A method for regulating the consistency of injector injection of each cylinder of an engine considering transient supply-injection disturbance

Abstract

The application aims to provide a method for regulating the consistency of injector injection of each cylinder of an engine considering transient supply-injection disturbance, belonging to the field of engines. First, the rail pressure curve, the injector inlet pressure curve, the driving current curve and the injection rate curve data under different supply time and injection time are obtained on a high-pressure fuel injection system test bench to construct a data set. Then, based on the data set, a Transformer time series model is established with the rail pressure, pressure and current curve as input and the injection rate as output, the injection rate is predicted online and the injection quantity is calculated through time integration. Next, through the programmable logic controller, the injection pulse width under different supply and injection time is feedback regulated to realize the accurate control of the injection quantity of single and multiple injections of each cylinder. The application effectively responds to the supply-injection disturbance under transient working condition, improves the consistency of the injection quantity of each cylinder, and improves the combustion efficiency and running stability of the engine.

Description

TECHNICAL FIELD

[0001] The present application relates to an engine control method, in particular, an injector control method. BACKGROUND

[0002] With the increasing demand for performance of modern engines, especially in the context of increasingly stringent emission regulations, the precise control of engine fuel injection system has become a key technology to improve engine performance and reduce emissions. High-pressure fuel injection system has been widely used due to its good injection pressure control ability and injection flexibility. However, during the actual operation of the engine, due to the influence of various factors such as transient operating condition changes, supply pressure fluctuations, injection pressure fluctuations, etc., there are certain differences in the injection amount and injection law of each cylinder injector. This injection inconsistency will directly affect the uniformity of engine combustion, and then have an adverse effect on the power, economy and emission performance of the engine.

[0003] Currently, the research on injection amount regulation mainly focuses on the control of injectors under static and steady state conditions, while the research on injector injection consistency under transient disturbance is relatively less. The traditional control method is difficult to accurately match the injection amount of each cylinder injector under dynamic conditions. SUMMARY

[0004] The purpose of the present application is to provide an engine cylinder injector injection consistency regulation method considering transient supply-injection disturbance, which can improve the injection consistency of each cylinder of the engine, reduce the injection deviation caused by transient supply-injection disturbance, and achieve the goals of high-efficiency combustion and low emission of the engine under different operating conditions.

[0005] The purpose of the present application is achieved as follows:

[0006] The engine cylinder injector injection consistency regulation method considering transient supply-injection disturbance of the present application is characterized by comprising the following steps:

[0007] (1) On a high-pressure fuel injection system test bench, obtain the rail pressure curve, the injector inlet pressure curve, the driving current curve, and the injection rate curve experimental data of each cycle under different supply time and injection time, and establish a data set;

[0008] (2) Construct a Transformer time series model with the rail pressure curve, the injector inlet pressure curve and the driving current curve as input, and the injection rate as output, train the model based on the data set to predict the injection rate online, and simultaneously integrate the time to output the injection amount;

[0009] (3) Embedding the Transformer model into the programmable logic controller, and programming the injection pulse width feedback control logic at different supply time and injection time in the controller, so that the control parameters of each cylinder injection strategy are optimized according to the deviation between the actual injection amount and the target injection amount of each cylinder;

[0010] (4) During the engine operation process, the rail pressure curve, the injector inlet pressure curve and the driving current curve of each cylinder are collected in real time every cycle, the Transformer time series model is constructed for online prediction of the injection amount of each cylinder, the deviation between the actual injection amount and the target injection amount of each cylinder is calculated, and the online closed-loop control of the injection amount of each cylinder is carried out based on the programmable logic controller.

[0011] The application can also include:

[0012] 1. The different supply time and injection time in step (1) include:

[0013] Three cases under single injection strategy: the supply time is before the injection time, the supply time is after the injection time and before the end of injection, and the supply time is after the injection time and after the end of injection;

[0014] Three cases under multiple injection strategy: the supply time is before the first injection time, the supply time is after the first injection time and before the end of the last injection, and the supply time is after the first injection time and after the end of the last injection.

[0015] 2. The construction method of the Transformer time series model in step (2) is:

[0016] (21) Divide the data into time segments of the same length, and each time segment is a sample; set the input time series The output time series Y inject =[Y inject,1 ,Y inject,2 ,...,Y inject,T ], wherein T represents the number of time steps, and the input time series and the output time series are normalized respectively;

[0017] (22) Create a position encoding to make the model perceive the information of the time step, and the position encoding vector PE is:

[0018]

[0019] In the formula, i is the dimension index, d model is the embedding dimension;

[0020] (23) Establish a Transformer encoder layer:

[0021] The long-term time dependence of the input time series data is captured by the multi-head self-attention layer:

[0022]

[0023] In the formula, Q, K, V are the query matrix, key matrix and value matrix corresponding to the input sequence respectively, d k is the attention dimension;

[0024] A feedforward neural network is established after each attention layer:

[0025] FFN(x)=max(0,xW1+b1)W2+b2

[0026] In the formula, x represents the input of each layer, W and b are the weight and bias of the feedforward neural network respectively, and subscripts 1 and 2 represent the first layer and the second layer in the feedforward network respectively;

[0027] Residual connection is added after each layer, and the outputs of the self-attention layer and the feedforward neural network layer are respectively:

[0028] Output=LayerNorm(x+Attention)

[0029] Output=LayerNorm(x+FFN)

[0030] (24) The loss function of the model is constructed:

[0031]

[0032] In the formula, The input time series is represented by the Transformer model to predict the output of the injector inlet pressure fluctuation; Y is the actual value of the injector inlet pressure fluctuation, and N represents the number of samples in the data set;

[0033] (25) For the six cases under the single injection strategy and the multiple injection strategy, the model is trained based on the data set to obtain six sets of Transformer model parameters to meet the prediction requirements under different conditions.

[0034] 3、The injection pulse width feedback control logic in step (3) is:

[0035] For single injection, the deviation between the actual injection amount and the target injection amount predicted by the Transformer model online in the current cycle is used to correct the injection pulse width:

[0036] PW(k)=k p ·e(k-1)+k i ·∫e(k)dk

[0037] In the formula, PW(k) is the injection pulse width at the kth injection, e(k-1) is the injection quantity deviation at the (k-1)th injection, k p , k i is a correction coefficient;

[0038] For multiple injections, the injection pulse width of each injection stage is corrected according to the deviation between the actual injection quantity and the target injection quantity of each injection stage predicted by the Transformer model in the current cycle:

[0039] PW m (k)=k p ·e m (k-1)+k i ·∫e m (k)dk

[0040] In the formula, the subscript m represents the injection stage.

[0041] 4. The predicted injection quantity of each cylinder in step (4) is specifically:

[0042] According to the engine supply time and the injection time of each cylinder, the Transformer model parameters suitable for the situation of each cylinder are adopted, the injection rate of each cylinder is predicted online based on the model, and the injection quantity is obtained by integration; for single injection, only the injection rate is predicted, and the injection quantity is obtained by integration; for multiple injections, the injection rate of each injection stage is predicted, and the injection quantity of each injection stage is obtained by integration.

[0043] The advantages of the present application are that the present application can predict the single injection rate and the multiple injection rate of each cylinder under different supply-injection disturbance conditions in real time, and the corresponding injection quantity; not only can the injection quantity under single injection of each cylinder be regulated, but also the injection quantity of each injection stage under multiple injection can be regulated, which significantly improves the injection consistency of the injectors of each cylinder. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 is a flowchart of the present application;

[0045] Figure 2 is an experimental schematic diagram of injection consistency regulation. DETAILED DESCRIPTION

[0046] The present application will be described in more detail below with examples combined with the accompanying drawings:

[0047] Combined Figures 1-2 , the engine cylinder injector injection consistency regulation method considering transient supply-injection disturbance of the present application includes the following steps:

[0048] Step S1, in the high-pressure fuel injection system experiment table, the rail pressure curve, the injector inlet pressure curve, the driving current curve and the injection rate curve experimental data of each cycle under different supply time and injection time are obtained, and the data set is established;

[0049] Step S2, a Transformer time series model taking the rail pressure curve, the injector inlet pressure curve and the driving current curve as input and taking the injection rate as output is constructed, and the data set is trained to enable the model to predict the injection rate online and simultaneously output the injection amount by time integration;

[0050] Step S3, the Transformer model is embedded in the programmable logic controller, and the injection pulse width feedback control logic programming under different supply time and injection time is programmed in the controller, so that it can optimize the control parameters of each cylinder injection strategy according to the deviation between the actual injection amount and the target injection amount of each cylinder;

[0051] Step S4, as shown in Figure 2 , in the actual engine working process, the rail pressure curve, the injector inlet pressure curve and the driving current curve of each cycle are collected in real time, the Transformer time series model is used to predict the injection amount of each cylinder online, the deviation between the actual injection amount and the target injection amount of each cylinder is calculated, and the programmable logic controller is used to perform online closed-loop control of the injection amount of each cylinder, thereby improving the injection consistency of each cylinder.

[0052] The different supply time and injection time in step S1 include:

[0053] Three cases under single injection strategy: supply time before injection time, supply time after injection time and before injection end, and supply time after injection time and after injection end;

[0054] Three cases under multiple injection strategy: supply time before the first injection time, supply time after the first injection time and before the last injection end, and supply time after the first injection time and after the last injection end.

[0055] The Transformer time series model constructed in step S2 is specifically:

[0056] Step (a), divide the data into time segments of the same length, and each time segment is a sample. Let the input time series and the output time series Y inject = [Y inject,1 , Y inject,2 ,…, Y inject,T ] where T represents the number of time steps. Normalize the input time series and output time series respectively.

[0057] Step (b), create a position encoding to make the model aware of the information of time step, the position encoding vector PE is calculated as:

[0058]

[0059] where i is the dimension index, d model is the embedding dimension.

[0060] Step (c), build the Transformer encoder layer, specifically:

[0061] Capture the long-term temporal dependence of input time series data through multi-head self-attention layer:

[0062]

[0063] where Q, K, V are the query matrix, key matrix and value matrix corresponding to the input sequence respectively, d k is the attention dimension.

[0064] After each attention layer, a feedforward neural network is built to improve the nonlinear representation ability of the model, which is represented as:

[0065] FFN(x)=max(0,xW1+b1)W2+b2 (3)

[0066] where x represents the input of each layer, W and b are the weight and bias of the feedforward neural network respectively, and subscripts 1 and 2 represent the first layer and the second layer in the feedforward network respectively.

[0067] On this basis, a residual connection is added after each layer to prevent gradient vanishing and improve training stability, and the outputs of the self-attention layer and the feedforward neural network layer are represented as:

[0068] Output=LayerNorm(x+Attention) (4)

[0069] Output=LayerNorm(x+FFN) (5)

[0070] Step (d), build the loss function of the model, which is used to evaluate the prediction error of the model, specifically:

[0071]

[0072] where represents the injector inlet pressure fluctuation predicted by the Transformer model; Y is the actual value of the injector inlet pressure fluctuation, and N represents the number of samples in the data set.

[0073] Step (e), based on the data set, model training is performed for six cases under single injection strategy and multiple injection strategy, and six sets of Transformer model parameters are obtained to meet the prediction requirements under different conditions.

[0074] The injection pulse width feedback control logic in step S3 is specifically:

[0075] For single injection, the injection pulse width is corrected according to the deviation between the actual injection amount and the target injection amount predicted by the Transformer model in the current cycle:

[0076] PW(k) = k p ·e(k-1) + k i ·∫e(k)dk (7)

[0077] In the formula, PW(k) is the injection pulse width at the kth injection, e(k-1) is the injection amount deviation at the (k-1)th injection, k p , k i is the correction coefficient.

[0078] For multiple injection, the injection pulse width of each injection stage is corrected according to the deviation between the actual injection amount and the target injection amount of each injection stage predicted by the Transformer model in the current cycle:

[0079] PW m (k) = k p ·e m (k-1) + k i ·∫e m (k)dk (8)

[0080] In the formula, subscript m represents the injection stage.

[0081] The prediction of the injection amount of each cylinder in step S4 is specifically:

[0082] According to the engine supply time and the injection time of each cylinder, the Transformer model parameters suitable for the condition of each cylinder are used to predict the injection rate of each cylinder based on the model, and the injection amount is obtained by integration. Among them, for single injection, only the injection rate is predicted, and the injection amount is obtained by integration; for multiple injection, the injection rate of each injection stage is predicted, and the injection amount of each injection stage is obtained by integration.

Claims

1. A method for regulating the consistency of the injection of each cylinder of an engine, taking into account transient supply-injection perturbations, characterized in that: The method comprises the following steps: (1) obtaining the rail pressure curve, the injector inlet pressure curve, the driving current curve and the injection rate curve of each cycle under different supply time and injection time on a high-pressure fuel injection system test bench, and establishing a data set; (2) constructing a Transformer time series model with the rail pressure curve, the injector inlet pressure curve and the driving current curve as inputs and the injection rate as output, training the model based on the data set to predict the injection rate online, and simultaneously integrating the time to output the injection amount; (3) embedding the Transformer model into a programmable logic controller, and programming the injection pulse width feedback control logic under different supply time and injection time in the controller to optimize the control parameters of the injection strategy of each cylinder according to the deviation between the actual injection amount and the target injection amount of each cylinder; (4) in the working process of the engine, the rail pressure curve, the injector inlet pressure curve and the driving current curve of each cycle are collected in real time, the injection amount of each cylinder is predicted online by using the constructed Transformer time series model, the deviation between the actual injection amount and the target injection amount of each cylinder is calculated, and the online closed-loop control of the injection amount of each cylinder is performed based on the programmable logic controller.

2. The method of claim 1, wherein the method further comprises: determining a target fuel injection quantity for each cylinder of the engine; and determining a target fuel injection quantity for each injector of the engine. The different supply time and injection time in step (1) include: Three cases under single injection strategy: the supply time is before the injection time, the supply time is after the injection time and before the end of injection, and the supply time is after the injection time and after the end of injection; Three cases under multi-injection strategy: the supply time is before the first injection time, the supply time is after the first injection time and before the end of the last injection, and the supply time is after the first injection time and after the end of the last injection.

3. The method of claim 2, wherein the method further comprises: determining a target fuel injection quantity for each cylinder of the engine; and determining a target fuel injection quantity for each injector of the engine based on the target fuel injection quantity for each cylinder of the engine and the determined fuel injection quantity for each injector of the engine. The construction method of the Transformer time series model in step (2) is: (21) dividing the data into time segments of the same length, each time segment as a sample; setting the input time sequence Output time sequence Y inject = [Y inject,1 , Y inject,2 ,..., Y inject,T ], where T represents the number of time steps, and the input time sequence and the output time sequence are normalized respectively; (22) create a position encoding to make the model aware of the information of the time step, and the position encoding vector PE is: where i is a dimension index, d model is an embedding dimension; (23) establish a Transformer encoder layer: Capture the long-term time dependence of the input time series data through a multi-head self-attention layer: In the formula, Q, K, and V are a query matrix, a key matrix, and a value matrix corresponding to an input sequence, respectively, d k is an attention dimension; After each attention layer, a feedforward neural network is established: FFN(x) = max(0, xW1 + b1)W2 + b2 In the formula, x represents the input of each layer, W and b are the weight and bias of the feedforward neural network, and subscripts 1 and 2 represent the first layer and the second layer in the feedforward network, respectively; Residual connection is added after each layer, and the outputs of the self-attention layer and the feedforward neural network layer are respectively: Output = LayerNorm(x + Attention) Output = LayerNorm(x + FFN) (24) construct the loss function of the model: In the formula, The input time series is characterized by the injector inlet pressure fluctuation predicted by the Transformer model; Y is the actual value of the injector inlet pressure fluctuation, and N represents the number of samples in the data set. (25) for the six cases under single injection strategy and multi-injection strategy, train the model based on the data set to obtain six sets of Transformer model parameters to meet the prediction requirements under different conditions.

4. The method of claim 2, wherein the method further comprises: determining a target fuel injection quantity for each cylinder of the engine; and determining a target fuel injection quantity for each injector of the engine. The injection pulse width feedback control logic in step (3) is: For single injection, the injection pulse width is corrected according to the deviation between the actual injection quantity and the target injection quantity predicted by the Transformer model online in the current cycle: PW(k) = k p • e(k - 1) + k i • ∫ e (k) dk In the formula, PW(k) is the injection pulse width at the kth injection, e(k-1) is the injection quantity deviation at the (k-1)th injection, k p , k i is a correction coefficient; For multiple injection, the injection pulse width of each injection stage is corrected according to the deviation between the actual injection quantity and the target injection quantity predicted by the Transformer model online in the current cycle: PW m (k) = k p · e m (k - 1) + k i · ∫ e m (k) dk In the formula, the subscript m represents the injection stage.

5. The method for controlling the injection consistency of each cylinder injector in an engine considering transient supply-injection disturbances according to claim 1, characterized in that: The prediction of the injection quantity of each cylinder in step (4) is specifically: Based on the engine supply time and the injection time of each cylinder, the Transformer model parameters suitable for the situation of each cylinder are adopted, the injection rate of each cylinder is predicted online based on the model, and the injection quantity is obtained by integration; for single injection, only the injection rate is predicted, and the injection quantity is obtained by integration; For multiple injection, the injection rate of each injection stage is predicted, and the injection quantity of each injection stage is obtained by integration.

Images