CN112364566B - Deduction prediction method based on typical time data characteristics - Google Patents
Deduction prediction method based on typical time data characteristics Download PDFInfo
- Publication number
- CN112364566B CN112364566B CN202011264280.9A CN202011264280A CN112364566B CN 112364566 B CN112364566 B CN 112364566B CN 202011264280 A CN202011264280 A CN 202011264280A CN 112364566 B CN112364566 B CN 112364566B
- Authority
- CN
- China
- Prior art keywords
- deduction
- typical time
- neural network
- data
- method based
- 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.)
- Active
Links
Images
Classifications
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
- G06N3/045—Combinations of 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)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Evolutionary Biology (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a deduction prediction method based on typical time data characteristics, which is characterized in that a characteristic extraction framework oriented to confrontation game process data decomposes a preprocessing process into three steps of data preliminary screening, combat typical time sequence slicing and data normalization, continuous data are subjected to discretization processing according to the variation gradient similarity of combat scores of various departments in the time sequence slicing process, and the problem that the traditional static data characteristic extraction method cannot be directly applied to the confrontation data characteristic extraction of a game is effectively solved.
Description
Technical Field
The invention relates to a dynamic data feature extraction method, in particular to a deduction prediction method based on typical time data features.
Background
Data feature extraction techniques, particularly those represented by the field of image recognition, have been applied to various aspects of our lives, such as building reconstruction and protection, remote sensing image analysis, urban planning, medical diagnosis, and the like.
The traditional image-oriented feature extraction algorithm mainly comprises three types: 1) HOG (histogram of oriented gradients) is a feature descriptor used for object detection in computer vision and image processing, and is used for constructing features by calculating and counting the gradient directions of local areas of an image; 2) LBP (local binary pattern) is an operator proposed in 1994 to describe local texture features of an image, and has the significant advantages of rotation invariance and gray scale invariance, and the main extracted features are the local texture features of the image; 3) the Haar-like features were first applied to face representation, which were combined into a feature template by three classes of features, edge, linear, central and diagonal.
In 2012, the feature extraction technology based on deep learning makes a breakthrough progress in various applications in the field of computer vision, and a plurality of deep learning network models such as AlexNet, VGG16, inclusion and the like appear successively, and these neural networks can well complete feature extraction and recognition of images. AlexNet is the first deep learning model that achieves significant effects in the universal dataset; the VGG16 is a model which is used most frequently in the fields of image classification and the like so far, and although the complexity of the model is not high, the practical effect is often better than that of a plurality of deep learning models with high complexity; the inclusion network is a more complex network model, and convolution features of different sizes can be learned in the same layer. For each deep learning model, two different ways can be respectively adopted to complete feature extraction and classification: firstly, an end-to-end mode is adopted, and the classification result of the image is directly obtained through a trained deep learning model (the accuracy performance index is given by a network model); secondly, extracting the full-connection characteristics of the trained deep learning model instead of directly obtaining the classification result, and completing the identification of the candidate region by adopting a KNN method.
In recent years, in the situation awareness for complex battlefields, the military strong countries represented by the united states have also gradually strengthened research and application of the intelligent feature extraction technology in the situation awareness. Deep Learning is adopted in Deep Learning method, more hidden useful features are extracted from a large amount of unlabelled sound, video, sensor and text data acquired from a battlefield, and the extracted useful features are used for pattern recognition and feature classification, association relationship mining, abnormal monitoring, event time relationship description and the like; the PPAML project aims to construct an intelligent learning machine, so that the intelligent learning machine can understand data, analyze results and reason association relation from uncertain information.
In summary, most of the current feature extraction technologies mainly aim at static data such as images and voices, and the feature extraction method provided by the invention mainly aims at data in a dynamic auxiliary command decision process in simulation deduction to realize the extraction of data features, and no literature report which is completely the same as the research content of the subject is found in non-patent documents and patent documents published at home and abroad at present.
Disclosure of Invention
The technical problem of the invention is solved: in the problems of the multi-disk analysis of process data and the feature extraction model, the key features playing a key role in a battle bureau are extracted from a large number of features in the face of massive game data in the simulation deduction process. The first problem is that unlike the feature extraction of traditional static data, the faced simulation deduction data has strong dynamic and game performance, the data may change in a large range at every moment, and as the deduction process proceeds, the weight of the data feature may change at any moment, and the original unimportant or less important information may become very important at the next moment or even determine the trend of the subsequent game, under which the traditional one-unchanged oriented static data extraction method becomes no longer applicable; secondly, in simulation deduction of a large-scale system level, the types and the number of the participating units are numerous and large, so that the feature dimension is very high, and how to accurately extract features under the condition of high dimension becomes difficult. The invention overcomes the defects of the prior art, provides an attention mechanism-based auxiliary command decision-making deduction data feature extraction method, and adopts an artificial intelligence technology to solve the problems.
The technical scheme of the invention is as follows:
a deduction prediction method based on typical time data characteristics comprises the following steps:
1) performing human-to-human, man-to-machine and machine-to-machine simulation deduction of multiple rounds according to the selected simulation deduction platform and the fighting scene to obtain simulation data of each round of fighting process, wherein the simulation data are the number, the positions and the states of fighting units;
2) storing and processing the simulation data of each countermeasure process in the step 1);
3) preliminarily screening the simulation data stored in the step 2), and extracting to obtain a plurality of preliminary key features, wherein the preliminary key features are state variables capable of influencing countermeasure results in the simulation data; and 3) the state variables are the hitting capacity of the weapons, the number of the weapons and the firepower configuration.
4) Defining typical time, and obtaining the preliminary key features corresponding to the typical time according to the preliminary key features in the step 3);
5) changing the number, position and state of the combat units; repeating the steps 1) to 4) n times to obtain typical time data corresponding to n times of countermeasure processes, and performing normalization processing on the typical time data corresponding to the n times of countermeasure processes to obtain input quantity of the neural network; meanwhile, obtaining the win and lose results corresponding to each part of confrontation process as the output quantity of the neural network;
6) establishing a feature extraction neural network model by adopting an attention mechanism;
7) training the feature extraction neural network model obtained in the step 6) according to the neural network input quantity and the neural network output quantity in the step 5) to obtain a trained feature extraction neural network model;
8) during actual combat deduction, secondary screening is carried out on the primary key features by using the trained feature extraction neural network model in the step 7) to obtain important key features;
and (4) predicting the actual combat deduction result by using the important key features obtained in the step (8) and the trained feature extraction neural network model in the step (7), and obtaining the output quantity of the neural network to be used as the prediction result of the actual combat deduction.
Step 4) the time length from the typical time to the starting time of the countermeasure process is equal to 4/5 to 8/9 of the total duration of the countermeasure process.
The 10 minutes before the war ending time is defined as a typical time.
n is a positive integer not less than 100.
The neural network model adopts a soft-attention mechanism network architecture.
And 7) training the feature extraction neural network model by adopting an error back propagation algorithm.
Step 4) also comprises the following steps: and dividing the countermeasure process into a front stage, a middle stage, a later stage or other scorching stages according to time, and determining a typical moment according to a division result.
And 5) the normalization processing method is any one of min-max normalization, 0 mean normalization or arctan function normalization.
Compared with the prior art, the invention has the advantages that:
1) the invention can complete the feature extraction of the dynamic game data by utilizing the designed feature extraction architecture and the feature extraction process;
2) the attention feature extraction neural network model adopted by the invention can finish extraction of key features aiming at typical moments of the deduction reply disk.
Drawings
FIG. 1 is a diagram of a feature extraction architecture for confrontational game process data designed in the present invention;
FIG. 2 is a flow chart of game data oriented feature extraction designed by the present invention;
FIG. 3 is a diagram of a attention mechanism feature extraction neural network model architecture used in the present invention.
Detailed Description
A feature extraction framework for fighting game process data is combined with a feature extraction neural network model of an attention mechanism to achieve key feature extraction for typical moments of a deduction reply disk.
As shown in figure 1, the feature extraction architecture for the data in the countermeasure game process decomposes a preprocessing process into three steps of data preliminary screening, combat typical time sequence slicing and data normalization, and discretizes continuous data according to the gradient similarity of the variation of combat scores of each office in the time sequence slicing process, so that the problem that the traditional static data feature extraction method cannot be directly applied to game countermeasure data feature extraction is effectively solved.
The method adopts the feature extraction neural network model of the attention mechanism, can assign the attention mechanism to the input data features according to the trained network weights facing the typical time of the deduction reply disk, thereby completing the extraction of key feature data through weight proportion values, can effectively predict the battle result according to the output of the network, and the accuracy of the prediction result can reflect the accuracy of the extraction of the key features to a certain extent.
The invention relates to a deduction prediction method based on typical time data characteristics, as shown in fig. 2, comprising the following steps:
1) according to the selected simulation deduction platform and the fighting scene, carrying out human-to-human, man-machine and machine-to-machine simulation deduction on of multiple rounds to obtain simulation data of each round of fighting process, such as the number, the positions, the states and the like of fighting units;
2) storing and processing the simulation data of each countermeasure process in the step 1);
3) preliminarily screening the simulation data stored in the step 2), and extracting to obtain a plurality of preliminary key features, wherein the preliminary key features are state variables which can influence a confrontation result in the simulation data, such as weapon hitting capacity, weapon quantity, firepower configuration and the like;
4) slicing the battle time sequence, defining time points from the battle preparation to the start of fire exchange and the like as typical moments, and obtaining the preliminary key features corresponding to the typical moments according to the preliminary key features in the step 3); in the embodiment of the present invention, 10 minutes before the war end time is defined as a typical time. The length of time from the typical time to the start time of the countermeasure process is equal to 4/5 to 8/9 of the total duration of the countermeasure process.
41) The slicing division can be performed by adopting the experiences summarized by human beings in the confrontation process, and dividing the battle according to time in the early stage, the middle stage, the later stage or other scorching stages;
42) typical time divisions can be made by data statistics methods.
The typical time division method for data statistics specifically comprises the following steps:
according to situation scoring result information formed in the process of the confrontation platform, gradient change statistics is carried out on the information, time is used as an abscissa, the change similarity of the situation scoring information is extracted, the similarity result can represent the fixed frequency of similar actions, and the places with the same frequency are time sliced, so that typical time division is completed.
5) Changing the number, position and state of the combat units; repeating the steps 1) to 4) n times to obtain typical time data corresponding to n times of countermeasure processes, and performing normalization processing on the typical time data corresponding to the n times of countermeasure processes to obtain input quantity of the neural network; meanwhile, obtaining the win and lose results corresponding to each part of confrontation process as the output quantity of the neural network; wherein n is a positive integer not less than 100;
6) establishing a feature extraction neural network model by adopting an attention mechanism, as shown in fig. 3;
61) taking the feature number obtained after the preliminary screening as network input (namely preliminary key features);
62) adopting a soft-attention mechanism network architecture;
63) and taking the predicted win or loss of the confrontation as network output.
7) Training the feature extraction neural network model obtained in the step 6) according to the neural network input quantity and the neural network output quantity in the step 5) to obtain a trained feature extraction neural network model; specifically, an error back propagation algorithm is adopted to train the neural network model, the training samples are human, man-machine or machine-machine confrontation data which are subjected to screening and normalization processing, and the trained feature extraction neural network model is obtained.
8) When actual combat deduction is carried out, secondary screening is carried out on the primary key features by using the trained feature extraction neural network model in the step 7) to obtain important key features (namely, key features corresponding to actual combat deduction); ultimately resulting in important key features such as aircraft position. The method comprises the following specific steps: and (3) obtaining weight distribution of the attention mechanism through forward propagation, taking the weight distribution proportion as a feature extraction result, and reselecting and training the features by adopting an iteration method when the extraction effect is not ideal. The iteration method specifically comprises the following steps: if the result of the war dispute cannot be effectively predicted by the feature extraction effect, the classified data is screened again, the range of the visual features is expanded properly until the key features extracted by the feature extraction model can correctly reflect the important influence degree on the war trend.
9) And (5) predicting the actual combat deduction result by using the important key features obtained in the step 8) and the trained feature extraction neural network model in the step 7), and obtaining the neural network output quantity as the prediction result of the actual combat deduction. The output quantity of the trained feature extraction network model is used as a prediction result of actual combat deduction, 0 and 1 represent two deduction results, and prediction of the deduction results is achieved.
Examples
An attention mechanism-based method for extracting characteristics of deduction data oriented to auxiliary command decision making comprises the following steps:
1) performing human-to-human, man-machine and machine-to-machine simulation deduction according to the selected simulation deduction platform and the fighting scene scenario to obtain deduction original data;
2) classifying and storing all combat entity units and state data which change constantly in the combat process according to types aiming at the countermeasure data of each office;
3) preliminarily screening the stored data for extracting key features from the data;
4) slicing the battle time sequence according to a typical process, and converting dynamic data into static data at a typical moment;
5) normalizing the obtained typical time data to enable the processed data to be suitable for being input as a neural network;
6) designing a characteristic extraction neural network model by adopting an attention mechanism;
7) according to the sample library, training an attention mechanism characteristic extraction neural network model;
8) carrying out feature extraction on the data by using the trained feature extraction model;
9) predicting the fight deduction result by using the trained feature extraction model;
10) verifying the prediction result in the verification set, returning to the step 3) if the prediction result does not meet the requirement, re-screening the data, and repeating the steps 4) to 9) until the prediction result meets the requirement after verification.
Step 3) a method for preliminarily screening the characteristics of the stored data, which comprises the following steps: envelope extraction is carried out on the data characteristics according to human experience, the data which is considered to possibly influence the war tendency is within the range, and the characteristics which are considered to be irrelevant or weakly relevant are removed.
The combat typical time slice dividing method of the step 4) comprises the following steps:
(41) the fight can be divided into the early stage, the middle stage, the later stage or other scorching stages according to the time by adopting the experiences summarized by the human beings in the fight process;
(42) typical time divisions may be made by data statistics methods.
According to situation scoring result information formed in the process of the confrontation platform, gradient change statistics is carried out on the information, time is used as an abscissa, the change similarity of the situation scoring information is extracted, the similarity result can represent the fixed frequency of similar actions, and the places with the same frequency are time sliced, so that typical time division is completed.
The method of step 5), comprising: in common normalization methods such as min-max normalization, 0-mean normalization, arctangent function normalization and the like, a normalization method suitable for network training is tried to be selected one by one for improving the training efficiency and the convergence rate of a network model.
The attention mechanism design feature extraction neural network model design method in step 6) comprises the following steps:
(61) taking the feature number obtained after the preliminary screening as network input;
(62) adopting a soft-attention mechanism network architecture;
(63) and taking the predicted competition success or failure as the network output.
And training the neural network model by adopting an error back propagation algorithm, wherein the training sample is human, man-machine or machine-machine confrontation data subjected to screening and normalization processing.
And (3) obtaining weight distribution of the attention mechanism through forward propagation, taking the weight distribution proportion as a feature extraction result, and reselecting and training the features by adopting an iteration method when the extraction effect is not ideal.
If the result of the feature extraction cannot effectively predict the outcome of the war dispute, the classified data is screened again, the range of the visual features is expanded properly until the key features extracted through the feature extraction model can correctly reflect the important influence degree on the war trend.
And outputting the trained feature extraction network model to be directly used as a deduction prediction result, wherein 0 and 1 represent two deduction results, and prediction of the deduction result is realized.
Those skilled in the art will appreciate that the details of the invention not described in detail in the specification are within the skill of those skilled in the art.
Claims (10)
1. A deduction prediction method based on typical time data characteristics is characterized by comprising the following steps:
1) performing human-to-human, man-to-machine and machine-to-machine simulation deduction of multiple rounds according to the selected simulation deduction platform and the fighting scene to obtain simulation data of each round of confrontation process;
2) storing and processing the simulation data of each part of countermeasure process in the step 1);
3) preliminarily screening the simulation data stored in the step 2), and extracting to obtain a plurality of preliminary key features, wherein the preliminary key features are state variables capable of influencing countermeasure results in the simulation data;
4) defining typical time, and obtaining the preliminary key features corresponding to the typical time according to the preliminary key features in the step 3);
5) changing the number, position and state of the combat units; repeating the steps 1) to 4) n times to obtain typical time data corresponding to n times of countermeasure processes, and performing normalization processing on the typical time data corresponding to the n times of countermeasure processes to obtain input quantity of the neural network; meanwhile, obtaining the win and lose results corresponding to each part of confrontation process as the output quantity of the neural network;
6) establishing a feature extraction neural network model by adopting an attention mechanism;
7) training the feature extraction neural network model obtained in the step 6) according to the neural network input quantity and the neural network output quantity in the step 5) to obtain a trained feature extraction neural network model;
8) during actual combat deduction, performing secondary screening on the primary key features by using the trained feature extraction neural network model in the step 7) to obtain important key features;
and (5) predicting the actual combat deduction result by using the important key features obtained in the step 8) and the trained feature extraction neural network model in the step 7), and obtaining the neural network output quantity as the prediction result of the actual combat deduction.
2. The deduction prediction method based on typical time data characteristics as claimed in claim 1, wherein the time length from the typical time to the start time of the countermeasure process in step 4) is equal to 4/5 to 8/9 of the total duration of the countermeasure process.
3. The deduction prediction method based on typical time data characteristics as claimed in claim 2, wherein 10 minutes before the war ending time is defined as a typical time.
4. The deductive prediction method based on the typical time data characteristics as claimed in any one of claims 1 to 3, wherein n is a positive integer not less than 100.
5. The deduction prediction method based on typical time data characteristics as claimed in claim 4, wherein the simulation data of step 1) is the number, position and status of combat units.
6. The deduction prediction method based on typical time data characteristics as claimed in claim 4, wherein the state variables of step 3) are weapon hitting ability, weapon quantity and firepower configuration.
7. The deduction prediction method based on typical time data characteristics as claimed in claim 1, wherein the neural network model adopts a soft-attention mechanism network architecture.
8. The deductive prediction method based on typical time data characteristics as claimed in claim 1, wherein step 7) trains the characteristic extraction neural network model by using an error back propagation algorithm.
9. The deduction prediction method based on typical time data characteristics as claimed in claim 1, wherein the step 4) further comprises: and dividing the countermeasure process into a front stage, a middle stage, a later stage or other scorching stages according to time, and determining a typical moment according to a division result.
10. The deductive prediction method based on typical time data characteristics as claimed in claim 1, wherein the normalization processing of step 5) is any one of min-max normalization, 0-mean normalization or arctan function normalization.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011264280.9A CN112364566B (en) | 2020-11-12 | 2020-11-12 | Deduction prediction method based on typical time data characteristics |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011264280.9A CN112364566B (en) | 2020-11-12 | 2020-11-12 | Deduction prediction method based on typical time data characteristics |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112364566A CN112364566A (en) | 2021-02-12 |
| CN112364566B true CN112364566B (en) | 2022-09-06 |
Family
ID=74514566
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011264280.9A Active CN112364566B (en) | 2020-11-12 | 2020-11-12 | Deduction prediction method based on typical time data characteristics |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112364566B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115222199B (en) * | 2022-05-31 | 2023-11-14 | 中国电子科技集团公司第五十二研究所 | Air combat key node determining method and device based on win-plant rate prediction |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109636699A (en) * | 2018-11-06 | 2019-04-16 | 中国电子科技集团公司第五十二研究所 | A kind of unsupervised intellectualized battle deduction system based on deeply study |
| CN109669415A (en) * | 2018-12-13 | 2019-04-23 | 宁波大学 | A kind of dynamic process monitoring method based on the analysis of structuring canonical variable |
-
2020
- 2020-11-12 CN CN202011264280.9A patent/CN112364566B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109636699A (en) * | 2018-11-06 | 2019-04-16 | 中国电子科技集团公司第五十二研究所 | A kind of unsupervised intellectualized battle deduction system based on deeply study |
| CN109669415A (en) * | 2018-12-13 | 2019-04-23 | 宁波大学 | A kind of dynamic process monitoring method based on the analysis of structuring canonical variable |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112364566A (en) | 2021-02-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110334741B (en) | Radar one-dimensional range profile identification method based on cyclic neural network | |
| CN107103171B (en) | Modeling method and device of machine learning model | |
| CN105975573B (en) | A kind of file classification method based on KNN | |
| CN110166484A (en) | A kind of industrial control system intrusion detection method based on LSTM-Attention network | |
| CN111126820B (en) | Method and system for preventing electricity stealing | |
| CN109934615B (en) | Product marketing method based on deep sparse network | |
| CN111598179A (en) | Power monitoring system user abnormal behavior analysis method, storage medium and equipment | |
| CN113259388B (en) | Network flow abnormity detection method, electronic equipment and readable storage medium | |
| CN113283590A (en) | Defense method for backdoor attack | |
| CN112364566B (en) | Deduction prediction method based on typical time data characteristics | |
| CN118332034B (en) | Data mining model construction method and system based on machine learning | |
| CN114912741A (en) | Effectiveness evaluation method and device for combat system structure and storage medium | |
| CN110705693A (en) | Unmanned aerial vehicle abnormal behavior recognition module and recognition method thereof | |
| CN108090905B (en) | The judgment method and system of producing line exception | |
| CN109934352B (en) | Automatic evolution method of intelligent model | |
| CN107729942A (en) | A kind of sorting technique of structured view missing data | |
| CN116416228A (en) | Method and system for detecting unknown defects of industrial products | |
| CN108805152A (en) | A kind of scene classification method and device | |
| CN111651930A (en) | Extreme learning machine-based flow field vortex region detection method | |
| CN116260565A (en) | Chip electromagnetic side channel analysis method, system and storage medium | |
| CN115840116A (en) | Switch cabinet partial discharge fault mode identification method based on GAN-CNN | |
| CN115883182A (en) | Method and system for improving network security situation element identification efficiency | |
| CN111816404B (en) | Demagnetization method and system | |
| CN112884069A (en) | Method for detecting confrontation network sample | |
| CN113887807A (en) | Robot game tactical prediction method based on machine learning and evidence theory |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |