Intelligent control method and system for mode conversion of turboramjet combined engine

Abstract

The present invention provides an intelligent control method and system for modal conversion of a turboramjet combined engine. A modal conversion control system of the turboramjet combined engine is constructed through the turboramjet combined engine, so that the engine can stably control the thrust reduction or fluctuation in different modes and modal conversion processes; a prediction model for the multi-objective performance of the combustion chamber is constructed, and an anti-disturbance control system is used to perform real-time anti-disturbance control on the parameters of fuel injection under high efficiency and high precision conditions, so that the combustion chamber performance is adaptively adjusted under factors such as complex nonlinear systems and external environmental uncertainties, and the combustion effect is optimized.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent control of modal conversion of combined engines, and in particular relates to a method and system for intelligent control of modal conversion of a turboramjet combined engine (TBCC). Background Art

[0002] High-speed power systems have always been an important factor in hypersonic flight, and the engine is the heart of the high-speed power system. Usually, when the flight Mach number is low, the turbine engine is the best power device, but as the total temperature of the incoming flow increases, the turbine is no longer suitable for high-speed use due to the temperature of the rotating parts; when the Mach number is high, the ramjet engine becomes the best power choice, with higher specific impulse performance. Therefore, the turboramjet combination engine is expected to become an important power device for hypersonic aircraft that take off horizontally from the ground.

[0003] At present, the smooth transition between turbine engines and ramjet engines is a key issue to ensure the normal operation of the combined engine power. Turbine engines and ramjet engines have different working states and characteristics, and there are nonlinear unstable changes when transitioning from one mode to another. This unstable change has a significant impact on the intake, combustion, and emissions of turbine engines and ramjet engines. Therefore, under the premise of ensuring that the combustion chamber performance remains optimal within the entire working envelope, how to enable the combined engine to smoothly switch between the turbine mode and the ramjet mode has become the current key breakthrough technology. Under the condition of ensuring the stable conversion of different modes, how to obtain the optimal injection angle, injection position, injection frequency, injection pressure, number of fuel supply paths and other injection parameters and perform intelligent regulation has become a difficulty in the project.

[0004] The existing mode conversion control and fuel injection control mainly use the adjustment motor or some basic control methods to set the control parameter setting value, thereby realizing the fuel injection and mode conversion. It cannot identify the combustion performance of the engine in real time online according to the changes in the external environment, and then actively adjust the combustion chamber control parameters under different mode working conditions. Moreover, the adjustment accuracy of the system dynamic characteristics is not high, and with the continuous increase of the conversion between flight modes and the Mach number, the anti-interference ability of the control system has been significantly reduced, and it cannot adaptively and smoothly adjust the conversion between modes, thereby meeting the comprehensive performance goals of the future turboramjet combined engine in the full envelope and full range. Summary of the invention

[0005] In view of the shortcomings existing in the prior art, the purpose of the present invention is to provide a method and system for intelligent control of modal conversion of a turboramjet combination engine, so as to realize modal conversion and adaptive control of the turboramjet combination engine.

[0006] To achieve the above-mentioned purpose of the invention, the technical solution of the present invention is as follows:

[0007] A method for intelligent control of mode conversion of a turboramjet engine comprises the following steps:

[0008] S1. Analyze the influence of the main controllable parameters of fuel injection on the combustion performance of the turboramjet engine under different modes and working conditions;

[0009] S2. Under different modes and working conditions of the turboramjet engine, the Latin hypercube sampling method is used to obtain sample points in the design space, and a data set is obtained through experiments;

[0010] S3, preprocessing the data set;

[0011] S4. Build a deep belief network to form an agent model for predicting combustion performance;

[0012] S5. Use immune-inspired multi-objective optimization algorithm combined with deep belief network agent model to find the solution of design variables with optimal combustion performance;

[0013] S6. Based on the global optimal variables obtained by the immune heuristic multi-objective algorithm, the anti-disturbance control algorithm is used to perform real-time anti-disturbance control;

[0014] S7. Use the thrust generated by combustion and the fusion results of the inlet data to decide the mode conversion.

[0015] As a preferred embodiment, step S1 is specifically as follows:

[0016] The influence of the main controllable parameters of fuel injection on the combustion performance of the turboramjet combined engine in different modes is analyzed. The main controllable parameters include the injection angle, injection position, injection frequency, injection pressure, and number of fuel supply paths of the fuel injection; the combustion performance includes the total pressure loss of the combustion chamber, thrust and combustion efficiency, and the mathematical relationship between the main controllable parameters of fuel injection and the combustion performance is further obtained.

[0017] As a preferred embodiment, step S2 is specifically as follows:

[0018] The five design variables of fuel injection, namely, injection angle, injection position, injection frequency, injection pressure, and number of fuel supply paths, are parameterized. The upper and lower bounds of each design variable are determined under multiple operating parameters of different modes, different Mach numbers, and different incoming flow conditions, and the Latin hypercube sampling method is used to obtain sample points in the design space.

[0019] As a preferred method, the sample points of the design space are obtained in step S2 by combining the Latin hypercube sampling method as follows:

[0020] (1) First, determine the number of samples M that need to be obtained;

[0021] (2) Divide the intervals of the five design variables into M segments;

[0022] (3) Randomly select a value from each of the M segments;

[0023] (4) Randomly combine the values ​​drawn from different design variables;

[0024] (5) Obtaining M groups of samples after sampling;

[0025] As a preferred method, the data set is obtained through experiments in step S2 specifically as follows:

[0026] Through ground wind tunnel tests, the total pressure loss, thrust and combustion efficiency are obtained, and the computational fluid dynamics (CFD) numerical simulation results are verified and corrected. Then, the data sample size is expanded through computational fluid dynamics (CFD) numerical simulation, and a data set with one-to-one correspondence between design variables and combustion performance is constructed.

[0027] As a preferred embodiment, step S3 is specifically as follows:

[0028] Analyze and preprocess the data set, screen and eliminate completely worthless data to ensure the quality of the basic data set. Completely worthless data includes abnormal data, redundant data, and data outside the range of design variables.

[0029] As a preferred embodiment, step S4 is specifically as follows:

[0030] A proxy model based on a deep belief network was built. Combined with the preprocessed data set, a mathematical model was constructed under different modes and working conditions with five design variables of the fuel injector, namely the injection angle, injection position, injection frequency, injection pressure, and number of fuel supply paths, and the total pressure loss, thrust and combustion efficiency of the combustion chamber as output to form a proxy model for predicting the combustion chamber performance.

[0031] As a preferred method, the deep belief network DBN "connects" several RBMs in series to form a DBN, wherein the hidden layer of the previous RBM is the visible layer of the next RBM, and the output of the previous RBM is the input of the next RBM; the input of the first layer is the design variables of the fuel injector, and the output is the combustion performance, i.e., the total pressure loss, thrust and combustion efficiency of the combustion chamber;

[0032] The deep belief network is a probabilistic generative model. Compared with the traditional discriminative neural network, the generative model is to establish a joint distribution between observed data and labels:

[0033] P(v,h 1 ,h 2 ,…,h N )=P(v|h 1 )P(h1 |h 2 )…P(h N-2 |h N-1 )P(h N-1 |h N )

[0034] RBM is a neural sensor consisting of a visible layer and a hidden layer. The neurons in the visible layer and the hidden layer are fully connected in both directions. In an RBM, the hidden layer neurons h j Probability of being activated:

[0035] P(h j |v)=σ(b j +∑ i W i,j v i )

[0036] Since RBM is bidirectionally connected, the probability that the visible layer neurons can be activated by the hidden layer neurons is:

[0037] P(v j |h)=σ(c i +∑ i W i,j h j )

[0038] Where P(·) represents probability and N is the number of layers. RBM has five parameters: h, v, c, b, W, where c, b, W are the corresponding weights and bias values, which can be obtained through learning; v is the input vector; h is the output vector; σ is the Sigmoid function;

[0039] Neurons in the same layer are independent, so the probability density also satisfies independence:

[0040]

[0041] When a piece of data is assigned to the display layer, RBM calculates the probability of each hidden layer neuron being turned on according to the probability formula of hidden layer neuron activation, takes a random number between 0 and 1 as the threshold, and neurons greater than the threshold are activated, otherwise they are not activated, that is:

[0042] h j =1, P(h j |x)≥μ;h j =0,P(h j |x)<μ

[0043] The update method of relevant weights in the deep belief network is:

[0044]

[0045] Among them, λ is the coefficient, μ represents the threshold, which is a random number between 0 and 1; N h Indicates the number of hidden layers; N v Indicates the number of input quantities;

[0046] The last layer of the deep belief network can be replaced by a BP neural network, that is, supervised learning; the input of the deep belief network is the important operating parameters of the turboramjet combined engine, namely, the injection angle, injection position, injection frequency, injection pressure, and fuel supply path number design variables of the fuel injection, and the output is the total pressure loss, thrust and combustion efficiency of the combustion chamber.

[0047] As a preferred embodiment, step S5 is specifically as follows:

[0048] In different modes and working conditions of the turboramjet engine, the agent model established by the deep belief network is used as the fitness function, and the immune-heuristic multi-objective optimization algorithm is used to find the injection angle, injection position, injection frequency, injection pressure, and number of fuel supply paths of the fuel injection that optimize the total pressure loss, thrust and combustion efficiency, that is, to find the optimal solution of the five design variables.

[0049] As a preferred embodiment, in step S5, the process of using the immune heuristic multi-objective optimization algorithm to find the optimal solution of the fuel injection design variables includes:

[0050] (1) Antigen identification, that is, understanding the problem to be optimized, conducting feasibility analysis on the problem, extracting prior knowledge, constructing a mathematical model that conforms to the input-output relationship, and formulating various constraints;

[0051] (2) Initial antibody population: The feasible solutions to the problem are represented as antibodies in the solution space by encoding, and an initial population is randomly generated in the solution space;

[0052] (3) Evaluate the affinity of each feasible solution in the population;

[0053] (4) Determine whether the algorithm termination condition is met; if the condition is met, terminate the algorithm optimization process and output the calculation result; otherwise, continue the optimization operation;

[0054] (5) Calculate antibody concentration and stimulation degree;

[0055] (6) Perform immune processing, including immune selection, cloning, mutation and clonal suppression;

[0056] Immune selection means: selecting high-quality antibodies based on the affinity and concentration calculation results of antibodies in the population and activating them;

[0057] Cloning: cloning and replicating the activated antibody to obtain several copies;

[0058] Mutation: that is, performing mutation operations on the cloned copies to cause affinity mutations;

[0059] Clone suppression: that is, reselect the compiled results to suppress antibodies with low affinity and retain variant results with high affinity.

[0060] (7) Refreshing the population, replacing the antibodies with lower excitation levels in the population with randomly generated new antibodies to form a new generation of antibodies, and then go to step (3).

[0061] As a preferred embodiment, step S6 is specifically as follows:

[0062] According to the immune-heuristic multi-objective algorithm, the optimal solution for a set of controllable variables is determined, namely, the injection angle, injection position, injection frequency, injection pressure, and number of fuel supply paths of the corresponding fuel injection under the optimal conditions of total pressure loss, thrust and combustion efficiency; an anti-disturbance controller is built to perform real-time adaptive control of the three design variables of injection angle, injection frequency and injection pressure to achieve adaptive injection with optimal combustion chamber performance.

[0063] As a preferred embodiment, in step S7:

[0064] The decision mode conversion is realized by the mode conversion control system, which includes four parts: engine cycle and working state selection, decision-making mechanism, different mode states, and thrust and inlet data fusion:

[0065] Engine cycle and operating state selection is obtained through the fusion of turboramjet thrust and inlet data; the obtained information is sent to the decision-making body;

[0066] The decision-making body will judge the current state and select the appropriate mode to work; according to the modal information and different working conditions, the deep belief network and immune heuristic multi-objective optimization algorithm are used to find the fuel injection parameters that can keep the engine combustion chamber at the optimal performance under the current mode and working conditions; further observe the thrust generated by the combined engine, that is, form a closed loop for the entire control system, and achieve the effect of real-time observation and adjustment;

[0067] Different mode states are used for different flight speeds. For example, when flying at low Mach numbers, the turbine engine is mainly used to provide power; when flying at high Mach numbers, the ramjet engine is used to provide power.

[0068] Inlet data fusion is used to detect the incoming flow conditions in real time, mainly including the Mach number, and then decide the engine mode selection. Among them, the fuel injection parameters that keep the engine combustion chamber at the best performance include injection angle, injection frequency and injection pressure. The three design variables of injection angle, injection frequency and injection pressure obtained by optimization are used as the expectations of the ADRC control system, and the ADRC control system is used to filter out interference and stabilize the injection parameters, so that the engine combustion chamber can always maintain the best performance in different modes.

[0069] The present invention also provides a turboramjet combined engine fuel injection control system based on active disturbance rejection control, which is applied to the above-mentioned intelligent control method, comprising:

[0070] The ADRC is used to control the fuel injector variables that are susceptible to interference and unstable during flight, that is, to complete the real-time control of the injection angle, injection frequency and injection pressure. The motor that controls the injection is susceptible to interference during flight, so the variables directly controlled by the motor are susceptible to interference and unstable variables.

[0071] The active disturbance rejection control system ADRC consists of three parts: tracking differentiator, extended state observer and nonlinear error feedback control; the tracking differentiator is used to track and differentiate the input signal of the system to obtain a stable system input signal; the extended state observer is used to estimate the real-time state of fuel injection and the total disturbance of the system; the nonlinear error feedback control is used to compensate the control rate according to the stable system input signal, the estimated real-time state and the total disturbance of the system, and generate the final control quantity of the fuel injector.

[0072] The beneficial effects of the present invention are as follows: the present invention constructs a turboramjet combined engine mode conversion control system through a turboramjet combined engine, so that the engine can stably control the thrust reduction or fluctuation in different modes and mode conversion processes; the constructed combustion chamber multi-objective performance prediction model uses an anti-disturbance control system to perform real-time anti-disturbance control on the parameters of fuel injection under high efficiency and high precision conditions, and the combustion chamber performance is adaptively adjusted under factors such as complex nonlinear systems and external environmental uncertainties, so as to achieve the best combustion effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 A schematic diagram of a flow chart of a method for intelligent control of mode conversion of a turboramjet combined engine according to the present invention;

[0074] Figure 2 It is a schematic structural diagram of a turboramjet combined engine fuel injection control system based on active disturbance rejection control according to the present invention;

[0075] Figure 3This is a schematic diagram of the structure of the deep belief network agent model of the present invention;

[0076] Figure 4 Schematic diagram of the network structure of the restricted Boltzmann machine of the present invention;

[0077] Figure 5 It is a schematic diagram of the flow chart of the immune heuristic multi-objective optimization algorithm of the present invention;

[0078] Figure 6 It is a schematic diagram of the structure of the turboramjet combined engine model of the present invention. DETAILED DESCRIPTION

[0079] The following describes the embodiments of the present invention through specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention.

[0080] Example 1

[0081] This embodiment provides a method for intelligently controlling mode conversion of a turboramjet engine, comprising the following steps:

[0082] S1. Analyze the influence of the main controllable parameters of fuel injection on the combustion performance of the turboramjet engine under different modes and working conditions;

[0083] S2. Under different modes and working conditions of the turboramjet engine, the Latin hypercube sampling method is used to obtain sample points in the design space, and a data set is obtained through experiments;

[0084] S3, preprocessing the data set;

[0085] S4. Build a deep belief network to form an agent model for predicting combustion performance;

[0086] S5. Use immune-inspired multi-objective optimization algorithm combined with deep belief network agent model to find the solution of design variables with optimal combustion performance;

[0087] S6. Based on the global optimal variables obtained by the immune heuristic multi-objective algorithm, the anti-disturbance control algorithm is used to perform real-time anti-disturbance control;

[0088] S7. Use the thrust generated by combustion and the fusion results of the inlet data to decide the mode conversion.

[0089] Embodiment 2:

[0090] The embodiment provides a method for intelligent control of mode conversion of a turboramjet combined engine, such as Figure 1 As shown, this embodiment specifically includes the following steps S1-S7:

[0091] S1. Analyze the influence of the main controllable parameters of fuel injection on the combustion performance of the turboramjet engine in different modes;

[0092] The specific implementation of step S1 is:

[0093] The influence of the main controllable parameters of fuel injection on the combustion performance of the turboramjet combined engine in different modes is analyzed. The main controllable parameters include the injection angle, injection position, injection frequency, injection pressure, and number of fuel supply paths of the fuel injection; the combustion performance includes the total pressure loss of the combustion chamber, thrust and combustion efficiency, and the mathematical relationship between the main controllable parameters of fuel injection and the combustion performance is further obtained.

[0094] S2. Under different modes of the turboramjet engine, the Latin hypercube sampling method is used to obtain sample points in the design space, and a data set is obtained through experiments;

[0095] The five design variables of fuel injection, including injection angle, injection position, injection frequency, injection pressure, and number of fuel supply paths, are parameterized. Under multiple operating parameters such as different modes, different Mach numbers, and different incoming flow conditions, the upper and lower bounds of each design variable are determined, and then the Latin hypercube sampling method is used to obtain sample points in the design space.

[0096] In step S2, the sample points of the design space are obtained by combining the Latin hypercube sampling method as follows:

[0097] (1) First, determine the number of samples M that need to be obtained;

[0098] (2) Divide the intervals of the five design variables into M segments;

[0099] (3) Randomly select a value from each of the M segments;

[0100] (4) Randomly combine the values ​​drawn from different design variables;

[0101] (5) Obtain M groups of samples after sampling.

[0102] In step S2, the data set is obtained through experiments as follows:

[0103] Through ground wind tunnel tests, the total pressure loss, thrust and combustion efficiency are obtained. The computational fluid dynamics (CFD) numerical simulation results are verified and corrected, and then the data sample size is expanded through computational fluid dynamics (CFD) numerical simulation to construct a data set with one-to-one correspondence between design variables and combustion performance.

[0104] S3, preprocessing the data set;

[0105] The specific implementation of step S3 is:

[0106] Analyze and preprocess the data set, filter and remove completely worthless data to ensure the quality of the basic data set. Completely worthless data includes abnormal data, redundant data, and data outside the design variable range.

[0107] S4. Build a deep belief network to form an agent model for predicting combustion performance;

[0108] The specific implementation of step S4 is as follows Figure 3 and Figure 4 As shown:

[0109] A proxy model based on a deep belief network was built. Combined with the preprocessed data set, a mathematical model was constructed under different modes and working conditions with five design variables of the fuel injector, including injection angle, injection position, injection frequency, injection pressure, and number of fuel supply paths, and the total pressure loss, thrust, and combustion efficiency of the combustion chamber as outputs to form a proxy model for predicting the combustion chamber performance.

[0110] A deep belief network (DBN) connects several RBMs in series to form a DBN, where the hidden layer of the previous RBM is the visible layer of the next RBM, and the output of the previous RBM is the input of the next RBM. The input of the first layer is the design variables of the fuel injector, and the output is the combustion performance, namely the total pressure loss, thrust and combustion efficiency of the combustion chamber.

[0111] The deep belief network is a probabilistic generative model. Compared with the traditional discriminative neural network, the generative model is to establish a joint distribution between observed data and labels:

[0112] P(v,h 1 ,h 2 ,…,h N )=P(v|h 1 )P(h 1 |h 2 )…P(h N-2 |h N-1 )P(h N-1 |h N )

[0113] RBM is a neural perceptron consisting of a visible layer and a hidden layer, with bidirectional full connections between the neurons in the visible layer and the hidden layer. j Probability of being activated:

[0114] P(h j |v)=σ(b j +∑i W i,j v i )

[0115] Since RBM is bidirectionally connected, the probability that the visible layer neurons can be activated by the hidden layer neurons is:

[0116] P(v j |h)=σ(c i +∑ i W i,j h j )

[0117] Where P(·) represents probability and N is the number of layers. RBM has five parameters: h, v, c, b, W, where c, b, W are the corresponding weights and bias values, which can be learned; v is the input vector; h is the output vector; and σ is the Sigmoid function.

[0118] Neurons in the same layer are independent, so the probability density also satisfies independence:

[0119]

[0120] When a piece of data is assigned to the display layer, RBM calculates the probability of each hidden layer neuron being turned on according to the probability formula of hidden layer neuron activation, takes a random number between 0 and 1 as the threshold, and neurons greater than the threshold are activated, otherwise they are not activated, that is:

[0121] h j =1, P(h j |x)≥μ;h j =0,P(h j |x)<μ

[0122] The update method of relevant weights in the deep belief network is:

[0123]

[0124] Among them, λ is the coefficient; μ represents the threshold, which is a random number between 0 and 1; N h Indicates the number of hidden layers; N v Indicates the number of input quantities;

[0125] The last layer of the deep belief network can be replaced by a BP neural network, i.e. supervised learning. The input of the deep belief network is the important operating parameters of the turboramjet engine, i.e. the injection angle, injection position, injection frequency, injection pressure, and fuel supply path design variables of the fuel injection, and the output is the total pressure loss, thrust, and combustion efficiency of the combustion chamber.

[0126] S5. Use immune-inspired multi-objective optimization algorithm combined with deep belief network agent model to find the solution of design variables with optimal combustion performance;

[0127] The specific implementation of step S5 is:

[0128] In different modes and working conditions of the turboramjet engine, the agent model established by the deep belief network is used as the fitness function, and the immune-heuristic multi-objective optimization algorithm is used to find the injection angle, injection position, injection frequency, injection pressure, and number of fuel supply paths of the fuel injection that optimize the total pressure loss, thrust and combustion efficiency, that is, to find the optimal solution of the five design variables.

[0129] In this embodiment, the process of finding the optimal solution set of fuel injection design variables using the immune heuristic multi-objective optimization algorithm is as follows: Figure 5 As shown, including:

[0130] (1) Antigen identification, that is, understanding the problem to be optimized, conducting feasibility analysis on the problem, extracting prior knowledge, constructing a mathematical model that conforms to the input-output relationship, and formulating various constraints;

[0131] (2) Initial antibody population: The feasible solutions to the problem are represented as antibodies in the solution space by encoding, and an initial population is randomly generated in the solution space.

[0132] (3) Perform affinity evaluation on each feasible solution in the population.

[0133] (4) Determine whether the algorithm termination condition is met; if the condition is met, terminate the algorithm optimization process and output the calculation result; otherwise, continue the optimization operation.

[0134] (5) Calculate antibody concentration and stimulation level.

[0135] (6) Perform immune processing, including immune selection, cloning, mutation and clonal suppression.

[0136] Immune selection: Select high-quality antibodies based on the affinity and concentration calculation results of antibodies in the population and activate them;

[0137] Cloning: clone and replicate the activated antibody to obtain several copies;

[0138] Mutation: Perform mutation operations on the cloned copies to cause affinity mutation;

[0139] Clone suppression: Reselect the compiled results to suppress antibodies with low affinity and retain variants with high affinity.

[0140] (7) Refreshing the population, replacing the antibodies with lower excitation levels in the population with randomly generated new antibodies to form a new generation of antibodies, and then go to step (3).

[0141] S6. Based on the global optimal variables obtained by the immune heuristic multi-objective algorithm, the automatic disturbance rejection control algorithm is used for real-time regulation;

[0142] In this embodiment, the specific implementation of step S6 is as follows Figure 2 As shown:

[0143] According to the immune-heuristic multi-objective algorithm, the optimal solution for a set of controllable variables is determined, namely, the injection angle, injection position, injection frequency, injection pressure, and number of fuel supply paths of the corresponding fuel injection under the optimal conditions of total pressure loss, thrust and combustion efficiency; an anti-disturbance controller is built to perform real-time adaptive control of the three design variables of injection angle, injection frequency and injection pressure to achieve adaptive injection with optimal combustion chamber performance.

[0144] S7. Use the thrust generated by combustion and the fusion results of the inlet data to decide the mode conversion.

[0145] In this embodiment, the specific implementation of step S7 is as follows Figure 2 As shown:

[0146] The decision mode conversion is realized by the mode conversion control system, which includes four parts: engine cycle and working state selection, decision-making mechanism, different mode states, and thrust and inlet data fusion:

[0147] Engine cycle and operating state selection is obtained through the fusion of turboramjet thrust and inlet data; the obtained information is sent to the decision-making body;

[0148] The decision-making body will judge the current state and select the appropriate mode to work; according to the modal information and different working conditions, the deep belief network and immune-heuristic multi-objective optimization algorithm are used to find the fuel injection parameters that can keep the engine combustion chamber at the optimal performance under the current mode and working conditions; further observe the thrust generated by the combined engine, that is, form a closed loop for the entire control system to achieve the effect of real-time observation and adjustment; different modal states are used for different flight speeds. For example, when flying at low Mach numbers, turbine engines are mainly used to provide power; when flying at high Mach numbers, ramjet engines are used to provide power. Inlet data fusion is used to detect the incoming flow conditions (mainly including Mach number) in real time, and then decide on the engine's mode selection. Among them, the fuel injection parameters that keep the engine combustion chamber at its best performance include injection angle, injection frequency and injection pressure. The three design variables of injection angle, injection frequency and injection pressure obtained by optimization are used as the expectations of the ADRC control system, and the ADRC control system is used to filter out interference and stabilize the injection parameters, so that the engine combustion chamber can always maintain the best performance under different modes.

[0149] Example 3

[0150] This embodiment provides a turboramjet combined engine fuel injection control system based on active disturbance rejection control, which is applied to the above-mentioned intelligent control method, including:

[0151] The ADRC is used to control the fuel injector variables that are susceptible to interference and unstable during flight, that is, to complete the real-time control of the injection angle, injection frequency and injection pressure. The motor that controls the injection is susceptible to interference during flight, so the variables directly controlled by the motor are susceptible to interference and unstable variables.

[0152] The active disturbance rejection control system ADRC consists of three parts: tracking differentiator, extended state observer and nonlinear error feedback control; the tracking differentiator is used to track and differentiate the input signal of the system to obtain a stable system input signal; the extended state observer is used to estimate the real-time state of fuel injection and the total disturbance of the system; the nonlinear error feedback control is used to compensate the control rate according to the stable system input signal, the estimated real-time state and the total disturbance of the system, and generate the final control quantity of the fuel injector.

[0153] Taking the fuel injection pressure v0 as an example, the fuel injection frequency and angle are also controlled using the same method and steps. The input of the ADRC system is the output v0 under the optimization system, and the output of the ADRC system is the final control amount U of the fuel injector injection pressure.

[0154] The input of the differential controller is the desired injection pressure v0, and the output is the first-order tracking differential signal v1 and the second-order tracking differential signal v2 of the input signal v0. The differential controller is:

[0155]

[0156]

[0157] Where r, v0 are parameters to be adjusted; h is the operation step; k is the number of sampling moments; v1 is the tracking input signal of the input signal v0 at time k, v2(k) is the first-order differential signal of v1(k); sign(.) is the sign function; fhan(.) is the fastest control comprehensive function; a0 is the boundary layer thickness, y, s y and a All are intermediate parameters.

[0158] The inputs of the extended state observer are: the actual injection pressure y of the fuel pulse injector system, the product of the control quantity U and the coefficient b0. The outputs are the observed injection pressure z1, the observed injection pressure change rate z2 and the observed total disturbance z3. Among them, the observed injection pressure can be regarded as the actual injection pressure; the observed injection pressure change rate can be regarded as the actual injection pressure change rate; the observed total disturbance is the total disturbance inside and outside the system. After dividing it by b0, it is subtracted from the state error feedback control law output u0 to obtain the final control quantity U of the fuel injection pressure given to the system:

[0159] U(k)=u0(k)-z3(k) / b0

[0160] Where k is the number of sampling moments; U(k) is the final control output of the ADRC subsystem at moment k; z3(k) is the total disturbance observed at moment k; and b0 is the compensation factor that determines the strength of compensation.

[0161] The extended state observer is:

[0162]

[0163] Where k is the number of sampling moments; z1(k) is the measured injection pressure at moment k; z2(k) is the observed injection pressure change rate at moment k; z3(k) is the total disturbance estimate of the internal and external disturbances of the injector at moment k; a1, a2, a3, β 01 ,β 02 and β 03 are adjustment parameters; δ is the length of the linear segment.

[0164] The nonlinear error feedback input is the injection pressure error e1 (tracking injection pressure v1-observing injection pressure z1) and the change rate error e2 (tracking injection pressure change rate v2-observing injection pressure change rate z2). The output control quantity U0 nonlinear error feedback is processed as follows:

[0165]

[0166] Where u0 is the error feedback control quantity; β1 and β2 are both adjustable parameters; e1 is the system injection pressure error, and e2 is the system injection pressure change rate error.

[0167] The turbine and ramjet engines are distributed side by side on the upper and lower parts of the aircraft body, with their own independent combustion chambers and nozzle convergence sections. The nozzle expansion section and the external compression part of the inlet duct are shared, and the internal compression part can be shared or independent depending on the design. The conversion and operation of the turbine mode, ramjet mode and common working mode are achieved by opening or closing the inlet adjustment ramp.

[0168] In turbine mode, the turbine channel is opened, and the airflow passes through the turbine engine to do work. At this time, the ramjet channel can be opened or closed. When the ramjet channel is in the open state, it can be used as a venting channel for excess air to reduce flight resistance. If necessary, a small amount of fuel can be sprayed into the ramjet combustion chamber (which can be regarded as a turbine ramjet dual mode) to generate appropriate thrust. In ramjet mode, the inlet regulating ramp is closed to prevent high-temperature air from entering the turbine engine channel and burning the compression components. At the same time, the regulating ramp plays the role of a compression ramp, and the airflow passes through the ramjet combustion chamber and the tail nozzle with an adjustable throat, and the ramjet engine does work.

[0169] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by a person of ordinary skill in the art without departing from the spirit and technical ideas disclosed by the present invention shall still be covered by the claims of the present invention.

Claims

1. A turboramjet engine mode conversion intelligent control method, characterized in that The steps include: S1. Analyze the influence of the main controllable parameters of fuel injection on the combustion performance of the turboramjet combined engine under different modes and working conditions; Analyze the influence of the main controllable parameters of fuel injection on the combustion performance of the turboramjet combined engine under different modes; the main controllable parameters include the injection angle, injection position, injection frequency, injection pressure, and number of fuel supply paths of the fuel injection; the combustion performance includes the total pressure loss of the combustion chamber, thrust and combustion efficiency, and further obtain the mathematical relationship between the main controllable parameters of fuel injection and the combustion performance; S2. Under different modes and working conditions of the turboramjet engine, the Latin hypercube sampling method is used to obtain sample points in the design space. Through ground wind tunnel tests, the total pressure loss, thrust and combustion efficiency are obtained, and the computational fluid dynamics (CFD) numerical simulation results are verified and corrected. Then, the data sample size is expanded through computational fluid dynamics (CFD) numerical simulation to construct a data set with one-to-one correspondence between design variables and combustion performance. S3, preprocessing the data set; S4. Build a deep belief network. The deep belief network DBN connects several RBMs in series to form a DBN, in which the hidden layer of the previous RBM is the visible layer of the next RBM, and the output of the previous RBM is the input of the next RBM; the input of the first layer is the design variables of the fuel injector, and the output is the combustion performance, that is, the total pressure loss, thrust and combustion efficiency of the combustion chamber; The deep belief network is a probabilistic generative model. Compared with the traditional discriminative neural network, the generative model is to establish a joint distribution between observed data and labels: P(v,h 1 ,h 2 ,…,h N )=P(v|h 1 )P(h 1 |h 2 )…P(h N-2 |h N-1 )P(h N-1 |h N ) RBM is a neural sensor consisting of a visible layer and a hidden layer. The neurons in the visible layer and the hidden layer are fully connected in both directions. In an RBM, the hidden layer neurons h j Probability of being activated: P(h j |v)=σ(b j +S i W i,j v i ) Since RBM is bidirectionally connected, the probability that the visible layer neurons can be activated by the hidden layer neurons is: P(v j |h)=σ(c i +S i W i,j h j ) Where P(·) represents probability, N is the number of layers; RBM has five parameters: h, v, c, b, W, where c, b, W are the corresponding weights and bias values, which can be obtained through learning; v is the input vector; h is the output vector; σ is the Sigmoid function; Neurons in the same layer are independent, so the probability density also satisfies independence: When a piece of data is assigned to the display layer, RBM calculates the probability of each hidden layer neuron being turned on according to the probability formula of hidden layer neuron activation, takes a random number between 0 and 1 as the threshold, and neurons greater than the threshold are activated, otherwise they are not activated, that is: h j =1,P(h j |x)≥μ;h j =0,P(h j |x)<μ The update method of relevant weights in the deep belief network is: Among them, λ is the coefficient; μ represents the threshold, which is a random number between 0 and 1; N h Indicates the number of hidden layers; N v Indicates the number of input quantities; The last layer of the deep belief network can be replaced with a BP neural network, i.e. supervised learning. The input of the deep belief network is the important operating parameters of the turboramjet engine, i.e. the injection angle, injection position, injection frequency, injection pressure, and fuel supply path number design variables of the fuel injection. The output is the total pressure loss, thrust, and combustion efficiency of the combustion chamber. A proxy model based on a deep belief network was built. Combined with the preprocessed data set, a mathematical model was constructed under different modes and working conditions with five design variables of the fuel injector, namely the injection angle, injection position, injection frequency, injection pressure, and number of fuel supply paths, and the total pressure loss, thrust, and combustion efficiency of the combustion chamber as outputs to form a proxy model for combustion chamber performance prediction. S5. Use immune-inspired multi-objective optimization algorithm combined with deep belief network agent model to find the solution of design variables with optimal combustion performance; S6. Based on the global optimal variables obtained by the immune heuristic multi-objective algorithm, the anti-disturbance control algorithm is used to perform real-time anti-disturbance control; S7. Utilize the thrust generated by combustion and the fusion result of the inlet data to make a decision on mode conversion. The mode conversion decision is realized by the mode conversion control system. The mode conversion control system includes four parts: engine cycle and working state selection, decision-making mechanism, different mode states, and thrust and inlet data fusion: Engine cycle and operating state selection is obtained through the fusion of turboramjet thrust and inlet data; the obtained information is sent to the decision-making body; The decision-making body will judge the current state and select the appropriate mode to work; according to the modal information and different working conditions, the deep belief network and immune heuristic multi-objective optimization algorithm are used to find the fuel injection parameters that can keep the engine combustion chamber at the optimal performance under the current mode and working conditions; further observe the thrust generated by the combined engine, that is, form a closed loop for the entire control system, and achieve the effect of real-time observation and adjustment; Different modal states are used for different flight speeds; Inlet data fusion is used to detect the incoming flow conditions in real time, mainly including the Mach number, and then decide the engine mode selection; Among them, the fuel injection parameters that enable the engine combustion chamber to maintain optimal performance include injection angle, injection frequency and injection pressure. The three design variables of injection angle, injection frequency and injection pressure obtained by optimization are used as the expectations of the ADRC control system, and the ADRC control system is used to filter out interference and stabilize the injection parameters, so that the engine combustion chamber can always maintain optimal performance under different modes.

2. The method for intelligent control of mode conversion of a turboramjet engine according to claim 1, characterized in that Step S2 is specifically as follows: The five design variables of fuel injection, namely, injection angle, injection position, injection frequency, injection pressure, and number of fuel supply paths, are parameterized. The upper and lower bounds of each design variable are determined under multiple operating parameters of different modes, different Mach numbers, and different incoming flow conditions, and the Latin hypercube sampling method is used to obtain sample points in the design space.

3. The method for intelligent control of mode conversion of a turboramjet engine according to claim 1, characterized in that In step S2, the sample points of the design space are obtained by combining the Latin hypercube sampling method as follows: (1) First, determine the number of samples M that need to be obtained; (2) Divide the intervals of the five design variables into M segments; (3) Randomly select a value from each of the M segments; (4) Randomly combine the values ​​drawn from different design variables; (5) Obtain M groups of samples after sampling.

4. The method for intelligent control of mode conversion of a turboramjet engine according to claim 1, characterized in that Step S3 is specifically as follows: Analyze and preprocess the data set, screen and eliminate completely worthless data to ensure the quality of the basic data set. Completely worthless data includes abnormal data, redundant data, and data outside the range of design variables.

5. The method for intelligent control of mode conversion of a turboramjet engine according to claim 1, characterized in that Step S5 is specifically as follows: In different modes and working conditions of the turboramjet engine, the agent model established by the deep belief network is used as the fitness function, and the immune-heuristic multi-objective optimization algorithm is used to find the injection angle, injection position, injection frequency, injection pressure, and number of fuel supply paths of the fuel injection that optimize the total pressure loss, thrust and combustion efficiency, that is, to find the optimal solution of the five design variables.

6. The method for intelligent control of mode conversion of a turboramjet engine according to claim 1, characterized in that In step S5, the process of finding the optimal solution of the fuel injection design variables using the immune heuristic multi-objective optimization algorithm includes: (1) Antigen identification, that is, understanding the problem to be optimized, conducting feasibility analysis on the problem, extracting prior knowledge, constructing a mathematical model that conforms to the input-output relationship, and formulating various constraints; (2) Initial antibody population: The feasible solutions to the problem are represented as antibodies in the solution space by encoding, and an initial population is randomly generated in the solution space; (3) Evaluate the affinity of each feasible solution in the population; (4) Determine whether the algorithm termination condition is met; if the condition is met, terminate the algorithm optimization process and output the calculation result; otherwise, continue the optimization operation; (5) Calculate antibody concentration and stimulation degree; (6) Perform immune processing, including immune selection, cloning, mutation and clonal suppression; Immune selection means: selecting high-quality antibodies based on the affinity and concentration calculation results of antibodies in the population and activating them; Cloning: cloning and replicating the activated antibody to obtain several copies; Mutation: that is, performing mutation operations on the cloned copies to cause affinity mutations; Clone suppression: reselect the compiled results to suppress antibodies with low affinity and retain variants with high affinity; (7) Refreshing the population, replacing the antibodies with lower excitation levels in the population with randomly generated new antibodies to form a new generation of antibodies, and then go to step (3).

7. The method for intelligent control of mode conversion of a turboramjet engine according to claim 1, characterized in that Step S6 is specifically as follows: According to the immune-heuristic multi-objective algorithm, the optimal solution for a set of controllable variables is determined, namely, the injection angle, injection position, injection frequency, injection pressure, and number of fuel supply paths of the corresponding fuel injection under the optimal conditions of total pressure loss, thrust and combustion efficiency; an anti-disturbance controller is built to perform real-time adaptive control of the three design variables of injection angle, injection frequency and injection pressure to achieve adaptive injection with optimal combustion chamber performance.

8. A turboramjet combined engine fuel injection control system based on active disturbance rejection control, applied to the intelligent control method according to any one of claims 1 to 7, characterized in that include: The ADRC is used to control the fuel injector variables that are susceptible to interference and unstable during flight, that is, to complete the real-time control of the injection angle, injection frequency and injection pressure. The motor that controls the injection is susceptible to interference during flight, so the variables directly controlled by the motor are susceptible to interference and unstable variables. The ADRC system consists of three parts: tracking differentiator, extended state observer and nonlinear error feedback control. The tracking differentiator is used to track and differentiate the input signal of the system to obtain a stable system input signal. The extended state observer is used to estimate the real-time state of fuel injection and the total disturbance of the system; The nonlinear error feedback control is used to compensate the control rate according to the system stable system input signal, the estimated real-time state and the total disturbance, and generate the final control amount of the fuel injector.

Images