Multi-objective optimization control method for fuel-fuel combined power plant based on double-layer structure
Through the multi-objective optimization control method of the two-layer structure, the fuel flow rate and propeller speed are optimized by using model prediction control and neural network model, and the problems of economic, maneuverability and stability optimization in the fuel-fuel joint power system control are solved, and the problem of economic, maneuverability and stability optimization in the fuel-fuel joint power system control is achieved. The overall performance of the system is improved.
Patent Information
- Application Number
- CN202210976781.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-08-15
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2042-08-15
AI Technical Summary
The existing fuel-fuel combined power system control method is difficult to effectively optimize the economy, maneuverability and stability of the system, especially in dynamic processes such as single-machine acceleration and deceleration, dual-machine acceleration and deceleration and power distribution, the response is lagging and maneuverability optimization. Large space.
The multi-objective optimization control method adopts a two-layer structure, including the system planning layer and the control execution layer, optimizes fuel flow and propeller speed through model prediction control algorithms and neural network models, identifys the best economic operation points and performs dynamic process control. Simplify constraints to improve system stability.
The economy, maneuverability and stability of the combustion-fuel combined power system are achieved, which shortens the time of the dynamic process and improves the overall performance of the system.
Smart Images

Figure CN115309050B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a gas turbine control method, in particular to a fuel-fuel combined power control method. Background Art
[0002] As a power device with high power density and flexible operation mode, the fuel-fuel combined power system is widely used in ship propulsion and is a device with great development potential. The structure of the fuel-fuel combined power system is complex, and common dynamic processes include: single-machine acceleration and deceleration, dual-machine acceleration and deceleration, and power distribution. How to effectively control the system while improving the overall performance of the system requires special attention.
[0003] The combined fuel-fuel power system mainly includes one engine and two propellers and two-stage and two propellers operation modes, and variable pitch propellers are often used to improve the maneuverability of the system. Therefore, there are multiple operating points in the system that can meet the corresponding target ship speed. How to select the most economical operating point and ensure that the system remains at the corresponding operating point during steady-state operation requires focused research. For the combined fuel-fuel power system, the existing control of the dynamic process is mainly based on proportional integral differential (PID) control. PID is a control algorithm based on feedback deviation. This control method makes the system response lag behind the external excitation of the system, and the maneuverability of the system has a large optimization space. As an optimal control algorithm, the model predictive control (MPC) algorithm can predict the future time domain operation trajectory of the system and realize the optimal control of the system by solving the quadratic programming online. It is a good way to improve the maneuverability of the system. Summary of the invention
[0004] The purpose of the present invention is to provide a multi-objective optimization control method for a fuel-fuel combined power plant based on a double-layer structure, which can realize the control of different dynamic processes such as single-machine acceleration and deceleration, dual-machine acceleration and deceleration and power distribution of a fuel-fuel combined power system, and at the same time optimize the economy, maneuverability and stability of the system.
[0005] The object of the present invention is achieved in that:
[0006] The multi-objective optimization control method of the fuel-fuel combined power plant based on the double-layer structure of the present invention is characterized by:
[0007] A two-layer control structure is introduced into the fuel-fuel combined power system. The two-layer control structure mainly includes: system planning layer and control execution layer. In the system planning layer, the economic optimal operating point under different working conditions is obtained based on the formulated economic optimal control strategy, and this operating point is transmitted to the control execution layer as a reference state. In the control execution layer, the model predictive control algorithm is used to control the dynamic process of single-machine acceleration and deceleration, multi-machine synchronous acceleration and deceleration, and power distribution, and the stability of the fuel-fuel combined power system is improved by explicitly constraining the fuel flow and propeller speed.
[0008] Establish steady-state models of the fuel-fuel combined power system under different operating modes, calculate the fuel flow rate and ship speed under the corresponding combination parameters by setting different pitch ratios and power turbine speeds, select the steady-state point that meets the ship speed conditions, and then select the steady-state point with the minimum fuel flow rate as the economical optimal operating point. Perform the above operations under different working conditions, and select the optimal operating point corresponding to each working condition as the economical optimal control strategy;
[0009] At the control execution layer, the model predictive control (MPC) algorithm is used to control different dynamic processes. Based on the mechanism model of the fuel-fuel combined power system, a prediction model with a linear variable parameter form composed of a series neural network is established. Based on the established prediction model, the MPC algorithm is designed for single-machine acceleration and deceleration, dual-machine acceleration and deceleration, and power distribution processes.
[0010] The long short-term memory neural network LSTM is used to predict the steady-state point corresponding to the current operating point, and then the artificial neural network ANN is used to predict the coefficient matrix through the parameters of the steady-state point predicted by LSTM, thereby obtaining a prediction model with a linear variable parameter form composed of LSTM and ANN in series;
[0011] The multivariable constraints of the fuel-fuel combined power system are simplified to the two-variable constraints of fuel flow and propeller speed, and the above two variables are explicitly constrained in the MPC algorithm.
[0012] The present invention may also include:
[0013] 1. In order to simplify the solution process of the MPC algorithm, a linear variable parameter model is used as the prediction model. The model expression is as follows:
[0014]
[0015] Wherein: nh1 is the high-pressure speed of engine 1, nh2 is the high-pressure speed of engine 2, nl1 is the low-pressure speed of engine 1, nl2 is the low-pressure speed of engine 2, np is the propeller speed, wf1 is the fuel flow rate of engine 1, wf2 is the fuel flow rate of engine 2, PR is the pitch ratio, subscript k represents the kth sampling moment, and subscript 0 represents the corresponding steady-state point.
[0016] 2. Perform small deviation linearization processing on the fuel-fuel combined power system at different steady-state points to obtain the state space model at different steady-state points, select nl10, nl20 and PR0 as scheduling parameters, take the scheduling parameters as input, and establish a long short-term memory neural network LSTM model to predict other steady-state point parameters, including: nl10, nl20, np0, wf10, wf20, N10 and N20; take the scheduling parameters and the parameters predicted based on the LSTM model as the input of the artificial neural network ANN, and predict the matrices A, B, C, and D, so as to obtain a prediction model in the form of a linear variable parameter model combined with the series neural network technology.
[0017] 3. Taking the propeller speed as the control target, the predicted control sequence and the target power sequence are combined into an objective function in the following form:
[0018] min J2(x(k), ΔU(k))=||Г y (Y np (k)-np ref )|| 2 +||Γ u ΔU(k)|| 2 +||Γ p (P1(k)-P2(k))|| 2
[0019] Where: np ref is the target propeller speed sequence, Y np (k) is the propeller speed sequence predicted by the prediction model based on the operating state of the system at step k, ΔU(k) is the system input increment to be solved, Γ y and Γ u is the weight matrix, and a power balance penalty term is added to the objective function. P1(k) is the predicted power of 1#GT, P2(k) is the predicted power of 2#GT, and Γ p is the penalty weight matrix.
[0020] 4. Take the fuel flow rate and propeller speed as constraints and organize the constrained objective function into the form of quadratic programming:
[0021]
[0022] stl≤C u ΔU(k)≤h
[0023] Where: H, G(k+1|k), C u is the coefficient matrix; l, h are the constraint boundary sequences;
[0024] The dual neural network method is used to solve the above quadratic programming problem, and the expression of the system input increment is obtained:
[0025] ΔU(k)=Pv-s
[0026] Where: v is a vector containing the predicted system operation information, P, s is a coefficient matrix.
[0027] The advantages of the present invention are:
[0028] 1. Based on the formulated economic optimal operation strategy, the optimal operation mode and optimal operation point of the system under the current ship speed command can be effectively identified.
[0029] 2. The prediction model established based on the series neural network can accurately identify the operating status of the fuel-fuel combined power system and effectively predict the operating trajectory of the system.
[0030] 3. The model predictive control (MPC) algorithm is used to control dynamic processes such as single-machine acceleration and deceleration, dual-machine acceleration and deceleration, and power distribution, which can shorten the time consumed by the dynamic process.
[0031] 4. By adopting a double-layer control structure, the economy, maneuverability and stability of the fuel-fuel combined power system can be improved simultaneously. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] Figure 1 It is the control principle diagram of the present invention;
[0033] Figure 2 Optimize flow charts for economic control strategies;
[0034] Figure 3 This is a comparison chart of the accuracy of the prediction model and the mechanism model;
[0035] Figure 4 The propeller speed variation curve of the present invention compared with the traditional PI algorithm during the acceleration and deceleration process of the dual engines. DETAILED DESCRIPTION
[0036] The present invention is described in more detail below with reference to the accompanying drawings:
[0037] Combination Figure 1-4The present invention adopts the following technical solution: a two-layer control structure is introduced into the fuel-fuel combined power system to achieve multi-objective performance optimization of the system. The control structure mainly includes: a system planning layer and a control execution layer. In the system planning layer, the economic optimal operating point under different working conditions is obtained based on the formulated economic optimal control strategy, and this operating point is transmitted to the control execution layer as a reference state. In the control execution layer, the model predictive control algorithm is used to control the dynamic processes such as single-machine acceleration and deceleration, multi-machine synchronous acceleration and deceleration, and power distribution, so as to improve the maneuverability of the dynamic process and improve the stability of the system by explicitly constraining the fuel flow and propeller speed.
[0038] A steady-state model of the fuel-fuel combined power system under different operating modes is established. The fuel flow rate and ship speed under the corresponding combination parameters are calculated by setting different pitch ratios and power turbine speeds. The steady-state point that meets the ship speed conditions is selected, and then the steady-state point with the minimum fuel flow rate is selected as the economical optimal operating point. The above operations are performed under different operating conditions, and the optimal operating point corresponding to each operating condition is selected as the economical optimal control strategy.
[0039] The model predictive control (MPC) algorithm is used in the control execution layer to control different dynamic processes. Based on the mechanism model of the power system, a prediction model with a linear variable parameter form composed of a series neural network is established. Based on the established prediction model, an MPC algorithm is designed for single-machine acceleration and deceleration, dual-machine acceleration and deceleration, and power distribution processes to improve the mobility of the dynamic process system.
[0040] Considering that the prediction model in the form of linear variable parameters is mainly composed of steady-state points and coefficient matrices, the present invention first uses a long short-term memory neural network (LSTM) to predict the steady-state point corresponding to the current operating point, and then uses an artificial neural network (ANN) to predict the coefficient matrix through the parameters of the steady-state points predicted by LSTM, thereby obtaining a prediction model in the form of linear variable parameters composed of LSTM and ANN in series.
[0041] In order to ensure the stable operation of the system while improving the economy and maneuverability of the system, the present invention proposes a double-layer control structure to perform multi-objective optimization control on the fuel-fuel combined power system. The control method of the invention can achieve effective control of the system's single-machine acceleration and deceleration, dual-machine acceleration and deceleration, and power distribution process, improve the system's economy based on the economic optimal control strategy, improve the system's maneuverability based on model predictive control, and improve the system's stability by constraining the fuel flow and propeller speed.
[0042] See also Figure 1 , the principle of the present invention is described in detail: the two-layer control structure mainly includes: system decision layer and control execution layer. In the system decision layer, the optimal economic control strategy (the optimization process of the control strategy is as follows) is executed. Figure 2As shown in the figure, the optimal economic operation mode and operation point are identified according to the ship speed command, and this operation point is used as the reference state of the control execution layer. In the control execution layer, the dynamic processes such as single-machine acceleration and deceleration, dual-machine acceleration and deceleration, and power distribution are controlled through the model predictive control (MPC) algorithm. Next, the dual-machine synchronous acceleration and deceleration process will be taken as an example for specific implementation.
[0043] Model predictive control (MPC) requires the use of a predictive model to identify the current operating state of the system and then predict the development trajectory of the system in a certain time domain in the future. Considering that the fuel-fuel combined power system has many input variables and a wide operating area, in order to ensure the accuracy and good generalization ability of the prediction model, the neural network is applied to the modeling process of the prediction model. In order to simplify the solution process of the MPC algorithm, a linear variable parameter model is used as the prediction model, and the model expression is as follows:
[0044]
[0045] Wherein: nh1 is the high-pressure speed of engine 1, nh2 is the high-pressure speed of engine 2, nl1 is the low-pressure speed of engine 1, nl2 is the low-pressure speed of engine 2, np is the propeller speed, wf1 is the fuel flow rate of engine 1, wf2 is the fuel flow rate of engine 2, PR is the pitch ratio, subscript k represents the kth sampling moment, and subscript 0 represents the corresponding steady-state point.
[0046] The fuel-fuel combined power system is linearized with small deviations at different steady-state points to obtain the state space model at different steady-state points. nl10, nl20 and PR0 are selected as scheduling parameters, and the scheduling parameters are used as input to establish a long short-term memory neural network (LSTM) model to predict other steady-state point parameters, including: nl10, nl20, np0, wf10, wf20, N10 and N20. The scheduling parameters and the parameters predicted based on the LSTM model are used as the input of the artificial neural network (ANN), and the matrices A, B, C, and D are predicted, thereby obtaining a prediction model in the form of a linear variable parameter model combined with the series neural network technology.
[0047] Taking the propeller speed as the control target, the prediction control sequence and the target power sequence are combined into an objective function in the following form:
[0048] min J2(x(k), ΔU(k))=||Г y (Y np (k)-np ref )|| 2 +||Γ u ΔU(k)|| 2 +||Γ p (P1(k)-P2(k))|| 2 Where: npref is the target propeller speed sequence, Y np (k) is the propeller speed sequence predicted by the prediction model based on the operating state of the system at step k, ΔU(k) is the system input increment to be solved, Γ y and Γ u is the weight matrix. To ensure the power balance of the two machines, a power balance penalty term is added to the objective function. P1(k) is the predicted power of 1#GT, P2(k) is the predicted power of 2#GT, and Γ p is the penalty weight matrix.
[0049] Taking fuel flow and propeller speed as constraints, the stable operation of the system can be ensured if these two variables change within a reasonable range. The constrained objective function is organized into the form of quadratic programming:
[0050]
[0051] stl≤C u ΔU(k)≤h
[0052] Where: H, G(k+1|k), C u is the coefficient matrix; l, h are the constraint boundary sequences.
[0053] The dual neural network method is used to solve the above quadratic programming problem, and the expression of the system input increment is obtained:
[0054] ΔU(k)=Pv-s
[0055] Where: v is a vector containing the predicted system operation information, P, s is a coefficient matrix.
[0056] like Figure 3 As shown, the prediction model in the form of linear variable parameters combined with a series neural network can well predict the operating characteristics of the system. The prediction model has a simple structure and high accuracy.
[0057] The model predictive control (MPC) algorithm is used to control the synchronous acceleration and deceleration process of the two machines, which can well ensure the synchronization of the power of the two machines and improve the maneuverability compared with the traditional PI control.
Claims
1. A multi-objective optimization control method for a fuel-fuel combined power plant based on a double-layer structure, characterized by: A two-layer control structure is introduced into the fuel-fuel combined power system. The two-layer control structure mainly includes: system planning layer and control execution layer. In the system planning layer, the economic optimal operating point under different working conditions is obtained based on the formulated economic optimal control strategy, and this operating point is transmitted to the control execution layer as a reference state. In the control execution layer, the model predictive control algorithm is used to control the dynamic process of single-machine acceleration and deceleration, multi-machine synchronous acceleration and deceleration, and power distribution, and the stability of the fuel-fuel combined power system is improved by explicitly constraining the fuel flow and propeller speed. Establish steady-state models of the fuel-fuel combined power system under different operating modes, calculate the fuel flow rate and ship speed under the corresponding combination parameters by setting different pitch ratios and power turbine speeds, select the steady-state point that meets the ship speed conditions, and then select the steady-state point with the minimum fuel flow rate as the economical optimal operating point. Perform the above operations under different working conditions, and select the optimal operating point corresponding to each working condition as the economical optimal control strategy; At the control execution layer, the model predictive control (MPC) algorithm is used to control different dynamic processes. Based on the mechanism model of the fuel-fuel combined power system, a prediction model with a linear variable parameter form composed of a series neural network is established. Based on the established prediction model, the MPC algorithm is designed for single-machine acceleration and deceleration, dual-machine acceleration and deceleration, and power distribution processes. The long short-term memory neural network LSTM is used to predict the steady-state point corresponding to the current operating point, and then the artificial neural network ANN is used to predict the coefficient matrix through the parameters of the steady-state point predicted by LSTM, thereby obtaining a prediction model with a linear variable parameter form composed of LSTM and ANN in series; The multivariable constraints of the fuel-fuel combined power system are simplified to the two-variable constraints of fuel flow and propeller speed, and the above two variables are explicitly constrained in the MPC algorithm.
2. The multi-objective optimization control method of a fuel-fuel combined power plant based on a double-layer structure according to claim 1 is characterized by: In order to simplify the solution process of the MPC algorithm, a linear variable parameter model is used as the prediction model. The model expression is as follows: Wherein: nh1 is the high-pressure speed of engine 1, nh2 is the high-pressure speed of engine 2, nl1 is the low-pressure speed of engine 1, nl2 is the low-pressure speed of engine 2, np is the propeller speed, wf1 is the fuel flow rate of engine 1, wf2 is the fuel flow rate of engine 2, PR is the pitch ratio, subscript k represents the kth sampling moment, and subscript 0 represents the corresponding steady-state point.
3. The multi-objective optimization control method of a fuel-fuel combined power plant based on a double-layer structure according to claim 1 is characterized in that: The fuel-fuel combined power system is linearized with small deviations at different steady-state points to obtain the state space models at different steady-state points. nl10, nl20 and PR0 are selected as scheduling parameters. The scheduling parameters are used as input to establish a long short-term memory neural network LSTM model to predict other steady-state point parameters, including: nl10, nl20, np0, wf10, wf20, N10 and N20. The scheduling parameters and the parameters predicted based on the LSTM model are used as input of the artificial neural network ANN to predict the matrices A, B, C, and D, thereby obtaining a prediction model in the form of a linear variable parameter model combined with the series neural network technology.
4. The multi-objective optimization control method of a fuel-fuel combined power plant based on a double-layer structure according to claim 1 is characterized by: Taking the propeller speed as the control target, the prediction control sequence and the target power sequence are combined into an objective function in the following form: min J2(x(k),ΔU(k))=||Γ y (Y np (k)-np ref )|| 2 +‖C u ΔU(k)‖ 2 +||C p (P1(k)-P2(k))|| 2 Where: np ref is the target propeller speed sequence, Y np (k) is the propeller speed sequence predicted by the prediction model based on the operating state of the system at step k, ΔU(k) is the system input increment to be solved, Γ y and Γ u is the weight matrix, and a power balance penalty term is added to the objective function. P1(k) is the predicted power of 1#GT, P2(k) is the predicted power of 2#GT, and Γ p is the penalty weight matrix.
5. The multi-objective optimization control method of a fuel-fuel combined power plant based on a double-layer structure according to claim 1 is characterized by: Taking the fuel flow rate and propeller speed as constraints, the constrained objective function is organized into the form of quadratic programming: stl≤C u ΔU(k)≤h in :H,G(k+1|k),C u is the coefficient matrix; l, h are the constraint boundary sequences; The dual neural network method is used to solve the above quadratic programming problem, and the expression of the system input increment is obtained: ΔU(k)=Pv-s Where: v is a vector containing the predicted system operation information, P, s is a coefficient matrix.
