CN116401756A - Solid rocket engine performance prediction method, prediction system, storage medium and equipment based on deep learning and data enhancement - Google Patents
Solid rocket engine performance prediction method, prediction system, storage medium and equipment based on deep learning and data enhancement Download PDFInfo
- Publication number
- CN116401756A CN116401756A CN202310176428.0A CN202310176428A CN116401756A CN 116401756 A CN116401756 A CN 116401756A CN 202310176428 A CN202310176428 A CN 202310176428A CN 116401756 A CN116401756 A CN 116401756A
- Authority
- CN
- China
- Prior art keywords
- rocket engine
- thrust
- solid rocket
- data set
- training
- 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
- 238000000034 method Methods 0.000 title claims abstract description 142
- 239000007787 solid Substances 0.000 title claims abstract description 103
- 238000013135 deep learning Methods 0.000 title claims abstract description 23
- 238000003860 storage Methods 0.000 title claims abstract description 11
- 238000012549 training Methods 0.000 claims abstract description 78
- 230000008569 process Effects 0.000 claims abstract description 72
- 238000012360 testing method Methods 0.000 claims abstract description 44
- 238000005259 measurement Methods 0.000 claims abstract description 37
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 28
- 238000007781 pre-processing Methods 0.000 claims abstract description 14
- 230000006870 function Effects 0.000 claims description 16
- 210000002569 neuron Anatomy 0.000 claims description 9
- 230000004913 activation Effects 0.000 claims description 7
- 238000004422 calculation algorithm Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 6
- 239000003380 propellant Substances 0.000 claims description 6
- 230000003044 adaptive effect Effects 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 description 6
- 238000013461 design Methods 0.000 description 6
- 238000012986 modification Methods 0.000 description 4
- 230000004048 modification Effects 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000004075 alteration Effects 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000012417 linear regression Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 235000015842 Hesperis Nutrition 0.000 description 1
- 235000012633 Iberis amara Nutrition 0.000 description 1
- 238000002679 ablation Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000002360 explosive Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000010008 shearing Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/15—Vehicle, aircraft or watercraft design
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Computer Hardware Design (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Aviation & Aerospace Engineering (AREA)
- Automation & Control Theory (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Testing Of Engines (AREA)
Abstract
The invention discloses a solid rocket engine performance prediction method, a prediction system, a storage medium and equipment based on deep learning and data enhancement, wherein the method comprises the following steps: constructing a thrust actual measurement data set in the ignition process of the solid engine; preprocessing a thrust actual measurement data set in the ignition process of the solid engine, classifying thrust data of each working condition, and obtaining a test data set and a training data set; building a deep convolutional neural network model, setting training super parameters, and then inputting a training data set into the deep convolutional neural network model for model training to obtain a trained model; and inputting the test data set into the trained model to obtain a solid rocket engine performance prediction result. The solid rocket engine performance prediction method, the prediction system, the storage medium and the device based on deep learning and data enhancement can overcome the low precision and limitation of small sample ground test data on the solid rocket engine performance prediction problem.
Description
Technical Field
The invention belongs to the technical field of intelligent automation technology and aircraft engine design, and particularly relates to a solid rocket engine performance prediction method, a prediction system, a storage medium and equipment based on deep learning and data enhancement.
Background
Solid rocket engines are one of the power systems widely applied to space vehicles such as missiles, rockets and the like, and the prediction of important parameters in the research and production processes of the solid rocket engines is one of the most central problems. By indicating the total stroke of the engine, key parameters are provided for the performance simulation of the engine, and the design efficiency and the design precision of the solid rocket engine can be improved. The commonly used regression prediction method comprises a least square linear regression method and a support vector regression prediction method.
In recent years, the explosive development of deep learning greatly benefits many research tasks, and significant progress has been made in various fields such as image classification, speech recognition, and the like. The deep neural network has the application advantages of rapid reasoning, low use threshold and the like, and can greatly promote the intelligent development of engine design research. Meanwhile, the deep learning model can effectively construct a complex mapping relation between high-dimensional data, and modeling efficiency and accuracy are improved. And the structure is simple and direct, the requirement on priori knowledge is less, and the industrial application is facilitated.
Deep neural networks are a model based on large data, requiring a large amount of diverse data as support. The ground test of the solid engine has high cost and can not fully simulate the flying state, the test and assessment are insufficient, the test sample is few, the current test mode can not meet the technical innovation and verification requirements of the engine, and sufficient data are difficult to collect aiming at each working condition, so when the quantity and the type of the ground test collected data can not meet the training requirements of the deep convolutional neural network, the most direct means are to enhance the data, including increasing the diversity of the data, expanding the data set and the like, in order to improve the generalization capability of the deep neural network for the ablation rate prediction method. However, the data enhancement method applied to images and voices is not suitable for sequence data having obvious physical meaning and mechanism characteristics. For example, when the methods such as overturn, shearing, symmetry and speed change are applied to pressure-time series data measured during the ground test of the solid engine, a pseudo sample which does not conform to the test curve of the general solid rocket engine can be obtained, and if the method is used for deep neural network learning, the characteristic extraction and learning processes can be influenced, so that the prediction accuracy is reduced. Therefore, from the working characteristics of the solid engine, the practical data sample of the solid engine with few and limited data is effectively expanded to improve the prediction accuracy.
Therefore, applying the deep learning and data enhancement method to solid rocket engine performance prediction becomes a problem to be solved urgently.
Disclosure of Invention
The invention aims to provide a solid rocket engine performance prediction method, a prediction system, a storage medium and equipment based on deep learning and data enhancement, so as to solve the problems.
The invention provides a solid rocket engine performance prediction method based on deep learning and data enhancement, which comprises the following steps:
s1: based on thrust-time sequence data of a plurality of processes of the solid rocket engine under a plurality of working conditions, which are collected by a thrust sensor in a ground test of the solid rocket engine, constructing a thrust actual measurement data set of the ignition process of the solid rocket engine;
s2: preprocessing the actual measurement data set of the thrust in the ignition process of the solid rocket engine, classifying the thrust data of each working condition in the preprocessed actual measurement data set of the thrust in the ignition process of the solid rocket engine, and obtaining a training sample and a testing sample, wherein the testing sample forms a testing data set, the training sample is respectively subjected to data enhancement to obtain a plurality of pseudo samples, and the training sample and the pseudo samples jointly form a training data set, wherein the data enhancement method is an adaptive Gaussian noise method or a random drift method;
s3: building a deep convolutional neural network model, setting training super parameters, and then inputting the training data set into the deep convolutional neural network model for model training to obtain a trained model;
s4: and inputting the test data set into the trained model to obtain a solid rocket engine performance prediction result.
Preferably, in S2, the preprocessing of the actual measurement data set of thrust during the ignition process of the solid rocket engine includes:
the time required for starting the igniter to work to the first point of the charging surface is recorded as the ignition delay time t of the propellant 1 The maximum thrust peak before the engine reaches the quasi-steady state working condition is recorded as an ignition thrust peak f max The time required from the start of the igniter to 80% of the ignition thrust peak value is recorded as a preset initial time t 2 ；
Intercepting the ignition delay time t of propellant 1 To a predetermined initial time t 2 Segment thrust data, as ignition process thrust, if no ignition delay time t is present 1 Then intercepting the thrust peak f greater than 0.2kN and less than ignition thrust in the whole process max 80% of the fraction of (a) is ignition process thrust;
and normalizing the time sequence to enable each process to have the same time index, and enabling the characteristic points of each process with different time histories at the same time index to be respectively corresponding.
Further preferably, in S3, inputting the training data set into the deep convolutional neural network model for model training includes the following steps:
s31: forward propagation is carried out on the training data set in batches, a loss function value is calculated, an average absolute error and a mean square error are adopted as the loss function, the absolute difference between the prediction and ground facts is measured, the calculation formula of the average absolute error and the mean square error is shown as a formula (1),
wherein L is MAE As average absolute errorDifference, L MSE Is mean square error, Y i As the i-th predicted value of the model,
a tag representing the performance of the solid rocket engine, n representing the number of predictions;
s32: training a deep convolutional neural network model by using a back propagation algorithm, and optimizing and adjusting parameters of the whole deep convolutional neural network model;
s33: steps S31-S32 are performed in a loop until the maximum number of model exercises is reached.
Further preferably, the back propagation algorithm is optimized using an Adam optimizer.
Further preferably, the maximum number of model exercises is 400, and the model converges after the model performs the maximum number of model exercises.
Further preferably, in S3, the constructed deep convolutional neural network model includes a convolutional layer and a plurality of fully connected layers, where the convolutional layer includes one-dimensional convolutional filters with a size of 3, a number of 64, and a stride of 1; after using an activation function on the feature map after one-dimensional convolution calculation, obtaining extracted features, wherein the extracted features are sequentially connected in a plurality of full-connection layers; the plurality of fully-connected layers includes a first fully-connected layer comprising 7936 neurons and a second fully-connected layer comprising 240 neurons; the last neuron is used as a predicted value of the performance of the solid rocket engine.
The invention also provides a solid rocket engine performance prediction system based on deep learning and data enhancement, which is used for implementing the method, and comprises the following steps:
the system comprises a solid rocket engine ignition process thrust actual measurement data set construction module, a solid rocket engine ignition process thrust actual measurement data set acquisition module and a solid rocket engine ignition process thrust actual measurement data set acquisition module, wherein the solid rocket engine ignition process thrust actual measurement data set construction module is used for constructing a solid rocket engine ignition process thrust actual measurement data set based on thrust-time sequence data of a plurality of processes under multiple working conditions of the solid rocket engine, collected by a thrust sensor in a solid rocket engine ground test;
the data preprocessing module is used for preprocessing the actual measurement data set of the thrust in the ignition process of the solid rocket engine;
sample classification module: the method comprises the steps of classifying thrust data of each working condition in a preprocessed actual measurement data set of the thrust of the solid rocket engine in the ignition process to obtain training samples and test samples, wherein the test samples form a test data set;
the data enhancement module is used for respectively carrying out data enhancement on the training samples to obtain a plurality of pseudo samples, and the pseudo samples and the training samples jointly form a training data set;
the deep convolutional neural network model building and training module is used for building a deep convolutional neural network model, setting training super parameters and carrying out model training through the training data set to obtain a trained model;
and the performance prediction module is used for inputting the test data set into the trained model to obtain a solid rocket engine performance prediction result.
The invention also provides an electronic device, comprising: the device comprises a processor and a memory, wherein a plurality of instructions are stored in the memory, and the processor is used for reading the instructions and executing the method.
The present invention also provides a computer readable storage medium having stored therein a plurality of instructions readable by a processor and performing the above-described method.
The method, the device, the electronic equipment and the computer readable storage medium provided by the invention have the following beneficial technical effects:
(1) The method overcomes the low precision and limitation of small sample ground test data in the problem of solid rocket engine performance prediction, pre-processes the data, enhances the data respectively, integrates the data together, puts the integrated data into a deep learning model, extracts the features through convolution operation, and puts the features together for analysis and prediction, thereby fully utilizing the characteristics of the data, having higher precision of performance prediction and becoming an effective tool for processing the collected data to predict the solid rocket engine performance;
(2) The similarity between the pseudo samples and the actually measured samples can be generated by changing the parameter control in the data enhancement method, and the partial overfitting condition in training caused by unbalance of the data set can be improved by controlling the quantity of the generated pseudo samples through the super parameter control. In addition, the robustness of the model to environmental noise can be enhanced, the prediction method with strong universality to different models and working conditions is researched, and theoretical basis and engineering guidance are provided for the design, manufacture and use of the solid rocket engine in engineering;
(3) The solid rocket engine performance based on deep learning and data enhancement is predicted, so that the complicated solving process of the traditional method is omitted, the high professionality in the field of solid rocket engines is not needed, and the robustness to environmental noise and the universality of different working conditions are higher;
(4) The traditional deep learning neural network is improved, so that the neural network is more suitable for predicting the performance of the solid rocket engine, and has superiority in predicting the performance of the solid rocket engine.
Drawings
FIG. 1 is a flow chart of a solid rocket engine performance prediction method based on deep learning and data enhancement provided by the invention;
FIG. 2 is a schematic diagram of ignition performance parameters of a solid engine;
FIG. 3 is an original ignition process thrust versus time graph;
FIG. 4 is a graph of ignition process thrust versus time after time normalization;
FIG. 5 is a comparative schematic diagram of model errors predicted by the performance of five solid rocket engines.
Detailed Description
The following describes in further detail the embodiments of the present invention with reference to the drawings and examples. The following examples are illustrative of the invention and are not intended to limit the scope of the invention.
As shown in fig. 1, the solid rocket engine performance prediction method based on deep learning and data enhancement comprises the following steps:
s1: based on thrust-time sequence data of a plurality of processes of the solid rocket engine under a plurality of working conditions, which are collected by a thrust sensor in a ground test of the solid rocket engine, constructing a thrust actual measurement data set of the ignition process of the solid rocket engine;
s2: preprocessing the actual measurement data set of the thrust in the ignition process of the solid rocket engine, classifying the thrust data of each working condition in the preprocessed actual measurement data set of the thrust in the ignition process of the solid rocket engine, and obtaining a training sample and a test sample, wherein the test sample forms a test data set, the training samples respectively perform data enhancement to obtain a plurality of pseudo samples, and the training samples and the pseudo samples jointly form a training data set;
the preprocessing of the actual measurement data set of the thrust in the ignition process of the solid rocket engine comprises the following steps:
the time required for starting the igniter to work to the first point of the charging surface is recorded as the ignition delay time t of the propellant 1 The maximum thrust peak before the engine reaches the quasi-steady state working condition is recorded as an ignition thrust peak f max The time required from the start of the igniter to 80% of the ignition thrust peak value is recorded as a preset initial time t 2 As shown in the solid engine ignition performance parameter diagram (fig. 2);
intercepting the ignition delay time t of propellant 1 To a predetermined initial time t 2 Segment thrust data, as ignition process thrust, if no ignition delay time t is present 1 Then intercepting the thrust peak f greater than 0.2kN and less than ignition thrust in the whole process max 80% of the fraction of (a) is ignition process thrust;
normalizing the time sequence to enable each process to have the same scale time index, wherein the method enables the characteristic points of each process with different time histories at the same time index to be respectively corresponding (as shown in fig. 3 and 4), and the specific normalization processing process is as follows:
wherein the input data sequence is expressed as x= [ x ] 1 ,x 2 ,…,x n ]，x i For the ith element, x, in the ith input sequence min And x max Respectively the minimum value and the maximum value of elements in the input sequence;
the data enhancement method is a self-adaptive Gaussian noise method or a random drift method, and the similarity between the generated pseudo sample and the actually measured sample is controlled by changing the size of the added Gaussian noise and the random drift amplitude. The number of generated pseudo samples can be controlled by changing the number of times of data enhancement through super parameters so as to improve the situation of partial overfitting in training caused by unbalance of a data set;
wherein, adaptive gaussian noise method (ANG): gaussian noise adaptively varying with the maximum difference of the data sequence, and directly applying random noise to the original data sequence, the input data sequence is expressed as x= [ x ] 1 ,x 2 ,…,x n ]，x aug Alpha as an enhanced data sequence gaus Is an adaptive Gaussian noise coefficient
x aug ＝x+α gaus G(x max -x min ),G～N(0,1) (4)
With alpha gaus The vibration amplitude of the raw data at each point increases, but the curve shape is not changed. Given different sizes of alpha gaus Different thrust curves can be obtained, and the same alpha gaus The AGN method is applied to the same sequence data for a plurality of times, so that different thrust curves can be obtained, and the diversity of the data is increased to a certain extent. For thrust data with different maximum thrust, the adaptive Gaussian noise can use the same ultrasonic parameter alpha gaus The same degree of disturbance is added to each group of thrust data, so that the device has good generalization.
Random drift method (RD): method for randomly and smoothly shifting the value of a data sequence from its original value by a maximum shift coefficient d max And (5) determining. Applying cubic spline interpolation to the data sequence between two points by randomly selecting 2 points on the data sequence, and then obtaining the data sequence by interpolation according to d max The product of the interpolation maximum value is normalized,superimposed on the incoming data sequence.
The random drifting method changes the number and the positions of standing points (points with the first derivative of 0) on the basis of original data, keeps the thrust peak value and the minimum thrust unchanged in a physical sense, and simultaneously meets the requirements of practical application and enriching the data diversity of the solid rocket engine.
S3: building a deep convolutional neural network model, setting training super parameters, and then inputting the training data set into the deep convolutional neural network model for model training to obtain a trained model;
specifically, a deep convolutional neural network is selected as a basic architecture, and the function of the input layer is to preprocess a data sequence with a length of n, which is shown as x= [ X ] 1 ,…,x t-1 ,x t ,…,x n ] T The data of the data is imported into a nerve network, the deep convolution layer extracts characteristics through the convolution layer, the dimension reduction mapping is carried out on the convolved data through the full connection layer, the mapped output is the predicted value, and the output and the activation function of the convolution layer are expressed as follows:
wherein: the output of the convolution layer is C i ；W i Is the weight; b i Is the deviation;
is a convolution operation symbol;
the leak ReLU function is a piecewise linear activation function that assigns a non-zero slope to each negative value, while positive values are unchanged. The function activation function is mainly to increase the nonlinear expression capacity of the training model by increasing nonlinear factors. The Leaky ReLU is used as an activation function, and all data input and output can be guaranteed to be tiny, so that the training network can realize continuous cyclic operation, key operation characteristics can be found better, and a training model is fitted.
Inputting the training data set into the deep convolutional neural network model for model training comprises the following steps:
s31: forward propagation is carried out on the training data set in batches, a loss function value is calculated, an average absolute error and a mean square error are adopted as the loss function, the absolute difference between the prediction and ground facts is measured, the calculation formula of the average absolute error and the mean square error is shown as a formula (1),
wherein L is MAE Is the average absolute error, L MSE Is mean square error, Y i As the i-th predicted value of the model,
a tag representing the performance of the solid rocket engine, n representing the number of predictions;
s32: training a deep convolutional neural network model by using a back propagation algorithm, and optimizing and adjusting parameters of the whole deep convolutional neural network model;
s33: steps S31-S32 are performed in a loop until the maximum number of model exercises is reached.
S4: and inputting the test data set into the trained model to obtain a solid rocket engine performance prediction result.
As an improvement of the technical scheme, the back propagation algorithm is optimized by adopting an Adam optimizer.
As an improvement of the technical scheme, the maximum model training frequency is 400 times, and the model converges after the model executes the maximum model training frequency. Preferably, a strategy with variable learning rate is adopted, and for the first 200 rounds of training, the learning rate is 0.00006 so as to ensure rapid optimization; the latter 200 rounds used a learning rate of 0.000001 to ensure stable convergence, and 10 consecutive experiments were performed on the same set of parameters, with the resulting errors averaged to reduce the effect of randomness.
As an improvement of the technical scheme, in S3, the constructed deep convolutional neural network model includes a convolutional layer and a plurality of fully connected layers, wherein the convolutional layer includes one-dimensional convolutional filters with a size of 3, a number of 64 and a stride of 1; after using an activation function on the feature map after one-dimensional convolution calculation, obtaining extracted features, wherein the extracted features are sequentially connected in a plurality of full-connection layers; the plurality of fully-connected layers includes a first fully-connected layer comprising 7936 neurons and a second fully-connected layer comprising 240 neurons; the last neuron is used as a predicted value of the performance of the solid rocket engine.
The invention also provides a solid rocket engine performance prediction system based on deep learning and data enhancement, which is used for implementing the method, and comprises the following steps:
the system comprises a solid rocket engine ignition process thrust actual measurement data set construction module, a solid rocket engine ignition process thrust actual measurement data set acquisition module and a solid rocket engine ignition process thrust actual measurement data set acquisition module, wherein the solid rocket engine ignition process thrust actual measurement data set construction module is used for constructing a solid rocket engine ignition process thrust actual measurement data set based on thrust-time sequence data of a plurality of processes under multiple working conditions of the solid rocket engine, collected by a thrust sensor in a solid rocket engine ground test;
the data preprocessing module is used for preprocessing the actual measurement data set of the thrust in the ignition process of the solid rocket engine;
sample classification module: the method comprises the steps of classifying thrust data of each working condition in a preprocessed actual measurement data set of the thrust of the solid rocket engine in the ignition process to obtain training samples and test samples, wherein the test samples form a test data set;
the data enhancement module is used for respectively carrying out data enhancement on the training samples to obtain a plurality of pseudo samples, and the pseudo samples and the training samples jointly form a training data set;
the deep convolutional neural network model building and training module is used for building a deep convolutional neural network model, setting training super parameters and carrying out model training through the training data set to obtain a trained model;
and the performance prediction module is used for inputting the test data set into the trained model to obtain a solid rocket engine performance prediction result.
The invention also provides an electronic device, comprising: the device comprises a processor and a memory, wherein a plurality of instructions are stored in the memory, and the processor is used for reading the instructions and executing the method.
The present invention also provides a computer readable storage medium having stored therein a plurality of instructions readable by a processor and performing the above-described method.
Examples
The method for predicting the performance of the solid rocket engine based on the deep learning and data enhancement method is characterized in that the thrust-time series data of the ignition process of 46 groups of solid rocket engines collected by a factory and total impact samples of the whole process are used as actual measurement data sets, the training data sets are shown in a table 1, and the total impact rapid prediction of the whole process under any working condition is completed, and the method specifically comprises the following steps:
1) The thrust-time series data is taken as input and the corresponding total process impulse data is taken as output;
2) Preprocessing the thrust-time sequence data, and normalizing the time coordinates;
3) Dividing training and testing samples, applying a data enhancement method to the training samples to obtain an extended training data set, and inputting a built deep neural network model for training to obtain a trained model;
4) And inputting the test sample into the trained model, and outputting the total impact predicted value of the whole process.
Table 1 training data set
The training total process prediction model is adopted to perform total process prediction, the obtained result is compared with the real total process, the mean absolute error and the mean square variance are calculated to verify the effectiveness of the method provided by the invention, and meanwhile, the method is compared with a least square linear regression method (LSLR) and a support vector regression prediction method (SVM) to verify the effectiveness of the method provided by the invention. The results obtained are shown in FIG. 5. As can be seen from the figure, the method provided by the invention can construct a total process prediction model with higher precision under the condition of the same training sample, and the effectiveness of the method is shown.
Compared with the existing method, the method has the advantages that the method is applied to various solid engine performance prediction cases, the artificial participation process is few, the design speed is high, and complex data from different models and different working conditions are integrated into feature learning, so that data fusion is realized, system information is enriched, the prediction model precision is continuously increased along with the cases, the solid rocket engine performance prediction requirement can be effectively met, and the method has good universality. In general, raw measurement data is used directly as input, without expertise in signal processing. The operation difficulty is low, the prediction result is obtained only by inputting the test sample into the proposed performance prediction model, the parameter identification scheme is automatically realized, and the applicability of the model in an actual industrial scene is improved to a great extent.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. It is therefore intended that the following claims be interpreted as including the preferred embodiments and all such alterations and modifications as fall within the scope of the invention. It will be apparent to those skilled in the art that various modifications and variations can be made to the present invention without departing from the spirit or scope of the invention. Thus, it is intended that the present invention also include such modifications and alterations insofar as they come within the scope of the appended claims or the equivalents thereof.
Claims (9)
1. The solid rocket engine performance prediction method based on deep learning and data enhancement is characterized by comprising the following steps of:
s1: based on thrust-time sequence data of a plurality of processes of the solid rocket engine under a plurality of working conditions, which are collected by a thrust sensor in a ground test of the solid rocket engine, constructing a thrust actual measurement data set of the ignition process of the solid rocket engine;
s2: preprocessing the actual measurement data set of the thrust in the ignition process of the solid rocket engine, classifying the thrust data of each working condition in the preprocessed actual measurement data set of the thrust in the ignition process of the solid rocket engine, and obtaining a training sample and a testing sample, wherein the testing sample forms a testing data set, the training sample is respectively subjected to data enhancement to obtain a plurality of pseudo samples, and the training sample and the pseudo samples jointly form a training data set, wherein the data enhancement method is an adaptive Gaussian noise method or a random drift method;
s3: building a deep convolutional neural network model, setting training super parameters, and then inputting the training data set into the deep convolutional neural network model for model training to obtain a trained model;
s4: and inputting the test data set into the trained model to obtain a solid rocket engine performance prediction result.
2. A solid rocket engine performance prediction method based on deep learning and data enhancement as recited in claim 1, wherein: s2, preprocessing the actual measurement data set of the thrust in the ignition process of the solid rocket engine comprises the following steps:
the time required for starting the igniter to work to the first point of the charging surface is recorded as the ignition delay time t of the propellant 1 The maximum thrust peak before the engine reaches the quasi-steady state working condition is recorded as an ignition thrust peak f max The time required from the start of the igniter to 80% of the ignition thrust peak value is recorded as a preset initial time t 2 ；
Intercepting the ignition delay time t of propellant 1 To a predetermined initial time t 2 Segment thrust data, as ignition process thrust, if no ignition delay time t is present 1 Then intercepting the thrust peak f greater than 0.2kN and less than ignition thrust in the whole process max 80% of the fraction of (a) is ignition process thrust;
and normalizing the time sequence to enable each process to have the same time index, and enabling the characteristic points of each process with different time histories at the same time index to be respectively corresponding.
3. A solid rocket engine performance prediction method based on deep learning and data enhancement as recited in claim 1, wherein: s3, inputting the training data set into the deep convolutional neural network model for model training, wherein the method comprises the following steps of:
s31: forward propagation is carried out on the training data set in batches, a loss function value is calculated, an average absolute error and a mean square error are adopted as the loss function, the absolute difference between the prediction and ground facts is measured, the calculation formula of the average absolute error and the mean square error is shown as a formula (1),
wherein L is MAE Is the average absolute error, L MSE Is mean square error, Y i The predicted value of the ith time of the model is Y, which represents the label of the performance of the solid rocket engine, and n represents the number of times of prediction;
s32: training a deep convolutional neural network model by using a back propagation algorithm, and optimizing and adjusting parameters of the whole deep convolutional neural network model;
s33: steps S31-S32 are performed in a loop until the maximum number of model exercises is reached.
4. A solid rocket engine performance prediction method based on deep learning and data enhancement as recited in claim 3, wherein: the back propagation algorithm is optimized by an Adam optimizer.
5. A solid rocket engine performance prediction method based on deep learning and data enhancement as recited in claim 3, wherein: the maximum number of model exercises is 400, and the model converges after the model performs the maximum number of model exercises.
6. A solid rocket engine performance prediction method based on deep learning and data enhancement as recited in claim 1, wherein: in S3, the built deep convolutional neural network model comprises a convolutional layer and a plurality of full-connection layers, wherein the convolutional layer comprises one-dimensional convolutional filters with the size of 3, the number of 64 and the stride of 1; after using an activation function on the feature map after one-dimensional convolution calculation, obtaining extracted features, wherein the extracted features are sequentially connected in a plurality of full-connection layers; the plurality of fully-connected layers includes a first fully-connected layer comprising 7936 neurons and a second fully-connected layer comprising 240 neurons; the last neuron is used as a predicted value of the performance of the solid rocket engine.
7. A solid rocket engine performance prediction system based on deep learning and data enhancement for implementing the method of any of claims 1-6, comprising:
the system comprises a solid rocket engine ignition process thrust actual measurement data set construction module, a solid rocket engine ignition process thrust actual measurement data set acquisition module and a solid rocket engine ignition process thrust actual measurement data set acquisition module, wherein the solid rocket engine ignition process thrust actual measurement data set construction module is used for constructing a solid rocket engine ignition process thrust actual measurement data set based on thrust-time sequence data of a plurality of processes under multiple working conditions of the solid rocket engine, collected by a thrust sensor in a solid rocket engine ground test;
the data preprocessing module is used for preprocessing the actual measurement data set of the thrust in the ignition process of the solid rocket engine;
sample classification module: the method comprises the steps of classifying thrust data of each working condition in a preprocessed actual measurement data set of the thrust of the solid rocket engine in the ignition process to obtain training samples and test samples, wherein the test samples form a test data set;
the data enhancement module is used for respectively carrying out data enhancement on the training samples to obtain a plurality of pseudo samples, and the pseudo samples and the training samples jointly form a training data set;
the deep convolutional neural network model building and training module is used for building a deep convolutional neural network model, setting training super parameters and carrying out model training through the training data set to obtain a trained model;
and the performance prediction module is used for inputting the test data set into the trained model to obtain a solid rocket engine performance prediction result.
8. An electronic device, comprising: a processor and a memory, wherein the memory has stored therein a plurality of instructions, the processor being for reading the instructions and performing the method of any of claims 1-6.
9. A computer-readable storage medium, characterized in that: the computer readable storage medium having stored therein a plurality of instructions readable by a processor and executable by the processor to perform the method of any of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310176428.0A CN116401756A (en) | 2023-02-28 | 2023-02-28 | Solid rocket engine performance prediction method, prediction system, storage medium and equipment based on deep learning and data enhancement |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310176428.0A CN116401756A (en) | 2023-02-28 | 2023-02-28 | Solid rocket engine performance prediction method, prediction system, storage medium and equipment based on deep learning and data enhancement |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116401756A true CN116401756A (en) | 2023-07-07 |
Family
ID=87014910
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310176428.0A Pending CN116401756A (en) | 2023-02-28 | 2023-02-28 | Solid rocket engine performance prediction method, prediction system, storage medium and equipment based on deep learning and data enhancement |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116401756A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116629143A (en) * | 2023-07-25 | 2023-08-22 | 东方空间技术(山东)有限公司 | Rocket simulation launching parameter interpretation method, computing equipment and storage medium |
| CN116680993A (en) * | 2023-07-31 | 2023-09-01 | 东方空间技术(山东)有限公司 | Rocket launching data processing method, device and equipment |
| CN116702334A (en) * | 2023-08-04 | 2023-09-05 | 中国人民解放军国防科技大学 | Sparse storage method for overall design case of solid engine |
| CN116772662A (en) * | 2023-07-17 | 2023-09-19 | 东方空间技术(山东)有限公司 | Rocket recovery sub-level landing leg control method, computing equipment and storage medium |
-
2023
- 2023-02-28 CN CN202310176428.0A patent/CN116401756A/en active Pending
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116772662A (en) * | 2023-07-17 | 2023-09-19 | 东方空间技术(山东)有限公司 | Rocket recovery sub-level landing leg control method, computing equipment and storage medium |
| CN116772662B (en) * | 2023-07-17 | 2024-04-19 | 东方空间技术(山东)有限公司 | Rocket recovery sub-level landing leg control method, computing equipment and storage medium |
| CN116629143A (en) * | 2023-07-25 | 2023-08-22 | 东方空间技术(山东)有限公司 | Rocket simulation launching parameter interpretation method, computing equipment and storage medium |
| CN116680993A (en) * | 2023-07-31 | 2023-09-01 | 东方空间技术(山东)有限公司 | Rocket launching data processing method, device and equipment |
| CN116702334A (en) * | 2023-08-04 | 2023-09-05 | 中国人民解放军国防科技大学 | Sparse storage method for overall design case of solid engine |
| CN116702334B (en) * | 2023-08-04 | 2023-10-20 | 中国人民解放军国防科技大学 | Sparse storage method for overall design case of solid engine |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116401756A (en) | Solid rocket engine performance prediction method, prediction system, storage medium and equipment based on deep learning and data enhancement | |
| CN101566829B (en) | Method for computer-aided open loop and/or closed loop control of a technical system | |
| CN111047085B (en) | Hybrid vehicle working condition prediction method based on meta-learning | |
| CN103559537B (en) | Based on the template matching method of error back propagation in a kind of out of order data stream | |
| CN111126132A (en) | Learning target tracking algorithm based on twin network | |
| CN110674965A (en) | Multi-time step wind power prediction method based on dynamic feature selection | |
| KR20190139539A (en) | A System of Searching the Channel Expansion Parameter for the Speed-up of Inverted Residual Block and the method thereof for low specification embedded system and the method thereof | |
| CN107563430A (en) | A kind of convolutional neural networks algorithm optimization method based on sparse autocoder and gray scale correlation fractal dimension | |
| CN109800517B (en) | Improved reverse modeling method for magnetorheological damper | |
| CN112578089B (en) | Air pollutant concentration prediction method based on improved TCN | |
| CN116244647A (en) | Unmanned aerial vehicle cluster running state estimation method | |
| CN111832911A (en) | Underwater combat effectiveness evaluation method based on neural network algorithm | |
| Gao et al. | Autonomous driving based on modified sac algorithm through imitation learning pretraining | |
| CN109886405A (en) | It is a kind of inhibit noise based on artificial neural network structure's optimization method | |
| CN116486285B (en) | Aerial image target detection method based on class mask distillation | |
| He et al. | Exploring linear feature disentanglement for neural networks | |
| CN117114053A (en) | Convolutional neural network model compression method and device based on structure search and knowledge distillation | |
| CN117763933A (en) | Solid rocket engine time sequence parameter prediction method and prediction system based on deep learning | |
| CN116432514A (en) | Interception intention recognition strategy simulation system and method for unmanned aerial vehicle attack and defense game | |
| CN111602145A (en) | Optimization method of convolutional neural network and related product | |
| CN116108891A (en) | Lightweight pruning method, device, terminal and computer readable storage medium based on Transformer | |
| US12020166B2 (en) | Meta-learned, evolution strategy black box optimization classifiers | |
| CN114495036A (en) | Vehicle track prediction method based on three-stage attention mechanism | |
| CN117933104B (en) | Solid attitude and orbit control engine gas regulating valve pressure correction method | |
| CN112906868A (en) | Behavior clone-oriented demonstration active sampling method |
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 |