Multi-objective control method for diesel engine combustion process based on data-driven

Abstract

A multi-objective control method for a diesel engine combustion process based on data drive in the technical field of internal combustion engines, comprising the following steps: initializing algorithm parameters, obtaining the injection timing, rail pressure, NOx emission and fuel consumption values ​​during the operation of the diesel engine in real time; obtaining control algorithm parameters in combination with a neural network algorithm according to the control error of NOx emission and fuel consumption at the current moment; obtaining the pseudo-Jacobian matrix at the current moment according to the minimization criterion function and the pseudo-Jacobian matrix estimation algorithm; obtaining the injection timing and rail pressure control quantity at the current moment according to the model-free adaptive algorithm formula and the control target weight rule based on the control system objective function and the pseudo-Jacobian estimation matrix, and implementing control on the diesel engine; performing neural network learning, obtaining the control error of NOx emission and fuel consumption of the diesel engine after implementing the control action, and adjusting the neural network parameters online according to the error back propagation algorithm to realize the adaptive adjustment of the control parameters. The present invention can meet the requirements of realizing multi-objective effective control of the diesel engine combustion process based on the real-time operation information of the diesel engine without modeling the diesel engine and the control structure.

Description

Technical Field

[0001] The present invention relates to a control method in the technical field of internal combustion engines, in particular to a data-driven multi-objective control method for a diesel engine combustion process that can achieve the goal of not needing to model a control system. Background Art

[0002] Diesel engines are widely used in ship propulsion and power plant systems due to their high thermal efficiency, good economy, easy starting, and adaptability to various engineering machinery. With the increasingly stringent emission regulations and the increasing requirements for fuel economy, this has promoted the development of engine electronic control technology, making the control of diesel injection parameters more flexible. However, diesel engine nitrogen oxide emissions and fuel consumption control are often coupled and mutually restricted. In order to determine the optimal control parameters of the diesel engine, a large number of test calibrations must be carried out, which not only consumes a lot of time and cost, but also cannot meet the increasingly complex diesel engine systems and the diverse needs of control objectives. Therefore, in order to achieve multi-objective control of fuel consumption and emissions under all operating conditions, it is particularly important to feedback the combustion process of the diesel engine and adjust the control parameters under multiple objectives according to real-time operating status information.

[0003] In the closed-loop feedback control of the combustion process, the control algorithms mainly include model predictive control and PID control. However, with the increase in the types of fuels and combustion methods of compression ignition engines, the mapping relationship between engine control parameters and combustion characteristics and performance parameters has changed, and a large amount of data needs to be re-acquired for model building. At the same time, when performing multi-objective control on the diesel engine combustion process, there is strong coupling and strong nonlinearity of multiple parameters, which greatly increases the application and design difficulty of the controller. Summary of the invention

[0004] In view of the deficiencies of the above-mentioned technologies, the present invention proposes a data-driven multi-objective control method for the diesel engine combustion process. Without relying on diesel engine modeling and controller structure design, the multi-objective control of the diesel engine combustion process is achieved by real-time capturing the relationship between the diesel engine injection timing, rail pressure and nitrogen oxide emissions and fuel consumption.

[0005] To solve the above technical problems, the present invention includes the following steps: step 1, algorithm parameter initialization, real-time acquisition of injection timing, rail pressure, nitrogen oxide emission and fuel consumption values ​​during the operation of the diesel engine; step 2, according to the control error of nitrogen oxide emission and fuel consumption at the current moment, the control algorithm parameters are obtained in combination with the neural network algorithm; step 3, according to the minimization criterion function and the pseudo-Jacobian matrix estimation algorithm, the pseudo-Jacobian matrix at the current moment is obtained; step 4, based on the control system objective function and the pseudo-Jacobian estimation matrix, the injection timing and rail pressure control amount at the current moment are obtained according to the model-free adaptive algorithm formula and the control target weight rule, and the diesel engine is controlled; step 5, neural network learning is performed to obtain the control error of nitrogen oxide emission and fuel consumption of the diesel engine after the control action is implemented, and the neural network parameters are adjusted online according to the error back propagation algorithm to realize the adaptive adjustment of the control parameters.

[0006] Furthermore, in the present invention, the control algorithm parameters are initialized to initialize the control algorithm parameters and the neural network parameters, wherein the control algorithm parameters include the pseudo-Jacobi estimation matrix μ, η, ρ, λ, neural network parameters include the weight coefficient w from the input layer to the hidden layer (2) and the weight coefficient w from the hidden layer to the output layer (3) .

[0007] Furthermore, in the present invention, the neural network algorithm is:

[0008]

[0009] Where: X(1) is the nitrogen oxide emission value of the diesel engine, in g / (kW·h); X(2) is the fuel consumption value of the diesel engine, in g / (kW·h); is the nitrogen oxide emission input value of the neural network, in g / (kW·h); is the fuel consumption input value of the neural network, in g / (kW·h); w (2) is the weight from the input layer to the hidden layer; O (2) is the output value of the hidden layer of the neural network; w (3) is the weight from the hidden layer to the output layer; ρ and η are algorithm parameters; Output value for the neural network; Output value for the neural network.

[0010] Furthermore, in the present invention, the minimization criterion function is:

[0011]

[0012] Where: k is the number of algorithm runs; Φ c(k) is the pseudo-Jacobian matrix of the system at time k; Δy(k) is the change in nitrogen oxide emissions and fuel consumption of the system at time k, in units of g / (kW·h); Δu(k-1) is the control change of the injection timing and rail pressure of the system at time k-1, in units of °CA and MPa, respectively; is the estimated value of the pseudo-Jacobian matrix of the system at time k-1; μ is the algorithm parameter.

[0013] Furthermore, in the present invention, the pseudo-Jacobian matrix estimation algorithm is:

[0014]

[0015] Where: k is the number of times the algorithm runs; is the pseudo-Jacobian estimation matrix of the system at time k; is the estimated value of the pseudo-Jacobian matrix of the system at time k-1; Δy(k) is the change in nitrogen oxide emissions and fuel consumption of the system at time k, in g / (kW·h); Δu(k-1) is the control change of the injection timing and rail pressure of the system at time k-1, in °CA and MPa respectively; μ and η are algorithm parameters.

[0016] Furthermore, in the present invention, the control system objective function is:

[0017] J(u(k))=[y * (k+1)-y(k+1)] T R[y * (k+1)-y(k+1)]+μ‖u(k)-u(k-1)‖ 2

[0018] Where: k is the number of algorithm runs; u(k) is the control amount of the system on the injection timing and rail pressure at time k, in units of °CA and MPa respectively; u(k-1) is the control amount of the system on the injection timing and rail pressure at time k-1, in units of °CA and MPa respectively; y * (k+1) is the control target value of the system's nitrogen oxide emissions and fuel consumption, in g / (kW·h); y(k+1) is the system's nitrogen oxide emissions and fuel consumption at time k+1, in g / (kW·h); R is the control target weight matrix; u is the algorithm parameter.

[0019] Furthermore, in the present invention, the model-free adaptive algorithm formula is:

[0020]

[0021] Where: k is the number of algorithm runs; u(k) is the control amount of the system on the injection timing and rail pressure at time k, in units of °CA and MPa respectively; u(k-1) is the control amount of the system on the injection timing and rail pressure at time k-1, in units of °CA and MPa respectively; y * (k+1) is the control target value of the system's nitrogen oxide emissions and fuel consumption, in g / (kW·h); y(k) is the system's nitrogen oxide emissions and fuel consumption at time k, in g / (kW·h); is the pseudo-Jacobian estimation matrix of the system at time k; R is the control target weight matrix; ρ and λ are algorithm parameters;

[0022] Among them, the control target weight rule is: during the fuel consumption control process, when the actual value of nitrogen oxide emissions is greater than the limit value, the control weight coefficients of nitrogen oxide emissions and fuel consumption are kept the same and are both 0.5; when the actual value of emissions is less than or equal to the limit value, the control weight coefficient of nitrogen oxide emissions is switched to 0, and the corresponding control weight is added to the control weight coefficient of fuel consumption, so that the fuel consumption control weight coefficient becomes larger to achieve multi-target control.

[0023] Furthermore, in the present invention, the error back propagation algorithm is:

[0024]

[0025] Where: k is the number of algorithm runs; u(k) is the control amount of the system on the injection timing and rail pressure at time k, in units of °CA and MPa respectively; Output algorithm parameter value of the neural network of the system at time k; is the neural network output algorithm parameter value of the system at time k; b is the algorithm correction term; λ is the algorithm parameter; is the pseudo-Jacobian estimation matrix of the system at time k; R is the control target weight matrix; y * (k+1) is the control target value of the system's nitrogen oxide emissions and fuel consumption, in g / (kW·h); y(k) is the system's nitrogen oxide emissions and fuel consumption values ​​at time k, in g / (kW·h); Δy(k) is the change in nitrogen oxide emissions and fuel consumption of the system at time k, in g / (kW·h); is the pseudo-Jacobian estimation matrix of the system at time k-1; Δu(k-1) is the control change of the injection timing and rail pressure of the system at time k-1, in units of °CA and MPa respectively; μ is the algorithm parameter.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention combines a neural network with a model-free adaptive control algorithm, and realizes multi-objective control of the diesel engine combustion process under all operating conditions by capturing the internal correlation among diesel engine nitrogen oxide emissions, fuel consumption, injection timing, and rail pressure in real time; the present invention can not only greatly reduce the workload of diesel engine control calibration, but also provide technical support for the control of compression ignition engines using new combustion modes and alternative fuels. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a flow chart of the control method of the present invention;

[0028] Figure 2 It is a multi-objective control effect diagram of nitrogen oxide emission and fuel consumption after adopting the control method of the present invention in a specific implementation case of the present invention. DETAILED DESCRIPTION

[0029] In order to make the contents of the present invention easier to understand, the technical solutions of the present invention are further explained below in combination with specific implementation methods. The examples are only used to illustrate the present invention, but the present invention is not limited to these contents. The operation methods without specific conditions in the following examples are usually carried out according to conventional conditions or according to the product instructions.

[0030] Example

[0031] Specific implementation examples include Figure 1 First, the algorithm parameters are initialized to obtain the injection timing, rail pressure, nitrogen oxide emissions and fuel consumption values ​​of the diesel engine in real time during operation. The control algorithm parameters are initialized to the control algorithm parameters and neural network parameters. The control algorithm parameters include the pseudo-Jacobian estimation matrix μ, η, ρ, λ, neural network parameters include the weight coefficient w from the input layer to the hidden layer (2) and the weight coefficient w from the hidden layer to the output layer (3) .

[0032] Then, according to the control error of nitrogen oxide emission and fuel consumption at the current moment, the control algorithm parameters are obtained through the following neural network algorithm formula.

[0033]

[0034] Where: X(1) is the nitrogen oxide emission value of the diesel engine, in g / (kW·h); X(2) is the fuel consumption value of the diesel engine, in g / (kW·h); is the nitrogen oxide emission input value of the neural network, in g / (kW·h); is the fuel consumption input value of the neural network, in g / (kW·h); w (2)is the weight from the input layer to the hidden layer; O (2) is the output value of the hidden layer of the neural network; w (3) is the weight from the hidden layer to the output layer; ρ and η are algorithm parameters; Output value for the neural network; Output value for the neural network.

[0035] Then, according to the minimization criterion function

[0036]

[0037] Provides a method for estimating the pseudo-Jacobian matrix of the system, and calculates the pseudo-Jacobian matrix at the current moment through the following formula:

[0038]

[0039] Where: k is the number of times the algorithm runs; is the pseudo-Jacobian estimation matrix of the system at time k; is the estimated value of the pseudo-Jacobian matrix of the system at time k-1; Δy(k) is the change in nitrogen oxide emissions and fuel consumption of the system at time k, in g / (kW·h); Δu(k-1) is the control change of the injection timing and rail pressure of the system at time k-1, in °CA and MPa respectively; μ and η are algorithm parameters.

[0040] Then, based on the control system objective function

[0041] J(u(k))=[y * (k+1)-y(k+1)] T R[y * (k+1)-y(k+1)]+μ‖u(k)-u(k-1)‖ 2 Solve the extreme point of the system input u(k), where: k is the number of algorithm runs; u(k) is the control amount of the system on the injection timing and rail pressure at time k, in units of °CA and MPa respectively; u(k-1) is the control amount of the system on the injection timing and rail pressure at time k-1, in units of °CA and MPa respectively; y * (k+1) is the control target value of the system's nitrogen oxide emissions and fuel consumption, in g / (kW·h); y(k+1) is the system's nitrogen oxide emissions and fuel consumption at time k+1, in g / (kW·h); R is the control target weight matrix; μ is the algorithm parameter.

[0042] Combined with the pseudo-Jacobian estimation matrix solved above, in the process of multi-objective control of fuel consumption, when the actual value of nitrogen oxide emissions is greater than the limit value, the control weight coefficients of nitrogen oxide emissions and fuel consumption are kept the same and are both 0.5; when the actual value of emissions is less than or equal to the limit value, the control weight coefficient of nitrogen oxide emissions is switched to 0, and the corresponding control weight is added to the control weight coefficient of fuel consumption to increase the fuel consumption control weight coefficient. The following model-free adaptive algorithm formula can be obtained, and the injection timing and rail pressure control amount at the current moment are obtained, and the diesel engine is controlled.

[0043]

[0044] Where: k is the number of algorithm runs; u(k) is the control amount of the system on the injection timing and rail pressure at time k, in units of °CA and MPa respectively; u(k-1) is the control amount of the system on the injection timing and rail pressure at time k-1, in units of °CA and MPa respectively; y * (k+1) is the control target value of the system's nitrogen oxide emissions and fuel consumption, in g / (kW·h); y(k) is the system's nitrogen oxide emissions and fuel consumption at time k, in g / (kW·h); is the pseudo-Jacobian estimation matrix of the system at time k; R is the control target weight matrix; ρ is the algorithm parameter; λ is the algorithm parameter; μ is the algorithm parameter.

[0045] Finally, neural network learning is performed to obtain the control errors of diesel engine nitrogen oxide emissions and fuel consumption after the control action is implemented. The neural network parameters are adjusted online according to the following error back propagation algorithm to achieve adaptive adjustment of the control parameters.

[0046]

[0047] Where: k is the number of algorithm runs; u(k) is the control amount of the system on the injection timing and rail pressure at time k, in units of °CA and MPa respectively; Output algorithm parameter value of the neural network of the system at time k; is the neural network output algorithm parameter value of the system at time k; b is the algorithm correction term; λ is the algorithm parameter; is the pseudo-Jacobian estimation matrix of the system at time k; R is the control target weight matrix; y * (k+1) is the control target value of the system's nitrogen oxide emissions and fuel consumption, in g / (kW·h); y(k) is the system's nitrogen oxide emissions and fuel consumption values ​​at time k, in g / (kW·h); Δy(k) is the change in nitrogen oxide emissions and fuel consumption of the system at time k, in g / (kW·h); is the pseudo-Jacobian estimation matrix of the system at time k-1; Δu(k-1) is the control change of the injection timing and rail pressure of the system at time k-1, in units of °CA and MPa respectively; μ is the algorithm parameter.

[0048] Taking the specific test results of a diesel engine as an example, under different loads, by dynamically adjusting the diesel engine nitrogen oxide emission limit, and adjusting the fuel consumption target value to adapt to the nitrogen oxide emission limit, after applying the control method of the present invention, the control changes of various parameters during the test process are as follows: Figure 2 As shown. Figure 2 It can be seen that during the control process, when the nitrogen oxide emission limit increases and the fuel consumption control target gradually decreases, the control system will adjust the injection timing and rail pressure in real time according to the control algorithm of the present invention, so that the actual fuel consumption can quickly and accurately follow the changes in the target fuel consumption. Although the nitrogen oxide emissions under partial load are overshot during the transient adjustment process, it can still achieve stable convergence to the limit. During the entire test process, the diesel engine speed always maintains stable operation. Based on the above control effect, the control algorithm proposed by the present invention realizes multi-objective control of diesel engine nitrogen oxide emissions and fuel consumption under all working conditions by adjusting the rail pressure and injection timing in real time.

[0049] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A data-driven multi-objective control method for diesel engine combustion process, characterized in that: The following steps are involved: Step 1: Initialize the algorithm parameters and obtain the injection timing, rail pressure, nitrogen oxide emissions and fuel consumption values ​​during the operation of the diesel engine in real time; Step 2, according to the control error of nitrogen oxide emission and fuel consumption at the current moment, the control algorithm parameters are obtained by combining the neural network algorithm; Step 3, according to the minimization criterion function and the pseudo-Jacobian matrix estimation algorithm, the pseudo-Jacobian matrix at the current moment is obtained; Step 4: Based on the control system objective function and pseudo-Jacobian estimation matrix, the injection timing and rail pressure control quantity at the current moment are obtained according to the model-free adaptive algorithm formula and the control target weight rule, and the diesel engine is controlled; Step 5: Perform neural network learning to obtain the control errors of diesel engine nitrogen oxide emissions and fuel consumption after implementing the control action, and adjust the neural network parameters online according to the error back propagation algorithm to achieve adaptive adjustment of the control parameters; Wherein, in step 2, the neural network algorithm is: Where: X(1) is the nitrogen oxide emission value of the diesel engine, in g / (kW·h); X(2) is the fuel consumption value of the diesel engine, in g / (kW·h); is the nitrogen oxide emission input value of the neural network, in g / (kW·h); is the fuel consumption input value of the neural network, in g / (kW·h); w (2) is the weight from the input layer to the hidden layer; O (2) is the output value of the hidden layer of the neural network; w (3) is the weight from the hidden layer to the output layer; ρ and η are algorithm parameters; Output value for the neural network; Output value for the neural network; In step three, the minimization criterion function is: Where: k is the number of algorithm runs; Φ c (k) is the pseudo-Jacobian matrix of the system at time k; Δy(k) is the change in nitrogen oxide emissions and fuel consumption of the system at time k, in units of g / (kW·h); Δu(k-1) is the control change of the injection timing and rail pressure of the system at time k-1, in units of °CA and MPa, respectively; is the estimated value of the pseudo-Jacobian matrix of the system at time k-1; μ is the algorithm parameter; In step three, the pseudo-Jacobian matrix estimation algorithm is: Where: k is the number of times the algorithm runs; is the pseudo-Jacobian estimation matrix of the system at time k; is the estimated value of the pseudo-Jacobian matrix of the system at time k-1; Δy(k) is the change in nitrogen oxide emissions and fuel consumption of the system at time k, in g / (kW·h); Δu(k-1) is the control change of the injection timing and rail pressure of the system at time k-1, in °CA and MPa respectively; μ and η are algorithm parameters; In step 4, the control system objective function is: J(u(k))=[y * (k+1)-y(k+1)] T R[y * (k+1)-y(k+1)]+μ‖u(k)-u(k-1)‖ 2 Where: k is the number of algorithm runs; u(k) is the control amount of the system on the injection timing and rail pressure at time k, in units of °CA and MPa respectively; u(k-1) is the control amount of the system on the injection timing and rail pressure at time k-1, in units of °CA and MPa respectively; y * (k+1) is the control target value of the system's nitrogen oxide emissions and fuel consumption, in g / (kW·h); y(k+1) is the system's nitrogen oxide emissions and fuel consumption at time k+1, in g / (kW·h); R is the control target weight matrix; μ is the algorithm parameter; In step 4, the model-free adaptive algorithm formula is: Where: k is the number of algorithm runs; u(k) is the control amount of the system on the injection timing and rail pressure at time k, in units of °CA and MPa respectively; u(k-1) is the control amount of the system on the injection timing and rail pressure at time k-1, in units of °CA and MPa respectively; y * (k+1) is the control target value of the system's nitrogen oxide emissions and fuel consumption, in g / (kW·h); y(k) is the system's nitrogen oxide emissions and fuel consumption at time k, in g / (kW·h); is the pseudo-Jacobian estimation matrix of the system at time k; R is the control target weight matrix; ρ and λ are algorithm parameters; In step five, the error back propagation algorithm is: Where: k is the number of algorithm runs; u(k) is the control amount of the system on the injection timing and rail pressure at time k, in units of °CA and MPa respectively; Output algorithm parameter value of the neural network of the system at time k; is the neural network output algorithm parameter value of the system at time k; b is the algorithm correction term; λ is the algorithm parameter; is the pseudo-Jacobian estimation matrix of the system at time k; R is the control target weight matrix; y * (k+1) is the control target value of the system's nitrogen oxide emissions and fuel consumption, in g / (kW·h); y(k) is the system's nitrogen oxide emissions and fuel consumption values ​​at time k, in g / (kW·h); Δy(k) is the change in nitrogen oxide emissions and fuel consumption of the system at time k, in g / (kW·h); is the pseudo-Jacobian estimation matrix of the system at time k-1; Δu(k-1) is the control change of the injection timing and rail pressure of the system at time k-1, in units of °CA and MPa respectively; μ is the algorithm parameter.

2. The data-driven multi-objective control method for diesel engine combustion process according to claim 1 is characterized in that In step 1, the control algorithm parameters are initialized to initialize the control algorithm parameters and the neural network parameters, wherein the control algorithm parameters include the pseudo-Jacobi estimation matrix μ, η, ρ, λ, neural network parameters include the weight coefficient w from the input layer to the hidden layer (2) and the weight coefficient w from the hidden layer to the output layer (3) .

3. The data-driven multi-objective control method for diesel engine combustion process according to claim 1 is characterized in that In step four, the control target weight rule is: in the multi-target control process of fuel consumption, when the actual value of nitrogen oxide emissions is greater than the limit value, the control weight coefficients of nitrogen oxide emissions and fuel consumption are kept the same and are both 0.5; when the actual value of emissions is less than or equal to the limit value, the control weight coefficient of nitrogen oxide emissions is switched to 0, and the corresponding control weight is added to the control weight coefficient of fuel consumption, so that the fuel consumption control weight coefficient becomes larger to achieve multi-target control.

Images