WO2021139004A1 - 基于自适应增强算法的涡扇发动机直接数据驱动控制方法 - Google Patents

基于自适应增强算法的涡扇发动机直接数据驱动控制方法 Download PDF

Info

Publication number
WO2021139004A1
WO2021139004A1 PCT/CN2020/080158 CN2020080158W WO2021139004A1 WO 2021139004 A1 WO2021139004 A1 WO 2021139004A1 CN 2020080158 W CN2020080158 W CN 2020080158W WO 2021139004 A1 WO2021139004 A1 WO 2021139004A1
Authority
WO
WIPO (PCT)
Prior art keywords
turbofan engine
data set
weight
error
2cor
Prior art date
Application number
PCT/CN2020/080158
Other languages
English (en)
French (fr)
Inventor
马艳华
丁男
孙希明
赵旭东
Original Assignee
大连理工大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 大连理工大学 filed Critical 大连理工大学
Priority to US17/052,756 priority Critical patent/US11788473B2/en
Publication of WO2021139004A1 publication Critical patent/WO2021139004A1/zh

Links

Images

Classifications

    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F02—COMBUSTION ENGINES; HOT-GAS OR COMBUSTION-PRODUCT ENGINE PLANTS
    • F02C—GAS-TURBINE PLANTS; AIR INTAKES FOR JET-PROPULSION PLANTS; CONTROLLING FUEL SUPPLY IN AIR-BREATHING JET-PROPULSION PLANTS
    • F02C9/00—Controlling gas-turbine plants; Controlling fuel supply in air- breathing jet-propulsion plants
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/20—Ensemble learning
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F05—INDEXING SCHEMES RELATING TO ENGINES OR PUMPS IN VARIOUS SUBCLASSES OF CLASSES F01-F04
    • F05D—INDEXING SCHEME FOR ASPECTS RELATING TO NON-POSITIVE-DISPLACEMENT MACHINES OR ENGINES, GAS-TURBINES OR JET-PROPULSION PLANTS
    • F05D2200/00—Mathematical features
    • F05D2200/10—Basic functions
    • F05D2200/11—Sum
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F05—INDEXING SCHEMES RELATING TO ENGINES OR PUMPS IN VARIOUS SUBCLASSES OF CLASSES F01-F04
    • F05D—INDEXING SCHEME FOR ASPECTS RELATING TO NON-POSITIVE-DISPLACEMENT MACHINES OR ENGINES, GAS-TURBINES OR JET-PROPULSION PLANTS
    • F05D2200/00—Mathematical features
    • F05D2200/10—Basic functions
    • F05D2200/12—Subtraction
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F05—INDEXING SCHEMES RELATING TO ENGINES OR PUMPS IN VARIOUS SUBCLASSES OF CLASSES F01-F04
    • F05D—INDEXING SCHEME FOR ASPECTS RELATING TO NON-POSITIVE-DISPLACEMENT MACHINES OR ENGINES, GAS-TURBINES OR JET-PROPULSION PLANTS
    • F05D2200/00—Mathematical features
    • F05D2200/10—Basic functions
    • F05D2200/13—Product
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F05—INDEXING SCHEMES RELATING TO ENGINES OR PUMPS IN VARIOUS SUBCLASSES OF CLASSES F01-F04
    • F05D—INDEXING SCHEME FOR ASPECTS RELATING TO NON-POSITIVE-DISPLACEMENT MACHINES OR ENGINES, GAS-TURBINES OR JET-PROPULSION PLANTS
    • F05D2200/00—Mathematical features
    • F05D2200/10—Basic functions
    • F05D2200/14—Division
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F05—INDEXING SCHEMES RELATING TO ENGINES OR PUMPS IN VARIOUS SUBCLASSES OF CLASSES F01-F04
    • F05D—INDEXING SCHEME FOR ASPECTS RELATING TO NON-POSITIVE-DISPLACEMENT MACHINES OR ENGINES, GAS-TURBINES OR JET-PROPULSION PLANTS
    • F05D2200/00—Mathematical features
    • F05D2200/20—Special functions
    • F05D2200/24—Special functions exponential
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F05—INDEXING SCHEMES RELATING TO ENGINES OR PUMPS IN VARIOUS SUBCLASSES OF CLASSES F01-F04
    • F05D—INDEXING SCHEME FOR ASPECTS RELATING TO NON-POSITIVE-DISPLACEMENT MACHINES OR ENGINES, GAS-TURBINES OR JET-PROPULSION PLANTS
    • F05D2270/00—Control
    • F05D2270/01—Purpose of the control system
    • F05D2270/02—Purpose of the control system to control rotational speed (n)
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F05—INDEXING SCHEMES RELATING TO ENGINES OR PUMPS IN VARIOUS SUBCLASSES OF CLASSES F01-F04
    • F05D—INDEXING SCHEME FOR ASPECTS RELATING TO NON-POSITIVE-DISPLACEMENT MACHINES OR ENGINES, GAS-TURBINES OR JET-PROPULSION PLANTS
    • F05D2270/00—Control
    • F05D2270/30—Control parameters, e.g. input parameters
    • F05D2270/304—Spool rotational speed
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F05—INDEXING SCHEMES RELATING TO ENGINES OR PUMPS IN VARIOUS SUBCLASSES OF CLASSES F01-F04
    • F05D—INDEXING SCHEME FOR ASPECTS RELATING TO NON-POSITIVE-DISPLACEMENT MACHINES OR ENGINES, GAS-TURBINES OR JET-PROPULSION PLANTS
    • F05D2270/00—Control
    • F05D2270/30—Control parameters, e.g. input parameters
    • F05D2270/306—Mass flow
    • F05D2270/3061—Mass flow of the working fluid
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F05—INDEXING SCHEMES RELATING TO ENGINES OR PUMPS IN VARIOUS SUBCLASSES OF CLASSES F01-F04
    • F05D—INDEXING SCHEME FOR ASPECTS RELATING TO NON-POSITIVE-DISPLACEMENT MACHINES OR ENGINES, GAS-TURBINES OR JET-PROPULSION PLANTS
    • F05D2270/00—Control
    • F05D2270/40—Type of control system
    • F05D2270/44—Type of control system active, predictive, or anticipative
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F05—INDEXING SCHEMES RELATING TO ENGINES OR PUMPS IN VARIOUS SUBCLASSES OF CLASSES F01-F04
    • F05D—INDEXING SCHEME FOR ASPECTS RELATING TO NON-POSITIVE-DISPLACEMENT MACHINES OR ENGINES, GAS-TURBINES OR JET-PROPULSION PLANTS
    • F05D2270/00—Control
    • F05D2270/70—Type of control algorithm
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F05—INDEXING SCHEMES RELATING TO ENGINES OR PUMPS IN VARIOUS SUBCLASSES OF CLASSES F01-F04
    • F05D—INDEXING SCHEME FOR ASPECTS RELATING TO NON-POSITIVE-DISPLACEMENT MACHINES OR ENGINES, GAS-TURBINES OR JET-PROPULSION PLANTS
    • F05D2270/00—Control
    • F05D2270/70—Type of control algorithm
    • F05D2270/709—Type of control algorithm with neural networks
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F05—INDEXING SCHEMES RELATING TO ENGINES OR PUMPS IN VARIOUS SUBCLASSES OF CLASSES F01-F04
    • F05D—INDEXING SCHEME FOR ASPECTS RELATING TO NON-POSITIVE-DISPLACEMENT MACHINES OR ENGINES, GAS-TURBINES OR JET-PROPULSION PLANTS
    • F05D2270/00—Control
    • F05D2270/70—Type of control algorithm
    • F05D2270/71—Type of control algorithm synthesized, i.e. parameter computed by a mathematical model
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F05—INDEXING SCHEMES RELATING TO ENGINES OR PUMPS IN VARIOUS SUBCLASSES OF CLASSES F01-F04
    • F05D—INDEXING SCHEME FOR ASPECTS RELATING TO NON-POSITIVE-DISPLACEMENT MACHINES OR ENGINES, GAS-TURBINES OR JET-PROPULSION PLANTS
    • F05D2270/00—Control
    • F05D2270/80—Devices generating input signals, e.g. transducers, sensors, cameras or strain gauges
    • F05D2270/803—Sampling thereof

Definitions

  • the invention belongs to the technical field of aeroengine control, and in particular relates to a direct data drive control method for a turbofan engine based on an adaptive enhancement algorithm.
  • Turbofan engines referred to as turbofan engines for short, are widely used in military and civilian fields due to their high propulsion efficiency and low fuel consumption.
  • the control system is a key component that directly determines its safety and performance.
  • the turbofan engine is a complex and strong nonlinear system, how to effectively control it has become a difficult problem.
  • the traditional turbofan engine control system adopts a control method that combines PID control algorithm and MIN/MAX switching. This method is conservative, with little performance improvement and slow dynamic response.
  • PID control algorithm PID control algorithm
  • MIN/MAX switching This method is conservative, with little performance improvement and slow dynamic response.
  • how to use advanced control technology to improve the conservativeness brought by the traditional control of turbofan engines and improve the safety and performance of turbofan engines Has important application significance.
  • turbofan engine control technology is mainly divided into model-based methods and data-driven methods.
  • model-based control method if the dynamic behavior of the controlled object is clearly defined, a controller can be designed to make the controlled object meet the required control requirements.
  • a complex thermodynamic system such as a turbofan engine
  • it is a data-driven control method which can directly use the controlled system or online or offline input and output data, instead of analyzing and using the mathematical model of the controlled object to design the controller.
  • the preset control target can be reached.
  • the data-based aviation turbofan engine control gets rid of the dependence on its precise model, and solves the problem of the turbofan engine when the precise mathematical model of the turbofan engine is not established or the mechanism model is difficult to establish.
  • the problem of achieving effective control there are the following problems: 1.
  • the generalization ability of a single model is weak, and it cannot completely control the full envelope range of complex nonlinear systems such as turbofan engines under different working conditions; 2.
  • the turbofan engine test data is relatively small, prone to the problem of data sample sparseness, which reduces the accuracy and generalization ability of the data-driven model. Therefore, in view of the above problems, it is of great significance to design a direct data-driven control method for turbofan engines that can solve the problems of weak generalization ability and sample sparseness.
  • the present invention proposes a turbofan engine direct data drive control method based on an adaptive enhancement algorithm.
  • the turbofan engine of the learner directly drives the controller with data.
  • the adaptive enhancement method not only improves the control accuracy, but also improves the generalization ability of the algorithm, and effectively solves the problem of sample sparsity.
  • the direct data drive control method of turbofan engine based on adaptive enhancement algorithm the steps are as follows:
  • Step 1 Establish a data set for the direct data-driven controller design of the turbofan engine
  • Step 1.2 ⁇ u represents the input of the turbofan engine, ⁇ y represents the output of the turbofan engine, ⁇ n 1cor (n) and ⁇ n 2cor (n) respectively represent the relative conversion speed changes of the low-pressure rotor and high-pressure rotor of the turbofan engine, ⁇ w f ( n) is the amount of change in the input fuel flow of the turbofan engine, defined
  • ⁇ y [ ⁇ n 2cor (1), ⁇ n 2cor (2),..., ⁇ n 2cor (n)] T
  • [ ⁇ u, ⁇ y] is the original data set designed by the direct data drive controller of the turbofan engine
  • Step 1.3 Use the relative converted speed n 2cor of the high-pressure rotor as the scheduling variable p (dimension equal to 1), and convert the scheduling variable p into [-1,1], as shown in the following formula:
  • n 2cor_max and n 2cor_min are the upper and lower limits of the relative speed n 2cor of the high-pressure rotor of the turbofan engine, respectively;
  • Step 2 Use mean value filling method and box plot analysis method to clean the data in the data set [ ⁇ u, ⁇ y], and fill in and eliminate data missing and abnormal data in the data set;
  • Step 3 Use the LSSVM algorithm to design the turbofan engine controller
  • Step 3.1 Using the random sampling method, use 80% of the data set as the training data set and 20% as the test data set;
  • the kernel function is expressed as follows:
  • t and k represent time t and k respectively
  • p(t) and p(k) represent the scheduling variables at time t and k
  • ⁇ is the initial hyperparameter radial basis width of the Gaussian kernel function, and ⁇ > 0;
  • Step 3.3 Establish the optimization problem of LSSVM:
  • ⁇ is the normal vector of the hyperplane
  • the hyperparameter ⁇ is the weight (dimension equal to 1) used to balance "finding the optimal hyperplane cost" and "the minimum deviation between the training set and the test set”
  • y i is The dependent variable after the control signal is given, e is the training error, b is the bias operator, and N is the number of samples in the training data set;
  • Step 3.4 Use the Gaussian kernel function in step 3.2 and solve the optimization problem in step 3.3 to obtain the LSSVM regression function, which is expressed as follows:
  • y lssvm is the output of the turbofan engine controller designed based on the LSSVM algorithm
  • ⁇ is the Lagrangian operator
  • b is the bias operator
  • N is the number of samples in the training data set
  • Step 4 Use the adaptive enhancement method and the output of the turbofan engine controller designed based on the LSSVM algorithm established in step 3 to construct the turbofan engine direct data drive controller based on the adaptive enhancement algorithm, and the parameters of the controller Make adjustments
  • the measured value of the change ⁇ n 2cor the basic learner in the adaptive enhancement algorithm uses the turbofan engine controller designed based on the LSSVM algorithm constructed in step 3, and gives the initial hyperparameter radial basis width ⁇ and weight ⁇ , And set epoch to the number of iterations of the basic learner;
  • the maximum error E k is expressed as follows:
  • Step 4.4 Calculate the relative error of the data samples in each training data set, using linear error, square error or exponential error, which are expressed as follows:
  • Step 4.5 Calculate the regression error rate e regression as shown in the following formula:
  • w ki is the weight of the data sample in the training data set obtained in the last iteration update
  • e ki is the relative error obtained in step 4.4;
  • Step 4.6 Calculate the weight coefficient weight k of the basic learner, as shown in the following formula:
  • Step 4.7 Update the sample weight distribution of the training data set, and adaptively adjust the initial hyperparameter radial basis width ⁇ according to the regression error rate, which is expressed as follows:
  • w ki is the weight coefficient of the i-th data sample at the k-th iteration
  • ⁇ k is the hyperparameter ⁇ at the k-th iteration
  • Step 4.8 Take the average of the predicted value yc generated in each iteration to obtain the final strong learner output y final , which is expressed as follows:
  • Step 5 Use the cross-validation method to determine the initial hyperparameter radial basis width ⁇ and weight ⁇ to satisfy the verification error of less than 0.1%.
  • the condition of ⁇ , ⁇ > ⁇ should be kept at all times, where ⁇ It is a small number not less than 0. If it does not match, the initial value is discarded, and the larger radial basis width ⁇ and weight ⁇ are selected as the initial values of the iteration to complete the direct data of the turbofan engine based on the adaptive enhancement algorithm Drive controller design.
  • the beneficial effect of the present invention the direct data drive control method of the turbofan engine based on the adaptive enhancement algorithm designed by the present invention.
  • Figure 1 is a flow chart of the control method of the present invention.
  • Figure 2 is a flow chart of the least squares support vector machine algorithm used in the present invention
  • Fig. 3 is a flowchart of the adaptive enhancement algorithm adopted by the present invention.
  • Figure 4 is a block diagram of the structure of a turbofan engine direct data drive controller based on an adaptive enhancement algorithm.
  • Step 1 Establish a data set for the direct data-driven controller design of the turbofan engine.
  • Step 1.2 ⁇ u represents the input of the turbofan engine, ⁇ y represents the output of the turbofan engine, ⁇ n 1cor (n) and ⁇ n 2cor (n) respectively represent the relative conversion speed changes of the low-pressure rotor and high-pressure rotor of the turbofan engine, ⁇ w f (n) is the amount of change in the input fuel flow of the turbofan engine, defined
  • ⁇ y [ ⁇ n 2cor (1), ⁇ n 2cor (2),..., ⁇ n 2cor (n)] T
  • [ ⁇ u, ⁇ y] is the data set designed by the direct data drive controller of the turbofan engine
  • Step 1.3 Use the relative converted speed n 2cor of the high-pressure rotor as the scheduling variable p (dimension equal to 1), and convert the scheduling variable p into [-1,1], as shown in the following formula
  • n 2cor_max and n 2cor_min are the upper and lower limits of the relative speed n 2cor of the high-pressure rotor of the turbofan engine, respectively;
  • Step 2 Use the mean value filling method and box plot analysis method to clean the data in the data set [ ⁇ u, ⁇ y], fill in and eliminate the data missing and abnormal data in the collected data set;
  • the LSSVM algorithm is used to design the turbofan engine controller as follows, and its structure diagram is shown in Figure 2:
  • Step 3.1 The training data obtained in step 1.2 is randomly sampled, using 80% of the total data as training data and 20% as test data;
  • p is the scheduling variable in the linear variable parameter model
  • t and k represent the time t and k respectively
  • p(t) and p(k) represent the scheduling variable at time t and k
  • ⁇ is the Gaussian kernel function Radial base width (belonging to hyperparameter), requires ⁇ >0;
  • Step 3.3 Establish an optimization problem:
  • ⁇ is the normal vector of the hyperplane
  • the hyperparameter ⁇ is the weight used to balance "finding the optimal hyperplane cost" and "the minimum deviation between the training set and the test set", and ⁇ > ⁇ is required, where ⁇ is not less than The smaller number of 0 (the dimension is equal to 1), y i is the dependent variable after the control signal is given, e is the training error, b is the bias operator, and N is the number of samples in the training data set.
  • Step 3.4 Use the Gaussian kernel function in step 3.2 and solve the optimization problem in step 3.3 to obtain the LSSSVM regression function, which is expressed as follows:
  • y lssvm is the output of the turbofan engine controller designed based on the LSSVM algorithm
  • ⁇ is the Lagrangian operator used in the solution process
  • N is the number of data samples used for training
  • the step 4 uses the adaptive enhancement method and the output of the turbofan engine controller designed based on the LSSVM algorithm established in step 3 to construct a turbofan engine direct data drive controller based on the adaptive enhancement algorithm , And adjust the parameters of the controller;
  • the measured value of the change ⁇ n 2cor the basic learner in the adaptive enhancement algorithm uses the turbofan engine controller designed based on the LSSVM algorithm constructed in step 3, and gives the initial hyperparameter radial basis width ⁇ and weight ⁇ , And set epoch to the number of iterations of the basic learner;
  • the maximum error E k is expressed as follows:
  • Step 4.4 Calculate the relative error of the data samples in each training data set, usually linear error, square error and exponential error (choose one when using), which are expressed as follows:
  • Step 4.5 Calculate the regression error rate e regression as shown in the following formula:
  • w ki is the weight of the data sample in the training data set obtained in the last iteration update, and e ki is the relative error obtained in step 4.4;
  • Step 4.6 Calculate the weight coefficient weight k of the basic learner, as shown in the following formula:
  • Step 4.7 Update the sample weight distribution of the training data set, and adaptively adjust the hyperparameter ⁇ according to the regression error rate, which is expressed as follows:
  • w ki is the weight coefficient of the i-th data sample at the k-th iteration
  • ⁇ k is the hyperparameter ⁇ at the k-th iteration
  • Step 4.8 Take the average of the predicted value y c generated in each iteration to obtain the final strong learner output y final , which is expressed as follows:
  • Step 5 Use the cross-validation method to determine the initial hyperparameter radial basis width ⁇ and weight ⁇ to satisfy the verification error of less than 0.1%.
  • the conditions of ⁇ , ⁇ > ⁇ should be kept at all times during the iteration process, here ⁇ is a small number not less than 0. If it does not match, the initial value is discarded, and the larger radial basis width ⁇ is selected as the initial value of the iteration to complete the direct data drive control of the turbofan engine based on the adaptive enhancement algorithm ⁇ Design.
  • the control absolute error and relative error comparison chart obtained by the controller shows that when the turbofan engine reaches a steady state, the absolute error is reduced by 95.8% compared with the original method after the lifting method is adopted, and the average relative error is reduced by 3.29% during the entire working time.
  • the turbofan engine direct data drive controller using the adaptive enhancement algorithm can speed up the time for the turbofan engine to reach the target speed, and can significantly reduce the control error and improve the control accuracy, which has obvious performance advantages.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Chemical & Material Sciences (AREA)
  • Software Systems (AREA)
  • Combustion & Propulsion (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • Medical Informatics (AREA)
  • Mechanical Engineering (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Feedback Control In General (AREA)

Abstract

一种基于自适应增强算法的涡扇发动机直接数据驱动控制方法，首先基于最小二乘支持向量机算法，设计涡扇发动机控制器，进一步，通过自适应增强算法改变训练样本的权重，从而构建成多个基本学习器组合成强学习器的涡扇发动机直接数据驱动控制器。相对于过去仅采用LSSVM的方案，通过使用自适应增强方法，不仅提高了控制精度，而且提升了算法的泛化能力，有效解决了样本稀疏性问题。

Description

基于自适应增强算法的涡扇发动机直接数据驱动控制方法 技术领域
本发明属于航空发动机控制技术领域，具体涉及一种基于自适应增强算法的涡扇发动机直接数据驱动控制方法。
背景技术
涡轮风扇发动机，简称为涡扇发动机，以其高推进效率，低燃油消耗率等优点广泛应用于军用和民用领域。控制系统作为涡扇发动机的大脑，是直接决定其安全和性能的关键部件。涡扇发动机作为一个复杂的强非线性系统，如何对其进行有效的控制成为了一个难题。传统的涡扇发动机控制系统采用PID控制算法和MIN/MAX切换相结合的控制方式，这种方式具有保守性，性能提升小，动态响应慢。与此同时，随着新兴的控制理论相继出现及在其他领域的验证和应用，如何采用先进的控制技术，改善涡扇发动机传统控制所带来的保守性，提高涡扇发动机的安全性和性能，具有重要的应用意义。
已有文献表明，涡扇发动机控制技术主要分为基于模型的方法和基于数据驱动的方法。首先，在基于模型的控制方法中，如果明确控制对象的动力学行为，便可以据此设计出控制器使被控对象达到所需的控制要求，而对于涡扇发动机这样一个复杂的热力学系统，很难达到理想的精确参数的模型，所以很难使涡扇发动机达到最优的控制效果。另一方面，则是基于数据驱动的控制方法，这种方法可以直接利用被控系统或在线或离线的输入输出的数据，而不用通过分析和使用被控对象的数学模型来设计控制器，同样可以达到预设的控制目标。因此相对于基于模型的控制方法，基于数据的航空涡扇发动机控制摆脱了对于其精确模型的依赖性，解决了在未建立涡扇发动机精确数学模型，或很难建立机理模型时对涡扇发动机实现有效控制的问题。然而，对于数据驱动的算法，存在以下问题：1、单一模型泛化能力较弱，不能完全对涡扇发动机这类复杂非线性系统不同工况下、全包线范围内进行控制；2、由于涡扇发动机试验数据相对较少，容易出现数据样本稀疏性问题，从而降低数据驱动模型的精度和泛化能力。因此，针对以上问题，设计一种能够解决泛化能力较弱和样本稀疏性问题的涡扇发动机的直接数据驱动控制方法，具有重大的意义。
发明内容
针对现有涡扇发动机控制方法存在的问题，本发明提出了一种基于自适应增强算法的涡扇发动机直接数据驱动控制方法。首先基于最小二乘支持向量机(Least Squares Support Vector Machine,LSSVM)算法，设计涡扇发动机控制器，进一步，通过自适应增强算法改变训练样本的权重，从而构建成多个基本学习器组合成强学习器的涡扇发动机直接数据驱动控制器。相对于过去仅采用LSSVM的方案，通过使用自适应增强方法，不仅提高了控制精度，而且提升了算法的泛化能力，并有效解决了样本稀疏性问题。
本发明的技术方案：
基于自适应增强算法的涡扇发动机直接数据驱动控制方法，步骤如下：
步骤1：建立涡扇发动机直接数据驱动控制器设计的数据集
步骤1.1：采集涡扇发动机运行过程中的控制信号，包括涡扇发动机的输入燃油流量w f(n)、低压转子的相对换算转速n 1cor(n)和高压转子的相对换算转速n 2cor(n)，n＝1,2,…表示第n个采 样周期；
步骤1.2：Δu表示涡扇发动机的输入，Δy表示涡扇发动机的输出，Δn 1cor(n)与Δn 2cor(n)分别表示涡扇发动机低压转子和高压转子的相对换算转速变化量，Δw f(n)为涡扇发动机输入燃油流量的变化量，定义
Δu＝[Δw f(1),Δw f(2),…,Δw f(n)] T
Δy＝[Δn 2cor(1),Δn 2cor(2),…,Δn 2cor(n)] T
则[Δu,Δy]为涡扇发动机直接数据驱动控制器设计的原始数据集；
步骤1.3：使用高压转子的相对换算转速n 2cor作为调度变量p(维数等于1)，将调度变量p转换到[-1,1]内，如下式所示：
Figure PCTCN2020080158-appb-000001
其中，n 2cor_max与n 2cor_min分别是涡扇发动机高压转子相对转速n 2cor的上限和下限；
步骤2：采用均值填补法与箱型图分析的方法对数据集[Δu,Δy]中的数据进行数据清洗，对于数据集中的数据缺失和数据异常进行填补与剔除；
步骤3：采用LSSVM算法，设计涡扇发动机控制器
步骤3.1：使用随机采样的方法，将数据集中的80％作为训练数据集，20％作为测试数据集；
步骤3.2：采用高斯核函数Ω＝K(p,t,k)将训练数据集从原空间映射到维数为z的高维特征空间中，实现训练数据集在z维特征空间的线性回归，核函数表示如下：
Figure PCTCN2020080158-appb-000002
其中，t和k是分别表示t时刻和k时刻，p(t)与p(k）表示t时刻和k时刻的调度变量，σ为高斯核函数初始的超参数径向基宽度，要求σ>0；
步骤3.3：建立LSSVM的优化问题：
Figure PCTCN2020080158-appb-000003
Figure PCTCN2020080158-appb-000004
其中，ω为超平面的法向量，超参数γ为用于平衡“寻找最优超平面花费算力”和“训练集与测试集偏差量最小”的权重(维数等于1)，y i为给予控制信号后的因变量，e为训练误差，b为偏置算子，N为训练数据集的样本数量；
步骤3.4：使用步骤3.2中的高斯核函数并求解步骤3.3中的优化问题得到LSSVM回归函数，表示如下：
Figure PCTCN2020080158-appb-000005
其中，y lssvm为基于LSSVM算法所设计的涡扇发动机控制器输出，α为拉格朗日算子，b为偏置算子，N为训练数据集的样本数量；
步骤4：使用自适应增强方法以及步骤3中建立的基于LSSVM算法所设计的涡扇发动机控制器输出，构建基于自适应增强算法的涡扇发动机直接数据驱动控制器，并对该控制器的参数进行调整
步骤4.1：训练数据集为T＝[Δu',Δy']，[Δu',Δy']为进行数据清洗后得到的涡扇发动机控制数据集，其中Δu'为控制信号，Δy'为高压转子变化量的测量值Δn 2cor，自适应增强算法中的基本学习器采用步骤3中构建的基于LSSVM算法所设计的涡扇发动机控制器，给出初始的超参数径向基宽度σ和权重γ，并设置epoch为基本学习器的迭代次数；
步骤4.2：初始化训练数据集的权重为D(1)＝(w 11,w 12,…,w 1N)，
Figure PCTCN2020080158-appb-000006
i＝1,2,…,N，w为每个训练数据集样本的权重；
步骤4.3：对于迭代次数k＝1,2,…,epoch，使用权重D(k)的训练数据集来训练，得到基本学习器G k(x)，计算训练数据集上的基本学习器预测的最大误差E k，表示如下：
Figure PCTCN2020080158-appb-000007
步骤4.4：计算每个训练数据集中数据样本的相对误差，采用线性误差、平方误差或指数误差，分别表示如下：
线性误差：
Figure PCTCN2020080158-appb-000008
平方误差：
Figure PCTCN2020080158-appb-000009
指数误差：
Figure PCTCN2020080158-appb-000010
步骤4.5：计算回归误差率e regression，如下式所示：
Figure PCTCN2020080158-appb-000011
其中，w ki为上一次迭代更新得到的训练数据集中数据样本的权重，e ki为步骤4.4得到的相对误差；
步骤4.6：计算基本学习器的权重系数weight k，如下式所示：
Figure PCTCN2020080158-appb-000012
步骤4.7：更新训练数据集的样本权重分布，并且根据回归误差率自适应调整初始的超参数径向基宽度σ，表示如下：
Figure PCTCN2020080158-appb-000013
σ k＝σ k-1-0.8*exp(-(weight k-weight k-1))
其中，w ki为第k次迭代时第i个数据样本的权重系数，σ k为第k次迭代时的超参数σ，
Figure PCTCN2020080158-appb-000014
为规范化算子；
步骤4.8：对于每次迭代产生的预测值yc取平均值，得到最终的强学习器输出y final，表示如下：
Figure PCTCN2020080158-appb-000015
步骤5：使用交叉验证的方法对于初始的超参数径向基宽度σ和权重γ进行确定使其满足验证误差低于0.1％，在迭代过程中要时刻保持σ,γ＞ζ的条件，这里ζ为不小于0的较小的数，如果不符合则丢弃该初始值，并选择更大的径向基宽度σ和权重γ作为迭代的初始值，完成基于自适应增强算法的涡扇发动机直接数据驱动控制器设计。
本发明的有益效果：通过本发明设计的基于自适应增强算法的涡扇发动机直接数据驱动控制方法。
附图说明
图1为本发明的控制方法流程图。
图2为本发明采用的最小二乘支持向量机算法流程图
图3为本发明采用的自适应增强算法流程图。
图4为基于自适应增强算法的涡扇发动机直接数据驱动控制器结构框图。
图5为在Δn 2cor＝88％，Δw f＝100工况下，基于单一LSSVM控制器与本发明基于自适应增强算法的涡扇发动机直接数据驱动控制器得到的控制效果对比图。
图6(a)和图6(b)分别为在Δn 2cor＝88％，Δw f＝100工况下，基于单一LSSVM控制器与本发明基于自适应增强算法的涡扇发动机直接数据驱动控制器得到的控制绝对误差与相对误差对比图。
具体实施方式
下面结合附图及技术方案对本发明实施做进一步详细说明。
本发明的控制方法流程图如图1所示，具体步骤如下：
步骤1：建立涡扇发动机直接数据驱动控制器设计的数据集。
步骤1.1：采集涡扇发动机运行过程中的控制信号，包括涡扇发动机的输入燃油流量w f(n)，低压转子的相对换算转速n 1cor(n)，高压转子的相对换算转速n 2cor(n)，n＝1,2,…表示第n个采样周期；
步骤1.2：Δu表示涡扇发动机的输入，Δy表示涡扇发动机的输出，Δn 1cor(n)与Δn 2cor(n)分别表示涡扇发动机低压转子与高压转子的相对换算转速的变化量，Δw f(n)为涡扇发动机输入燃油流量的变化量，定义
Δu＝[Δw f(1),Δw f(2),…,Δw f(n)] T
Δy＝[Δn 2cor(1),Δn 2cor(2),…,Δn 2cor(n)] T
则[Δu,Δy]为涡扇发动机直接数据驱动控制器设计的数据集；
步骤1.3：使用高压转子的相对换算转速n 2cor作为调度变量p(维数等于1)，将调度变量p转换到[-1,1]内，如下式所示
Figure PCTCN2020080158-appb-000016
其中，n 2cor_max与n 2cor_min分别是涡扇发动机高压转子相对转速n 2cor的上限和下限；
步骤2：采用均值填补法与箱型图分析的方法对数据集[Δu,Δy]中的数据进行数据清洗， 对于采集得到的数据集中的数据缺失和数据异常进行填补和剔除；
所述步骤3中采用LSSVM算法，设计涡扇发动机控制器步骤如下，其结构框图如图2：
步骤3.1：由步骤1.2得到的训练数据，使用随机采样的方法，将总数据量的80％作为训练数据，20％作为测试数据；
步骤3.2：采用高斯核函数Ω＝K(p,t,k)将训练数据集从原空间映射到维数为z的高维特征空间中，实现训练数据集在z维特征空间的线性回归，表示如下：
Figure PCTCN2020080158-appb-000017
其中，p为线性变参数模型中的调度变量，t和k是分别表示t时刻和k时刻，p(t)与p(k)表示t时刻和k时刻的调度变量，σ为高斯核函数的径向基宽度(属于超参数)，要求σ>0；
步骤3.3：建立优化问题：
Figure PCTCN2020080158-appb-000018
Figure PCTCN2020080158-appb-000019
其中ω为超平面的法向量，超参数γ为用于平衡“寻找最优超平面花费算力”和“训练集与测试集偏差量最小”的权重，要求γ>ζ，这里ζ为不小于0的较小的数(维数等于1)，y i为给予控制信号后的因变量，e为训练误差，b为偏置算子，N为训练数据集的样本数量。
步骤3.4：使用步骤3.2中的高斯核函数并求解步骤3.3中的优化问题得到LSSSVM回归函数，表示如下：
Figure PCTCN2020080158-appb-000020
其中y lssvm为基于LSSVM算法所设计的涡扇发动机控制器输出，α为求解过程中使用的拉格朗日算子，N为用于训练的数据样本数量；
如图3所示，所述步骤4中使用自适应增强方法以及步骤3中建立的基于LSSVM算法所设计的涡扇发动机控制器输出，构建基于自适应增强算法的涡扇发动机直接数据驱动控制器，并对该控制器的参数进行调整；
步骤4.1：输入训练样本T＝[Δu',Δy']，[Δu',Δy']为进行数据清洗后得到的涡扇发动机控制数据集，其中Δu'为控制信号，Δy'为高压转子转速变化量的测量值Δn 2cor，自适应增强算法中的基本学习器采用步骤3中构建的基于LSSVM算法所设计的涡扇发动机控制器，给出初始的超参数径向基宽度σ和权重γ，并设置epoch为基本学习器的迭代次数；
初始设置σ＝30，γ＝10，设置基本学习器的迭代次数epoch＝10。
步骤4.2：初始化训练集的权重为D(1)＝(w 11,w 12,…,w 1N)，
Figure PCTCN2020080158-appb-000021
i＝1,2,…,N，w为每个数据样本的权重；
步骤4.3：对于迭代次数k＝1,2,…,epoch，使用权重D(k)的训练数据集来训练，得到基本学习器G k(x)，计算训练数据集上的基本学习器预测的最大误差E k，表示如下：
Figure PCTCN2020080158-appb-000022
步骤4.4：计算每个训练数据集中数据样本的相对误差，通常采用线性误差、平方误差和指数误差(使用时选择一种即可)，分别表示如下：
线性误差：
Figure PCTCN2020080158-appb-000023
平方误差：
Figure PCTCN2020080158-appb-000024
指数误差：
Figure PCTCN2020080158-appb-000025
步骤4.5：计算回归误差率e regression，如下式所示：
Figure PCTCN2020080158-appb-000026
其中w ki为上一次迭代更新得到的训练数据集中数据样本的权重，e ki为步骤4.4得到的相对误差；
步骤4.6：计算基本学习器的权重系数weight k，如下式所示：
Figure PCTCN2020080158-appb-000027
步骤4.7：更新训练数据集的样本权重分布，并且根据回归误差率自适应调整超参数σ，表示如下：
Figure PCTCN2020080158-appb-000028
σ k＝σ k-1-0.8*exp(-(weight k-weight k-1))
其中w ki为第k次迭代时第i个数据样本的权重系数，σ k为第k次迭代时的超参数σ，
Figure PCTCN2020080158-appb-000029
为规范化算子。
步骤4.8：对于每次迭代产生的预测值y c取平均值，得到最终的强学习器输出y final，表示如下：
Figure PCTCN2020080158-appb-000030
步骤5：使用交叉验证的方法对于初始的超参数径向基宽度σ和权重γ进行确定，使其满足验证误差低于0.1％，在迭代过程中要时刻保持σ,γ＞ζ的条件，这里ζ为不小于0的较小的数，如果不符合则丢弃该初始值，并选择更大的径向基宽度σ作为迭代的初始值，完成基于自适应增强算法的涡扇发动机直接数据驱动控制器设计。
图5为在高压转子Δn 2cor＝88％，Δw f＝100工况下，基于单一LSSVM控制器与本发明基于自适应增强算法的涡扇发动机直接数据驱动控制器得到的控制效果对比图，可以看出本发明控制器加快了响应时间并且降低了超调。
图6(a)和图6(b)分别为在高压转子Δn 2cor＝88％，Δw f＝100工况下，基于单一LSSVM控制器与本发明基于自适应增强算法的涡扇发动机直接数据驱动控制器得到的控制绝对误差与相 对误差对比图，当涡扇发动机工作状态达到稳态后采用提升方法后绝对误差相对原始方法降低了95.8％，全工作时间内平均相对误差降低了3.29％。
综上可见，使用自适应增强算法的涡扇发动机直接数据驱动控制器可以加快涡扇发动机到目标转速的时间，并且可以显著降低控制误差，提高控制精度，具有明显的性能优势。

Claims (1)

  1. 一种基于自适应增强算法的涡扇发动机直接数据驱动控制方法，其特征在于，步骤如下：
    步骤1：建立涡扇发动机直接数据驱动控制器设计的数据集
    步骤1.1：采集涡扇发动机运行过程中的控制信号，包括涡扇发动机的输入燃油流量w f(n)、低压转子的相对换算转速n 1cor(n)和高压转子的相对换算转速n 2cor(n)，n＝1,2,…表示第n个采样周期；
    步骤1.2：Δu表示涡扇发动机的输入，Δy表示涡扇发动机的输出，Δn 1cor(n)与Δn 2cor(n)分别表示涡扇发动机低压转子和高压转子的相对换算转速变化量，Δw f(n)为涡扇发动机输入燃油流量的变化量，定义
    Δu＝[Δw f(1),Δw f(2),…,Δw f(n)] T
    Δy＝[Δn 2cor(1),Δn 2cor(2),…,Δn 2cor(n)] T
    则[Δu,Δy]为涡扇发动机直接数据驱动控制器设计的原始数据集；
    步骤1.3：使用高压转子的相对换算转速n 2cor作为调度变量p(维数等于1)，将调度变量p转换到[-1,1]内，如下式所示：
    Figure PCTCN2020080158-appb-100001
    其中，n 2cor_max与n 2cor_min分别是涡扇发动机高压转子相对转速n 2cor的上限和下限；
    步骤2：采用均值填补法与箱型图分析的方法对数据集[Δu,Δy]中的数据进行数据清洗，对于数据集中的数据缺失和数据异常进行填补与剔除；
    步骤3：采用LSSVM算法，设计涡扇发动机控制器
    步骤3.1：使用随机采样的方法，将数据集中的80％作为训练数据集，20％作为测试数据集；
    步骤3.2：采用高斯核函数Ω＝K(p,t,k)将训练数据集从原空间映射到维数为z的高维特征空间中，实现训练数据集在z维特征空间的线性回归，核函数表示如下：
    Figure PCTCN2020080158-appb-100002
    其中，t和k是分别表示t时刻和k时刻，p(t)与p(k)表示t时刻和k时刻的调度变量，σ为高斯核函数初始的超参数径向基宽度，要求σ>0；
    步骤3.3：建立LSSVM的优化问题：
    Figure PCTCN2020080158-appb-100003
    Figure PCTCN2020080158-appb-100004
    其中，ω为超平面的法向量，超参数γ为用于平衡“寻找最优超平面花费算力”和“训练集与测试集偏差量最小”的权重，y i为给予控制信号后的因变量，e为训练误差，b为偏置算子，N为训练数据集的样本数量；
    步骤3.4：使用步骤3.2中的高斯核函数并求解步骤3.3中的优化问题得到LSSVM回归函数，表示如下：
    Figure PCTCN2020080158-appb-100005
    其中，y lssvm为基于LSSVM算法所设计的涡扇发动机控制器输出，α为拉格朗日算子，b为偏置算子，N为训练数据集的样本数量；
    步骤4：使用自适应增强方法以及步骤3中建立的基于LSSVM算法所设计的涡扇发动机控制器输出，构建基于自适应增强算法的涡扇发动机直接数据驱动控制器，并对该控制器的参数进行调整
    步骤4.1：训练数据集为T＝[Δu',Δy']，[Δu',Δy']为进行数据清洗后得到的涡扇发动机控制数据集，其中Δu'为控制信号，Δy'为高压转子变化量的测量值Δn 2cor，自适应增强算法中的基本学习器采用步骤3中构建的基于LSSVM算法所设计的涡扇发动机控制器，给出初始的超参数径向基宽度σ和权重γ，并设置epoch为基本学习器的迭代次数；
    步骤4.2：初始化训练数据集的权重为D(1)＝(w 11,w 12,…,w 1N)，
    Figure PCTCN2020080158-appb-100006
    i＝1,2,…,N，w为每个训练数据集样本的权重；
    步骤4.3：对于迭代次数k＝1,2,…,epoch，使用权重D(k)的训练数据集来训练，得到基本学习器G k(x)，计算训练数据集上的基本学习器预测的最大误差E k，表示如下：
    Figure PCTCN2020080158-appb-100007
    步骤4.4：计算每个训练数据集中数据样本的相对误差，采用线性误差、平方误差或指数误差，分别表示如下：
    线性误差：
    Figure PCTCN2020080158-appb-100008
    平方误差：
    Figure PCTCN2020080158-appb-100009
    指数误差：
    Figure PCTCN2020080158-appb-100010
    步骤4.5：计算回归误差率e regression，如下式所示：
    Figure PCTCN2020080158-appb-100011
    其中，w ki为上一次迭代更新得到的训练数据集中数据样本的权重，e ki为步骤4.4得到的相对误差；
    步骤4.6：计算基本学习器的权重系数weight k，如下式所示：
    Figure PCTCN2020080158-appb-100012
    步骤4.7：更新训练数据集的样本权重分布，并且根据回归误差率自适应调整初始的超参数径向基宽度σ，表示如下：
    Figure PCTCN2020080158-appb-100013
    σ k＝σ k-1-0.8*exp(-(weight k-weight k-1)).
    其中，w ki为第k次迭代时第i个数据样本的权重系数，σ k为第k次迭代时的超参数σ，
    Figure PCTCN2020080158-appb-100014
    为规范化算子；
    步骤4.8：对于每次迭代产生的预测值y c取平均值，得到最终的强学习器输出y final，表示如下：
    Figure PCTCN2020080158-appb-100015
    步骤5：使用交叉验证的方法对于初始的超参数径向基宽度σ和权重γ进行确定使其满足验证误差低于0.1％，在迭代过程中要时刻保持σ,γ＞ζ的条件，这里ζ为不小于0的较小的数，如果不符合则丢弃该初始值，并选择更大的径向基宽度σ和权重γ作为迭代的初始值，完成基于自适应增强算法的涡扇发动机直接数据驱动控制器设计。
PCT/CN2020/080158 2020-01-07 2020-03-19 基于自适应增强算法的涡扇发动机直接数据驱动控制方法 WO2021139004A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/052,756 US11788473B2 (en) 2020-01-07 2020-03-19 Adaptive boosting algorithm-based turbofan engine direct data-driven control method

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010013151.6A CN111219257B (zh) 2020-01-07 2020-01-07 基于自适应增强算法的涡扇发动机直接数据驱动控制方法
CN202010013151.6 2020-01-07

Publications (1)

Publication Number Publication Date
WO2021139004A1 true WO2021139004A1 (zh) 2021-07-15

Family

ID=70832264

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/080158 WO2021139004A1 (zh) 2020-01-07 2020-03-19 基于自适应增强算法的涡扇发动机直接数据驱动控制方法

Country Status (3)

Country Link
US (1) US11788473B2 (zh)
CN (1) CN111219257B (zh)
WO (1) WO2021139004A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2022061871A1 (zh) * 2020-09-28 2022-03-31 大连理工大学 一种基于混合自适应差分进化的航空发动机模型迭代算法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102411305A (zh) * 2011-11-11 2012-04-11 南京航空航天大学 单旋翼直升机/涡轴发动机综合抗扰控制系统设计方法
US20150345403A1 (en) * 2014-06-02 2015-12-03 United Technologies Corporation Model-Based Optimal Control For Stall Margin Limit Protection in an Aircraft Engine
CN108375474A (zh) * 2018-01-26 2018-08-07 大连理工大学 一种航空发动机过渡态关键性能参数预测方法
CN110348079A (zh) * 2019-06-25 2019-10-18 南京航空航天大学 基于机器学习算法的涡轴发动机稳态开环控制设计方法

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR100275375B1 (ko) * 1997-11-10 2000-12-15 이중구 가스터빈 제어 방법
US7620461B2 (en) * 2007-06-26 2009-11-17 General Electric Company Systems and methods for using a combustion dynamics tuning algorithm with a multi-can combustor
CN102945324A (zh) * 2012-11-13 2013-02-27 江苏科技大学 无刷直流电机的多模型最小二乘支持向量机建模方法
CN103501148A (zh) * 2013-09-24 2014-01-08 江苏大学 一种无轴承永磁同步电机无径向位移传感器运行控制方法
CN107292326A (zh) * 2016-03-31 2017-10-24 阿里巴巴集团控股有限公司 一种模型的训练方法和装置
CN109162813B (zh) * 2018-10-09 2019-10-11 南京航空航天大学 一种基于迭代学习修正的航空发动机智能转速控制方法
CN109441644B (zh) * 2018-12-11 2021-01-05 大连理工大学 一种基于自抗扰理论的涡扇发动机稳态过渡态多变量控制方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102411305A (zh) * 2011-11-11 2012-04-11 南京航空航天大学 单旋翼直升机/涡轴发动机综合抗扰控制系统设计方法
US20150345403A1 (en) * 2014-06-02 2015-12-03 United Technologies Corporation Model-Based Optimal Control For Stall Margin Limit Protection in an Aircraft Engine
CN108375474A (zh) * 2018-01-26 2018-08-07 大连理工大学 一种航空发动机过渡态关键性能参数预测方法
CN110348079A (zh) * 2019-06-25 2019-10-18 南京航空航天大学 基于机器学习算法的涡轴发动机稳态开环控制设计方法

Also Published As

Publication number Publication date
US20210348567A1 (en) 2021-11-11
CN111219257A (zh) 2020-06-02
CN111219257B (zh) 2022-07-22
US11788473B2 (en) 2023-10-17

Similar Documents

Publication Publication Date Title
US20210201155A1 (en) Intelligent control method for dynamic neural network-based variable cycle engine
CN111042928B (zh) 一种基于动态神经网络的变循环发动机智能控制方法
CN104776446B (zh) 一种锅炉燃烧优化控制方法
CN109162813B (zh) 一种基于迭代学习修正的航空发动机智能转速控制方法
CN108803336B (zh) 一种航空发动机自适应lqg/ltr控制器设计方法
CN110348079B (zh) 基于机器学习算法的涡轴发动机稳态开环控制设计方法
CN114675535B (zh) 一种基于强化学习的航空发动机过渡态寻优控制方法
CN105867121B (zh) 用于自旋恢复的直升机非线性预测控制方法及装置
CN110579962B (zh) 基于神经网络的涡扇发动机推力预测方法及控制器
CN110219736A (zh) 基于非线性模型预测控制的航空发动机直接推力控制方法
WO2021139004A1 (zh) 基于自适应增强算法的涡扇发动机直接数据驱动控制方法
CN108196443A (zh) 变循环发动机的非线性预测控制设计方法
CN111679576B (zh) 一种基于改进确定性策略梯度算法的变循环发动机控制器设计方法
Zheng et al. Double fuzzy pitch controller of wind turbine designed by genetic algorithm
Kumar et al. Inverse design of airfoils using convolutional neural network and deep neural network
CN114167717B (zh) 一种基于改进pso-模糊pid的火电机组deh转速控制方法
Xiao et al. Research on the robustness of the roots waste heat power generation system
CN116577993B (zh) 涡轴发动机神经网络自适应控制方法及装置
CN115600429B (zh) 一种航空发动机机载自适应模型的参数估计方法
CN113867151B (zh) 一种基于sdnn-mpc的燃-燃联合动力装置负荷分配控制方法
CN114723104B (zh) 一种风电机组机舱电气设备能耗优化综合建模方法及系统
CN113282004B (zh) 基于神经网络的航空发动机线性变参数模型建立方法
Wang et al. Gain scheduling controller of the aero-engine based on LPV model
CN114943187B (zh) 一种基于神经网络的阀门特性曲面构造方法
CN116561711A (zh) 一种有效风速软测量方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20911742

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20911742

Country of ref document: EP

Kind code of ref document: A1