A method for identification and predictive control of turboshaft engine based on MRR-KELM

Abstract

The application belongs to the technical field of helicopter control, and aims at solving the problem of poor generalization of the prediction model constructed for a turboshaft engine, and provides a turboshaft engine identification and prediction control method based on MRR-KELM, which adopts the MRR-KELM algorithm combined with the Huber loss function and the adaptive regularization strategy, suppresses the noise interference and the influence of abnormal values, greatly reduces the prediction error of the constructed turboshaft engine prediction model, and can realize dynamic mapping through the Gaussian kernel and adaptive adjustment of the working condition parameters; the algorithm introduces the regularization term and the robust loss function, combines the kernel function mapping capability, significantly improves the generalization performance and calculation efficiency of the model under complex working conditions such as small samples and noise interference, and further embeds the turboshaft engine prediction model into a nonlinear model predictive control framework, realizes high-precision stable control of the power turbine speed through rolling optimization and feedback correction.

Description

Technical Field

[0001] This invention belongs to the field of helicopter control technology, specifically relating to a turboshaft engine identification and predictive control method based on MRR-KELM. Background Technology

[0002] As the core power unit of helicopters, turboshaft engines need to maintain stable turbine speeds under complex flight conditions, which places extremely high demands on the accuracy and real-time performance of control system modeling. Traditional component-level modeling methods rely on aerodynamic and thermodynamic equations, resulting in complex modeling and high computational costs. System identification techniques, through data-driven modeling, are gradually becoming an important research direction. Among existing technologies, Least Squares Support Vector Machines (LS-SVM) perform excellently in small-sample regression, but their solution process involves linear equation operations, leading to a significant increase in computational complexity with increasing data volume. Furthermore, they are sensitive to noise, making it difficult to meet the real-time control requirements of engines. In addition, LS-SVM models suffer from poor sparsity and parameter tuning relies on experience, resulting in insufficient generalization ability in dynamic flight environments. Therefore, a novel identification algorithm with high accuracy, low computational cost, and strong robustness is needed to support efficient predictive control of turboshaft engines. Summary of the Invention

[0003] In order to solve at least one of the above-mentioned technical problems in the prior art, the present invention provides a turboshaft engine identification and predictive control method based on MRR-KELM.

[0004] This invention is achieved using the following technical solution: a method for identification and predictive control of a turboshaft engine based on MRR-KELM, comprising the following steps:

[0005] Step S1: Based on the MRR-KELM algorithm, the Huber loss function is introduced to calculate the weighted error, and an improved MRR-KELM algorithm model is constructed. At the same time, an adaptive regularization strategy is combined to balance the complexity and generalization ability of the MRR-KELM algorithm model.

[0006] Step S2: Identify the dynamic characteristics of the turboshaft engine based on the improved MRR-KELM algorithm model, and establish a turboshaft engine prediction model;

[0007] Step S3: Embed the turboshaft engine prediction model into a nonlinear model prediction control framework, establish a performance index function with constant power turbine speed and minimum fuel consumption as optimization objectives, set the upper limit of fuel flow and speed fluctuation constraints, use the sequential quadratic programming method to solve the optimal fuel input sequence in the future time domain, and obtain the predicted speed of the power turbine based on the optimal fuel input sequence.

[0008] Step S4: Compare the predicted speed with the actual speed of the turboshaft engine in real time to obtain the power turbine speed error, and dynamically update the model parameters of the turboshaft engine prediction model based on the proportional-integral term of the power turbine speed error to correct the control command.

[0009] Step S5: Apply the first control increment issued by the updated turboshaft engine prediction model to the actual control system, and advance the control in the time domain.

[0010] Preferably, the adaptive regularization strategy in step S1 includes:

[0011] The width of the Gaussian kernel is dynamically adjusted based on the flight altitude and Mach number.

[0012] Regularization coefficients were determined by combining cross-validation and Bayesian optimization.

[0013] A support vector sparsity strategy is adopted to select dominant feature vectors with a cumulative energy ratio of over 95%, and the MRR-KELM algorithm model is compressed.

[0014] Preferably, step S1 further includes:

[0015] The output layer weights are quickly solved by initializing the hidden layer weights randomly and using a closed-form analytical solution.

[0016] Preferably, the closed-form analytical solution is:

[0017]

[0018] In the formula, It is the output weight matrix, representing the connection weights between the hidden layer and the output layer in the improved MRR-KELM algorithm model; It is the output matrix of the hidden layer, containing the result of processing each training sample in the hidden layer; It is the expected output matrix, representing the expected output for each training sample.

[0019] Preferably, the turboshaft engine prediction model in step S2 is:

[0020]

[0021] In the formula, It is the output sequence that predicts multiple future times within the time domain, used for multi-step prediction with rolling optimization; It is the input data sequence of the prediction model.

[0022] Preferably, step S3 further includes:

[0023] Acquire sensor data corresponding to the current and historical fuel flow rate, required torque, gas turbine speed, power turbine speed, rotor torque, and turbine stage temperature of the turboshaft engine; and use the above sensor data as input to the turboshaft engine prediction model.

[0024] Preferably, step S5 further includes:

[0025] The MRR-KELM algorithm model is updated using online incremental learning, while the training dataset is updated in real time using a sliding window strategy.

[0026] The network structure of the MRR-KELM algorithm model is retained, and the output weights are adjusted only according to the updated model parameters, while the kernel width and regularization parameters are fine-tuned.

[0027] Preferably, before step S1, the method further includes:

[0028] Multi-source sensor data of the turboshaft engine are collected and preprocessed using sliding window noise reduction and operating condition segment normalization.

[0029] Compared with the prior art, the beneficial effects of the present invention are:

[0030] This invention employs the MRR-KELM algorithm combined with the Huber loss function and adaptive regularization strategy to suppress noise interference and outlier effects, significantly reducing the prediction error of the constructed turboshaft engine prediction model. Simultaneously, it can adaptively adjust operating parameters through Gaussian kernel dynamic mapping. By introducing regularization terms and robust loss functions, combined with kernel function mapping capabilities, this algorithm significantly improves the model's generalization performance and computational efficiency under complex operating conditions such as small samples and noise interference. Furthermore, the turboshaft engine prediction model is embedded into a nonlinear model predictive control framework, achieving high-precision and stable control of the power turbine speed through rolling optimization and feedback correction. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 A schematic diagram of the simulation verification process for the MRR-KELM algorithm method in the identification of turboshaft engine systems.

[0033] Figure 2 A schematic diagram of the framework for embedding nonlinear model predictive control into the predictive model of a turboshaft engine;

[0034] Figure 3 A schematic diagram of a nonlinear model predictive control framework for an engine is provided in this application;

[0035] Figure 4 This is a schematic diagram illustrating the recursive nature of an engine prediction model provided in this application. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other implementation methods obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] It should be noted that the structures, proportions, sizes, etc., shown in the accompanying drawings of this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed in the specification, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportional relationships, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should fall within the scope of the technical content disclosed in the present invention. It should be noted that in this specification, relational terms such as "first" and "second" are only used to distinguish one entity from several other entities, and do not necessarily require or imply any actual relationship or order between these entities.

[0038] This invention relates to a nonlinear model predictive engine control method based on the improved robust kernel extreme learning machine (MRR-KELM), aiming to solve the problems of complex modeling and high computational cost associated with traditional component-level modeling methods for turboshaft engines under complex operating conditions, which rely on aerodynamic and thermodynamic equations.

[0039] The core of the nonlinear model predictive engine control method is to embed the turboshaft engine prediction model into the nonlinear model predictive control framework to construct a prediction model. This prediction model performs rolling optimization control within a certain control time domain. However, in actual engineering, it is impossible to find a prediction model that is completely consistent with the controlled object.

[0040] Due to the complexity of component-based modeling of aero-engines, engine modeling using system identification methods has seen significant development. System identification is a scientific method that studies how to build mathematical models describing unknown systems using offline or online data.

[0041] like Figure 1 As shown, this embodiment of the invention provides a turboshaft engine identification and predictive control method based on MRR-KELM, including the following steps:

[0042] Step S1: Based on the MRR-KELM algorithm, the Huber loss function is introduced to calculate the weighted error, and an improved MRR-KELM algorithm model is constructed. At the same time, an adaptive regularization strategy is combined to balance the complexity and generalization ability of the MRR-KELM algorithm model.

[0043] Optionally, before step S1, the method further includes: collecting sensor data from the turboshaft engine and preprocessing it using sliding window noise reduction and operating condition segment normalization.

[0044] In this embodiment, data acquisition and preprocessing are performed on the turboshaft engine. This embodiment uses sensor data from the GE-T700 turboshaft engine under various flight conditions as an example, including parameters such as flight altitude, forward velocity, gas turbine speed, power turbine speed, fuel flow rate, and output torque. High-precision sensors are used for data acquisition, covering the entire flight envelope of the turboshaft engine, encompassing different flight phases such as hovering, forward flight, climb, and descent. The sampling frequency is set to 10Hz to ensure that the data reflects the engine's rapid dynamic changes.

[0045] In this embodiment, the data preprocessing steps are as follows:

[0046] 1. Median filtering is used to remove noise points from the data, and the size of the filtering window is dynamically adjusted according to the degree of data change.

[0047] 2. Normalize each parameter to map the numerical range to the [0,1] interval, thereby improving the convergence speed and stability of model training.

[0048] 3. The processed data is divided into training and test sets in a 7:3 ratio. Stratified sampling is used to ensure that both the training and test sets can reflect the operating characteristics of the turboshaft engine under different operating conditions.

[0049] Optionally, the adaptive regularization strategy includes: dynamically adjusting the Gaussian kernel width based on flight altitude and Mach number; determining the regularization coefficients through cross-validation and Bayesian optimization; employing a support vector sparsification strategy to select dominant feature vectors with a cumulative energy ratio of over 95%; and compressing the MRR-KELM algorithm model.

[0050] In this embodiment, the training set data is input into the MRR-KELM algorithm. The input data includes historical sequences of control variables such as fuel flow rate and required torque, and the output data consists of corresponding sequences of controlled variables such as turbine speed and output torque. The data sequence length is determined to be 20 sampling points to ensure that the model can capture the dynamic characteristics of the engine.

[0051] During training, the kernel width and regularization parameters are optimized using cross-validation. Batch gradient descent is employed to train the model. In each training iteration, the prediction error of the model on the current training data is calculated, and the network's output weights are adjusted via backpropagation. Simultaneously, the kernel width and regularization parameters are dynamically adjusted to minimize the model's prediction error and improve its generalization performance.

[0052] In this embodiment, the size of the MRR-KELM algorithm model is compressed to the range that the embedded controller can handle.

[0053] Optionally, it also includes: rapidly solving the output layer weights by initializing the hidden layer weights randomly and using a closed-form analytical solution.

[0054] In this embodiment, random hidden layer weight initialization refers to the process of randomly assigning values ​​to the weights of the hidden layers before training begins in the MRR-KELM algorithm. This initialization method helps break the symmetry of the neural network, preventing different neurons from learning the same features, thereby improving the model's expressive power and generalization ability.

[0055] After the hidden layer weights are initialized, the output layer weights can be quickly solved using a closed-form analytical solution.

[0056] The expression for the closed-form analytical solution is as follows:

[0057]

[0058] in: is the output weight matrix, representing the connection weights between the hidden layer and the output layer; H is the output matrix of the hidden layer, containing the result of each training sample after processing in the hidden layer; T is the expected output matrix, representing the expected output of each training sample;

[0059] In this embodiment, the Huber loss function, also known as the smoothed L1 loss function, is a commonly used loss function in regression tasks. It combines the advantages of mean squared error (MSE) and mean absolute error (MAE).

[0060] The Huber loss function is a commonly used loss function in regression tasks, combining the advantages of mean squared error (MSE) and mean absolute error (MAE). Specifically, the Huber loss function uses MSE when the error is small, making it insensitive to noise; while it uses MAE when the error is large, effectively suppressing the influence of outliers. Combining it with the MRR-KELM algorithm model can better handle noisy data and outliers during training, improving the model's robustness and prediction accuracy.

[0061] The improved MRR-KELM algorithm model structure is as follows:

[0062] Input Layer: The input layer receives multi-source sensor data from the turboshaft engine, including current and historical fuel flow, required torque, gas turbine speed, power turbine speed, rotor torque, and interstage temperature data. This sensor data, after preprocessing, serves as the input feature vector for the improved MRR-KELM algorithm model.

[0063] Hidden layers: Hidden layers consist of multiple neurons, with the input of each neuron connected to the feature vector of the input layer. The weights of the hidden layers are randomly initialized before training begins, and the activation function is a non-linear function (such as a sigmoid function) to introduce non-linear mapping capabilities.

[0064] Output Layer: Based on the output of the hidden layers and the expected output from the training data, the output layer quickly solves for the output weight matrix using a closed-form analytical solution. The prediction results of the output layer include controlled variables such as the turbine speed and output torque.

[0065] Model Training and Optimization: During the training of the MRR-KELM algorithm model, the Huber loss function is introduced to calculate the weighted error, constructing an improved MRR-KELM algorithm model. The kernel width and regularization parameters are optimized and adjusted using cross-validation, and the MRR-KELM algorithm model is trained using batch gradient descent. The specific steps are as follows:

[0066] ① Initialize the hidden layer weights;

[0067] ② Calculate the hidden layer output matrix;

[0068] ③ The prediction error is calculated using the Huber loss function;

[0069] ④ Update the output weights using batch gradient descent;

[0070] ⑤ Dynamically adjust the kernel function width and regularization parameters to minimize the model's prediction error and improve its generalization performance.

[0071] Through this model structure and training strategy, the improved MRR-KELM algorithm model can better handle the dynamic characteristics of turboshaft engines, thereby improving the accuracy and robustness of turboshaft engine prediction models.

[0072] To further optimize the model, we adopted an adaptive regularization strategy. The Gaussian kernel width was dynamically adjusted based on flight altitude and Mach number to adapt to data distributions under different operating conditions. Regularization coefficients were jointly determined through cross-validation and Bayesian optimization to ensure that the improved MRR-KELM algorithm model could fit the data well during training while avoiding overfitting, thereby improving the model's generalization ability. Furthermore, a support vector sparsity strategy was employed, selecting dominant feature vectors with a cumulative energy percentage of over 95%, and the MRR-KELM algorithm model was compressed to reduce model complexity and computational cost, making it more suitable for embedded controller applications.

[0073] During model training, we also employed an online incremental learning method. A sliding window strategy was used to update the training dataset in real time, preserving the network structure of the MRR-KELM algorithm model. Only the output weights and kernel width and regularization parameters were adjusted based on the updated model parameters. This ensures that the MRR-KELM algorithm model can quickly adapt to new data, maintaining good predictive performance, while also reducing the computational resources and time required for updating the MRR-KELM algorithm model.

[0074] The improved MRR-KELM model combines the Huber loss function with the MRR-KELM algorithm. Through the aforementioned model structure and training strategy, it effectively enhances the accuracy and robustness of the turboshaft engine prediction model constructed based on the improved MRR-KELM algorithm. In practical applications, the turboshaft engine prediction model can better meet the control requirements of turboshaft engines under complex operating conditions, providing reliable support for the nonlinear model predictive control framework.

[0075] Step S2: Identify the dynamic characteristics of the turboshaft engine based on the MRR-KELM algorithm model and establish a turboshaft engine prediction model.

[0076] In this embodiment, KELM provides a solution for identifying engine prediction models through a system due to its strong generalization ability, fast computation speed, and high algorithm stability. To identify a high-precision turboshaft engine KELM model, a large amount of training data is required, resulting in data redundancy and complexity. When training a model using the standard KELM algorithm, the number of hidden layers in the KELM corresponds one-to-one with the training sample data, making it difficult to guarantee the generalization ability of KELM. Therefore, it is necessary to select sample data from the training samples that are beneficial to model identification, eliminate redundant sample data, and simplify the model structure to improve the model's generalization ability.

[0077] In this embodiment, during the construction of the turboshaft engine prediction model based on the MRR-KELM algorithm model, the structural design of the turboshaft engine prediction model is first performed. The input layer receives preprocessed turboshaft engine sensor data, including historical sequences of sensor data corresponding to fuel flow, required torque, gas turbine speed, power turbine speed, rotor torque, and turbine stage temperature, as input feature vectors to capture the dynamic characteristics of the engine. The hidden layer uses a Gaussian kernel function as a nonlinear mapping function to map the input data to a high-dimensional feature space. The number of nodes is determined by cross-validation, the weights of the hidden layer are randomly initialized before training, and a nonlinear activation function is used. The output layer outputs the predicted power turbine speed, output torque, and other controlled variables, and the output weights are quickly solved using a closed-form analytical solution.

[0078] During the model training and optimization phases, the hidden layer weights are randomly initialized before training begins. Then, the hidden layer output matrix is ​​calculated based on the input data and the randomly initialized weights. When constructing the improved MRR-KELM model, the Huber loss function is introduced to calculate the prediction error. Simultaneously, cross-validation and Bayesian optimization are jointly used to determine the regularization coefficients, dynamically adjusting the kernel width and regularization parameters. Furthermore, a support vector sparsity strategy is employed to select dominant feature vectors, eliminate redundant sample data to simplify the model structure, and compress the MRR-KELM model size to a level manageable by an embedded controller.

[0079] Then, model validation and optimization are performed. The trained MRR-KELM model is validated using test data, and the root mean square error and coefficient of determination are calculated. The model is updated every 100 hours using an online incremental learning method. When retraining with new data, the original network structure and most parameters are retained, with only the output weights adjusted and the kernel width and regularization parameters fine-tuned. If the MRR-KELM model is found to have low prediction accuracy under specific conditions, a parameter optimization method based on particle swarm optimization is used for local optimization.

[0080] Finally, there is model embedding and control. For example... Figure 2 , Figure 3As shown, a nonlinear model predictive control framework for the engine is constructed by embedding the trained MRR-KELM model into a nonlinear model predictive control (NMPC) framework. Optimization objectives and constraints are set, and the optimal fuel input sequence is solved using the sequential quadratic programming (SQP) method. A feedback correction loop is introduced into NMPC, and an external disturbance is applied to the controlled object (i.e., the target helicopter) to obtain the actual rotational speed of the target helicopter. The predicted rotational speed is then compared with the actual rotational speed in real time. Based on the error correction control command, the model parameters are dynamically updated to ensure high-precision and stable control of the power turbine speed.

[0081] Optionally, it also includes: acquiring historical sequences of sensor data corresponding to fuel flow rate, required torque, gas turbine speed, power turbine speed, rotor torque, and turbine stage temperature of the turboshaft engine; and using the aforementioned sensor data as input to the turboshaft engine prediction model.

[0082] In this embodiment, the MRR-KELM algorithm model is used to identify the turboshaft engine prediction model, and the model input of the turboshaft engine prediction model is the fuel flow rate. Required torque, gas turbine speed Power turbine speed Rotor torque and turbine stage temperature The historical sequence of parameters is output as the corresponding sensor data at the current moment. , , , At this point, the un-iterated turboshaft engine prediction model can be described as:

[0083]

[0084] In the formula, This is the single-step output at the current moment. ; Inputs to the engine nonlinear model predictive control framework: ; The time index for the current moment; This refers to the length of the time window, which is the number of time steps in the historical data. This represents the transpose of a matrix.

[0085] In this embodiment, to ensure good model accuracy of the turboshaft engine prediction model while reducing its structural complexity, a suitable model structure needs to be selected. Since the engine dynamic response can be simplified to a second-order response process, the length of the control time window can be chosen. Since the controller solves the optimization problem with a constant turbine speed as the objective function through rolling optimization, ensuring that the input is always the optimal value at the next time step, the prediction time domain... Therefore, it can be followed Figure 4 The scheme illustrated in the recursive diagram of the prediction model is based on the fuel level at the current moment. Future prediction time domain fuel sequence of steps The corresponding historical input and output information are used to obtain the iterative prediction model for the un-iterated turboshaft engine. Predictive model of turboshaft engine after step:

[0086]

[0087] In the formula, It is the output sequence that predicts multiple future times within the time domain, used for multi-step prediction with rolling optimization; It is the input sequence of the turboshaft engine prediction model.

[0088] A predictive model for a turboshaft engine, based on the MRR-KELM algorithm, is used to predict the engine's output over a certain time domain in the future. The model's inputs include current and historical fuel flow rates and required torque, while its outputs include key parameters such as turbine speed and output torque. This predictive model allows for the advance estimation of engine behavior under different control inputs, providing a foundation for rolling optimization.

[0089] Step S3: Embed the turboshaft engine prediction model into a nonlinear model prediction control framework, establish a performance index function with constant power turbine speed and minimum fuel consumption as optimization objectives, set an upper limit for fuel flow and speed fluctuation constraints, use a sequential quadratic programming method to solve for the optimal fuel input sequence in the future time domain, and obtain the predicted speed based on the optimal fuel input sequence.

[0090] In this embodiment, the MRR-KELM model is embedded into a nonlinear model predictive control framework. By rolling optimization of the fuel input sequence in the future time domain, multivariate coordinated control of fuel flow and power turbine speed is achieved. An integrated feedback correction mechanism is used to compare the predicted output with the actual value in real time, dynamically update the model parameters to suppress accumulated errors, ensure that the power turbine speed tracking accuracy is better than ±0.5%, shorten the dynamic response delay to less than 10ms, and reduce fuel consumption by more than 15% when the load changes abruptly.

[0091] In this embodiment, rolling optimization is a core component of NMPC. For example... Figure 4As shown, within each control time domain, based on the current system state and the turboshaft engine prediction model, the optimal control action sequence is sought to ensure that the system achieves predetermined performance targets and satisfies constraints within a future period. The specific steps are as follows:

[0092] 1. Define performance indicators: Set control objectives, such as maintaining a constant turbine speed or reducing fuel consumption, and translate them into mathematical performance indicator functions. These typically include terms such as minimizing tracking error and minimizing changes in control variables. The specific formula for the performance indicator function is as follows:

[0093]

[0094] in: Is The actual output of the timing system, which here represents the power turbine speed of the turboshaft engine; It is a moment The system's desired output, i.e., the target power turbine speed; express Constantly control the increment of input, that is, the change in fuel flow; It is a weighting coefficient used to balance the relationship between tracking error and changes in control quantity; It is the length of the predicted time domain, i.e., the number of future time steps considered in rolling optimization.

[0095] 2. Constraint setting: Determine various constraints during system operation, such as the upper limit of fuel flow and the range of speed fluctuation, to ensure that the engine operates under safe and stable conditions.

[0096] 3. Optimization Algorithm Selection: The Sequential Quadratic Programming (SQP) optimization algorithm is used to solve the rolling optimization problem. Under the constraint conditions, the control input sequence that minimizes the performance index is sought.

[0097] 4. Rolling Implementation: At each sampling time, based on the latest state information of the current nonlinear model predictive control framework, prediction and optimization are performed again, and the first control increment obtained from the optimization is applied to the actual control system. Subsequently, the control time domain is rolled forward by one sampling cycle, and the above process is repeated to achieve real-time optimized control of the engine.

[0098] Step S4: Compare the predicted speed with the actual speed of the turboshaft engine in real time, and dynamically update the model parameters of the turboshaft engine prediction model based on the proportional-integral term correction command of the power turbine speed error.

[0099] In this embodiment, since the predictive control of the turboshaft engine predictive model depends on the accuracy of the model parameters, and the actual system contains factors such as model uncertainty and external disturbances, it is crucial to introduce a feedback correction stage in the NMPC. Feedback correction feeds back the error between the actual system output and the turboshaft engine predictive model output to the control system, adjusting the turboshaft engine predictive model parameters and control input in real time, thereby improving control accuracy and system stability.

[0100] In this embodiment, the NMPC feedback correction module is designed using a correction strategy based on model estimation error correction control commands. The strategy of correcting controller commands will not change the NMPC structure. Meanwhile, in order to eliminate the steady-state error of the control system, the control commands are corrected based on the proportional integral term of the error between the power turbine speed output and the actual engine speed corresponding to the prediction model.

[0101] Step S5: Apply the first control increment issued by the updated turboshaft engine prediction model to the actual control system, and advance the control in the time domain.

[0102] Optionally, it also includes: updating the MRR-KELM algorithm model using online incremental learning, while updating the training dataset in real time using a sliding window strategy; retaining the network structure of the MRR-KELM algorithm model, adjusting the output weights and fine-tuning the kernel width and regularization parameters only according to the updated model parameters.

[0103] In this embodiment, the MRR-KELM algorithm is used for parameter setting. A Gaussian kernel function is selected as the kernel function of the MRR-KELM algorithm, and its width parameter is initially set to 1.5. The regularization parameter is initially set to 0.1 to balance the model's fitting ability and generalization ability. The number of hidden layer nodes in the MRR-KELM network is set to 50, and the sigmoid activation function is used.

[0104] The training set data is input into the MRR-KELM algorithm. The input data includes historical sequences of control variables such as fuel flow rate and required torque. The output data consists of corresponding sequences of controlled variables such as turbine speed and output torque. The data sequence length is set to 20 sampling points to ensure that the model can capture the dynamic characteristics of the engine.

[0105] During training, the kernel width and regularization parameters are optimized using cross-validation. Batch gradient descent is employed to train the model. In each training iteration, the prediction error of the model on the current training data is calculated, and the network's output weights are adjusted via backpropagation. Simultaneously, the kernel width and regularization parameters are dynamically adjusted to minimize the model's prediction error and improve its generalization performance.

[0106] After approximately 2-3 hours of training, the MRR-KELM algorithm model achieved a root mean square error of 0.048 on the training set. The kernel width obtained during training was optimized to 1.8, and the regularization parameter was optimized to 0.05.

[0107] The trained MRR-KELM algorithm model was validated using test data. The root mean square error of the MRR-KELM algorithm model on the test set was calculated to be 0.042, and the coefficient of determination was [missing information]. The value reached 0.983, indicating that the MRR-KELM algorithm model has high prediction accuracy and good generalization ability.

[0108] During the experimental simulation, it was found that the MRR-KELM algorithm model had relatively low prediction accuracy under low-speed, high-load conditions. Therefore, local optimization of the MRR-KELM algorithm model was performed, focusing on adjusting the kernel width and regularization parameter values ​​under this condition. A parameter optimization method based on particle swarm optimization was used to retrain the MRR-KELM algorithm model on the training data under this condition. After optimization, the root mean square error of the MRR-KELM algorithm model on the test set was reduced to 0.038. The value was increased to 0.987, and the model performance was significantly improved.

[0109] During the actual operation of the turboshaft engine, new operational data is collected every 100 hours. An incremental learning method is used to update the MRR-KELM model, adding the new data to the training set and retraining the MRR-KELM algorithm model. In the incremental learning process, the original MRR-KELM algorithm model's network structure and most parameters are retained, with only the output weights adjusted. The kernel width and regularization parameters are fine-tuned based on the characteristics of the new data. Incremental learning significantly reduces training time, typically completing the process within 30 minutes to 1 hour.

[0110] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for identification and predictive control of a turboshaft engine based on MRR-KELM, characterized in that, Includes the following steps: Step S1: Based on the MRR-KELM algorithm, the Huber loss function is introduced to calculate the weighted error, and an improved MRR-KELM algorithm model is constructed. At the same time, an adaptive regularization strategy is combined to balance the complexity and generalization ability of the MRR-KELM algorithm model. The adaptive regularization strategy mentioned in step S1 includes: The width of the Gaussian kernel is dynamically adjusted based on the flight altitude and Mach number. Regularization coefficients were determined by combining cross-validation and Bayesian optimization. A support vector sparsity strategy is adopted to select the dominant feature vectors with a cumulative energy ratio of more than 95%, and the MRR-KELM algorithm model is compressed. Also includes: The output layer weights are quickly solved by initializing the hidden layer weights randomly and using a closed-form analytical solution. Step S2: Identify the dynamic characteristics of the turboshaft engine based on the improved MRR-KELM algorithm model, and establish a turboshaft engine prediction model; Step S3: Embed the turboshaft engine prediction model into a nonlinear model prediction control framework, establish a performance index function with constant power turbine speed and minimum fuel consumption as optimization objectives, set the upper limit of fuel flow and speed fluctuation constraints, use the sequential quadratic programming method to solve the optimal fuel input sequence in the future time domain, and obtain the predicted speed of the power turbine based on the optimal fuel input sequence. Step S4: Compare the predicted speed with the actual speed of the turboshaft engine in real time to obtain the power turbine speed error, and dynamically update the model parameters of the turboshaft engine prediction model based on the proportional-integral term of the power turbine speed error to correct the control command. Step S5: Apply the first control increment issued by the updated turboshaft engine prediction model to the actual control system, and advance the control in the time domain.

2. The method for identification and predictive control of a turboshaft engine based on MRR-KELM according to claim 1, characterized in that, The closed-form analytical solution is: In the formula, It is the output weight matrix, representing the connection weights between the hidden layer and the output layer in the improved MRR-KELM algorithm model; It is the output matrix of the hidden layer, containing the result of processing each training sample in the hidden layer; It is the expected output matrix, representing the expected output for each training sample.

3. The turboshaft engine identification and predictive control method based on MRR-KELM according to claim 1, wherein the turboshaft engine prediction model in step S2 is: In the formula, It is the output sequence that predicts multiple future times within the time domain, used for multi-step prediction with rolling optimization; It is the input data sequence of the prediction model.

4. The turboshaft engine identification and predictive control method based on MRR-KELM according to claim 1, step S3 further includes: Acquire sensor data corresponding to the current and historical fuel flow, required torque, gas turbine speed, power turbine speed, rotor torque, and turbine stage temperature of the turboshaft engine; The aforementioned sensor data is then used as input to the turboshaft engine prediction model.

5. The turboshaft engine identification and predictive control method based on MRR-KELM according to claim 1, step S5 further includes: The MRR-KELM algorithm model is updated using online incremental learning, while the training dataset is updated in real time using a sliding window strategy. The network structure of the MRR-KELM algorithm model is retained, and the output weights are adjusted only according to the updated model parameters, while the kernel width and regularization parameters are fine-tuned.

6. The turboshaft engine identification and predictive control method based on MRR-KELM according to claim 1, further comprising, before step S1: Sensor data from the turboshaft engine is collected and preprocessed using sliding window noise reduction and operating condition segment normalization.

Images