Gas turbine data driving rotating speed self-adaptive control method based on DQN algorithm

Abstract

The invention discloses a rotating speed self-adaptive control method for gas turbine data driving based on a DQN algorithm, and aims to solve the problem that a traditional gas turbine control method has high dependence on a model, and provides a tight-format dynamic linearization model of a gas turbine under a large state in an online mode. And a gas turbine cascade model-free adaptive controller combined with a DQN algorithm is further designed, an outer loop is a PID control loop of the rotating speed of a power turbine, and an inner loop is a gas turbine rotating speed MFAC loop containing limitation protection multi-path switching. A cascade inner loop control parameter setting and outer loop instruction tracking scheme based on a DQN algorithm is adopted, the state is designed to be overshoot and error integration, and the action is set to be four controller parameters. According to the method, the instruction rotating speed can be well tracked in gas turbine interference and performance degradation scenes, and higher robust stability is achieved for fuel flow and load.

Description

Technical Field

[0001] The present invention belongs to the field of propulsion system control, and in particular relates to a gas turbine data-driven speed adaptive control method based on a DQN algorithm. Background Art

[0002] The performance of the control system of a gas turbine under variable operating conditions determines the economy and safety of its corresponding power unit. Among them, the gas turbine speed control system is an important component of gas turbine control. The working state of the load system will affect the working state of the gas turbine. During the operation of the gas turbine, the most basic goal of the control system is to stably operate at a certain speed to provide the required load while meeting the operating reliability. To achieve the control goal, traditional gas turbine control generally adopts cascade double-loop control, and most of them still use conventional PI controllers. However, due to the strong nonlinearity and time-varying nature of gas turbines, the PI parameters adjusted under a certain operating condition may not be applicable to all working states of the gas turbine. Moreover, since the traditional PI control law design has a strong dependence on the model, the accuracy of the gas turbine model will greatly limit the improvement of control quality. The system uncertainty caused by environmental changes, performance degradation, etc. will also make the control performance less than the expected results. Using data-driven methods to achieve adaptive control of gas turbines is another feasible solution.

[0003] In the data-driven control method, in 1994, Professor Hou Zhongsheng proposed Model Free Adaptive Control (MFAC). With the rapid development of computer technology, in-depth research has been carried out on the MFAC method in recent years, but the research on the application of model-free adaptive control algorithm to gas turbine control is relatively small, and there are still many problems to be solved in the speed control of gas turbines. On the one hand, the single-loop control scheme of model-free adaptive control is a basic speed control scheme, which is widely used due to its simple structure, but it cannot meet the control quality requirements under multiple working conditions at the same time, and there is a problem of slow response to internal disturbances of the engine. On the other hand, there are multiple parameters in the model-free adaptive control algorithm that need to be adjusted. For a new research object under different working conditions and in the face of environmental changes and performance degradation, it needs to be debugged repeatedly before it can be applied, and the operation is complicated.

[0004] The cascade dual-loop control scheme reduces the control system's dependence on the model. When faced with system uncertainties caused by environmental changes and performance degradation, it can run quickly and stably at a certain speed. In response to the blindness of the control parameter adjustment of the inner-loop model-free adaptive controller, a control parameter tuning method based on the Deep Q Network (DQN) was developed to reduce the blindness of the control parameter adjustment and speed up online learning, thereby improving the rapidity of the system and optimizing the dynamic performance indicators. Summary of the invention

[0005] The technical problem to be solved by the present invention is to overcome the problem that traditional gas turbine control methods have a strong dependence on models, and to provide a gas turbine data-driven speed adaptive control method based on the DQN algorithm, so as to reduce the control system's dependence on the model and speed up the online learning speed, thereby improving the rapidity of the system. When facing system uncertainties caused by environmental changes, performance degradation, etc., the system can run quickly and stably at a certain speed to maintain control quality.

[0006] The present invention adopts the following technical solution, comprising the following steps:

[0007] Step A), based on the input and output data of the gas turbine, a compact data model of the inner loop of the gas turbine controller is established online and in real time using a dynamic linearization method;

[0008] Step B), designing a data-driven series hybrid control scheme based on a compact data model, wherein the outer loop of the controller is PID control, with the power turbine speed as the controlled signal, and the gas turbine high-pressure rotor speed command is calculated according to the deviation between the command value and the current actual power turbine speed value; the inner loop takes the gas turbine high-pressure speed as the controlled variable, and uses MFAC control to adjust the fuel flow according to the deviation between the current gas turbine speed and the given value; and the DQN algorithm is used to self-learn and adjust the control parameters of the MFAC;

[0009] In step C), facing the uncertain interference of input and output data of the gas turbine in large states and multiple operating conditions, the series hybrid control scheme and DQN algorithm in step B) are adopted to realize adaptive control of performance degradation within the life cycle, and further verify the control performance and robustness.

[0010] As a gas turbine data-driven speed adaptive control method based on the DQN algorithm, the specific steps of step A) are as follows:

[0011] Step A1), linearizing the step response of the nonlinear model of the gas turbine by the least square fitting method, obtaining the high pressure speed and the power turbine speed curve, thereby obtaining the transfer function G2(s) of the gas turbine gas generator and the load system transfer function G3(s);

[0012] In step A2), a compact data model is established online using a dynamic linearization method based on the output and input data of the gas turbine, taking into account the dynamic characteristics of the actuator.

[0013] A data-driven speed adaptive control method for a gas turbine based on a DQN algorithm, wherein the transfer function of the gas generator and the load system identified in step A1) is:

[0014]

[0015] s is the Laplace operator;

[0016] Furthermore, the specific steps of step A2) are:

[0017] The inner control system of the cascade control loop is described by the following discrete-time single-input single-output nonlinear system:

[0018] y(k+1)=f(y(k),··,y(kn y ),u(k),··,u(kn u )) (3)

[0019] Among them, u(k) and y(k) are the gas turbine input and output at time k, respectively. For gas turbine speed control, it represents the fuel flow rate W f and high pressure rotor speed Ng;n y ,n u is the unknown order of the system.

[0020] Based on the output and input data of the gas turbine, a compact dynamic linear data model of the gas turbine at the current operating point is established using the dynamic linearization method:

[0021] Δy(k+1)=φ(k)Δu(k) (4)

[0022] Among them, Δu(k) is the change of fuel flow, Δy is the change of the speed of the high-pressure rotor of the gas turbine, which is the controlled variable of the inner loop, and φ(k)∈R is the pseudo partial derivative PPD of the system.

[0023] Based on the dynamic linearization model established above, the control law of the compact dynamic linearization model-free adaptive control is obtained by introducing the step size factor after finding the extreme value of the criterion function:

[0024]

[0025] Among them, λ>0 is the penalty factor, which can limit the change of the control amount u(k), and indirectly limit the change of the pseudo partial derivative, ρ∈(0,1] is the step size factor of the control rate, y ris the expected output. The PPD estimation without matrix inversion is further given:

[0026]

[0027] Among them, η∈(0,2] is the step size factor of PPD estimation, μ>0 is the weight factor, and these two factors are adjustable during the control process. And for application considerations, the pseudo partial derivative is reset when the pseudo partial derivative estimate is obviously unreasonable: if sign(φ(k))≠sign(φ(1)), or |φ(k)|≤ε, or |Δu(k-1)|≤ε, then φ(k)=φ(1). Among them, ε is an arbitrarily small constant.

[0028] The data-driven speed adaptive control method for a gas turbine based on a DQN algorithm described in the present invention, the specific steps of step B) are as follows:

[0029] Step B1), assume the transfer function of the actuator to be a first-order inertia link:

[0030]

[0031] The s-domain transfer functions in equations (1), (2), and (7) are transformed into z-transforms suitable for discrete-time systems to express the relationship between the input and output of the system.

[0032] Step B2), design the limit protection controller module, adopt MIN-MAX switching logic, to prevent the engine from overheating and overpressure, control the high pressure compressor outlet total pressure P3 and the low pressure turbine outlet total temperature T5 not to exceed the limit value. The limit protector is used to calculate the limit parameter to maintain within the limit value, and the limit protection controller adopts PI control. The fuel flow obtained by the limit protector and the fuel flow obtained by the main loop model-free adaptive controller are selected for high / low, and then the fuel flow that finally acts on the gas turbine is obtained through the actuator.

[0033] Step B3), design a data-driven series hybrid control scheme, the outer loop is a PID controller, and the inner loop is a compact model-free adaptive controller designed in step A) with over-limit protection.

[0034] Step B4), use the DQN algorithm to self-learn and adjust the control parameters of the MFAC, and the input of the outer loop PID controller is the difference between the power turbine speed command value and the actual value:

[0035] e(t)=Npr(t)-Np(t) (8)

[0036] Add e(t) to the state parameter and set the state S t=[maxe(t),∫e(t)dt], and limit it so that the gas turbine always operates within the safe range; set the action to the four parameters A of the inner loop MFAC controller t =[eita,miu,rou,lamda], eita, miu, rou, lamda correspond to the step size factor η of PPD estimation, weight factor μ, step size factor ρ of control rate, penalty factor λ respectively. Four controller parameters are selected as output actions from the action set through a greedy strategy. Each action that may be selected is assigned its corresponding execution probability, and then all possible actions are tried to be completed. The range of the designed model-free adaptive controller parameters must meet the following constraints: eita∈(0,2],miu>0,rou∈(0,1],lamda>0, so that the gas turbine control system remains stable;

[0037] The reward function is set to be the sum of the two state quantities:

[0038] R t =maxe(t)+∫e(t)dt (9)

[0039] The smaller the expected reward value, the better, that is, the overshoot and steady-state performance are guaranteed at this time; the gas turbine outputs the power turbine speed, calculates the difference with the command value and then calculates the reward function, stores the state, action and reward in the memory data playback library, calculates the loss function required for the update, randomly extracts part of the experience data from the memory library for training, selects actions based on the Epsilon-Greedy strategy, and outputs it for MFAC control law design and CFDL dynamic linearization calculation.

[0040] The data-driven speed adaptive control method for a gas turbine based on a DQN algorithm described in the present invention comprises the following specific steps in step C1):

[0041] The data-driven series hybrid control scheme designed in step B) is simulated under rated operating conditions. At the beginning of the simulation, it needs to run for a certain time until it stabilizes. Therefore, the system simulation process first maintains operation at a certain fuel amount for 10 seconds and then connects to the controller to control the fuel amount. The expected power turbine speed signal of the simulation process is a step-by-step jump from 95% of the rated speed to the rated speed in the 10th second. The actuator transfer function G1(s) and the load system transfer function G3(s) are subjected to two interference conditions: the gain deviation is twice and the phase lag does not exceed 60 degrees. The stability of the series hybrid control scheme based on the DQN algorithm and the effectiveness of tracking are verified.

[0042] Further, the specific scheme of step C2) is:

[0043] Under the design point conditions, open-loop control is performed to observe the impact of the five rotating parts of the low-pressure compressor, high-pressure compressor, high-pressure turbine, low-pressure turbine and power turbine under no degradation, low degradation level and medium degradation level in different working modes on the performance of the gas turbine. The same simulation process is used for the three cases. The data-driven series hybrid control scheme designed in step B) is simulated at three working conditions: 100% working condition, 90% working condition and 80% working condition, respectively, to verify the adaptive maintenance effect of the control loop designed in the gas turbine under large-state multi-working conditions and different levels of performance degradation.

[0044] The present invention has the following beneficial effects:

[0045] (1) The present invention only uses the input and output of the gas turbine, so that the controller parameters do not depend on the model, and the internal parameters are adjusted online adaptively. Since the response speed and control quality of the control loop depend on the parameter selection of the MFAC controller, a control parameter adjustment method based on the deep Q network algorithm is developed to reduce the blindness of the control parameter adjustment.

[0046] (2) Compared with the traditional control, the present invention has stronger anti-interference and self-adaptation capabilities. It can better track the command speed under different degrees of performance degradation under multiple operating points and has stronger robust stability to fuel flow and load. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is the structure diagram of MFAC cascade control scheme;

[0048] Figure 2 is a schematic diagram of the inner loop control system;

[0049] Figure 3 is the fuel flow rate W after MFAC and DQN algorithm adjustment f Simulation results;

[0050] Figure 4 It is the simulation result of the power turbine speed Np after MFAC and DQN algorithm tuning;

[0051] Figure 5 The fuel flow rate W of the control method and PI control designed by the present invention when the actuator phase is pulled out of phase f Simulation results;

[0052] Figure 6 It is the simulation result of the power turbine speed Np of the control method designed by the present invention and PI control when the actuator phase is pulled off;

[0053] Figure 7 The fuel flow rate W of the control method and PI control designed by the present invention when the phase of the controlled object is pulled off f Simulation results;

[0054] Figure 8 It is the simulation result of the power turbine speed Np of the control method designed by the present invention and PI control when the phase of the controlled object is pulled out;

[0055] Fig. 9 is the fuel flow rate W of the control method designed by the present invention when different degrees of degradation occur under the 1.0 working condition f Simulation results;

[0056] Fig.10 It is the simulation result of the power turbine speed Np of the control method designed by the present invention when different degrees of degradation occur under the 1.0 working condition;

[0057] Fig.11 is the fuel flow rate W of the control method designed by the present invention when different degrees of degradation occur under the 0.9 working condition f Simulation results;

[0058] Fig.12 It is the simulation result of the power turbine speed Np of the control method designed by the present invention when different degrees of degradation occur under the 0.9 working condition;

[0059] Fig.13 is the fuel flow rate W of the control method designed by the present invention when different degrees of degradation occur under the 0.8 working condition f Simulation results;

[0060] Fig.14 It is the simulation result of the power turbine speed Np of the control method designed by the present invention when different degrees of degradation occur under the 0.8 working condition. DETAILED DESCRIPTION

[0061] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings.

[0062] A certain type of gas turbine used in the present invention has main components including a low-pressure compressor, a high-pressure compressor, a combustion chamber, a high-pressure turbine, a low-pressure turbine, a power turbine and an exhaust device. The nonlinear mathematical model of the engine is established based on the C language using the component method, and then encapsulated into a dynamic link library form and called in the MALTAB environment for digital simulation verification.

[0063] The present invention proposes a gas turbine data driven speed adaptive control method based on DQN algorithm, the structure diagram of which is shown in FIG. Figure 1 As shown in the figure, Npr is the power turbine speed command value; Np is the power turbine speed actual value; Ng is the gas turbine high pressure rotor speed actual value; W fis the fuel flow rate, d1 is the disturbance of the fuel flow rate, and d2 is the disturbance of the load; G(s) with different subscripts represent the transfer functions of different components or system modules, G1(s) is the transfer function of the actuator, that is, the transfer function from the command calculated by the controller to the fuel flow rate. G2(s) is the transfer function from the fuel flow rate to the high-pressure rotor speed of the gas generator, G3(s) is the transfer function from the high-pressure rotor speed of the gas generator to the power turbine rotor speed, C1(s) is the inner loop MFAC controller, and C2(s) is the outer loop PID controller, which specifically includes the following steps:

[0064] Step A), based on the input and output data of the gas turbine, a compact data model of the inner loop of the gas turbine controller is established online and in real time using a dynamic linearization method;

[0065] Step A1), linearize the step response of the nonlinear model of the gas turbine using the least squares fitting method to obtain the high pressure speed and power turbine speed curves, thereby obtaining the transfer function G2(s) of the gas turbine gas generator and the load system transfer function G3(s):

[0066]

[0067] s is the Laplace operator;

[0068] Step A2), taking into account the dynamic characteristics of the actuator, based on the output and input data of the gas turbine, a compact data model is established online using a dynamic linearization method. The inner loop control system of the cascade control loop is described by the following discrete-time single-input single-output nonlinear system:

[0069] y(k+1)=f(y(k),··,y(kn y ),u(k),··,u(kn u )) (3)

[0070] Among them, u(k) and y(k) are the gas turbine input and output at time k, respectively. For gas turbine speed control, it represents the fuel flow rate W f and high pressure rotor speed Ng;n y ,n u is the unknown order of the system.

[0071] Based on the output and input data of the gas turbine, a compact dynamic linear data model of the gas turbine at the current operating point is established using the dynamic linearization method:

[0072] Δy(k+1)=φ(k)Δu(k) (4)

[0073] Among them, Δu(k) is the change of fuel flow, Δy is the change of the speed of the high-pressure rotor of the gas turbine, which is the controlled variable of the inner loop, and φ(k)∈R is the pseudo partial derivative PPD of the system.

[0074] For the application of compact format model-free adaptive control in general nonlinear systems, the controlled object must meet the following assumptions:

[0075] Assumption 1: Except for a finite number of time points, f(·..·) is y +2) The partial derivatives of the control input variables are continuous.

[0076] Assumption 2: Except for finite time points, the generalized Lipschitz condition is satisfied, that is, for any k1≠k2, k1, k2≥0 and u(k1)≠u(k2), ||y(k1+1)-y(k2+1)||≤b||u(k1)-u(k2)||, where y(k i +1)=f(y(k i ),...,y(k i -n y ),u(k i ),...,u(k i -n u )), i=1,2; b>0 is a constant, that is, a bounded input energy change should produce a bounded output energy change in the system.

[0077] Based on the dynamic linearization model established above, the control law of the compact dynamic linearization model-free adaptive control is obtained by introducing the step size factor after finding the extreme value of the criterion function:

[0078]

[0079] Among them, λ>0 is the penalty factor, which can limit the change of the control amount u(k), and indirectly limit the change of the pseudo partial derivative, ρ∈(0,1] is the step size factor, y r is the expected output. The PPD estimation without matrix inversion is further given:

[0080]

[0081] Among them, η∈(0,2] is the step size factor, μ>0 is the weight factor, and these two factors are adjustable during the control process. And for application considerations, the pseudo partial derivative is reset when the pseudo partial derivative estimate is obviously unreasonable: if sign(φ(k))≠sign(φ(1)), or |φ(k)|≤ε, or |Δu(k-1)|≤ε, then φ(k)=φ(1). Among them, ε is an arbitrarily small constant.

[0082] Step B), designing a data-driven series hybrid control scheme based on a compact data model, wherein the outer loop of the controller is PID control, with the power turbine speed as the controlled signal, and the gas turbine high-pressure rotor speed command is calculated according to the deviation between the command value and the current actual power turbine speed value; the inner loop takes the gas turbine high-pressure speed as the controlled variable, and uses MFAC control to adjust the fuel flow according to the deviation between the current gas turbine speed and the given value; and the DQN algorithm is used to self-learn and adjust the control parameters of the MFAC;

[0083] Step B1), assume the transfer function of the actuator to be a first-order inertia link:

[0084]

[0085] The s-domain transfer functions in equations (1), (2), and (7) are transformed into z-transforms suitable for discrete-time systems to express the relationship between the input and output of the system.

[0086] Step B2), the control target is to change the amount of fuel so that the power turbine speed can continuously track the command value. Design a limit protection controller module, and adopt MIN-MAX switching logic. To prevent the engine from overheating and overpressure, control the high-pressure compressor outlet total pressure P3 and the low-pressure turbine outlet total temperature T5 not to exceed the limit value. The limit protector is used to calculate so that the limit parameter is maintained within the limit value, and the limit protection controller adopts PI control. The fuel flow obtained by the limit protector and the fuel flow obtained by the main loop model-free adaptive controller are selected for high / low, and then passed through the actuator to obtain the final fuel flow acting on the gas turbine;

[0087] Step B3), design a data-driven series hybrid control scheme, the outer loop is a PID controller, and the inner loop is a compact model-free adaptive controller designed in step A) with over-limit protection;

[0088] Step B4), use the DQN algorithm to self-learn and adjust the control parameters of MFAC. In the DQN algorithm, two

[0089] The value function approximation network has the same structure, where the target network The parameter w in - It will be fixed in a batch training and used to generate the target Q value to calculate the TD target y j , thereby updating the required loss function (y j -Q(S j ,A j ;w)) 2The online network Q(s,a;w) is used to evaluate the strategy, and the network parameters are updated in each iteration. After the online network is updated a certain number of times, the latest network weight parameters are directly used to update the target network as the fixed parameters of the next round of target network.

[0090] The specific solution steps are: the state S of the environment perceived by the agent t , take random actions, interact with the environment for a period of time, and execute action A according to the greedy strategy t And observe the reward R t , and the new state S t+1 , and then take the empirical data of a period of time [S t ,A t ,R t ,S t+1 ] is stored in the experience playback library, and then a small sample set [S j ,A j ,R j ,S j+1 ] is used to update the online network. The state of the next moment is input to the target network. The maximum value is selected from the Q value of each action to calculate the TD target and determine whether it is the terminal state. After updating the online network a certain number of times, for (y j -Q(S j ,A j ;w)) 2 Use the gradient descent method to update the online network parameters, and continue to repeat the above steps, transforming the current environment state into a new state until the maximum number of cycles is reached and the result is output.

[0091] In the DQN algorithm, state S t It directly reflects the gas turbine cascade control target, that is, under the premise of meeting the operational reliability, the power turbine speed of the gas turbine can still work stably at a certain speed. The input of the outer loop PID controller is the difference between the power turbine speed command value and the actual value:

[0092] e(t)=Npr(t)-Np(t) (8)

[0093] Add e(t) to the state parameter and set the state S t =[maxe(t),∫e(t)dt], and limit it so that the gas turbine always operates within the safe range; set the action to the four parameters A of the inner loop MFAC controller t=[eita,miu,rou,lamda], four controller parameters are selected as output actions from the action set through a greedy strategy, and each possible action is assigned its corresponding execution probability, and then all possible actions are tried to be completed, and the range of the designed model-free adaptive controller parameters must meet the following constraints: eita∈(0,2],miu>0,rou∈(0,1],lamda>0, so that the gas turbine control system remains stable.

[0094] In addition, the reward R t Whether the design is reasonable will directly affect the convergence effect and control accuracy of the neural network. The reward condition and the control target must also have a corresponding relationship to effectively guide network training. This design improves the reward function and sets the reward function as the sum of two state quantities:

[0095] R t =maxe(t)+∫e(t)dt (9)

[0096] The smaller the expected reward value, the better, that is, at this time, the overshoot and steady-state performance are guaranteed. The overall structure diagram of the inner loop control system is as follows: Figure 2 As shown. The calculation process is: based on the gas turbine input data W f Get the output data Ng; estimate the pseudo partial derivative PPD according to the formula, perform tight format dynamic linearization online, and build a tight format dynamic linearization data model at the current working point of the gas turbine; calculate the fuel flow of the gas turbine according to the CFDL control law, and apply the control input to the gas turbine. In the process of control law design and dynamic linearization, it takes many attempts to adjust the four parameters involved, so the control parameter setting is added. The gas turbine outputs the power turbine speed, calculates the difference with the command value and then calculates the reward function, stores the state, action and reward in the memory data playback library, sets the experience library to store 5000 sets of data, calculates the loss function required for updating, randomly extracts part of the experience data from the memory library for training, sets the batch number to 1000, selects actions based on the Epsilon-Greedy strategy, and outputs for MFAC control law design and CFDL dynamic linearization calculation.

[0097] Step C), facing the uncertain interference of input and output data of the gas turbine in large states and multiple operating conditions, the series hybrid control scheme and DQN algorithm in step B) are used to realize adaptive control of performance degradation within the life cycle, and further verify the control performance and robustness;

[0098] Step C1), based on the strong dependence of the gas turbine on the model and the slow response to internal disturbances, a data-driven series hybrid control scheme is designed to conduct robustness analysis under multiple operating conditions for the disturbances of gain and phase deviation at two different positions of the actuator and the controlled object;

[0099] Step C2) Open-loop control is performed under the design point conditions to observe the impact of rotating components on the gas turbine performance under different operating modes without degradation, low degradation level, and medium degradation level; facing the large state and multiple operating points of the gas turbine, the adaptive maintenance effect of the designed control loop is verified when facing different levels of performance degradation of the gas turbine.

[0100] The four parameters involved in the model-free adaptive controller are obtained through parameter tuning of the DQN algorithm. In addition to the need to design multiple hyperparameters involved in the DQN algorithm, other training parameter settings are shown in Table 1. The initial controller parameters are set to: eita = 1, miu = 2, rou = 0.5, lamda = 2; after tuning: eita = 0.81, miu = 1.92, rou = 0.41, lamda = 2.11, therefore, the parameter settings of the MFAC controller are shown in Table 2.

[0101] Table 1 Training parameter settings

[0102]

[0103] Table 2 MFAC controller parameter settings

[0104]

[0105] The simulation was carried out under 100% rated working conditions, the simulation time was 25 seconds, and the sampling step was set to 0.025s to verify the command tracking ability of the MFAC controller for the power turbine rotor speed in the cascade control. At the beginning of the simulation, it needs to run for a certain time until it stabilizes, so the system simulation process first maintains operation at a certain fuel amount for 10s before connecting to the controller to control the fuel amount control. The expected power turbine speed signal of the simulation process is a step from 95% rated speed to rated speed in the 10th second, and the changes in fuel amount and power turbine speed simulated in the whole process of the system are observed. In order to analyze the anti-interference effect of the MFAC controller, the cascade control process with the inner loop as a PI controller and the inner loop as a CFDL-MFAC controller are simulated and compared. For the two disturbances of fuel flow and load, the phases of the two transfer functions G1(s) and G3(s) are pulled out, and the parameter settings of the inner loop as a PI controller are shown in Table 3 (subscript 1 is the inner loop PI controller parameter, subscript 2 is the outer loop PI controller parameter). When the actuator phase is pulled out, the fuel flow simulation results are as follows Figure 5 , the power turbine speed simulation results are as follows Figure 6As shown; when the controlled object phase is pulled off, the fuel flow simulation results are as follows Figure 7 , the power turbine speed simulation results are as follows Figure 8 As shown; it should be noted that the simulation environment runs on an Intel Core i7-12700 CPU, and the system operation results are given after normalization.

[0106] Table 3 PI controller parameter settings

[0107]

[0108] The simulation results show that when the actuator phase is pulled off, the response adjustment time of the power turbine speed is 7s, the overshoot is 0.6%, and there is no steady-state error; when the controlled object phase is pulled off, the response adjustment time of the power turbine speed is 8.8s, the overshoot is 0.5%, and there is no steady-state error; and compared with the traditional PI control law method, when the fuel flow and load are disturbed, the PI controller will vibrate, and the balance between the system response speed and the vibration problem depends on the parameter selection, and the controller parameters need to be designed in detail to obtain better control quality. Therefore, compared with the traditional PI control law method, MFAC has better anti-interference effect, which proves that the model-free adaptive control algorithm can improve the control quality and automatically compensate for the system performance changes caused by the uncertainty of the process object parameters and environment, thereby verifying the stability and tracking effectiveness of the MFAC solution based on CFDL.

[0109] Gas turbine compressors and turbines may experience temporary or permanent performance degradation when they continue to operate under harsh conditions such as high temperature, high pressure and polluted environment. The causes and forms of failures are different, but they can all be characterized by changes in the flow coefficient and efficiency coefficient of the components. The thermal efficiency and output power of the gas turbine decrease, thus affecting the economic feasibility and safety of the gas turbine power plant. In order to quantitatively represent the degree to which the component characteristics deviate from the design state, a set of health parameters are defined to measure the performance degradation of each engine component.

[0110] In order to ensure that the control system designed for the gas turbine under rated conditions can still maintain good dynamic and steady-state performance after degradation, it is necessary to study the adaptive maintenance method of control performance. By adding health parameters to the gas turbine component-level model, the performance changes after degradation are simulated. i = 1, 2, 3, 4, 5 represent the efficiency coefficients of the five rotating components of the low-pressure compressor, high-pressure compressor, high-pressure turbine, low-pressure turbine and power turbine respectively: [SE1, SE2, SE3, SE4, SE5]. The specific setting values ​​of the health parameters for the three degradation conditions are shown in Table 4.

[0111] Table 4 Gas turbine health parameter settings (%)

[0112]

[0113] Different degrees of performance degradation will affect the control of the gas turbine and make it unable to perform as expected. Based on the cascade dual-loop control scheme designed above, the impact of components on gas turbine performance under no degradation, low degradation level, and medium degradation level is observed, so as to verify the robustness of model-free adaptive control on gas turbine speed control under different degradation scenarios. The performance degradation of the gas turbine is simulated by pulling the efficiency parameters of the five rotating components to simulate the performance degradation of the gas turbine. The cascade control process of the MFAC controller is simulated under three operating conditions: 1.0 condition (output power is rated power), 0.9 condition (output power is 90% of rated power), and 0.8 condition (output power is 80% of rated power). Figures 9 to 14 As shown, the adaptive maintenance effect of the designed control loop in the face of gas turbine performance degradation is verified.

[0114] The simulation results show that, in the face of performance degradation, under the action of the model-free adaptive controller, the power turbine speed can basically be maintained at the command value at three operating points, and the power turbine speed can be controlled to the command value within 7s, without steady-state error, and the overshoot is within 1%. Compared with the prior art, the technical solution of the present invention has the following beneficial effects: The overall simulation process of the present invention shows that, considering the degradation of gas turbine performance, the MFAC control has a smaller overshoot than the PID controller at each operating point under interference, without jitter, and shows better dynamic performance and robustness.

Claims

1. A gas turbine data-driven speed adaptive control method based on DQN algorithm, characterized by: The following steps are involved: Step A), based on the input and output data of the gas turbine, a compact data model of the inner loop of the gas turbine controller is established online and in real time using a dynamic linearization method; Step B), designing a data-driven series hybrid control scheme based on a compact data model, wherein the outer loop of the controller is PID control, with the power turbine speed as the controlled signal, and the gas turbine high-pressure rotor speed command is calculated according to the deviation between the command value and the current actual power turbine speed value; the inner loop takes the gas turbine high-pressure speed as the controlled variable, and uses MFAC control to adjust the fuel flow according to the deviation between the current gas turbine speed and the given value; and the DQN algorithm is used to self-learn and adjust the control parameters of the MFAC; In step C), facing the uncertain interference of input and output data of the gas turbine in large states and multiple operating conditions, the series hybrid control scheme and DQN algorithm in step B) are adopted to realize adaptive control of performance degradation within the life cycle, and further verify the control performance and robustness.

2. A gas turbine data-driven speed adaptive control method based on DQN algorithm as claimed in claim 1, characterized in that: The specific steps of step A) are as follows: Step A1), linearizing the step response of the nonlinear model of the gas turbine by the least square fitting method, obtaining the high pressure speed and the power turbine speed curve, thereby obtaining the transfer function G2(s) of the gas turbine gas generator and the load system transfer function G3(s); Step A2), taking into account the dynamic characteristics of the actuator, based on the gas turbine output and input data, a compact data model is established online using a dynamic linearization method.

3. A gas turbine data-driven speed adaptive control method based on DQN algorithm as claimed in claim 2, characterized in that: The transfer function of the gas generator and the load system identified in step A1) is: s is the Laplace operator.

4. A gas turbine data-driven speed adaptive control method based on DQN algorithm as claimed in claim 2, characterized in that: The specific steps of step A2) include: The inner control system of the cascade control loop is described by the following discrete-time single-input single-output nonlinear system: y(k+1)=f(y(k),··,y(k-n y ),u(k),··,u(k-n u )) (3) Among them, u(k) and y(k) are the gas turbine input and output at time k, respectively. For gas turbine speed control, it represents the fuel flow rate W f and high pressure rotor speed Ng;n y ,n u is the unknown order of the system; Based on the output and input data of the gas turbine, a compact dynamic linear data model of the gas turbine at the current operating point is established using the dynamic linearization method: Δy(k+1)=φ(k)Δu(k) (4) Among them, Δu(k) is the change of fuel flow, Δy is the change of the speed of the high-pressure rotor of the gas turbine controlled by the inner loop, and φ(k)∈R is the pseudo partial derivative PPD of the system; Based on the dynamic linearization data model established above, the control law of compact dynamic linearization model-free adaptive control is obtained by introducing the step size factor after finding the extreme value of the criterion function: Among them, λ>0 is the penalty factor, which can limit the change of the control amount u(k) and indirectly limit the change of the pseudo partial derivative, ρ∈(0,1] is the step size factor of the control rate, y r is the expected output; furthermore, the PPD estimation without matrix inversion operation is given: Among them, η∈(0,2] is the step size factor of PPD estimation, μ>0 is the weight factor, and these two factors are adjustable during the control process; and for application considerations, the pseudo partial derivative is reset when the pseudo partial derivative estimate is obviously unreasonable: if sign(φ(k))≠sign(φ(1)), or |φ(k)|≤ε, or |Δu(k-1)|≤ε, then φ(k)=φ(1), where ε is a constant.

5. A gas turbine data-driven speed adaptive control method based on DQN algorithm as claimed in claim 3, characterized in that: The specific steps of designing a data-driven series hybrid control scheme based on a compact data model in step B) are: Step 1), assume the transfer function of the actuator to be a first-order inertia link: The s-domain transfer functions in equations (1), (2), and (7) are transformed into z-transforms suitable for discrete-time systems to express the relationship between the input and output of the system; Step 2), design a limit protection controller module, adopt MIN-MAX switching logic, to prevent the engine from over-temperature and over-pressure, control the high-pressure compressor outlet total pressure P3 and the low-pressure turbine outlet total temperature T5 not to exceed the limit value; the limit parameter is maintained within the limit value through the limit protector calculation, and the limit protection controller adopts PI control; the fuel flow obtained by the limit protector and the fuel flow obtained by the main loop model-free adaptive controller are selected for high / low, and then the fuel flow that finally acts on the gas turbine is obtained through the actuator; Step 3), design a data-driven series hybrid control scheme, the outer loop is a PID controller, and the inner loop is a compact model-free adaptive controller designed in step A) with over-limit protection.

6. A gas turbine data-driven speed adaptive control method based on DQN algorithm as claimed in claim 4, characterized in that: The specific steps of using the DQN algorithm to self-learn and adjust the control parameters of the MFAC in step B) include: The input of the outer loop PID controller is the difference between the power turbine speed command value and the actual value: e(t)=Npr(t)-Np(t) (8) Add e(t) to the state parameter and set the state S t =[maxe(t),∫e(t)dt], and limit it so that the gas turbine always operates within the safe range; set the action to the four parameters A of the inner loop MFAC controller t =[eita,miu,rou,lamda], eita, miu, rou, lamda correspond to the step size factor η of PPD estimation, the weight factor μ, the step size factor ρ of the control rate, and the penalty factor λ, respectively. Four controller parameters are selected from the action set as output actions through a greedy strategy. Each action that may be selected is assigned its corresponding execution probability, and then all possible actions are tried to be completed. The range of the designed model-free adaptive controller parameters must meet the following constraints: eita∈(0,2],miu>0,rou∈(0,1],lamda>0, so that the gas turbine control system remains stable; The reward function is set to be the sum of the two state quantities: R t =maxe(t)+∫e(t)dt (9) The smaller the expected reward value, the better, that is, the overshoot and steady-state performance are guaranteed at this time; the gas turbine outputs the power turbine speed, calculates the difference with the command value and then calculates the reward function, stores the state, action and reward in the memory data playback library, calculates the loss function required for the update, randomly extracts part of the experience data from the memory library for training, selects actions based on the Epsilon-Greedy strategy, and outputs it for MFAC control law design and CFDL dynamic linearization calculation.

7. A gas turbine data-driven speed adaptive control method based on DQN algorithm as claimed in claim 1, characterized in that: The specific steps of step C) are as follows: Step C1), performing robustness analysis under multiple working conditions on the disturbance of gain and phase deviation at two different positions of the actuator and the controlled object; Step C2) Open-loop control is performed under the design point conditions to observe the impact of rotating components on the gas turbine performance under different operating modes without degradation, low degradation level, and medium degradation level; facing the large state and multiple operating points of the gas turbine, the adaptive maintenance effect of the designed control loop is verified when facing different levels of performance degradation of the gas turbine.

8. A gas turbine data-driven speed adaptive control method based on DQN algorithm as claimed in claim 7, characterized in that: The specific steps of step C1) are: The data-driven series hybrid control scheme designed in step B) is simulated under rated operating conditions. At the beginning of the simulation, it needs to run for a certain time until it stabilizes. Therefore, the system simulation process first maintains operation at a certain fuel amount for 10 seconds and then connects to the controller to control the fuel amount. The expected power turbine speed signal of the simulation process is a step-by-step jump from 95% of the rated speed to the rated speed in the 10th second. The actuator transfer function G1(s) and the load system transfer function G3(s) are subjected to two interference conditions: the gain deviation is twice and the phase lag does not exceed 60 degrees. The stability of the series hybrid control scheme based on the DQN algorithm and the effectiveness of tracking are verified.

9. A gas turbine data-driven speed adaptive control method based on DQN algorithm as claimed in claim 7, characterized in that: The specific steps of step C2) are: Open-loop control is performed under design point conditions to observe the impact of five rotating components, namely, low-pressure compressor, high-pressure compressor, high-pressure turbine, low-pressure turbine and power turbine, on the performance of the gas turbine under no degradation, low degradation level and medium degradation level in different working modes. The same simulation process is used for the three cases. The data-driven series hybrid control scheme designed in step B) is simulated at three operating points, namely, 100% operating condition, 90% operating condition and 80% operating condition, to verify the adaptive maintenance effect of the control loop designed for the gas turbine under large-state and multi-operating conditions and facing different levels of performance degradation.

Images