CN113867151A - SDNN-MPC-based fuel-fuel combined power plant load distribution control method - Google Patents
SDNN-MPC-based fuel-fuel combined power plant load distribution control method Download PDFInfo
- Publication number
- CN113867151A CN113867151A CN202111207951.2A CN202111207951A CN113867151A CN 113867151 A CN113867151 A CN 113867151A CN 202111207951 A CN202111207951 A CN 202111207951A CN 113867151 A CN113867151 A CN 113867151A
- Authority
- CN
- China
- Prior art keywords
- rotating speed
- sdnn
- fuel
- state
- load distribution
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
- G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
Abstract
The invention aims to provide a load distribution control method of a fuel-fuel combined power plant based on an SDNN-MPC (software development neural network-MPC), which comprises the steps of collecting the rotating speed of a high-pressure turbine, the rotating speed of a low-pressure turbine and the rotating speed of a propeller of each gas turbine, taking a linear variable parameter model (LPV) as a prediction model, identifying the current running state of a system, and further predicting the power transfer track of the system at the future time; solving a quadratic plan with constraint on line by adopting a simplified dual neural network algorithm (SDNN), and finding out the optimal fuel flow of the combustion engine at the current moment; based on the rolling optimization design, the operation is repeated in each control step length, and the optimization control of the whole load distribution control is realized. The invention can realize the balance of the power between different gas turbines of the system in any proportion, and has good rapidity and better stability of the rotating speed of the propeller in the process of power transfer.
Description
Technical Field
The invention relates to an engine control method, in particular to a gas turbine control method.
Background
In recent years, with the deep development of marine resources and the high importance of marine interests in various countries in the world, the technology in the field of ships has been rapidly developed. The combined combustion-combustion power device is widely applied to ship propulsion as a power form with high power, good maneuverability and flexible working mode. For a combined combustion-combustion power plant, the control system affects the overall operating performance of the plant.
The existing load distribution control strategy mainly includes: the parallel mode, the master-slave mode and the parallel power feedback mode are realized mainly based on Proportional Integral Derivative (PID) no matter which control strategy is adopted, and control parameter optimization is carried out on the basis of the PID even if the PID is improved. The PID algorithm eliminates errors through feedback. This control lags its regulation behind the powertrain. If the PID parameters are not set reasonably, the phenomenon of power oscillation occurs, and the power cannot be distributed in proportion. During the power distribution process, the operating state of the system generally fluctuates greatly. Because PID can not restrain relevant variables, the stability of the rotating speed of the propeller can not be ensured. Model Predictive Control (MPC) is an algorithm that combines the system identification theory, optimal control theory and adaptive control theory, and has been widely studied in the fields of petrochemical industry and aerospace. With the development of control algorithms and hardware technologies, the method is gradually applied to the field of gas turbine control.
Disclosure of Invention
The invention aims to provide a fuel-combustion combined power plant load distribution control method based on an SDNN-MPC, which can simultaneously realize power distribution in any proportion among combustion engines of the fuel-combustion combined power plant.
The purpose of the invention is realized as follows:
the invention discloses a fuel-fuel combined power device load distribution control method based on an SDNN-MPC, which is characterized by comprising the following steps: collecting the high-pressure turbine rotating speed, the low-pressure turbine rotating speed and the propeller rotating speed of each gas turbine, identifying the current running state of the system by taking a linear variable parameter model (LPV) as a prediction model, and further predicting the power transfer track of the system at the future time; solving a quadratic plan with constraint on line by adopting a simplified dual neural network algorithm (SDNN), and finding out the optimal fuel flow of the combustion engine at the current moment; based on the rolling optimization design, the operation is repeated in each control step length, and the optimization control of the whole load distribution control is realized.
The present invention may further comprise:
1. the specific steps of establishing the LPV model as a prediction model are as follows:
the combustion-combustion combined power system, namely a COGAG system, is a typical nonlinear time-varying system, a state space model of corresponding steady-state points is established by carrying out small deviation linearization at different steady-state operating points, the high-pressure turbine rotating speed of each gas turbine in the load distribution process is selected as a scheduling parameter, the connection between the state models of the different steady-state points is established, a linear variable parameter (LPV) model of the combustion-combustion combined power system is established and used as a prediction model, and the LPV model is in the following form:
wherein: nh1、nh2The high-pressure turbine rotating speed of different combustion engines; a, B, C and D are coefficient matrixes; x is the number of0A state vector that is a steady-state point; y is0An output vector that is a steady-state point; x is the number ofkState vector at time k; y iskIs the output vector at the k moment; deltaxk=xk-x0;δyk=yk-y0;δuk=uk-u0。
2. The online solution of quadratic programming with constraints by using a simplified dual neural network algorithm, namely, SDNN, specifically includes:
the simplified constraint conditions are a fuel flow sequence and a propeller rotating speed sequence, and the expression is as follows:
s.t.l≤CuΔU(k)≤h
wherein: Δ U (k) is a fuel flow sequence; H. g (k +1| k), CuIs a coefficient matrix; l and h are constraint boundary sequences;
converting the target function into a dual form, and designing a neural network based on the Kuntake (KKT) condition, wherein the neural network is expressed as:
wherein: ε is a scaling factor; z is a neuron; j (z) is a nonlinear activation function; e, P and s are coefficient arrays; i is a unit array;
solving a fuel flow sequence of an objective function by adopting a simplified dual neural network:
ΔU(k)=Pv-s
wherein: v is a vector containing information of the predicted system operation.
The invention has the advantages that:
1. the current operation state of the fuel-combustion combined power system can be accurately identified based on a linear variable parameter (LPV) model, and then the operation track of the system is predicted.
2. Because the Model Predictive Control (MPC) is designed by combining the Simplified Dual Neural Network (SDNN) with the simplified constraint condition, the execution efficiency of the controller is improved, and the method is better suitable for fast time-varying systems such as a fuel-combustion combined power system.
3. Compared with the traditional PID control, the SDNN-MPC control method can reduce the power transfer time and ensure the stability of the rotating speed of the propeller during the load transfer process.
Drawings
FIG. 1 is a schematic diagram of the present invention;
FIG. 2 is a comparison of the accuracy of the LPV model and the mechanistic model;
FIG. 3 is a graph of power characteristic variation compared to a conventional PI algorithm during load distribution according to the present invention;
FIG. 4 is a graph of the change in propeller speed compared to a conventional PI algorithm during load distribution according to the present invention;
fig. 5 is a comparison of different algorithm execution times.
Detailed Description
The invention will now be described in more detail by way of example with reference to the accompanying drawings in which:
with reference to fig. 1-5, the combustion-combustion combined power system is a nonlinear system, and performs small deviation linearization on different steady-state operating points of the system, and obtains a coefficient matrix of a state space model by a least square fitting method, thereby establishing state space equations near the different steady-state points. Selecting the high-pressure turbine rotating speeds of different combustion engines as scheduling parameters, identifying the current operating state of the system by adopting a bivariate identification method, and establishing a linear variable parameter model, wherein the model expression is as follows:
wherein: nh1,nh2The high-pressure turbine rotating speed of different combustion engines; a, B, C and D are coefficient matrixes, and the coefficient values are related to the high-pressure rotating speed; x is the number of0A state vector that is a steady-state point; y is0An output vector that is a steady-state point; x is the number ofkIs k at
State vectors of the scales; y iskIs the output vector at the k moment; deltaxk=xk-x0;δyk=yk-y0;
δuk=uk-u0。
And (3) designing a Model Predictive Control (MPC) algorithm by using a linear variable parameter (LPV) model as a prediction model. Taking the difference of the state space equations of the adjacent operating points, the following expression can be obtained:
Δxk+1=AΔxk+BΔuk
Δyk+1=CΔxk+1+DΔuk
wherein: Δ xk=xk-xk-1,Δuk=uk-uk-1,Δyk+1=yk+1-yk。
Will be Δ xk+1Substituting the above equation, by recursion, the matrix form can be obtained as follows:
wherein: y (k +1| k) [ Y ]k+1,yk+2,…,yk+p]TAnd the prediction sequence of the output power is expressed, m is the length of a control time domain, and p is the length of a prediction time domain.
And (3) with power allocation as a target, forming an objective function by the predicted control sequence and the target power sequence, wherein the form is as follows:
J(x(k),ΔU(k))=||Γy(Y(k+1|k)-R(k+1))||2+||ΓuΔU(k)||2
wherein: r (k +1) is a target power sequence; Δ U (k) is a fuel flow sequence; gamma-shapedy,ΓuIs a weight matrix.
The constraint conditions for stable operation of the fuel-fuel combined power system comprise: the surge margin of the gas compressor, the outlet temperature of the combustion chamber, the lean oil-rich flameout fuel-air ratio, the rotating speed of the power turbine and the like. The characteristics of the modeling analysis system can find that once the operating ranges of the fuel flow and the rotating speed of the power turbine are reasonable, other boundary conditions can be met. The invention takes the fuel flow and the propeller rotating speed (the propeller rotating speed and the power turbine rotating speed have a transmission ratio coefficient proportional relation) as constraint conditions. The objective function is arranged into a quadratic programming form:
s.t.l≤CuΔU(k)≤h
wherein: h, G (k +1| k), CuIs a coefficient matrix; l and h are constraint boundary sequences.
Converting the quadratic programming into a dual form, establishing a simplified dual neural network based on the KKT condition, wherein the neural network model can be expressed as:
wherein: ε is a scaling factor; z is a neuron; j (z) is a nonlinear activation function; e, P and s are coefficient arrays; and I is a unit array.
Further, solving a fuel flow sequence of the objective function by adopting a simplified dual neural network:
ΔU(k)=Pv-s
wherein: v is a vector containing information of the predicted system operation.
The method comprises the steps of collecting the rotating speed of a high-pressure compressor, the rotating speed of a low-pressure compressor and the rotating speed of a propeller of each combustion engine from a combustion-combustion combined power device to serve as state vector elements, and taking the output power of different combustion engines as output vector elements. The LPV model is adopted as the prediction model, so that the accuracy is good.
Solving the MPC using a Simplified Dual Neural Network (SDNN) may improve algorithm execution efficiency. The designed control method can realize the power distribution of the system in any proportion by changing the power distribution proportion coefficient.
Claims (3)
1. A fuel-combustion combined power plant load distribution control method based on an SDNN-MPC is characterized by comprising the following steps: collecting the high-pressure turbine rotating speed, the low-pressure turbine rotating speed and the propeller rotating speed of each gas turbine, identifying the current running state of the system by taking a linear variable parameter model (LPV) as a prediction model, and further predicting the power transfer track of the system at the future time; solving a quadratic plan with constraint on line by adopting a simplified dual neural network algorithm (SDNN), and finding out the optimal fuel flow of the combustion engine at the current moment; based on the rolling optimization design, the operation is repeated in each control step length, and the optimization control of the whole load distribution control is realized.
2. The SDNN-MPC based fuel-combustion combined power plant load distribution control method as set forth in claim 1, wherein: the specific steps of establishing the LPV model as a prediction model are as follows:
the combustion-combustion combined power system, namely a COGAG system, is a typical nonlinear time-varying system, a state space model of corresponding steady-state points is established by carrying out small deviation linearization at different steady-state operating points, the high-pressure turbine rotating speed of each gas turbine in the load distribution process is selected as a scheduling parameter, the connection between the state models of the different steady-state points is established, a linear variable parameter (LPV) model of the combustion-combustion combined power system is established and used as a prediction model, and the LPV model is in the following form:
wherein: nh1、nh2The high-pressure turbine rotating speed of different combustion engines; a, B, C and D are coefficient matrixes; x is the number of0A state vector that is a steady-state point; y is0An output vector that is a steady-state point; x is the number ofkState vector at time k; y iskIs the output vector at the k moment; deltaxk=xk-x0;δyk=yk-y0;δuk=uk-u0。
3. The SDNN-MPC based fuel-combustion combined power plant load distribution control method as set forth in claim 1, wherein: the online solution of quadratic programming with constraints by using a simplified dual neural network algorithm, namely, SDNN, specifically includes:
the simplified constraint conditions are a fuel flow sequence and a propeller rotating speed sequence, and the expression is as follows:
s.t.l≤CuΔU(k)≤h
wherein: Δ U (k) is a fuel flow sequence; H. g (k +1| k), CuIs a coefficient matrix; l and h are constraint boundary sequences;
converting the target function into a dual form, and designing a neural network based on the Kuntake (KKT) condition, wherein the neural network is expressed as:
wherein: ε is a scaling factor; z is a neuron; j (z) is a nonlinear activation function; e, P and s are coefficient arrays; i is a unit array;
solving a fuel flow sequence of an objective function by adopting a simplified dual neural network:
ΔU(k)=Pv-s
wherein: v is a vector containing information of the predicted system operation.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111207951.2A CN113867151A (en) | 2021-10-18 | 2021-10-18 | SDNN-MPC-based fuel-fuel combined power plant load distribution control method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111207951.2A CN113867151A (en) | 2021-10-18 | 2021-10-18 | SDNN-MPC-based fuel-fuel combined power plant load distribution control method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113867151A true CN113867151A (en) | 2021-12-31 |
Family
ID=78999947
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111207951.2A Pending CN113867151A (en) | 2021-10-18 | 2021-10-18 | SDNN-MPC-based fuel-fuel combined power plant load distribution control method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113867151A (en) |
-
2021
- 2021-10-18 CN CN202111207951.2A patent/CN113867151A/en active Pending
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103732895B (en) | Sliding mode controller, and internal combustion engine system control apparatus | |
US20160108732A1 (en) | Switch gain scheduled explicit model predictive control of diesel engines | |
US9952566B2 (en) | Method for controlling and/or regulating a technical system in a computer-assisted manner | |
CN111679574B (en) | Variable-cycle engine transition state optimization method based on large-scale global optimization technology | |
Wang et al. | Multiple model predictive functional control for marine diesel engine | |
Bing et al. | Wide-range model predictive control for aero-engine transient state | |
Iwadare et al. | Multi-variable air-path management for a clean diesel engine using model predictive control | |
CN113867151A (en) | SDNN-MPC-based fuel-fuel combined power plant load distribution control method | |
Aissat et al. | Identification of two-shaft gas turbine variables using a decoupled multi-model approach with genetic algorithm | |
CN111894752B (en) | Model predictive control algorithm-based diesel engine VGT-EGR control method | |
Bai et al. | The nonlinear single controller of DGEN380 aero engine design | |
Lutambo et al. | Turbofan engine modelling and control design using linear quadratic regulator (LQR) | |
So et al. | Temperature control of a regasification system for LNG-fuelled marine engines using nonlinear control techniques | |
Kiakojoori et al. | Dynamie neural networks for jet engine degradation prediction and prognosis | |
Esfahani et al. | Designing and Optimizing Multi-Objective Turbofan Engine Control Algorithm Using Min-Max Method | |
Panda et al. | Sliding mode control of gas turbines using multirate-output feedback | |
Cerri et al. | Inverse methodologies for actual status recognition of gas turbine components | |
CN111443595B (en) | Aeroengine gain scheduling controller based on health degradation | |
Lv et al. | Intelligent ammonia precooling control for TBCC mode transition based on neural network improved equilibrium manifold expansion model | |
CN114415506B (en) | Design method of dual-mode tracking prediction control system of aero-engine | |
Lutz et al. | Ukf-based constrained extremum-seeking control with application to a large-bore gas engine | |
Hu et al. | Data-driven model predictive control of Air-fuel Ratio for PFISI engine | |
CN113807024B (en) | Gas turbine optimal dynamic working point selection method based on proxy model | |
Delvar | Internal Combustion Engine Control Based on CFM Strategy | |
Chen et al. | Study on boiler combustion optimization based on sparse least squares support vector machine |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |