Turboshaft engine rapid dynamic response control method and system

Abstract

The invention discloses a rapid dynamic response control method for a turboshaft engine. According to the method, a gas turbine rotation acceleration cascade control structure is used for controlling the turboshaft engine; a feed-forward compensation form with predicted required power as a feed-forward compensation amount is introduced into an inner ring fuel flow instruction control position of a gas turbine rotation acceleration cascade control structure; introducing a feed-forward compensation form which takes a required power prediction error as a feed-forward compensation amount into an outer ring gas turbine rotation acceleration instruction control position of the gas turbine rotation acceleration cascade control structure; and a recursive least square algorithm is used to carry out adaptive adjustment on feed-forward compensation parameters of the two loops. The invention further discloses a rapid dynamic response control system for the turboshaft engine. Compared with the prior art, the technical scheme can effectively improve the response speed of the engine control system.

Description

Technical Field

[0001] The present invention relates to a control method for a turboshaft engine, and more particularly to a rapid dynamic response control method and system for a turboshaft engine, belonging to the technical field of aero-engine control. Background Art

[0002] As the core power device of a helicopter, control performances such as the response speed and the stability of power output of a turboshaft engine are crucial for the flight safety, maneuverability, and operation reliability of the helicopter. Especially in high-load change scenarios such as mode conversion, variable rotor speed, and high-g maneuver flight, the performance of the turboshaft engine control system plays a key role in the rapid response ability and stability of the whole machine. To ensure that the helicopter always maintains excellent flight performance under the above complex flight conditions, it is necessary to optimize the operation performance and adaptability of the turboshaft engine through a reasonably designed control system.

[0003] Under the traditional design framework, the control systems of the helicopter and the engine are usually designed independently after clarifying the overall performance indicators and cross-linking signals. This design method can meet the basic performance requirements of the system, but a certain margin is usually reserved in the independent design process, and the comprehensive performance potential of the helicopter / engine cannot be fully exerted. In addition, the response speed of the engine control system under the traditional design framework to changes in power demand is slow, resulting in a significant time-delay effect in the engine power supply, thereby reducing the dynamic flight performance of high-speed helicopters.

[0004] In this context, the propulsion system of high-speed helicopters faces more complex power scheduling requirements. High-speed helicopters involve complex processes such as multi-power distribution and rotor speed changes. The operating range of the engine has been significantly expanded, further enhancing the dynamic coupling effect of the propulsion system. Especially during the process of changing rotor speed, due to the real-time adjustment of the power requirements of the rotor, tail rotor, and propulsion propeller, the difficulty of dynamically matching the engine load increases, posing higher requirements for the response speed of the control system. Entering the 21st century, the US military launched a multi-stage research program on the rapid response control of the propulsion system for the next-generation Black Hawk helicopter and advanced turboshaft engines, and successively proposed methods based on transient torque feedforward control and on-board adaptive composite power prediction control. Zagranski et al. optimized the engine control performance by predicting the total torque changes of the helicopter's main rotor and tail rotor in the flight control system and converting them into changes in the engine acceleration and deceleration rates (Zagranski R D, Niebanck R D. Rotor torque anticipator. US Patents, EP1310646A2, 2003.). European countries such as the UK and France have successively proposed the "Integrated Flight and Engine Control (IFEC)" program and the "Advanced Power System and Engine Control (APSEC)" program for engine rapid response control methods to promote the development of engine rapid response control technology. Cai C proposed a multi-variable cascade control architecture for turboshaft engines. Among them, the outer loop calculates the output torque of the power turbine according to the total rotor pitch command combined with the rotor load dynamic model, thereby reducing the deviation of the power turbine speed; the inner loop is based on the gas generator dynamic model and adjusts the fuel flow rate and the angle of the adjustable guide vane of the compressor in real time according to the difference between the load torque and the calculated torque of the power turbine. This control architecture not only improves the dynamic response performance of the engine but also optimizes the transition process during the spin-in and recovery phases (Cai C, Crowley T J, Meisner R P. Cascaded multi-variable control system for a turboshaft engine. US Patents, US10113487B2, 2018.). Existing methods mainly rely on fixed power prediction models and lack real-time adaptive adjustment capabilities, so it is difficult to accurately capture the rapidly changing power requirements of high-speed helicopters under variable rotor speed conditions.

[0005] However, the conventional engine cascade PID control method based on rotor collective pitch feedforward compensation lacks accurate prediction of power demand, resulting in insufficient response tracking ability at the engine end and a decline in engine control performance. Therefore, the control system of a variable-speed turboshaft engine needs to enhance its perception and response capabilities to dynamic power changes, thereby improving the response speed of the engine control system and meeting the high-quality control requirements of the propulsion system for high-speed helicopters. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide a fast dynamic response control method for a turboshaft engine, which can effectively improve the response speed of the engine control system.

[0007] The present invention specifically adopts the following technical solutions to solve the above technical problems:

[0008] A fast dynamic response control method for a turboshaft engine uses a gas turbine rotational acceleration cascade control structure to control the turboshaft engine; a feedforward compensation form with the predicted required power as the feedforward compensation amount is introduced at the inner-loop fuel flow command control position of the gas turbine rotational acceleration cascade control structure, and a feedforward compensation form with the required power prediction error as the feedforward compensation amount is introduced at the outer-loop gas turbine rotational acceleration command control position of the gas turbine rotational acceleration cascade control structure; and the Recursive Least Square (RLS) algorithm is used to adaptively adjust the feedforward compensation parameters of the two loops.

[0009] Preferably, the use of the Recursive Least Square algorithm to adaptively adjust the feedforward compensation parameters of the two loops is specifically as follows:

[0010] Step 1: Calculate the gain vector K(t) = [K1(t), K2(t)] at the current time t for the two feedforward positions of the inner loop and the outer loop according to the required power prediction error e(t):

[0011]

[0012] where P(t - 1) represents the covariance matrix at time t - 1; λ is the forgetting factor that controls the weight update speed; φ(t) is the input vector, ρ i is the regularization coefficient; i = 1 and i = 2 respectively represent the inner loop and the outer loop;

[0013] Step 2: Use the gain vector K(t) and the required power prediction error e(t) to update the gain adjustment parameter θ i (t):

[0014] θ1(t) = θ1(t - 1) + K(t)e(t)

[0015]

[0016] where k ξ is the regularization ratio coefficient;

[0017] Step 3. Update the covariance matrix P(t):

[0018]

[0019] Step 4. Obtain the feedforward gains of the inner loop and the outer loop as follows:

[0020] u f1 (t) = θ1(t)P T,P (t)

[0021] u f2 (t) = θ2(t)e(t)

[0022] where P T,P (t) is the required power for prediction.

[0023] Preferably, the required power for prediction is obtained based on a neural network model pre-trained by the following method: First, take all measurable cross-linking parameters of the helicopter and the engine as the candidate feature set, and use the combination of the Random Forest (RF) method and the Recursive Feature Elimination (RFE) method to screen the candidate feature set to obtain the optimal cross-linking parameter feature combination most relevant to the required power; then, use the obtained optimal cross-linking parameter feature combination as the input feature and the required power as the expected output to train the neural network model.

[0024] More preferably, the neural network model is a BP neural network model.

[0025] Preferably, the gas turbine rotational acceleration cascade control structure further includes a limit protection controller, and the control output of the limit protection controller is introduced into the outer loop gas turbine rotational acceleration command control position or the inner loop fuel flow command control position of the gas turbine rotational acceleration cascade control structure to participate in the main loop limit protection.

[0026] Based on the same inventive concept, the following technical solutions can also be obtained:

[0027] A turbo-shaft engine fast dynamic response control system includes a gas turbine rotational acceleration cascade control structure for controlling the turbo-shaft engine; the control system further includes:

[0028] A required power model for predicting the required power of the helicopter and generating the required power for prediction;

[0029] A required power feedforward module is used to introduce, at the fuel flow command control position of the inner loop in the gas turbine rotational acceleration cascade control structure, a feedforward compensation form that predicts the required power as the feedforward compensation amount, and to introduce, at the gas turbine rotational acceleration command control position of the outer loop in the gas turbine rotational acceleration cascade control structure, a feedforward compensation form that uses the required power prediction error as the feedforward compensation amount;

[0030] An RLS gain adaptive adjustment module is used to adaptively adjust the feedforward compensation parameters of the two loops using the recursive least squares algorithm.

[0031] Preferably, the adaptive adjustment of the feedforward compensation parameters of the two loops using the recursive least squares algorithm is specifically as follows:

[0032] Step 1: Calculate the gain vector K(t)=[K1(t), K2(t)] at the current time t for the two feedforward positions of the inner loop and the outer loop according to the required power prediction error e(t):

[0033]

[0034] where P(t - 1) represents the covariance matrix at time t - 1; λ is the forgetting factor that controls the weight update speed; φ(t) is the input vector, ρ i is the regularization coefficient; i = 1 and i = 2 respectively represent the inner loop and the outer loop;

[0035] Step 2: Use the gain vector K(t) and the required power prediction error e(t) to update the gain adjustment parameter θ i (t):

[0036] θ1(t)=θ1(t - 1)+K(t)e(t)

[0037]

[0038] where k ξ is the regularization ratio coefficient;

[0039] Step 3: Update the covariance matrix P(t):

[0040]

[0041] Step 4: Obtain the feedforward gains of the inner loop and the outer loop as follows:

[0042] u f1 (t)=θ1(t)P T,P (t)

[0043] u f2 (t)=θ2(t)e(t)

[0044] Among them, P T,P (t) is the predicted required power.

[0045] Preferably, the required power model is a neural network model pre-trained by the following method: First, take all measurable cross-linking parameters of the helicopter and the engine as the set of candidate features, and use a combination of the random forest method and the recursive feature elimination method to screen the set of candidate features to obtain the optimal cross-linking parameter feature combination most relevant to the required power; Then, use the obtained optimal cross-linking parameter feature combination as the input feature and the required power as the expected output to train the neural network model.

[0046] More preferably, the neural network model is a BP neural network model.

[0047] Preferably, the gas turbine rotational acceleration cascade control structure further includes a limit protection controller, and the control output of the limit protection controller is introduced into the outer loop gas turbine rotational acceleration command control position or the inner loop fuel flow command control position of the gas turbine rotational acceleration cascade control structure to participate in the main loop limit protection.

[0048] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0049] Based on the existing gas turbine rotational acceleration cascade control structure, the present invention respectively introduces feedforward compensation with the predicted required power and the required power prediction error as the feedforward compensation amounts at the inner loop fuel flow command control and the outer loop gas turbine rotational acceleration command control positions, and combines the recursive least squares algorithm with the required power prediction model to adaptively adjust the feedforward compensation amounts of the inner and outer control loops according to the system parameter changes, avoiding the occurrence of steady-state errors and effectively improving the response speed of the engine control system;

[0050] The present invention further screens all cross-linking control signals of the helicopter and the turboshaft engine by using a combination of the random forest method and the recursive feature elimination method (abbreviated as the RF-REF method) to obtain the optimal cross-linking parameter feature combination, which can effectively improve the prediction performance and calculation efficiency of the required power model, avoid the situation that the model contribution degree of the autocorrelated parameters is lower than the actual value due to the strong autocorrelation between some parameters of the helicopter, and thus improve the calculation accuracy and modeling efficiency of the required power model. Description of the Drawings

[0051] Figure 1 is a schematic structural principle diagram of the turboshaft engine fast dynamic response control system of the present invention;

[0052] Figure 2 is a schematic diagram of the RF importance evaluation process;

[0053] Figure 3 It is a schematic diagram of the RF - RFE feature screening process;

[0054] Figure 4 It is a heat map of the importance index of cross - linking parameter features;

[0055] Figure 5 It is a schematic diagram of the cross - validation accuracy of the number of features;

[0056] Figure 6 It is a schematic diagram of the extraction result of cross - linking feature parameters based on RF - RFE;

[0057] Figure 7 It is a structure diagram of a BP neural network;

[0058] Figure 8 It is a comparison chart of the relative test error of the required power model based on RF - RFE;

[0059] Figure 9 It is a schematic diagram of the feed - forward position of the cascade control system based on the gas turbine rotational acceleration;

[0060] Figure 10 It is the comparison result of the fast dynamic response control method of the present invention and other existing methods in the low - speed mode; among them, (a) is the comparison chart of the gas turbine speed, (b) is the comparison chart of the power turbine speed, (c) is the comparison chart of the power turbine inlet temperature, and (d) is the comparison chart of the fuel flow rate;

[0061] Figure 11 It is the comparison result of the fast dynamic response control method of the present invention and other existing methods in the transition mode; among them, (a) is the comparison chart of the gas turbine speed, (b) is the comparison chart of the power turbine speed, (c) is the comparison chart of the power turbine inlet temperature, and (d) is the comparison chart of the fuel flow rate;

[0062] Figure 12 It is the comparison result of the fast dynamic response control method of the present invention and other existing methods in the high - speed mode; among them, (a) is the comparison chart of the gas turbine speed, (b) is the comparison chart of the power turbine speed, (c) is the comparison chart of the power turbine inlet temperature, and (d) is the comparison chart of the fuel flow rate. Detailed implementation manners

[0063] Aiming at the deficiencies of the existing technology, the solution idea of the present invention is to introduce a feed-forward compensation with predicted required power and predicted required power prediction error as feed-forward compensation amounts respectively at the inner-loop fuel flow command control and the outer-loop gas turbine rotational acceleration command control positions on the basis of the existing gas turbine rotational acceleration cascade control structure. And through the recursive least squares algorithm combined with the required power prediction model, the feed-forward compensation amounts of the inner and outer control loops are adaptively adjusted according to the changes of system parameters, avoiding the occurrence of steady-state errors and effectively improving the response speed of the engine control system.

[0064] The technical solution proposed by the present invention is specifically as follows:

[0065] A control method for the fast dynamic response of a turboshaft engine uses a gas turbine rotational acceleration cascade control structure to control the turboshaft engine; a feed-forward compensation form with predicted required power as the feed-forward compensation amount is introduced at the inner-loop fuel flow command control position of the gas turbine rotational acceleration cascade control structure, and a feed-forward compensation form with the predicted required power prediction error as the feed-forward compensation amount is introduced at the outer-loop gas turbine rotational acceleration command control position of the gas turbine rotational acceleration cascade control structure; and the recursive least squares algorithm is used to adaptively adjust the feed-forward compensation parameters of the two loops.

[0066] A turboshaft engine fast dynamic response control system includes a gas turbine rotational acceleration cascade control structure for controlling the turboshaft engine; the control system further includes:

[0067] A required power model for predicting the required power of the helicopter and generating a predicted required power;

[0068] A required power feed-forward module for introducing a feed-forward compensation form with predicted required power as the feed-forward compensation amount at the inner-loop fuel flow command control position of the gas turbine rotational acceleration cascade control structure, and introducing a feed-forward compensation form with the predicted required power prediction error as the feed-forward compensation amount at the outer-loop gas turbine rotational acceleration command control position of the gas turbine rotational acceleration cascade control structure;

[0069] An RLS gain adaptive adjustment module for using the recursive least squares algorithm to adaptively adjust the feed-forward compensation parameters of the two loops.

[0070] For the convenience of public understanding, taking the turboshaft engine control system designed for a coaxial high-speed helicopter as an example, the technical solution of the present invention will be described in detail with reference to the accompanying drawings:

[0071] The turboshaft engine control system of the coaxial high-speed helicopter in this embodiment has a structure as Figure 1As shown, it includes a gas turbine rotational acceleration cascade control structure with inner and outer double loops, as well as a required power model, a required power feedforward module, and an RLS gain adaptive adjustment module. Among them, the required power model is used to predict the required power of a coaxial high-speed helicopter and generate a predicted required power. The required power feedforward module is used to introduce a feedforward compensation with the predicted required power as the feedforward compensation amount at the inner-loop fuel flow command control position of the gas turbine rotational acceleration cascade control structure, and introduce a feedforward compensation with the required power prediction error as the feedforward compensation amount at the outer-loop gas turbine rotational acceleration command control position of the gas turbine rotational acceleration cascade control structure. The RLS gain adaptive adjustment module is used to adaptively adjust the feedforward compensation parameters of the two loops using the recursive least squares (RLS) algorithm. In addition, a limit protection controller is added to the gas turbine rotational acceleration cascade control structure in this embodiment, and the control output of the limit protection controller is introduced into the outer-loop gas turbine rotational acceleration command control position of the gas turbine rotational acceleration cascade control structure (it can also be introduced into the inner-loop fuel flow command control position).

[0072] The principles of each module introduced in the above turboshaft engine control system are further described in detail below:

[0073] (1) Required power model:

[0074] Due to the strong autocorrelation between some cross-linked parameters of high-speed helicopters and turboshaft engines, the model contribution degree may be low during feature selection and analysis. Therefore, selecting appropriate feature parameters as input variables is crucial for establishing a required power prediction model for high-speed helicopters. The present invention proposes a feature parameter extraction method combining random forest feature importance evaluation and recursive feature elimination, and evaluates the importance of all cross-linked parameters of the helicopter and the engine to screen out the optimal feature combination most relevant to the required power.

[0075] This method uses a random forest (RF) to evaluate parameter importance and combines a recursive feature elimination (REF) method to optimize the feature set to improve the accuracy and generalization ability of the prediction model.

[0076] First, for all 25 cross-linked parameters of high-speed helicopters and engines (longitudinal velocity V x , lateral velocity V y , vertical velocity V z , roll angle φ, pitch angle θ, yaw angle ψ, lateral position, longitudinal position, vertical position, total rotor pitch θ0, differential total pitch θ c , longitudinal cyclic pitch θ 1s , lateral cyclic pitch θ 1c , elevator deflection angle δ e , elevator differential deflection angle δ a, rudder deflection angle δ r , pitch angle θ of the thrust propeller p , rotor speed Ω MR , rotational speed Ω of the thrust propeller P , square of the longitudinal speed cube of the longitudinal speed , square of the lateral speed cube of the lateral speed , square of the vertical speed cube of the vertical speed ) are normalized to ensure a balanced feature distribution and eliminate the influence of inconsistent parameter ranges on the evaluation function. Subsequently, the random forest method is used to calculate the feature importance index and analyze its correlation with the power of the upper and lower rotors, the power of the thrust propeller, and the total required power of the high-speed helicopter.

[0077] The importance evaluation process of the random forest method is as Figure 2 shown. First, the coaxial high-speed helicopter dataset is randomly resampled to obtain multiple data subsets, and multiple decision trees are constructed by randomly splitting nodes. The optimal decision tree classification result is determined through voting. To obtain the importance index of the input features, based on the optimal classification model, the out-of-bag data is used to add noise to the input features, and its correlation with the target output is determined to obtain the feature importance index of all cross-linked parameters.

[0078] After initially evaluating the importance of each feature in all training sets through the RF method, it is used as an iterative classifier, and irrelevant or redundant features are gradually removed through the RFE method to determine the optimal cross-linked parameter feature combination. Subsequently, k-fold cross-validation is used to minimize the generalization error. The feature screening process of RF-RFE is as Figure 3 shown. The specific steps are as follows:

[0079] 1. Set the initial feature set of RF {X1, X2, …, X M}, and use the RF model to train the initial feature set. Calculate the initial importance index S i (X (k) ) of all features through the out-of-bag data;

[0080] 2. Sort the features in descending order according to the feature importance index, and calculate the classification accuracy P(X (k) ) of the current model in combination with k-fold cross-validation. Gradually remove the feature with the smallest importance and the least impact on the model performance to obtain a new feature subset {X1, X2, …, X M-1};

[0081] 3. Retrain the random forest model using the pruned feature subset, and calculate the classification accuracy P(X (k+1));

[0082] 4. By comparing the changes in cross-validation accuracy before and after removing features |P(X (k+1) )-P(X (k) )|, evaluate the impact of removing features on model performance;

[0083] 5. If the cross-validation accuracy before and after removing a feature drops by more than the preset threshold ε, stop the elimination process, output the previous feature subset, and use it as the optimal feature subset; otherwise, return to Step 2 and continue to iterate the process until the accuracy threshold is met or the number of features reaches the preset minimum value.

[0084] Feature extraction through RF-RFE can effectively prevent the degradation of model performance during feature elimination, ensure the stability and generalization ability of feature selection, and provide an important tool for cross-linking parameter analysis of high-speed helicopter engine integrated systems.

[0085] The present invention uses the RF-RFE method to gradually eliminate the cross-linking features that have little impact on the power required by the high-speed helicopter, and finally obtains the optimal feature set. The importance index of all 25 feature parameters is as follows Figure 4 As shown, the cross-validation accuracy of all feature numbers is as follows Figure 5 As shown in . The initial number of features is 25. After removing irrelevant or redundant features, the cross-validation accuracy change with a feature number of 10 remains within the threshold. Figure 6 As shown, the optimal cross-linking features finally determined are ranked as follows: θ p , V z , δ a ,θ0, θ, ψ, V x , which is used as the optimal input combination of the required power prediction model.

[0086] Based on the optimal cross-link feature combination screened by RF-RFE, a power requirement prediction model of BP neural network offline training is established. The model input is the optimized feature set, and the output is the required power of the upper and lower rotors and the required power of the thrust propeller of the high-speed helicopter. All input parameters can be measured by sensors.

[0087] The BP neural network mapping expression can be expressed as:

[0088]

[0089] like Figure 7As shown in the figure, the BP neural network adopts a three-layer structure, including an input layer, a hidden layer, and an output layer. The hidden layer contains 15 neurons and uses the Tansig activation function, while the output layer uses the Purelin function. The training data is from the simulation platform of the high-speed helicopter propulsion system, and the data is normalized. 1000 groups of data are selected, 80% of which is used for training and 20% for testing. The k = 5-fold cross-validation is adopted, and the error threshold ε is set to 0.02.

[0090] Compare the relative test errors of the required power prediction models constructed before and after RF-RFE screening for high-speed helicopters. The results are as Figure 8 shown. The calculation formula for the relative error of the output parameters of the BP neural network is

[0091] In Figure 8 , the relative errors of the output parameters of the required power model based on RF-RFE are all less than 0.8%. Except for a very small number of points, the errors of most samples are within 0.25%. Compared with the required power model without RF-RFE screening, the required power model based on RF-RFE still maintains a high prediction accuracy and avoids overfitting of the prediction model.

[0092] (2). RLS gain adaptive adjustment module, required power feedforward module:

[0093] The inner and outer loop feedforward links designed for the gas turbine rotational acceleration cascade control structure of the present invention are at positions 1 and 2 as shown in Figure 9 . To further improve the control performance of the double-loop required power feedforward, the present invention uses the recursive least squares algorithm to adaptively adjust the feedforward link of the control system to achieve the optimal configuration of the double-loop feedforward parameters. Through the coordinated control of the inner and outer double loops, the feedforward adaptive compensation is realized according to the system parameter changes, further improving the system robustness and control response speed.

[0094] Set the gain adjustment parameter vector θ = [θ1, θ2] for the inner and outer loop feedforward controls, and the initial covariance matrix P(0) = αI, where α is a relatively large constant used to improve the initial update rate, and I is the identity matrix.

[0095] The predicted required power P T,P (t) of the required power model based on RF-RFE and the actual required power P T (t) The error e(t) between them is expressed as:

[0096] e(t) = P T,P (t) - P T (t)

[0097] First, calculate the gain vector K(t) = [K1(t), K2(t)] of the current two feedforward positions according to the required power prediction error e(t):

[0098]

[0099] where P(t - 1) represents the covariance matrix; λ is the forgetting factor that controls the speed of weight update; φ(t) is the input vector, to increase the system's perception ability of dynamic characteristics; in addition, introduce the regularization coefficient ρ i to limit the amplitude of gain adjustment, ensure the system stability under the action of nonlinear compensation, and balance the compensation effect and the system response speed.

[0100] Then, use the gain vector K(t) and the error e(t) to update the gain adjustment parameter θ i (t):

[0101] θ1(t) = θ1(t - 1) + K(t)e(t)

[0102]

[0103] where k ξ is the regularization ratio coefficient, dynamically adjusts the weight coefficient through the error change rate, ensures that the weight coefficient tends to zero at steady state, makes the feedforward compensation amount of the outer loop zero at steady state, and avoids generating steady-state error in the outer loop compensation.

[0104] Subsequently, update the covariance matrix P(t):

[0105]

[0106] Through the above recursive process, RLS realizes the adaptive update of the feedforward gain coefficient.

[0107] According to the dynamic response characteristics of the inner and outer loops, introduce the predicted power P T,P as the feedforward compensation amount of the inner loop, and introduce the required power prediction error e(t) as the feedforward compensation amount of the outer loop. The feedforward gains of the double loops can be further expressed as:

[0108] u f1 (t) = θ1(t)P T,P (t)

[0109] u f2 (t) = θ2(t)e(t)

[0110] The above RLS-based dual-loop adaptive feedforward control method can dynamically adjust the feedforward compensation amount according to system disturbances by coordinating the feedforward effects of the inner and outer loops in the gas turbine rotational acceleration cascade control, thereby improving the control response speed and dynamic tracking performance, and significantly enhancing the anti-disturbance ability of the gas turbine rotational acceleration cascade control system during the feedforward control process. The RLS method effectively avoids the over-compensation phenomenon during the feedforward control caused by the large change in the power required by the high-speed helicopter, thus effectively avoiding the large fluctuations generated by the system.

[0111] To ensure that the RLS-based dual-loop adaptive feedforward control method always keeps the engine operating within the safe operating range during complex maneuvers, a limit protection controller is introduced at the control position of the outer-loop gas turbine rotational acceleration command or the inner-loop fuel flow command in the gas turbine rotational acceleration cascade control structure. This controller can monitor the key parameters of the engine in real time and actively adjust the control system to limit and protect high-risk operating conditions, ensuring the stable operation of the engine under complex flight conditions and extending its service life.

[0112] To verify the control effect of the above turboshaft engine fast dynamic response control system in different modes of high-speed helicopters, flight maneuvers are carried out in the low-speed mode, transition mode, and high-speed mode respectively, and the simulation verification of the engine fast response control method is carried out, and compared with the existing collective pitch feedforward control and the feedforward control method based on torque difference. All parameters are normalized according to the design point data.

[0113] Figure 10 This is the comparison result of the fast dynamic response control method (adaptive feedforward) of the present invention and other existing methods in the low-speed mode. As the flight speed increases continuously, the power demand of the coaxial dual rotors of the high-speed helicopter increases. Due to the certain response delay of the turboshaft engine, the power turbine speed drops instantaneously. At this time, the power required by the coaxial dual rotors still dominates. Figure 10(a)-(f) in the figure indicate that the engine fast dynamic response control method based on power demand adaptive feedforward of the present invention can effectively suppress the overshoot and droop of the power turbine speed. At the same time, compared with torque difference feedforward, the maximum droop amount is reduced by 36.5%, and the maximum overshoot amount is reduced by 39.9%. In addition, during the maneuver of a high-speed helicopter, the power demand of the rotor changes rapidly. The fast dynamic response control method based on power demand adaptive feedforward enables the engine to respond more quickly and has better dynamic control quality. Due to the lag in measuring the engine output torque, the large inertia of the rotor, and the aerodynamic coupling between the gas turbine and the power turbine, the fuel flow cannot be compensated in time by the torque difference feedforward correction command. However, the power demand adaptive feedforward inputs multiple cross-linked parameters of the high-speed helicopter and the engine into the prediction model. Using the predicted power and its change as the feedforward quantity, the RLS adaptive adjustment feedforward compensation quantity is adopted to correct the command and the fuel flow command online, so as to perform feedforward in two control loops, and thus more rapid compensation can be achieved. This method improves the control response speed and dynamic tracking performance, and significantly enhances the anti-disturbance ability of the engine control system based on the gas turbine rotational acceleration cascade during flight.

[0114] Figure 11 The figure shows the comparison results of the fast dynamic response control method of the present invention and other existing methods in the transition mode. As the flight speed increases, in order to avoid the tip stall of the rotor blade, the rotor speed needs to be reduced, and the power turbine speed drops synchronously. At the same time, the control efficiency of the rotor collective pitch weakens, and it is necessary to increase the collective pitch of the thrust paddle and reduce the rotor collective pitch to maintain the current lift and thrust constant. During this process, the proportion of the power demand of the coaxial dual rotors gradually decreases, while the proportion of the power demand of the thrust paddle gradually increases, and the total power demand of the high-speed helicopter increases accordingly. At this time, relying only on the rotor collective pitch is no longer sufficient to accurately characterize the change of the power demand of the high-speed helicopter, resulting in a reduction in the control performance based on collective pitch feedforward. As shown in Figure 11 (e) in the figure, during the accelerating climb maneuver starting at 134 seconds, the rotor speed continues to decrease, and the power turbine speed experiences an instantaneous droop due to the lag in the response of the turboshaft engine. At this time, the changes in the collective pitch and torque difference can no longer accurately reflect the change of the power demand of the high-speed helicopter, and it is difficult to meet the dynamic control requirements of the engine. In contrast, the fast dynamic response control method based on power demand adaptive feedforward achieves a faster power response through power demand adaptive feedforward compensation. The droop amount of the power turbine speed is reduced by 28.2%, and the fuel flow changes more smoothly.

[0115] Figure 12 The figure shows the comparison results of the fast dynamic response control method of the present invention and other existing methods in the high-speed mode. In the high-speed mode, the flight speed reaches 120 m / s, the coaxial dual rotors are in the unloaded state, the forward thrust is provided by the thrust paddle, and the power demand of the high-speed helicopter is mainly determined by the power demand of the thrust paddle. As shown in Figure 12As shown, the power turbine speed is reduced to 80%. After 190 seconds, the high-speed helicopter accelerates its climb by adjusting the pitch of the propulsive propeller and decelerates after 210 seconds. During this period, the power required by the high-speed helicopter changes drastically, resulting in a significant fluctuation in the engine output torque. However, due to the mechanical coupling and gas path coupling between the turboshaft engine and the high-speed helicopter, the delayed response of the engine causes significant overshoot and droop in the power turbine speed of the engine. Compared with the torque difference feedforward control method, the fast dynamic response control method based on the power required adaptive feedforward significantly improves the response speed of the turboshaft engine to power changes, reducing the maximum droop of the relative speed of the power turbine by 30.1% and the maximum overshoot by 17.5%. As Figure 12 shown in (a) and (d) of , the fast dynamic response control method based on the power required adaptive feedforward of the present invention can keep the engine running stably and reliably throughout the working process, demonstrating better steady-state and dynamic performance, as well as excellent robustness.

Claims

1. A rapid dynamic response control method for a turboshaft engine, which controls the turboshaft engine using a gas turbine rotational acceleration cascade control structure; characterized in that, A feedforward compensation form with the predicted required power as the feedforward compensation amount is introduced at the inner-loop fuel flow command control position of the gas turbine rotational acceleration cascade control structure, and a feedforward compensation form with the required power prediction error as the feedforward compensation amount is introduced at the outer-loop gas turbine rotational acceleration command control position of the gas turbine rotational acceleration cascade control structure; and the recursive least squares algorithm is used to adaptively adjust the feedforward compensation parameters of the two loops.

2. The rapid dynamic response control method for a turboshaft engine according to claim 1, wherein The use of the recursive least squares algorithm to adaptively adjust the feedforward compensation parameters of the two loops is specifically as follows: Step 1: Calculate the gain vector K(t)=[K1(t), K2(t)] of the two feedforward positions of the inner loop and the outer loop at the current time t according to the required power prediction error e(t): Wherein, P(t - 1) represents the covariance matrix at the moment of t - 1; λ is a forgetting factor for controlling the weight update speed; φ(t) is an input vector, ρ i is a regularization coefficient; i = 1 and i = 2 respectively represent the inner loop and the outer loop; Step 2: Update the gain adjustment parameter θ of the feedforward compensation by using the gain vector K(t) and the required power prediction error e(t). i (t): θ1(t)=θ1(t - 1)+K(t)e(t) where k ξ is the regularization ratio coefficient; Step 3: Update the covariance matrix P(t): Step 4: Obtain the feedforward gains of the inner loop and the outer loop as follows: u f1 u(t) = θ1(t)P T,P (t) u f2 (t) = θ2(t)e(t) Among them, P T,P (t) is the predicted required power.

3. The rapid dynamic response control method of the turboshaft engine according to claim 1, characterized in that The predicted required power is obtained based on a neural network model pre-trained by the following method: First, all measurable cross-linking parameters of the helicopter and the engine are used as the candidate feature set, and the random forest method and the recursive feature elimination method are combined to screen the candidate feature set to obtain the optimal cross-linking parameter feature combination most relevant to the required power; Then, the obtained optimal cross-linking parameter feature combination is used as the input feature, and the required power is used as the expected output to train the neural network model.

4. The helicopter power requirement prediction model according to claim 3, wherein The neural network model is a BP neural network model.

5. The rapid dynamic response control method for a turboshaft engine according to claim 1, characterized in that, A limit protection controller is further included in the gas turbine rotational acceleration cascade control structure, and the control output of the limit protection controller is introduced into the outer-loop gas turbine rotational acceleration command control position or the inner-loop fuel flow command control position of the gas turbine rotational acceleration cascade control structure to participate in the main loop limit protection.

6. A rapid dynamic response control system for a turboshaft engine, including a gas turbine rotational acceleration cascade control structure for controlling the turboshaft engine; characterized in that, The control system further includes: A required power model for predicting the required power of the helicopter and generating a predicted required power; A required power feedforward module for introducing a feedforward compensation form with the predicted required power as the feedforward compensation amount at the inner-loop fuel flow command control position of the gas turbine rotational acceleration cascade control structure, and introducing a feedforward compensation form with the required power prediction error as the feedforward compensation amount at the outer-loop gas turbine rotational acceleration command control position of the gas turbine rotational acceleration cascade control structure; An RLS gain adaptive adjustment module for using the recursive least squares algorithm to adaptively adjust the feedforward compensation parameters of the two loops.

7. The rapid dynamic response control system for a turboshaft engine according to claim 6, wherein The use of the recursive least squares algorithm to adaptively adjust the feedforward compensation parameters of the two loops is specifically as follows: Step 1: Calculate the gain vector K(t)=[K1(t), K2(t)] of the two feedforward positions of the inner loop and the outer loop at the current time t according to the required power prediction error e(t): Wherein, P(t - 1) represents the covariance matrix at time t - 1; λ is a forgetting factor for controlling the weight update speed; φ(t) is the input vector, ρ i is the regularization coefficient; i = 1 and i = 2 respectively represent the inner loop and the outer loop; Step 2: Update the gain adjustment parameter θ of the feedforward compensation by using the gain vector K(t) and the required power prediction error e(t): i (t): θ1(t)=θ1(t - 1)+K(t)e(t) where k ξ is the regularization ratio coefficient; Step 3: Update the covariance matrix P(t): Step 4: Obtain the feedforward gains of the inner loop and the outer loop as follows: u f1 u(t) = θ1(t)P T,P (t) u f2 (t) = θ2(t)e(t) Among them, P T,P (t) is the predicted required power.

8. The rapid dynamic response control system of a turboshaft engine according to claim 6, characterized in that, The required power model is a neural network model pre-trained by the following method: First, all measurable cross-linking parameters of the helicopter and the engine are used as the set of candidate features, and the random forest method is combined with the recursive feature elimination method to screen the set of candidate features, so as to obtain the optimal cross-linking parameter feature combination most relevant to the required power; Then, using the obtained optimal cross-linking parameter feature combination as the input feature and the required power as the expected output, the neural network model is trained.

9. The rapid dynamic response control system of the turboshaft engine according to claim 8, characterized in that The neural network model is a BP neural network model.

10. The rapid dynamic response control system of the turboshaft engine according to claim 6, characterized in that, The gas turbine rotational acceleration cascade control structure further includes a limit protection controller, and the control output of the limit protection controller is introduced into the outer loop gas turbine rotational acceleration command control position or the inner loop fuel flow command control position of the gas turbine rotational acceleration cascade control structure to participate in the main loop limit protection.

Images