CN113094804A - Method for predicting specific impulse performance of solid rocket engine with incremental learning capability - Google Patents
Method for predicting specific impulse performance of solid rocket engine with incremental learning capability Download PDFInfo
- Publication number
- CN113094804A CN113094804A CN202010015966.8A CN202010015966A CN113094804A CN 113094804 A CN113094804 A CN 113094804A CN 202010015966 A CN202010015966 A CN 202010015966A CN 113094804 A CN113094804 A CN 113094804A
- Authority
- CN
- China
- Prior art keywords
- rocket engine
- neural network
- solid rocket
- specific impulse
- impulse performance
- Prior art date
- Legal status (The legal status 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 status listed.)
- Pending
Links
- 239000007787 solid Substances 0.000 title claims abstract description 59
- 238000000034 method Methods 0.000 title claims abstract description 25
- 238000013528 artificial neural network Methods 0.000 claims abstract description 63
- 238000012549 training Methods 0.000 claims abstract description 28
- 238000013461 design Methods 0.000 claims abstract description 13
- 239000003380 propellant Substances 0.000 claims abstract description 5
- 239000013598 vector Substances 0.000 claims description 38
- 238000010606 normalization Methods 0.000 claims description 6
- 238000004422 calculation algorithm Methods 0.000 claims description 5
- 230000000295 complement effect Effects 0.000 claims description 5
- 238000013507 mapping Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 abstract description 8
- 230000008569 process Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 235000015842 Hesperis Nutrition 0.000 description 1
- 235000012633 Iberis amara Nutrition 0.000 description 1
- 238000000342 Monte Carlo simulation Methods 0.000 description 1
- 241000392831 Pharyngostrongylus lambda Species 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000002860 competitive effect Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method for predicting the specific impulse performance of a solid rocket engine with incremental learning capability, and relates to a method for rapidly predicting the specific impulse performance of the solid rocket engine according to characteristic parameters of the initial design of the solid rocket engine. Firstly, selecting a solid rocket engine successfully tested on the ground as a typical sample, extracting characteristic parameters and specific impulse performance data designed by the solid rocket engine, inputting the data into a Fuzzy ART-BP neural network for training and learning, and finishing the training and learning after a preset error is reached, wherein the characteristic parameters comprise: propellant density, working time, average working pressure, average burning speed, nozzle throat diameter, average expansion ratio and expansion half angle. The characteristic parameters of the newly designed solid rocket engine are input into a Fuzzy ART-BP neural network, and the specific impulse performance of the solid rocket engine can be rapidly calculated. The method can save complicated charge design, internal trajectory calculation and the like of the solid rocket engine, quickly and simply calculate the specific impulse performance of the solid rocket engine, obviously improve the capability of a system engineer in quickly iterating the overall scheme of the solid rocket, and can be used for quickly predicting the specific impulse performance of the solid rocket engine.
Description
Technical Field
The invention relates to the field of rocket engines, in particular to a method for predicting specific impulse performance of a solid rocket engine.
Background
The solid rocket engine has the advantages of simple structure, no maintenance, low guarantee requirement and the like, is suitable for boosting power sources of solid carrier rockets and missiles, and has wide application scenes;
the specific impulse performance of the solid rocket engine is an important technical index for reflecting the energy of the propellant used by the engine and the perfection degree of the internal working process of the engine. At present, in the initial stage of solid rocket engine design, the specific impulse performance prediction mainly comprises the following steps: 1. correcting the theoretical specific impulse of the newly designed solid rocket engine by using the correction coefficient obtained from the previous test; 2. the specific impulse performance is estimated by calculating various losses. The methods have the defects of large sample data and poor prediction accuracy, and the mutual influence among design characteristic parameters is not considered, so that the expected specific impulse design effect of the solid rocket engine cannot be achieved, and the overall scheme of the solid carrier rocket/missile is repeated or not closed;
therefore, for the general designers of the solid carrier rocket/missile, complicated solid rocket engine charging design, internal trajectory calculation and empirical parameters need to be omitted without depending on the support of a solid power professional post, the specific impulse performance of the solid rocket engine is quickly calculated by a simple, effective and accurate method, the quick iteration of the general scheme of the solid rocket/missile is supported, and the efficiency of the demonstration of the scheme of the solid rocket/missile can be obviously improved.
Disclosure of Invention
The invention aims to provide a method for predicting the specific impulse performance of a solid rocket engine, which can effectively shorten the development time of the solid rocket engine and improve the performance prediction precision;
a method for predicting the specific impulse performance of a solid rocket engine with incremental learning capability is characterized by comprising the following steps:
(1) selecting a solid rocket engine successfully tested on the ground as a typical training sample, and extracting designed characteristic parameters of the solid rocket engine, wherein the characteristic parameters comprise: propellant density, working time, average working pressure, average burning speed, nozzle throat diameter, average expansion ratio and expansion half angle;
(2) normalizing the obtained design parameters of the solid rocket engine to be used as an input layer input vector of a Fuzzy ART-BP neural network, normalizing the corresponding specific impulse performance to be used as an output layer output vector of the Fuzzy ART-BP neural network, and setting the number of hidden layer nodes to be 8-12; classifying the training samples by using Fuzzy ART, then respectively training and learning the classified training samples by using a BP neural network, and when the convergence error of the BP neural network is less than 10e-5, the BP neural network is finally converged, the training and learning of the training samples are finished, and the classification number, the weight and the threshold of the learning samples are automatically recorded;
(3) the characteristic parameters of the newly designed solid rocket engine are led into a Fuzzy ART-BP neural network, the Fuzzy ART neural network firstly classifies the characteristic parameters by utilizing the stored classification weight, then nonlinear mapping prediction calculation is carried out by utilizing the classified characteristic parameters stored by the BP neural network, and the specific impulse performance data of the newly designed solid rocket engine can be rapidly calculated for the total rapid iteration solid carrying general scheme of the rocket (rocket) total;
compared with the prior art, the method for predicting the specific impulse performance of the solid rocket engine has the following beneficial effects:
the neural network algorithm can perform self-adaptive incremental learning on the specific impulse performance of the untrained and learned solid rocket engine, realize off-line learning capability, enhance the generalization of the neural network and simultaneously enable the prediction result of the specific impulse performance to be more practical; the specific impulse performance of the solid rocket engine can be rapidly calculated, and a total rapid iteration scheme of the solid rocket/missile is provided.
Drawings
FIG. 1 is a flow chart of fast calculation of specific impulse performance of a solid rocket engine;
FIG. 2 is a diagram of the learning and training process of a Fuzzy ART-BP neural network; .
Detailed Description
The invention relates to a method for predicting the specific impulse performance of a solid rocket engine with incremental learning capability, which can be used for rapidly calculating the specific impulse performance of the solid rocket engine, and a rapid calculation flow chart is shown in figure 1. FIG. 2 is a diagram showing the learning and training process of Fuzzy ART-BP neural network;
the method is characterized in that an artificial neural network is used for predicting the specific impulse performance of the solid rocket engine, the problems that the BP neural network does not have the incremental learning capability are solved to a certain extent, the actual application has more samples and larger data difference, incremental learning of input data cannot be realized by adopting one neural network, the convergence is poor, and the prediction precision is low, therefore, according to the combination of an input vector and a certain rule, the whole sample is divided into a plurality of independent sub-sample sets, a multiple sub BP neural network, namely a Fuzzy ART-BP mixed neural network, is established, the adaptive classification function of the Fuzzy ART neural network is fully utilized, and the BP neural network is utilized to carry out nonlinear mapping on the samples classified by the Fuzzy ART. The method is introduced into the prediction of the specific impulse performance of the solid rocket engine, the specific impulse performance of the engine can be predicted more accurately, and the method is suitable for the prediction of the specific impulse performance of various solid rocket engines;
the Fuzzy ART-BP neural network algorithm is trained in a guiding learning mode, and the main idea is as follows: inputting a learned sample vector, firstly carrying out self-adaptive competitive clustering through a Fuzzy ART neural network, modifying the weight of the class to which the input vector belongs, then repeatedly adjusting and training the weight and the threshold of the hybrid neural network by using a back propagation algorithm BP according to the class to which the input vector belongs, enabling the actually output vector to be as close to the expected output vector as possible, finishing training when the sum of squares of errors of a network output layer is smaller than a specified error, storing the weight and the threshold of the network at the moment, and finishing training and learning of the Fuzzy ART-BP neural network;
the specific steps are as follows:
(1) input parameter normalization
Normalizing the characteristic parameters of the solid rocket engine design to obtain an input vector P = (a)1,a2,…,an）TDesired output vector T =(s)1,s2, …,sq) T；
(2) Input vector complement
For input vector P = (a)1,a2,…,an）TPerforming complement operation, specifically:
I=(a,ac)=(a1,…an,…,ac n)
wherein, ai c=1-ai ，1≤i≤n；
(3) Fuzzy ART neural network initialization
Bottom-up weight vector W to class layer nodes of Fuzzy ART neural networkjiAnd the top-down weight vector WijInitializing, namely giving a random number within (-1, 1);
(4) calculating input vector matching degree
The Fuzzy ART neural network carries out the input vector I after the complement and the weight vector W of each kind unit storedijGoodness of fit, Tj(I) The vector to which class the input vector I belongs is determined. Aiming at any node j of the category, a selection function of the Fuzzy ART neural network on the input vector is as shown in the formula;
Tj(I)=I.∧Wj/t+|Wj|
in which the blurring and operation are represented, i.e. (P. Lambda. Q)i=min(P i,Q i) Is L |1Norm t is a selection parameter t>0，0≤j≤m。
The Fuzzy ART neural network will select the node J of the category on the principle of "winner is king" so that the winner will have the maximum activation value TjAs shown in formula (I);
Tj=max(Tj:j=1,2,…,m)
m is the number of nodes of the category;
calculating to obtain a maximum activation value through the above formula, and calculating the matching degree k of the input vector, wherein the calculation is shown as the following formula;
k=|I∧Wj|/|I|
comparing the input vector matching degree k obtained by the calculation with an alert threshold rho of a Fuzzy ART neural network, if k is more than or equal to rho at the moment, the Fuzzy ART neural network is in a resonance state, giving the input vector I to a node j of the category, and adjusting the corresponding Fuzzy ART neural network weight, wherein the weight adjustment is carried out according to the following formula;
Wj new=η(I∧Wj old)+(1-η) Wj old
(5) establishing BP neural network of belonged class J number
Establishing BP neural network of corresponding category for category number J to which input vector I belongs, and initializing BP neural network of category J, namely giving each connection weight Wi1j1And Vti1Threshold value thetai1And YtGiving a random number in (-1, 1), and finally setting a convergence error epsilon of the BP neural network algorithm;
(6) class J-numbered BP neural network training learning
Inputting vector P = (a)1,a2,…,an）TDesired output vector T =(s)1,s2, …,sq) TThe samples of (1) are input into a BP neural network of the class J, and an actual output vector T = (T) is calculatedr1,t r2, …,trq) TCalculating an error E between the expected output vector and the actual expected output vector, wherein if the error E is smaller than a set convergence error epsilon as shown in the formula, the class J number BP neural network training and learning are finished, the iteration is finished, and the weight and the threshold are saved; otherwise, after modifying the weight and the threshold, continuously judging whether the error E is smaller than the set convergence error;
selecting the next training sample to input into a Fuzzy ART-BP neural network, and continuing the steps (1) - (6) until all the training samples are trained;
the above process can be represented by the learning training process diagram of figure 2 Fuzzy ART-BP neural network;
specific examples are as follows: firstly, extracting design characteristic parameters and vacuum specific impulse of a solid rocket engine which is successfully tried, wherein the characteristic parameters comprise: density of propellant (1795 kg/m)3) Working time(70 s), average working pressure (7.2 MPa), average burning speed (7.1 mm/s), nozzle throat diameter (phi 154 mm), average expansion ratio (41) and expansion half angle (15 degrees); outputting parameters: vacuum pulse 285s, because the data units collected are not consistent, in order to speed up the convergence of the training network, the data must be processed to [0, 1 ]]Normalization processing, wherein the normalization processing on the characteristic parameters respectively comprises the following steps: 0.1795, 0.7, 0.72, 0.71, 0.154, 0.41, 0.15; the output parameters are normalized as follows: 0.285. the learning rate of the Fuzzy ART neural network is 0.18, and the warning threshold value is 0.94; the learning rate of the BP neural network is taken as 0.44, the momentum factor is taken as 0.91, and the expected error is taken as 10 e-3. Because the accumulated data is limited, Monte Carlo simulation is adopted to simulate more data and provide the data for a Fuzzy ART-BP neural network to train and learn;
inputting characteristic parameters after normalization of a newly designed solid rocket engine: 0.1795, 0.71, 0.7, 0.156, 0.42, 0.15, until a Fuzzy ART-BP neural network is used for carrying out vacuum specific impulse prediction, wherein the prediction value of the vacuum specific impulse is 286.6s, the solid rocket engine designed according to the parameters is subjected to ground hot test run, the back calculation vacuum specific impulse is 285.6s, and the prediction precision reaches 0.35%;
because the solid rocket engine belongs to the customized design, the engine data of similar parameters is less, the change of design parameters is larger, and if the traditional BP neural network is used for predicting the specific impulse performance, the prediction precision can be reduced along with the increase of training samples;
compared with a BP neural network, the Fuzzy ART-BP neural network can perform self-adaptive classification learning on input data, and then perform BP neural network training learning on the classified data, so that the network is more targeted on the training data, the degree of destroying a memorized mode after new data learning is reduced as much as possible, the network prediction accuracy is improved, and the network has the ability of incremental learning.
Claims (5)
1. A method for predicting the specific impulse performance of a solid rocket engine with incremental learning capability comprises the following steps:
(1) selecting a solid rocket engine successfully tested on the ground as a typical sample, and extracting designed characteristic parameters of the solid rocket engine;
(2) taking the obtained characteristic parameters of the solid rocket engine as input vectors of a Fuzzy ART-BP neural network, taking corresponding specific impulse performance as output vectors of the Fuzzy ART-BP neural network, training and learning the neural network, finishing the training and learning of the neural network after the network reaches a preset error, and automatically recording the class number, weight and threshold of a learning sample;
(3) and (3) introducing the characteristic parameters of the newly designed solid rocket engine into a Fuzzy ART-BP neural network, and calculating the specific impulse performance data of the newly designed solid rocket engine.
2. A method of predicting the specific impulse performance of a solid-rocket engine with incremental learning capabilities as recited in claim 1; the characteristic parameters of the design include: propellant density, working time, average working pressure, average burning speed, nozzle throat diameter, average expansion ratio and expansion half angle.
3. The method of predicting specific impulse performance of a solid-rocket engine with incremental learning capabilities of claim 1: and inputting the characteristic parameters into the Fuzzy ART-BP neural network for normalization data processing.
4. The method of predicting specific impulse performance of a solid-rocket engine with incremental learning capabilities of claim 1: and aiming at the solid rocket engine design characteristic parameters successful in ground trial and the corresponding specific impulse performance, taking the characteristic parameters and the corresponding specific impulse performance as input and output vectors of the Fuzzy ART-BP neural network, and performing nonlinear mapping learning and incremental learning.
5. The method of predicting specific impulse performance of a solid-rocket engine with incremental learning capabilities of claim 1: when the Fuzzy ART-BP neural network algorithm is trained, a tutor learning mode is adopted, and the specific mode is as follows:
(1) input parameter normalization
Normalizing the characteristic parameters designed by the solid rocket engine to obtain input vectors;
(2) input vector complement
Performing complement operation on the input vector;
(3) fuzzy ART neural network initialization
Initializing a bottom-up weight vector and a top-down weight vector of a classification layer node of a Fuzzy ART neural network, namely endowing a (-1, 1) internal random number;
(4) calculating the matching degree of the input vector;
(5) establishing a BP neural network with a category J number;
(6) class K numbered BP neural network training learning
And (4) selecting the next training sample and inputting the next training sample into a Fuzzy ART-BP neural network, and continuing the steps (1) to (6) until all the training samples are trained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010015966.8A CN113094804A (en) | 2020-01-08 | 2020-01-08 | Method for predicting specific impulse performance of solid rocket engine with incremental learning capability |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010015966.8A CN113094804A (en) | 2020-01-08 | 2020-01-08 | Method for predicting specific impulse performance of solid rocket engine with incremental learning capability |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113094804A true CN113094804A (en) | 2021-07-09 |
Family
ID=76663211
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010015966.8A Pending CN113094804A (en) | 2020-01-08 | 2020-01-08 | Method for predicting specific impulse performance of solid rocket engine with incremental learning capability |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113094804A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7411543B1 (en) * | 2004-08-13 | 2008-08-12 | Lockheed Martin Corporation | Maximum-likelihood rocket identifier |
| CN102779287A (en) * | 2012-05-24 | 2012-11-14 | 北京工业大学 | Ink key opening forecasting method having increment type learning capacity |
| WO2015172560A1 (en) * | 2014-05-16 | 2015-11-19 | 华南理工大学 | Central air conditioner cooling load prediction method based on bp neural network |
| CN106930865A (en) * | 2017-02-24 | 2017-07-07 | 湖北航天技术研究院总体设计所 | The high-energy solid rocket engine that a kind of temperature wide is used |
-
2020
- 2020-01-08 CN CN202010015966.8A patent/CN113094804A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7411543B1 (en) * | 2004-08-13 | 2008-08-12 | Lockheed Martin Corporation | Maximum-likelihood rocket identifier |
| CN102779287A (en) * | 2012-05-24 | 2012-11-14 | 北京工业大学 | Ink key opening forecasting method having increment type learning capacity |
| WO2015172560A1 (en) * | 2014-05-16 | 2015-11-19 | 华南理工大学 | Central air conditioner cooling load prediction method based on bp neural network |
| CN106930865A (en) * | 2017-02-24 | 2017-07-07 | 湖北航天技术研究院总体设计所 | The high-energy solid rocket engine that a kind of temperature wide is used |
Non-Patent Citations (3)
| Title |
|---|
| 张宇星等: "基于神经网络的固体火箭发动机比冲性能的预示研究", 《工程力学》 * |
| 徐玲玲等: "一种改进的模糊ART神经网络学习算法", 《计算机工程与应用》 * |
| 王晔等: "基于Matlab和BP神经网络的固体火箭发动机比冲性能的预测", 《内蒙古科技与经济》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109767759B (en) | Method for establishing CLDNN structure applied to end-to-end speech recognition | |
| Islam et al. | A constructive algorithm for training cooperative neural network ensembles | |
| CN111047085B (en) | Hybrid vehicle working condition prediction method based on meta-learning | |
| US20220108185A1 (en) | Inverse and forward modeling machine learning-based generative design | |
| CN110837223A (en) | Combustion optimization control method and system for gas turbine | |
| CN104616072B (en) | A kind of method of the raising glutamic acid fermentation production concentration based on range optimization | |
| CN109284662B (en) | Underwater sound signal classification method based on transfer learning | |
| CN112149883A (en) | Photovoltaic power prediction method based on FWA-BP neural network | |
| CN115689070B (en) | Energy prediction method for optimizing BP neural network model based on monarch butterfly algorithm | |
| CN103559537A (en) | Template matching method based on error back propagation in out-of-order data streams | |
| CN116401756A (en) | Solid rocket engine performance prediction method, prediction system, storage medium and equipment based on deep learning and data enhancement | |
| Wang et al. | Framework of nacelle inverse design method based on improved generative adversarial networks | |
| CN115422995A (en) | Intrusion detection method for improving social network and neural network | |
| CN114880806A (en) | New energy automobile sales prediction model parameter optimization method based on particle swarm optimization | |
| CN107016239B (en) | steam turbine valve flow characteristic analysis method | |
| CN112132469A (en) | Reservoir group scheduling method and system based on multi-population cooperative particle swarm algorithm | |
| CN111832911A (en) | Underwater combat effectiveness evaluation method based on neural network algorithm | |
| CN103149878A (en) | Self-adaptive learning system of numerical control machine fault diagnosis system in multi-agent structure | |
| CN109886405A (en) | It is a kind of inhibit noise based on artificial neural network structure's optimization method | |
| CN113821025A (en) | Mobile robot path planning method for optimizing heuristic function through neural network | |
| CN113094804A (en) | Method for predicting specific impulse performance of solid rocket engine with incremental learning capability | |
| CN116070498A (en) | Handheld rapid damage assessment device | |
| CN116681154A (en) | Photovoltaic power calculation method based on EMD-AO-DELM | |
| Lu et al. | Apso-based optimization algorithm of lstm neural network model | |
| CN113222256A (en) | Stock prediction method and system based on MPSO-LSTM model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20210709 |