High-pressure fuel injection system injection quantity fuzzy PID closed-loop control method based on physical mechanism-data driving fusion

Abstract

The invention aims to provide a high-pressure fuel injection system injection quantity fuzzy PID closed-loop control method based on physical mechanism-data driving fusion, belongs to the field of engine fuel injection, and aims to provide a high-pressure fuel injection system injection quantity fuzzy PID closed-loop control method based on physical mechanism-data driving fusion. Physical mechanisms of monotonicity loss and initial condition loss in the dynamic injection process are used as constraints, and a Transform time sequence neural network model is obtained through training; and the difference value and the difference value percentage of the predicted injection quantity and the target injection quantity are calculated, and real-time closed-loop control over the next injection quantity is achieved by superimposing injection pulse width control signals output by the fuzzy PID controller. By combining a physical mechanism and data driving, the problems that a pure data driving model is insufficient in generalization ability and poor in interpretability are solved, the injection rate and the injection amount can be accurately predicted on line in real time under the complex working condition of the system, and closed-loop control over the injection amount is achieved.

Description

Technical Field

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

[0002] High-pressure fuel injection technology is one of the important core technologies in the field of internal combustion engines such as dual-fuel engines and low-carbon fuel engines. It stores fuel in a volume chamber in a high-pressure form and injects it through a precisely controlled injector, thereby achieving fine control of the combustion process. As engines develop towards high thermal efficiency and ultra-high injection pressure, the transient dynamic response characteristics of the fuel injection system are becoming more and more complex. This makes the precise control of fuel injection, especially the consistency of the injection amount of each cylinder and the system stability, become the difficulty and focus of the design and control of the fuel injection system.

[0003] Traditional fuel injection systems are controlled mostly using open-loop control or closed-loop control based on low-dimensional feedback variables such as speed and torque. However, fuel injection is affected by nonlinear factors such as pressure fluctuations and fluid inertia effects. This leads to severe limitations on the control accuracy based on static models or linear controllers, making it impossible to accurately adjust the injection volume in a highly dynamic environment, making it difficult to meet the needs of modern engines.

[0004] With the widespread application of machine learning technology, data-driven fuel injection prediction models have gradually become a research hotspot. These models can predict the injection amount to a certain extent by learning and training with a large amount of historical data. However, pure data-driven neural networks have long been overly dependent on data sampling, and robustness and generalization are difficult to guarantee. In addition, the model lacks a deep understanding of the physical mechanism and has poor model interpretability. Secondly, pure data-driven models may show poor generalization and robustness when facing operating conditions outside the training data or abnormal conditions, and cannot effectively cope with complex transient changes or extreme operating conditions of the system. Therefore, it is difficult to achieve high-precision, real-time injection amount prediction and control in the high dynamic environment of the fuel injection system by relying solely on data-driven methods. Summary of the invention

[0005] The purpose of the present invention is to provide a fuzzy PID closed-loop control method for the injection quantity of a high-pressure fuel injection system based on physical mechanism-data driven fusion, which can solve the problems of the injection law and injection quantity being unable to be accurately predicted online in real time and the injection quantity being unable to be accurately closed-loop controlled.

[0006] The object of the present invention is achieved in that:

[0007] The present invention provides a fuzzy PID closed-loop control method for injection quantity of a high-pressure fuel injection system based on physical mechanism-data driven fusion, which is characterized by comprising the following steps:

[0008] (1) An injector is provided, the inlet of the injector is connected to a high-pressure fuel pipe, a single-shot injection instrument is installed below the injector, a current clamp is clamped on the wire of the high-speed solenoid valve of the injector, a pressure sensor is installed at the connection between the injector and the high-pressure fuel pipe, the current clamp, the pressure sensor and the single-shot injection instrument are all connected to a data acquisition unit, the excitation current signal of the high-speed solenoid valve and the pressure signal at the inlet of the injector are measured, the injection rate of the injector is measured using the single-shot injection instrument, and a data set of the excitation current, the injector inlet pressure and the injection rate under different rail pressure conditions and injection pulse width conditions is obtained, and the data in the data set is batch processed and normalized;

[0009] (2) Construct the Transformer time series neural network model framework, use the structural loss and regression loss as the basic loss functions of the network model, use the initial condition loss and monotonicity loss as the physical mechanism loss functions of the network model, and use the sum of the basic loss function and the physical mechanism loss function as the total loss function of the network model;

[0010] (3) Based on the data set established in step (1), the Transformer time series neural network model is trained so that the model can predict the injection rate at the same time based on the excitation current and the injector inlet pressure time series data. After each round of training, it is checked whether the error is lower than the preset 5% threshold. If the error meets the requirement, the iteration is stopped. Otherwise, the training iteration is continued until the error between the predicted injection rate and the measured injection rate meets the threshold requirement.

[0011] (4) Based on the excitation current signal and the pressure signal at the injector inlet measured in real time by the high-pressure fuel injection system, the Transformer time series neural network model established in step (3) is used to predict the injection rate in real time, and the real-time injection amount of each injection is obtained by time integration; the real-time injection amount is compared with the target injection amount, and the injection amount deviation and the injection amount deviation percentage are calculated by fuzzy rules; the fuzzy controller fuzzifies the deviation and the deviation percentage according to the defined fuzzy rule base, outputs the fuzzy control quantity through fuzzy reasoning, and obtains the parameter adjustment quantity of the PID controller through the defuzzification process; and the injection pulse width control signal output by the fuzzy PID controller is superimposed to complete the real-time closed-loop control of the injection pulse width of the next injection.

[0012] The present invention may also include:

[0013] 1. In step (1), the rail pressure conditions include all pressure conditions between the idle rail pressure and the rated pressure with an interval of 10 MPa; the injection pulse width conditions include all injection pulse width conditions between the minimum injection start pulse width and the rated pulse width with an interval of 0.2 ms.

[0014] 2. The data batch normalization in step (1) is specifically as follows:

[0015]

[0016] In the formula, x * is the processed data; represents the mean of unnormalized data; represents the standard deviation of the data; β and γ are the translation parameter and scale parameter respectively; ε is the margin used to prevent division by zero.

[0017] 3. Step (2) specifically includes:

[0018] (21) Define the input data format. The time series data of the excitation current and the injector inlet pressure are used as input features, namely X1 and X2, respectively. The input format is (B, T, F), where B is the batch size, T is the sampling step, and F is the number of input features. X1 and X2 are combined to form a complete input matrix

[0019] (22) Create a position code and add it to the input feature to provide time sequence information for the input data. The position code vector PE is calculated as:

[0020]

[0021] Where i is the dimension index, d model is the embedding dimension;

[0022] (23) Establish the Transformer encoder layer, specifically:

[0023] Capture the long-term temporal dependencies of input time series data through multi-head self-attention layers:

[0024]

[0025] Where Q, K, and V are query matrix, key matrix, and value matrix, respectively. k It is the attention dimension;

[0026] A feedforward neural network is established after each attention layer, which is expressed as:

[0027] FFN(x)=max(0,XW1+b1)W2+b2

[0028] Where W and b are the weight and bias of the feedforward neural network, respectively, and the subscripts 1 and 2 represent the first and second layers in the feedforward network, respectively;

[0029] On this basis, residual connections are added after each layer, and the outputs of the self-attention layer and the feedforward neural network layer are expressed as:

[0030] Output=LayerNorm(X+Attention)

[0031] Output = LayerNorm (X + FFN)

[0032] (24) Construct the total loss function of the model, specifically:

[0033] Construct regression loss E r , used to evaluate the prediction error of the model:

[0034]

[0035] In the formula, Characterize the input excitation current array X1 and the input injector inlet pressure array X2 to predict the output injection rate through the Transformer model; y i is the actual value of the injection rate;

[0036] Constructing the structure loss E s :

[0037]

[0038] In the formula, ω and β are the weight and bias of the Transformer model respectively. is the number of layers of the model architecture, and n is the number of nodes in each layer;

[0039] Construct the initial condition loss E i :

[0040]

[0041] Where, t p-down is the initial moment of the ejector inlet pressure drop, v sound is the speed of sound in the pipeline, L is the distance from the ball valve to the pressure sensor;

[0042] Constructing the monotonicity loss E m :

[0043]

[0044] In the formula, m i is the monotonicity factor;

[0045] In summary, the total loss function of the model is:

[0046] E=E r +λ s E s +λ i E i +λ m E m

[0047] In the formula, λs is the weight coefficient of the result loss, λ i is the weight coefficient of the initial condition loss, λ m is the weight coefficient of monotonicity loss.

[0048] 4. Step (4) is specifically:

[0049] (41) The real-time measured excitation current and injector inlet pressure time series data are input into the Transformer model, and then the output injection rate Q is calculated. rate (t), the total injection quantity Q of each injection is calculated by time integration total :

[0050] Q total =∫Q rate (t)dt

[0051] The error e and error percentage δ between the real-time predicted injection amount and the target injection amount are calculated in real time:

[0052]

[0053] (42) The fuzzy controller is designed, including input fuzzification, definition of fuzzy rule base, fuzzy reasoning and defuzzification. The fuzzy controller obtains the parameter adjustment amount of PID according to the injection amount deviation e and the deviation percentage δ. The adjustment formula of the fuzzy PID controller output is:

[0054] u(t)=(K p0 +ΔK p )·e+(K i0 +ΔK i )∫e dt+(K d0 +ΔK d )·δ

[0055] In the formula, K p0 , K i0 and K d0 is the initial PID parameter, ΔK p , ΔK i and ΔK d is the parameter adjustment amount calculated by the fuzzy controller;

[0056] (43) By comparing the real-time predicted injection amount with the target injection amount, the fuzzy PID controller outputs the control signal T of the next injection pulse width according to the calculated PID adjustment amount. inject :

[0057] T inject =T inject,0 +u(t)

[0058] Where, T inject,0is the initial injection pulse width, T inject As the next injection pulse width, real-time closed-loop control of the injection amount is achieved.

[0059] The advantages of the present invention are:

[0060] (1) The present invention combines the advantages of physical mechanism and data-driven, which not only utilizes the interpretability and robustness of physical mechanism, but also enhances the prediction accuracy of data-driven model in complex nonlinear systems, thus overcoming the problems of insufficient generalization ability and poor interpretability of pure data-driven model.

[0061] (2) The present invention only needs to install a pressure sensor at the injector inlet, and can use the timing data of the excitation current and the injector inlet pressure to realize real-time online prediction of the injection pattern and injection amount without destroying the overall structure of the engine injector and the combustion chamber, thereby improving the reliability and durability of the system.

[0062] (3) The present invention combines the fuzzy PID control algorithm to automatically adjust the parameters of the PID controller by fuzzifying the injection quantity deviation and deviation percentage, thereby realizing adaptive adjustment of the injection pulse width and ensuring the closed-loop control accuracy of the system on the injection quantity under complex working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 is a flow chart of the present invention;

[0064] Figure 2 It is a diagram of the experimental device of the present invention;

[0065] Figure 3 This is the Transformer time series neural network model framework diagram;

[0066] Figure 4 This is the fuzzy PID control block diagram. DETAILED DESCRIPTION

[0067] The present invention is described in more detail below with reference to the accompanying drawings:

[0068] Combination Figure 1-4 ,like Figure 1 As shown, the present invention provides a fuzzy PID closed-loop control method for injection quantity of a high-pressure fuel injection system based on physical mechanism-data driven fusion, comprising the following steps:

[0069] Step S1, such as Figure 2As shown, a current clamp is clamped on the wire of the injector high-speed solenoid valve, and a pressure sensor is installed at the connection between the injector and the high-pressure fuel pipe to measure the excitation current signal of the high-speed solenoid valve and the pressure signal at the injector inlet. In addition, a single injection instrument is used to measure the injection rate of the injector. The excitation current signal, the injector inlet pressure signal and the injection rate are stored in the host computer through the data acquisition unit, and the data in the data set are batch processed and normalized to obtain data sets of the excitation current, injector inlet pressure and injection rate under different rail pressure conditions and injection pulse width conditions.

[0070] Step S2, construct the Transformer time series neural network model framework, take the structural loss and regression loss as the basic loss functions of the network model, and according to the one-dimensional pipeline fluid pressure fluctuation theory, take the initial condition loss and monotonicity loss as the physical mechanism loss function of the network model, and take the sum of the basic loss function and the physical mechanism loss function as the total loss function of the network model.

[0071] Step S3, training the Transformer time series neural network model based on the data set established in step S1, so that the model can predict the injection rate at the same time based on the excitation current and injector inlet pressure time series data. After each round of training, check whether the error is lower than the preset 5% threshold. If the error meets the requirement, stop the iteration, otherwise continue the training iteration until the error between the predicted injection rate and the measured injection rate meets the threshold requirement.

[0072] Step S4, based on the excitation current signal measured in real time by the high-pressure fuel injection system and the pressure signal at the injector inlet, the Transformer time series neural network model established in step S3 is used to predict the injection rate in real time, and the real-time injection amount of each injection is obtained by time integration. The real-time injection amount is compared with the target injection amount, and the injection amount deviation and the injection amount deviation percentage are calculated by fuzzy rules. The fuzzy controller fuzzifies the deviation and the deviation percentage according to the defined fuzzy rule base, outputs the fuzzy control amount through fuzzy reasoning, and obtains the parameter adjustment amount of the PID controller through the defuzzification process. By superimposing the injection pulse width control signal output by the fuzzy PID controller, the real-time closed-loop control of the injection pulse width of the next injection is completed.

[0073] The data set operating conditions in step S1 are specifically as follows: the rail pressure conditions include all rail pressure conditions between the idle rail pressure and the rated rail pressure with an interval of 10 MPa; the injection pulse width conditions include all injection pulse width conditions between the minimum start-injection pulse width and the rated pulse width with an interval of 0.2 ms.

[0074] The data batch normalization in step S1 is specifically as follows:

[0075]

[0076] In the formula, x * is the processed data; represents the mean of unnormalized data; Represents the data standard deviation; β and γ are the translation parameter and scale parameter, respectively, which are optimized by back propagation during training; ε is the remainder used to prevent division by zero.

[0077] Specifically, step S2 is as follows: Figure 3 As shown, including:

[0078] Step S201, define the input data format, the time series data of the excitation current and the injector inlet pressure are used as input features, respectively X1 (excitation current) and X2 (injector inlet pressure), the input format is (B, T, F), where B is the batch size, T is the sampling step, and F is the number of input features. Combine X1 and X2 to form a complete input matrix

[0079] Step S202: Create a position code and add it to the input feature to provide time sequence information for the input data. The position code vector PE is calculated as:

[0080]

[0081] Where i is the dimension index, d model is the embedding dimension.

[0082] Step S203, specifically, establish a Transformer encoder layer:

[0083] Capture the long-term temporal dependencies of input time series data through multi-head self-attention layers:

[0084]

[0085] Where Q, K, and V are query matrix, key matrix, and value matrix, respectively. k It is the attention dimension.

[0086] A feedforward neural network is established after each attention layer to improve the nonlinear representation ability of the model, which is expressed as:

[0087] FFN(x)=max(0,XW1+b1)W2+b2 (4)

[0088] Where W and b are the weight and bias of the feedforward neural network, respectively, and the subscripts 1 and 2 represent the first and second layers in the feedforward network, respectively.

[0089] On this basis, a residual connection is added after each layer to prevent the gradient from disappearing and improve the training stability. The outputs of the self-attention layer and the feedforward neural network layer are expressed as:

[0090] Output=LayerNorm(X+Attention) (5)

[0091] Output = LayerNorm (X + FFN) (6)

[0092] Step S204, constructing the total loss function of the model, specifically:

[0093] Construct regression loss E r , used to evaluate the prediction error of the model:

[0094]

[0095] In the formula, Characterize the input excitation current array X1 and the input injector inlet pressure array X2 to predict the output injection rate through the Transformer model; y i is the actual value of the injection rate.

[0096] Constructing the structure loss E s , used to reduce overfitting of the model:

[0097]

[0098] In the formula, ω and β are the weight and bias of the Transformer model respectively. is the number of model architecture layers, and n is the number of nodes in each layer.

[0099] Construct the initial condition loss E i , used to reduce overfitting of the model:

[0100]

[0101] Where, t p-down is the initial moment of the ejector inlet pressure drop, v sound is the speed of sound in the pipeline, and L is the distance from the ball valve to the pressure sensor.

[0102] Constructing the monotonicity loss E m , used to meet the actual physical mechanism and increase the interpretability of the model:

[0103]

[0104] In the formula, m i is the monotonicity factor, which is equal to 1 if the physical monotonicity is positive and equal to -1 if the physical monotonicity is negative.

[0105] An embodiment of the monotonicity rule is as follows:

[0106] (a) Predicted monotonic relationship between injection rate and pressure fluctuation. When the needle valve cannot be fully opened, and the rail end compression wave is not reflected to the measurement point, the injection rate before the inflection point of the injection rate rising stage is monotonically opposite to the injector inlet pressure, and the injection rate after the inflection point is monotonically the same as the injector inlet pressure; when the rail end compression wave is reflected to the measurement point, before the inflection point of the injection rate rising stage, the injection rate is monotonically opposite to the injector inlet pressure, and during the period between the inflection point and the reflection point, the injection rate is monotonically the same as the injector inlet pressure, and after the reflection point, the injection rate is monotonically opposite to the injector inlet pressure. When the needle valve can be fully opened, before the rail end compression wave is reflected to the measurement point, the injection rate is monotonically opposite to the injector inlet pressure, and during the period between this moment and the inflection point of the injection rate closing stage, the injection rate is monotonically the same as the injector inlet pressure; after the inflection point of the injection rate closing stage, the injection rate is monotonically opposite to the injector inlet pressure.

[0107] (b) Monotonic relationship between the predicted injection rate and the actual injection rate. The derivative of the predicted injection rate between two consecutive points has the same monotonicity as the derivative of the actual injection rate between two consecutive points.

[0108] In summary, the total loss function of the model is:

[0109] E=E r +λ s E s +λ i E i +λ m E m (11)

[0110] In the formula, λ s is the weight coefficient of the result loss, λ i is the weight coefficient of the initial condition loss, λ m is the weight coefficient of monotonicity loss.

[0111] Step S4, such as Figure 4 As shown, specifically:

[0112] Step S401, input the real-time measured excitation current and injector inlet pressure time series data into the Transformer model, and then calculate the output injection rate Q rate (t), the total injection quantity Q of each injection is calculated by time integration total :

[0113] Q total =∫Q rate (t)dt (12)

[0114] The deviation e and deviation percentage δ between the real-time predicted injection quantity and the target injection quantity are calculated in real time:

[0115]

[0116] Step S402, designing a fuzzy controller, including:

[0117] Input fuzzification, specifically, converting the deviation and deviation percentage between the real-time predicted injection quantity and the target injection quantity into a fuzzy quantity; defining a fuzzy rule base, specifically, defining fuzzy rules according to the properties of the fuzzy quantity; fuzzy reasoning, specifically, reasoning the input fuzzy quantity based on the fuzzy rule base to obtain the fuzzy control quantity; defuzzification, specifically, converting the fuzzy control quantity into the PID controller parameter adjustment quantity. Finally, the fuzzy controller obtains the PID parameter adjustment quantity according to the injection quantity deviation e and the deviation percentage δ. The adjustment formula of the fuzzy PID controller output is:

[0118] u(t)=(K p0 +ΔK p )·e+(K i0 +ΔK i )∫e dt+(K d0 +ΔK d )·δ (14)

[0119] In the formula, K p0 , K i0 and K d0 is the initial PID parameter, ΔK p , ΔK i and ΔK d It is the parameter adjustment amount calculated by the fuzzy controller.

[0120] Step S403, by comparing the real-time predicted injection amount with the target injection amount, the fuzzy PID controller outputs a control signal T of the next injection pulse width according to the calculated PID adjustment amount. inject :

[0121] T inject =T inject,0 +u(t) (15)

[0122] Where, T inject,0 is the initial injection pulse width. inject As the next injection pulse width, real-time closed-loop control of the injection amount is achieved.

Claims

1. A fuzzy PID closed-loop control method for injection quantity of a high-pressure fuel injection system based on physical mechanism-data driven fusion, characterized by: The steps include: (1) An injector is provided, the inlet of the injector is connected to a high-pressure fuel pipe, a single-shot injection instrument is installed below the injector, a current clamp is clamped on the wire of the high-speed solenoid valve of the injector, a pressure sensor is installed at the connection between the injector and the high-pressure fuel pipe, the current clamp, the pressure sensor and the single-shot injection instrument are all connected to a data acquisition unit, the excitation current signal of the high-speed solenoid valve and the pressure signal at the inlet of the injector are measured, the injection rate of the injector is measured using the single-shot injection instrument, and a data set of the excitation current, the injector inlet pressure and the injection rate under different rail pressure conditions and injection pulse width conditions is obtained, and the data in the data set is batch processed and normalized; (2) Construct the Transformer time series neural network model framework, use the structural loss and regression loss as the basic loss functions of the network model, use the initial condition loss and monotonicity loss as the physical mechanism loss functions of the network model, and use the sum of the basic loss function and the physical mechanism loss function as the total loss function of the network model; (3) Based on the data set established in step (1), the Transformer time series neural network model is trained so that the model can predict the injection rate at the same time based on the excitation current and the injector inlet pressure time series data. After each round of training, it is checked whether the error is lower than the preset 5% threshold. If the error meets the requirement, the iteration is stopped. Otherwise, the training iteration is continued until the error between the predicted injection rate and the measured injection rate meets the threshold requirement. (4) Based on the excitation current signal and the pressure signal at the injector inlet measured in real time by the high-pressure fuel injection system, the Transformer time series neural network model established in step (3) is used to predict the injection rate in real time, and the real-time injection amount of each injection is obtained by time integration; the real-time injection amount is compared with the target injection amount, and the injection amount deviation and the injection amount deviation percentage are calculated by fuzzy rules; the fuzzy controller fuzzifies the deviation and the deviation percentage according to the defined fuzzy rule base, outputs the fuzzy control quantity through fuzzy reasoning, and obtains the parameter adjustment quantity of the PID controller through the defuzzification process; and the injection pulse width control signal output by the fuzzy PID controller is superimposed to complete the real-time closed-loop control of the injection pulse width of the next injection.

2. According to claim 1, a fuzzy PID closed-loop control method for injection quantity of a high-pressure fuel injection system based on physical mechanism-data driven fusion is characterized by: In step (1), the rail pressure conditions include all pressure conditions between the idle rail pressure and the rated pressure with an interval of 10 MPa; the injection pulse width conditions include all injection pulse width conditions between the minimum injection start pulse width and the rated pulse width with an interval of 0.2 ms.

3. The method for fuzzy PID closed-loop control of injection quantity of a high-pressure fuel injection system based on physical mechanism-data driven fusion according to claim 1 is characterized by: The data batch normalization in step (1) is specifically as follows: In the formula, x * is the processed data; represents the mean of unnormalized data; represents the data standard deviation; β and γ are translation parameters and scale parameters respectively; ε is a margin used to prevent division by zero.

4. The method for controlling the injection quantity of a high-pressure fuel injection system based on a fuzzy PID closed-loop control of a physical mechanism-data driven fusion according to claim 1 is characterized by: Step (2) specifically includes: (21) Define the input data format. The time series data of the excitation current and the injector inlet pressure are used as input features, namely X1 and X2, respectively. The input format is (B, T, F), where B is the batch size, T is the sampling step, and F is the number of input features. X1 and X2 are combined to form a complete input matrix (22) Create a position code and add it to the input feature to provide time sequence information for the input data. The position code vector PE is calculated as: Where i is the dimension index, d model is the embedding dimension; (23) Establish the Transformer encoder layer, specifically: Capture the long-term temporal dependencies of input time series data through multi-head self-attention layers: Where Q, K, and V are query matrix, key matrix, and value matrix, respectively. k It is the attention dimension; A feedforward neural network is established after each attention layer, which is expressed as: FFN(x)=max(0,XW1+b1)W2+b2 Where W and b are the weight and bias of the feedforward neural network, respectively, and the subscripts 1 and 2 represent the first and second layers in the feedforward network, respectively; On this basis, residual connections are added after each layer, and the outputs of the self-attention layer and the feedforward neural network layer are expressed as: Output=LayerNorm(X+Attention) Output = LayerNorm (X + FFN) (24) Construct the total loss function of the model, specifically: Construct regression loss E r , used to evaluate the prediction error of the model: In the formula, Characterize the input excitation current array X1 and the input injector inlet pressure array X2 to predict the output injection rate through the Transformer model; y i is the actual value of the injection rate; Constructing the structure loss E s : In the formula, ω and β are the weight and bias of the Transformer model respectively. is the number of layers of the model architecture, and n is the number of nodes in each layer; Construct the initial condition loss E i : Where, t p-down is the initial moment of the ejector inlet pressure drop, v sound is the speed of sound in the pipeline, L is the distance from the ball valve to the pressure sensor; Constructing the monotonicity loss E m : In the formula, m i is the monotonicity factor; In summary, the total loss function of the model is: E=E r +λ s AND s +λ i AND i +λ m AND m In the formula, λ s is the weight coefficient of the result loss, λ i is the weight coefficient of the initial condition loss, λ m is the weight coefficient of monotonicity loss.

5. The method of fuzzy PID closed-loop control of injection quantity of high-pressure fuel injection system based on physical mechanism-data driven fusion according to claim 1 is characterized in that: Step (4) is specifically as follows: (41) The real-time measured excitation current and injector inlet pressure time series data are input into the Transformer model, and then the output injection rate Q is calculated. rate (t), the total injection quantity Q of each injection is calculated by time integration total : Q total =∫Q rate (t)dt The error e and error percentage δ between the real-time predicted injection amount and the target injection amount are calculated in real time: (42) The fuzzy controller is designed, including input fuzzification, definition of fuzzy rule base, fuzzy reasoning and defuzzification. The fuzzy controller obtains the parameter adjustment amount of PID according to the injection amount deviation e and the deviation percentage δ. The adjustment formula of the fuzzy PID controller output is: u(t)=(K p0 +ΔK p )·e+(K i0 +ΔK i )∫e dt+(K d0 +ΔK d )·δ In the formula, K p0 , K i0 and K d0 is the initial PID parameter, ΔK p , ΔK i and ΔK d is the parameter adjustment amount calculated by the fuzzy controller; (43) By comparing the real-time predicted injection amount with the target injection amount, the fuzzy PID controller outputs the control signal T of the next injection pulse width according to the calculated PID adjustment amount. inject : T inject =T inject,0 +u(t) Where, T inject,0 is the initial injection pulse width, T inject As the next injection pulse width, real-time closed-loop control of the injection amount is achieved.

Images