​of ; Set the time window length T and divide the data into sequence inputs: (22) Construct a model structure framework and use a recurrent neural network to model the time dependency of the input sequence: h t =RNN(X(t),h t-1 ) In the formula, h t is the hidden state at the current moment; according to the hidden state, the probability distribution of the output injection rate In the formula, μ t , σ t are the mean and standard deviation of the injection rate, respectively, which are calculated from the hidden state through the fully connected layer: Where W μ , W σ is the model weight, b μ 、b σ is the model bias; the negative log-likelihood loss function is used to quantify the prediction accuracy of the model: (23) Construct a 4th-order Runge-Kutta algorithm optimization framework based on multi-point estimation: Where k1, k2, k3, and k4 are the initial estimate, the second estimate, the third estimate, and the fourth estimate, respectively; With the above estimated values, the final predicted value is updated as: (24) The DeepAR model is trained based on the dataset, and optimized using the fourth-order Runge–Kutta algorithm at each time step to obtain the trained and optimized DeepAR model.

3. The method for online prediction of the amount of pilot oil for two micro-injections of a dual-fuel engine considering thermal effects according to claim 1 is characterized by: The process of establishing the prediction model of the fuel temperature at the nozzle outlet in step (3) is as follows: According to the nozzle structure parameters, calculate the adiabatic number Ad: Where D and L are the lengths of the nozzle diameter, Pr is the Prandtl number, Re is the Reynolds number, and Nu is the Nuschel number; Based on the Joule-Thomson fuel heat generation theory, calculate the fuel temperature T at the nozzle outlet section out : In the formula, C p is the specific heat capacity at constant pressure, β is the thermal expansion coefficient, is the average temperature of the nozzle inlet and outlet, ρ is the fuel density, p in 、p out are the inlet and outlet pressures, T in is the nozzle inlet temperature, v out is the fuel flow rate at the nozzle outlet, and δ represents the temperature lost by heat transfer.

4. The method for online prediction of the amount of pilot oil for two micro-injections of a dual-fuel engine considering thermal effects according to claim 1 is characterized by: The correction method of the injection amount in step (4) is: Where M inj,1 、M inj,2 are the corrected injection amount of the first micro-injection and the second micro-injection, ρ(T out ) is the fuel density at the nozzle outlet, t0 and t1 are the start and end times of the first micro-spray, and t1 and t2 are the start and end times of the second micro-spray.

Images