To address the confrontation decision-making issues in multi-round air combat,a dynamic game decision method is proposed based on decision tree for the confrontation of unmanned aerial vehicle(UAV)air combat.Based on ...To address the confrontation decision-making issues in multi-round air combat,a dynamic game decision method is proposed based on decision tree for the confrontation of unmanned aerial vehicle(UAV)air combat.Based on game the-ory and the confrontation characteristics of air combat,a dynamic game process is constructed including the strategy sets,the situation information,and the maneuver decisions for both sides of air combat.By analyzing the UAV’s flight dyna-mics and the both sides’information,a payment matrix is estab-lished through the situation advantage function,performance advantage function,and profit function.Furthermore,the dynamic game decision problem is solved based on the linear induction method to obtain the Nash equilibrium solution,where the decision tree method is introduced to obtain the optimal maneuver decision,thereby improving the situation advantage in the next round of confrontation.According to the analysis,the simulation results for the confrontation scenarios of multi-round air combat are presented to verify the effectiveness and advan-tages of the proposed method.展开更多
System upgrades in unmanned systems have made Unmanned Aerial Vehicle(UAV)-based patrolling and monitoring a preferred solution for ocean surveillance.However,dynamic environments and large-scale deployments pose sign...System upgrades in unmanned systems have made Unmanned Aerial Vehicle(UAV)-based patrolling and monitoring a preferred solution for ocean surveillance.However,dynamic environments and large-scale deployments pose significant challenges for efficient decision-making,necessitating a modular multiagent control system.Deep Reinforcement Learning(DRL)and Decision Tree(DT)have been utilized for these complex decision-making tasks,but each has its limitations:DRL is highly adaptive but lacks interpretability,while DT is inherently interpretable but has limited adaptability.To overcome these challenges,we propose the Adaptive Interpretable Decision Tree(AIDT),an evolutionary-based algorithm that is both adaptable to diverse environmental settings and highly interpretable in its decision-making processes.We first construct a Markov decision process(MDP)-based simulation environment using the Cooperative Submarine Search task as a representative scenario for training and testing the proposed method.Specifically,we use the heat map as a state variable to address the issue of multi-agent input state proliferation.Next,we introduce the curiosity-guiding intrinsic reward to encourage comprehensive exploration and enhance algorithm performance.Additionally,we incorporate decision tree size as an influence factor in the adaptation process to balance task completion with computational efficiency.To further improve the generalization capability of the decision tree,we apply a normalization method to ensure consistent processing of input states.Finally,we validate the proposed algorithm in different environmental settings,and the results demonstrate both its adaptability and interpretability.展开更多
To solve the multi-class fault diagnosis tasks, decision tree support vector machine （DTSVM）, which combines SVM and decision tree using the concept of dichotomy, is proposed. Since the classification performance of...To solve the multi-class fault diagnosis tasks, decision tree support vector machine （DTSVM）, which combines SVM and decision tree using the concept of dichotomy, is proposed. Since the classification performance of DTSVM highly depends on its structure, to cluster the multi-classes with maximum distance between the clustering centers of the two sub-classes, genetic algorithm is introduced into the formation of decision tree, so that the most separable classes would be separated at each node of decisions tree. Numerical simulations conducted on three datasets compared with ＂one-against-all＂ and ＂one-against-one＂ demonstrate the proposed method has better performance and higher generalization ability than the two conventional methods.展开更多
Define and theory of autocorrelation decision tree （ADT） is introduced. In spatial data mining, spatial parallel query are very expensive operations. A new parallel algorithm in terms of autocorrelation decision tre...Define and theory of autocorrelation decision tree （ADT） is introduced. In spatial data mining, spatial parallel query are very expensive operations. A new parallel algorithm in terms of autocorrelation decision tree is presented. And the new method reduces CPU- and I/O-time and improves the query efficiency of spatial data. For dynamic load balancing, there are better control and optimization. Experimental performance comparison shows that the improved algorithm can obtain a optimal accelerator with the same quantities of processors. There are more completely accesses on nodes. And an individual implement of intelligent information retrieval for spatial data mining is presented.展开更多
A high-speed comer detection algorithm based on fuzzy ID3 decision tree was proposed. In the algorithm, the Bresenham circle with 3-pixel radius was used as the test mask, overlapping the candidate comers with the nuc...A high-speed comer detection algorithm based on fuzzy ID3 decision tree was proposed. In the algorithm, the Bresenham circle with 3-pixel radius was used as the test mask, overlapping the candidate comers with the nucleus. Connected pixels on the circle were applied to compare the intensity value with the nucleus, with the membership function used to give the fuzzy result. The pixel with maximum information gain was chosen as the parent node to build a binary decision tree. Thus, the comer detector was derived. The pictures taken in Fengtai Railway Station in Beijing were used to test the method. The experimental results show that when the number of pixels on the test mask is chosen to be 9, best result can be obtained. The comer detector significantly outperforms existing detector in computational efficiency without sacrificing the quality and the method also provides high performance against Poisson noise and Gaussian blur.展开更多
In many decision making tasks,the features and decision are ordinal.Several ordinal classification learning algorithms have been developed in recent years,it is shown that these algorithms are sensitive to noisy sampl...In many decision making tasks,the features and decision are ordinal.Several ordinal classification learning algorithms have been developed in recent years,it is shown that these algorithms are sensitive to noisy samples and do not work in real-world applications.In this work,we propose a new measure of feature quality, called rank mutual information.Then,we design an ordinal decision tree(REOT) construction technique based on rank mutual information.The theoretic and experimental analysis shows that the proposed algorithm is effective.展开更多
Developmental and reproductive toxicity(DART)endpoint entails a toxicological assessment of all developmental stages and reproductive cycles of an organism.In silico tools to predict DART will provide a method to asse...Developmental and reproductive toxicity(DART)endpoint entails a toxicological assessment of all developmental stages and reproductive cycles of an organism.In silico tools to predict DART will provide a method to assess this complex toxicity endpoint and will be valuable for screening emerging pollutants as well as for m anaging new chemicals in China.Currently,there are few published DART prediction models in China,but many related research and development projects are in progress.In 2013,WU et al.published an expert rule-based DART decision tree(DT).This DT relies on known chemical structures linked to DART to forecast DART potential of a given chemical.Within this procedure,an accurate DART data interpretation is the foundation of building and expanding the DT.This paper excerpted case studies demonstrating DART data curation and interpretation of four chemicals(including 8-hydroxyquinoline,3,5,6-trichloro-2-pyridinol,thiacloprid,and imidacloprid)to expand the existing DART DT.Chemicals were first selected from the database of Solid Waste and Chemicals Management Center,Ministry of Ecology and Environment(MEESCC)in China.The structures of these 4 chemicals were analyzed and preliminarily grouped by chemists based on core structural features,functional groups,receptor binding property,metabolism,and possible mode of actions.Then,the DART conclusion was derived by collecting chemical information,searching,integrating,and interpreting DART data by the toxicologists.Finally,these chemicals were classified into either an existing category or a new category via integrating their chemical features,DART conclusions,and biological properties.The results showed that 8-hydroxyquinoline impacted estrous cyclicity,s exual organ weights,and embryonal development,and 3,5,6-trichloro-2-pyridinol caused central nervous system(CNS)malformations,which were added to an existing subcategory 8e(aromatic compounds with multi-halogen and nitro groups)of the DT.Thiacloprid caused dystocia and fetal skeletal malformation,and imidacloprid disrupted the endocrine system and male fertility.They both contain 2-chloro-5-methylpyridine substituted imidazolidine c yclic ring,which were expected to create a new category of neonicotinoids.The current work delineates a t ransparent process of curating toxicological data for the purpose of DART data interpretation.In the presence of sufficient related structures and DART data,the DT can be expanded by iteratively adding chemicals within the a pplicable domain of each category or subcategory.This DT can potentially serve as a tool for screening emerging pollutants and assessing new chemicals in China.展开更多
It is easy for teenagers to view pornographic pictures on social networks. Many researchers have studied the detection of real pornographic pictures, but there are few studies on those that are artificial. In this wor...It is easy for teenagers to view pornographic pictures on social networks. Many researchers have studied the detection of real pornographic pictures, but there are few studies on those that are artificial. In this work, we studied how to detect artificial pornographic pictures, especially when they are on social networks. The whole detection process can be divided into two stages： feature selection and picture detection. In the feature selection stage, seven types of features that favour picture detection were selected. In the picture detection stage, three steps were included. 1） In order to alleviate the imbalance in the number of artificial pornographic pictures and normal ones, the training dataset of artificial pornographic pictures was expanded. Therefore, the features which were extracted from the training dataset can also be expanded too. 2） In order to reduce the time of feature extraction, a fast method which extracted features based on the proportionally scaled picture rather than the original one was proposed. 3） Three tree models were compared and a gradient boost decision tree （GBDT） was selected for the final picture detection. Three sets of experimental results show that the proposed method can achieve better recognition precision and drastically reduce the time cost of the method.展开更多
Continuous and stable tracking of the ground maneuvering target is a challenging problem due to the complex terrain and high clutter. A collaborative tracking method of the multisensor network is presented for the gro...Continuous and stable tracking of the ground maneuvering target is a challenging problem due to the complex terrain and high clutter. A collaborative tracking method of the multisensor network is presented for the ground maneuvering target in the presence of the detection blind zone(DBZ). First, the sensor scheduling process is modeled within the partially observable Markov decision process(POMDP) framework. To evaluate the target tracking accuracy of the sensor, the Fisher information is applied to constructing the reward function. The key of the proposed scheduling method is forecasting and early decisionmaking. Thus, an approximate method based on unscented sampling is presented to estimate the target state and the multi-step scheduling reward over the prediction time horizon. Moreover, the problem is converted into a nonlinear optimization problem, and a fast search algorithm is given to solve the sensor scheduling scheme quickly. Simulation results demonstrate the proposed nonmyopic scheduling method(Non-MSM) has a better target tracking accuracy compared with traditional methods.展开更多
Studies on ballistic penetration to laminates is complicated,but important for design effective protection of structures.Experimental means of study is expensive and can often be dangerous.Numerical simulation has bee...Studies on ballistic penetration to laminates is complicated,but important for design effective protection of structures.Experimental means of study is expensive and can often be dangerous.Numerical simulation has been an excellent supplement,but the computation is time-consuming.Main aim of this thesis was to develop and test an effective tool for real-time prediction of projectile penetrations to laminates by training a neural network and a decision tree regression model.A large number of finite element models were developed;the residual velocities of projectiles from finite element simulations were used as the target data and processed to produce sufficient number of training samples.Study focused on steel 4340tpolyurea laminates with various configurations.Four different 3D shapes of the projectiles were modeled and used in the training.The trained neural network and decision tree model was tested using independently generated test samples using finite element models.The predicted projectile velocity values using the trained machine learning models are then compared with the finite element simulation to verify the effectiveness of the models.Additionally,both models were trained using a published experimental data of projectile impacts to predict residual velocity of projectiles for the unseen samples.Performance of both the models was evaluated and compared.Models trained with Finite element simulation data samples were found capable to give more accurate predication,compared to the models trained with experimental data,because finite element modeling can generate much larger training set,and thus finite element solvers can serve as an excellent teacher.This study also showed that neural network model performs better with small experimental dataset compared to decision tree regression model.展开更多
In machine learning,randomness is a crucial factor in the success of ensemble learning,and it can be injected into tree-based ensembles by rotating the feature space.However,it is a common practice to rotate the featu...In machine learning,randomness is a crucial factor in the success of ensemble learning,and it can be injected into tree-based ensembles by rotating the feature space.However,it is a common practice to rotate the feature space randomly.Thus,a large number of trees are required to ensure the performance of the ensemble model.This random rotation method is theoretically feasible,but it requires massive computing resources,potentially restricting its applications.A multimodal genetic algorithm based rotation forest(MGARF)algorithm is proposed in this paper to solve this problem.It is a tree-based ensemble learning algorithm for classification,taking advantage of the characteristic of trees to inject randomness by feature rotation.However,this algorithm attempts to select a subset of more diverse and accurate base learners using the multimodal optimization method.The classification accuracy of the proposed MGARF algorithm was evaluated by comparing it with the original random forest and random rotation ensemble methods on 23 UCI classification datasets.Experimental results show that the MGARF method outperforms the other methods,and the number of base learners in MGARF models is much fewer.展开更多
Sheet metal is widely used on auto-bodies, plane-bodies and metal furniture, etc. For instance, a typical auto-body commonly consists of hundreds of sheet metal stamping parts. Because of its complexity of structure a...Sheet metal is widely used on auto-bodies, plane-bodies and metal furniture, etc. For instance, a typical auto-body commonly consists of hundreds of sheet metal stamping parts. Because of its complexity of structure and manufacturing process, auto-bodies inevitably have geometrical variation results from a number of different sources, such as the geometrical variation of stamping parts, the transformation of assembly process parameters and even the improper design concept. As more than 30% quality defects of an auto-body are born from the dimensional deviation of Body-In-White originated during the manufacturing process, effective diagnosis and control of dimensional faults are essential to the continuous improvement of the quality of vehicles. Especially during the period of new car launching or model changing when the assembly process was changed and adjusted frequently. For continuously improving the quality of modern cars, rapid dimensional variation causes identification becomes a challenging but essential work. In this paper, main variation causes of auto-body was firstly been cataloged and analyzed, then, a dimensional variation diagnostic reasoning and decision approach was developed through the combination of data mining and knowledge discovery techniques. This approach is driven by variation pattern identification which can be discovered from the dispersive, isolated massive measured data: Correlation Analysis (CA) and Maximal Tree (MT) methods were applied to extract the large variation group from massive multidimensional measured data, while multivariate statistical analysis (MSA) approach was used to discovery the principle variation pattern. A Decision Tree (DT) approach based on the knowledge of product and assembly process was developed to fulfill the "Hypothesis and Validation" characterized variation causes reasoning procedure. An practical application case with sudden and severe dimension variation on rear end panel in up/down direction was analyzed and successfully solved aided by the devloped variation diagnostic method, which have proved that the approach is effective and efficient.展开更多
The present article outlines progress made in designing an intelligent information system for automatic management and knowledge discovery in large numeric and scientific databases, with a validating application to th...The present article outlines progress made in designing an intelligent information system for automatic management and knowledge discovery in large numeric and scientific databases, with a validating application to the CAST-NEONS environmental databases used for ocean modeling and prediction. We describe a discovery-learning process (Automatic Data Analysis System) which combines the features of two machine learning techniques to generate sets of production rules that efficiently describe the observational raw data contained in the database. Data clustering allows the system to classify the raw data into meaningful conceptual clusters, which the system learns by induction to build decision trees, from which are automatically deduced the production rules.展开更多
A general classification algorithm of neural networks is unable to obtain satisfied results because of the uncertain problems existing among the features in moot classification programs, such as interaction. With new ...A general classification algorithm of neural networks is unable to obtain satisfied results because of the uncertain problems existing among the features in moot classification programs, such as interaction. With new features constructed by optimizing decision trees of examples, the input of neural networks is improved and an optimized classification algorithm based on neural networks is presented. A concept of dispersion of a classification program is also introduced too in this paper. At the end of the paper, an analysis is made with an example.展开更多
基金supported by the Major Projects for Science and Technology Innovation 2030(2018AAA0100805).
文摘To address the confrontation decision-making issues in multi-round air combat,a dynamic game decision method is proposed based on decision tree for the confrontation of unmanned aerial vehicle(UAV)air combat.Based on game the-ory and the confrontation characteristics of air combat,a dynamic game process is constructed including the strategy sets,the situation information,and the maneuver decisions for both sides of air combat.By analyzing the UAV’s flight dyna-mics and the both sides’information,a payment matrix is estab-lished through the situation advantage function,performance advantage function,and profit function.Furthermore,the dynamic game decision problem is solved based on the linear induction method to obtain the Nash equilibrium solution,where the decision tree method is introduced to obtain the optimal maneuver decision,thereby improving the situation advantage in the next round of confrontation.According to the analysis,the simulation results for the confrontation scenarios of multi-round air combat are presented to verify the effectiveness and advan-tages of the proposed method.
文摘System upgrades in unmanned systems have made Unmanned Aerial Vehicle(UAV)-based patrolling and monitoring a preferred solution for ocean surveillance.However,dynamic environments and large-scale deployments pose significant challenges for efficient decision-making,necessitating a modular multiagent control system.Deep Reinforcement Learning(DRL)and Decision Tree(DT)have been utilized for these complex decision-making tasks,but each has its limitations:DRL is highly adaptive but lacks interpretability,while DT is inherently interpretable but has limited adaptability.To overcome these challenges,we propose the Adaptive Interpretable Decision Tree(AIDT),an evolutionary-based algorithm that is both adaptable to diverse environmental settings and highly interpretable in its decision-making processes.We first construct a Markov decision process(MDP)-based simulation environment using the Cooperative Submarine Search task as a representative scenario for training and testing the proposed method.Specifically,we use the heat map as a state variable to address the issue of multi-agent input state proliferation.Next,we introduce the curiosity-guiding intrinsic reward to encourage comprehensive exploration and enhance algorithm performance.Additionally,we incorporate decision tree size as an influence factor in the adaptation process to balance task completion with computational efficiency.To further improve the generalization capability of the decision tree,we apply a normalization method to ensure consistent processing of input states.Finally,we validate the proposed algorithm in different environmental settings,and the results demonstrate both its adaptability and interpretability.
基金supported by the National Natural Science Foundation of China (60604021 60874054)
文摘To solve the multi-class fault diagnosis tasks, decision tree support vector machine （DTSVM）, which combines SVM and decision tree using the concept of dichotomy, is proposed. Since the classification performance of DTSVM highly depends on its structure, to cluster the multi-classes with maximum distance between the clustering centers of the two sub-classes, genetic algorithm is introduced into the formation of decision tree, so that the most separable classes would be separated at each node of decisions tree. Numerical simulations conducted on three datasets compared with ＂one-against-all＂ and ＂one-against-one＂ demonstrate the proposed method has better performance and higher generalization ability than the two conventional methods.
文摘Define and theory of autocorrelation decision tree （ADT） is introduced. In spatial data mining, spatial parallel query are very expensive operations. A new parallel algorithm in terms of autocorrelation decision tree is presented. And the new method reduces CPU- and I/O-time and improves the query efficiency of spatial data. For dynamic load balancing, there are better control and optimization. Experimental performance comparison shows that the improved algorithm can obtain a optimal accelerator with the same quantities of processors. There are more completely accesses on nodes. And an individual implement of intelligent information retrieval for spatial data mining is presented.
基金Project(J2008X011) supported by the Natural Science Foundation of Ministry of Railway and Tsinghua University,China
文摘A high-speed comer detection algorithm based on fuzzy ID3 decision tree was proposed. In the algorithm, the Bresenham circle with 3-pixel radius was used as the test mask, overlapping the candidate comers with the nucleus. Connected pixels on the circle were applied to compare the intensity value with the nucleus, with the membership function used to give the fuzzy result. The pixel with maximum information gain was chosen as the parent node to build a binary decision tree. Thus, the comer detector was derived. The pictures taken in Fengtai Railway Station in Beijing were used to test the method. The experimental results show that when the number of pixels on the test mask is chosen to be 9, best result can be obtained. The comer detector significantly outperforms existing detector in computational efficiency without sacrificing the quality and the method also provides high performance against Poisson noise and Gaussian blur.
基金supported by National Natural Science Foundation of China under Grant 60703013 and 10978011Key Program of National Natural Science Foundation of China under Grant 60932008+1 种基金National Science Fund for Distinguished Young Scholars under Grant 50925625China Postdoctoral Science Foundation.
文摘In many decision making tasks,the features and decision are ordinal.Several ordinal classification learning algorithms have been developed in recent years,it is shown that these algorithms are sensitive to noisy samples and do not work in real-world applications.In this work,we propose a new measure of feature quality, called rank mutual information.Then,we design an ordinal decision tree(REOT) construction technique based on rank mutual information.The theoretic and experimental analysis shows that the proposed algorithm is effective.
文摘Developmental and reproductive toxicity(DART)endpoint entails a toxicological assessment of all developmental stages and reproductive cycles of an organism.In silico tools to predict DART will provide a method to assess this complex toxicity endpoint and will be valuable for screening emerging pollutants as well as for m anaging new chemicals in China.Currently,there are few published DART prediction models in China,but many related research and development projects are in progress.In 2013,WU et al.published an expert rule-based DART decision tree(DT).This DT relies on known chemical structures linked to DART to forecast DART potential of a given chemical.Within this procedure,an accurate DART data interpretation is the foundation of building and expanding the DT.This paper excerpted case studies demonstrating DART data curation and interpretation of four chemicals(including 8-hydroxyquinoline,3,5,6-trichloro-2-pyridinol,thiacloprid,and imidacloprid)to expand the existing DART DT.Chemicals were first selected from the database of Solid Waste and Chemicals Management Center,Ministry of Ecology and Environment(MEESCC)in China.The structures of these 4 chemicals were analyzed and preliminarily grouped by chemists based on core structural features,functional groups,receptor binding property,metabolism,and possible mode of actions.Then,the DART conclusion was derived by collecting chemical information,searching,integrating,and interpreting DART data by the toxicologists.Finally,these chemicals were classified into either an existing category or a new category via integrating their chemical features,DART conclusions,and biological properties.The results showed that 8-hydroxyquinoline impacted estrous cyclicity,s exual organ weights,and embryonal development,and 3,5,6-trichloro-2-pyridinol caused central nervous system(CNS)malformations,which were added to an existing subcategory 8e(aromatic compounds with multi-halogen and nitro groups)of the DT.Thiacloprid caused dystocia and fetal skeletal malformation,and imidacloprid disrupted the endocrine system and male fertility.They both contain 2-chloro-5-methylpyridine substituted imidazolidine c yclic ring,which were expected to create a new category of neonicotinoids.The current work delineates a t ransparent process of curating toxicological data for the purpose of DART data interpretation.In the presence of sufficient related structures and DART data,the DT can be expanded by iteratively adding chemicals within the a pplicable domain of each category or subcategory.This DT can potentially serve as a tool for screening emerging pollutants and assessing new chemicals in China.
基金Projects(61573380,61303185) supported by the National Natural Science Foundation of ChinaProjects(2016M592450,2017M612585) supported by the China Postdoctoral Science FoundationProjects(2016JJ4119,2017JJ3416) supported by the Hunan Provincial Natural Science Foundation of China
文摘It is easy for teenagers to view pornographic pictures on social networks. Many researchers have studied the detection of real pornographic pictures, but there are few studies on those that are artificial. In this work, we studied how to detect artificial pornographic pictures, especially when they are on social networks. The whole detection process can be divided into two stages： feature selection and picture detection. In the feature selection stage, seven types of features that favour picture detection were selected. In the picture detection stage, three steps were included. 1） In order to alleviate the imbalance in the number of artificial pornographic pictures and normal ones, the training dataset of artificial pornographic pictures was expanded. Therefore, the features which were extracted from the training dataset can also be expanded too. 2） In order to reduce the time of feature extraction, a fast method which extracted features based on the proportionally scaled picture rather than the original one was proposed. 3） Three tree models were compared and a gradient boost decision tree （GBDT） was selected for the final picture detection. Three sets of experimental results show that the proposed method can achieve better recognition precision and drastically reduce the time cost of the method.
基金supported by the National Defense Pre-Research Foundation of China(0102015012600A2203)。
文摘Continuous and stable tracking of the ground maneuvering target is a challenging problem due to the complex terrain and high clutter. A collaborative tracking method of the multisensor network is presented for the ground maneuvering target in the presence of the detection blind zone(DBZ). First, the sensor scheduling process is modeled within the partially observable Markov decision process(POMDP) framework. To evaluate the target tracking accuracy of the sensor, the Fisher information is applied to constructing the reward function. The key of the proposed scheduling method is forecasting and early decisionmaking. Thus, an approximate method based on unscented sampling is presented to estimate the target state and the multi-step scheduling reward over the prediction time horizon. Moreover, the problem is converted into a nonlinear optimization problem, and a fast search algorithm is given to solve the sensor scheduling scheme quickly. Simulation results demonstrate the proposed nonmyopic scheduling method(Non-MSM) has a better target tracking accuracy compared with traditional methods.
文摘Studies on ballistic penetration to laminates is complicated,but important for design effective protection of structures.Experimental means of study is expensive and can often be dangerous.Numerical simulation has been an excellent supplement,but the computation is time-consuming.Main aim of this thesis was to develop and test an effective tool for real-time prediction of projectile penetrations to laminates by training a neural network and a decision tree regression model.A large number of finite element models were developed;the residual velocities of projectiles from finite element simulations were used as the target data and processed to produce sufficient number of training samples.Study focused on steel 4340tpolyurea laminates with various configurations.Four different 3D shapes of the projectiles were modeled and used in the training.The trained neural network and decision tree model was tested using independently generated test samples using finite element models.The predicted projectile velocity values using the trained machine learning models are then compared with the finite element simulation to verify the effectiveness of the models.Additionally,both models were trained using a published experimental data of projectile impacts to predict residual velocity of projectiles for the unseen samples.Performance of both the models was evaluated and compared.Models trained with Finite element simulation data samples were found capable to give more accurate predication,compared to the models trained with experimental data,because finite element modeling can generate much larger training set,and thus finite element solvers can serve as an excellent teacher.This study also showed that neural network model performs better with small experimental dataset compared to decision tree regression model.
基金Project(61603274)supported by the National Natural Science Foundation of ChinaProject(2017KJ249)supported by the Research Project of Tianjin Municipal Education Commission,China。
文摘In machine learning,randomness is a crucial factor in the success of ensemble learning,and it can be injected into tree-based ensembles by rotating the feature space.However,it is a common practice to rotate the feature space randomly.Thus,a large number of trees are required to ensure the performance of the ensemble model.This random rotation method is theoretically feasible,but it requires massive computing resources,potentially restricting its applications.A multimodal genetic algorithm based rotation forest(MGARF)algorithm is proposed in this paper to solve this problem.It is a tree-based ensemble learning algorithm for classification,taking advantage of the characteristic of trees to inject randomness by feature rotation.However,this algorithm attempts to select a subset of more diverse and accurate base learners using the multimodal optimization method.The classification accuracy of the proposed MGARF algorithm was evaluated by comparing it with the original random forest and random rotation ensemble methods on 23 UCI classification datasets.Experimental results show that the MGARF method outperforms the other methods,and the number of base learners in MGARF models is much fewer.
基金supported by the National Natural Science Foundation of China(6137301761402241+4 种基金615722606157226161472192)the Scientific&Technological Support Project of Jiangsu Province(BE2014718BE2015702)
文摘Sheet metal is widely used on auto-bodies, plane-bodies and metal furniture, etc. For instance, a typical auto-body commonly consists of hundreds of sheet metal stamping parts. Because of its complexity of structure and manufacturing process, auto-bodies inevitably have geometrical variation results from a number of different sources, such as the geometrical variation of stamping parts, the transformation of assembly process parameters and even the improper design concept. As more than 30% quality defects of an auto-body are born from the dimensional deviation of Body-In-White originated during the manufacturing process, effective diagnosis and control of dimensional faults are essential to the continuous improvement of the quality of vehicles. Especially during the period of new car launching or model changing when the assembly process was changed and adjusted frequently. For continuously improving the quality of modern cars, rapid dimensional variation causes identification becomes a challenging but essential work. In this paper, main variation causes of auto-body was firstly been cataloged and analyzed, then, a dimensional variation diagnostic reasoning and decision approach was developed through the combination of data mining and knowledge discovery techniques. This approach is driven by variation pattern identification which can be discovered from the dispersive, isolated massive measured data: Correlation Analysis (CA) and Maximal Tree (MT) methods were applied to extract the large variation group from massive multidimensional measured data, while multivariate statistical analysis (MSA) approach was used to discovery the principle variation pattern. A Decision Tree (DT) approach based on the knowledge of product and assembly process was developed to fulfill the "Hypothesis and Validation" characterized variation causes reasoning procedure. An practical application case with sudden and severe dimension variation on rear end panel in up/down direction was analyzed and successfully solved aided by the devloped variation diagnostic method, which have proved that the approach is effective and efficient.
文摘The present article outlines progress made in designing an intelligent information system for automatic management and knowledge discovery in large numeric and scientific databases, with a validating application to the CAST-NEONS environmental databases used for ocean modeling and prediction. We describe a discovery-learning process (Automatic Data Analysis System) which combines the features of two machine learning techniques to generate sets of production rules that efficiently describe the observational raw data contained in the database. Data clustering allows the system to classify the raw data into meaningful conceptual clusters, which the system learns by induction to build decision trees, from which are automatically deduced the production rules.
文摘A general classification algorithm of neural networks is unable to obtain satisfied results because of the uncertain problems existing among the features in moot classification programs, such as interaction. With new features constructed by optimizing decision trees of examples, the input of neural networks is improved and an optimized classification algorithm based on neural networks is presented. A concept of dispersion of a classification program is also introduced too in this paper. At the end of the paper, an analysis is made with an example.
