Sea-battlefield Situation Assessment based on Improved Decision Tree

Transcription

1 Applied Mechanics and Materials Submitted: ISSN: , Vol. 574, pp Accepted: doi: / Online: Trans Tech Publications, Switzerland Sea-battlefield Situation Assessment based on Improved Decision Tree Yong-tao YU 1, a, Ying Ding 2 1 Naval Academy of Armament, Beijing, china, Beijing Sino-Ocean JIYE Propetry Management CO., LTD, Beijing, china, a y_yt128@aliyun.com Keywords: Decision tree algorithm Sea-battlefield Situation assessment Abstract. How to efficiently evaluate the dynamic changing sea-battlefield is the key of command decision. According to research sea-battlefield situation assessment based on improved decision tree algorithm based on derived attributes. First is based on the decision tree algorithm to establish the sea-battlefield situation assessment initial decision tree. Second are dynamically generated derivative branches set based on derived attributes. Once again, it can be grafting and pruning derivative branches to form the sea-battlefield situation assessment derived decision tree model. Finally, it may use a derived dynamically decision tree model assess sea-battlefield in different time slices. Introduction Sea-battlefield situation assessment is the basic means of commanders cognition battlefield, is an important support of commanders make decisions. Because sea-battlefield is broad, complex and changing environment, combat unit types is complicated and contains a lot of uncertainties and probabilistic object (such as the various combat formations).so sea-battlefield assessment related to a variety of uncertainties. As technology advances, new types of weapons emerging cause the types of sea-battle object increases, while situational awareness means continue to enrich cause the information of sea-battlefield situation is expanding. Therefore sea-battlefield assessment is a dynamic process. In addition a number of different types of sea-battle objects performing their operations and achieving their operational intentions. Sea-battlefield situation assessment requires parallel processing the deep impact of the uncertainty caused by the massive nonlinear potential and random phenomenon. By multi-attribute decision tree algorithm is selected for the purpose of large amounts of data classification, information extraction potential decisions for dealing with complex nonlinear input/output relationship problems. The sea-battlefield situation assessment is typical of complex nonlinear input/output relationship problems. But as time goes on, the types and properties of sea-battle objects may continue to appear and die, fighting intentions and actions of sea-battle objects are also subject to frequent changes occur. As the model has been fixed before assessment, the conventional decision tree algorithm is difficult to describe these new types of objects or new evaluation attribute. It must be give up the decision tree model has been established, re-establish the decision tree model for every change of the sea-battlefield. So we improved decision tree algorithm by introducing derived attributes. In the situation assessment process, we may be subject to changes in sea battle, based on the derived attributes immediate to establish new derivative branches, used to characterize the dynamic classification rules and potential relationships. By grafting and pruning derivative branches on the initial decision tree, form a new tree, dynamic and parallel assessment of sea battlefield. All rights reserved. No part of contents of this paper may be reproduced or transmitted in any form or by any means without the written permission of Trans Tech Publications, (# , Pennsylvania State University, University Park, USA-18/09/16,15:38:50)

2 640 Recent Research on Mechanical Engineering, Mechatronics and Automation Basic Concept Decision Tree Algorithm Goal of decision tree algorithm is the rules in cases never deduce classification rules, and manifested in the form of a tree structure diagram. Decision tree obey the rules of recursion. And it can to determine the following branches depending on the value for the property, and ultimately gets the decision tree classification conclusions on the leaf nodes. Therefore, the path from the tree root began arriving along each leaf node corresponds with a classification rules. A decision tree corresponds to a set of expression rules. Decision Tree Algorithm Based on Derived Attributes Essence of improved decision tree algorithm based on derived attributes is according to dynamic evaluated objects, constantly introducing new branches based on derived attributes in traditional decision tree model, the formation of a new decision tree after pruning. Its use process: The first is to establish the initial decision tree model. The initial decision tree model can be applied to traditional methods (such as ID3 and C4.5 algorithms) to build. Determine the situation Assessment properties of the tree root, based on the rest of the property and recursive algorithm grown child nodes, situation assessment of all property until the application. The second is creating derivative branches set based on the derived attributes. Derivative branches can be divided into two categories. A class called new value branch. Causes of such branches are abundant the expected amount of information of original nodes. Layers of classification rules that are fixed, but the increase branch of the original layer. Another called new attribute branches. Causes of such branches are to increase the node which decisions tree does not have. That classification rules to increase the number of layers, but the same branch of the original layer. Create derivative branches to determine the attribute of root nodes, the application of derived attributes, recursively growing child nodes of branches. The third is to determine grafted points of derivative branches set. It can match the attributes on each node of decision tree and root nodes of derivative branches. If the property is consistent, these nodes can be identified as grafted point of derivatives branch. The fourth is grafted derivative branches, pruning decision tree, forming a new decision tree. It can be added to the corresponding derivative branches in the graft points, the formation of a derived decision tree. Because there may be a plurality of grafted points on the same classified branch, it must determine the original branches and derivative branches, trim overgrown branches. The judgment rule usually is calculate classification accuracy, information gain or support of original branches and grafted branches. It can be pruned the branches with low accuracy, little information gain or small degree of support. Sea-battlefield Situation Assessment Model Based on Improved Decision Tree Algorithm Building sea-battlefield situation assessment model based on improved decision tree algorithm. The first is the historical experience of the learning process. The historical experience is stored in the decision tree. With the dynamic change in sea-battlefield situation, the new rules immediately grafted onto the original tree and trim accordingly, constantly updated tree. Sea-battlefield situation assessment is searching for a dynamic update of the decision tree, using the existing decision-making experience. Selecting Property of Decision Tree Sea-battlefield situation assessment is closely related in time and geography, type of troops, the relative position, relative distance and combat operations and other factors. Select the decision tree attribute should contain sufficient information to ensure that the basis on as succinct as possible. Performance of weapons determines the threat of the submarine. Based on an in-depth analysis, we

3 Applied Mechanics and Materials Vol selected two indicators of the weapons comprehensive performance factors as property of decision tree, namely R (IFF spacing, characterization submarine weapons range) and A (the azimuth of submarine, characterization submarine weapon firing angle). At t=0 time slice of the sea-battlefield, the submarine equipped with two types of weapons, acoustic homing torpedo (AHT) and wire-guided torpedo (). Specific assessment attributes values as shown in Table 1. Table 1. t=0 relationship between assessment attributes and new weapon systems Acoustic homing torpedo AHT Wire-guided torpedo Establish the Initial Decision Tree R (IFF spacing) Effective range A (the azimuth of submarine) Firing angle ± First, determine the root nodes of the decision tree. According to information gain values and information gain ratio of decision attributes, attributes IFF spacing (R) has the maximum rate of information gain, can be identified as the root nodes of the decision tree. Second is to comply with the recursive rules generated tree. Select the azimuth of submarine (A) as an internal nodes, according to IFF spacing (R) of the desired amount of information to build decision branches. Finally, the tactics of ships against submarine identified as the terminal nodes of decision tree. Eventually build sea-battlefield situation assessment initial decision tree model is shown in Fig.1. Fig. 1. Sea-battlefield situation assessment initial decision tree model Fig. 2. Sea-battle field situation assessment derivative branches set

4 642 Recent Research on Mechanical Engineering, Mechatronics and Automation Creating Derivative Branches Set When the dynamic evolution of a new generation of sea-battlefield situation, it can be generated derivative branches according to the derived attributes of situation assessment. At t=1 time slice of the sea-battlefield, new submarine weapons appeared that submarine-to-ship missile (SSM) and backwash homing torpedo (). It can be rich the desired amount of information of original properties node IFF spacing (R) and the azimuth of submarine (A) to generate corresponding new value derivative branches. Specific weapon systems values as shown in Table 2. Table 2. t=1 relationship between assessment attributes and new weapon systems R (IFF spacing) Effective range A (the azimuth of submarine) Firing angle Submarine-to-ship missile SSM Backwash homing torpedo 0-25 ±90±30 In the basis on t=1 time slice, t=2 time slices of sea-battlefield enhance target detection means that with precise directional sonar ability to obtain accurate information about the submarine depth. It can be increasing the new node of H (the depth of submarine), to generate the corresponding new properties derivative branches. Specific assessment attributes values as shown in Table 3. Table 3. t=2 relationship between new assessment attributes and weapon systems H (the depth of submarine ) Shooting depth Submarine-to-sh ip missile SSM Wire-guided torpedo Backwash homing torpedo Acoustic homing torpedo AHT According to derived attributes generated derivative branches set is shown in Fig.2. Grafted Derivative Branches Set It can be grafted derivative branches set based on derived attributes on sea-battlefield situation assessment initial decision tree, as shown in Fig.3. Fig. 3. Sea-battlefield situation assessment derived decision tree model after grafting

5 Applied Mechanics and Materials Vol Pruning Derived Decision Tree Based on the classification accuracy, information gain or support, it can pruning the sea-battlefield assessments derived tree, as shown in Fig.4. Fig. 4. Sea-battlefield situation assessment derived decision tree model after pruning Sea-battlefield Situation Assessment Application Examples Below we take submarine threat of sea-battlefield situation assessment as an example to illustrate the application of decision tree algorithm based on derived attributes. Here chose the 13 sea-battlefield samples for assessment. Specific sample information and situation assessment conclusions values as shown in Table 4.

6 644 Recent Research on Mechanical Engineering, Mechatronics and Automation Table 4. Sample information and situation assessment conclusions Sample information Assessment conclusions No R A H IFF The azimuth of The depth of t=0 t=1 t=2 spacing submarine submarine No threat No threat No threat SSM No threat No threat AHT No threat AHT AHT SSM AHT SSM AHT SSM SSM SSM SSM AHT No threat SSM AHT SSM SSM AHT Analysis sea-battlefield situation assessment derived decision tree model to assess value available: For sample 6. At t=0 time slice, IFF spacing(r) and the azimuth of submarine(a) meet firing conditions of wire-guided torpedo() and acoustic homing torpedo(aht), to determine the submarine impose wire-guided torpedo() and acoustic homing torpedo(aht) attack. Assessment conclusion is that the ship be taken to avoid wire-guided torpedo () and acoustic homing torpedo (AHT) maneuver. At t=1 time slice, submarine weapon types increased. But IFF spacing(r) does not meet firing conditions of submarine-to-ship missile (SSM), the azimuth of submarine (A) does not meet firing conditions of backwash homing torpedo (), to determine the submarine still impose wire-guided torpedo() and acoustic homing torpedo(aht) attack. Threat remains unchanged. Assessment conclusion reaffirm the ship be taken to avoid the wire-guided torpedo and acoustic homing torpedo maneuver. At t=2 time slice, target detection means richen. The depth of submarine (H) does not meet firing conditions of wire-guided torpedo (), to determine the submarine can impose acoustic homing torpedo (AHT) attack. The threat is reduced. Assessment conclusion is that the ship just takes avoid acoustic homing torpedo (AHT) maneuver. For sample 3. At t=0 time slice, IFF spacing(r) and the azimuth of submarine(a) meet firing conditions of wire-guided torpedo(), to determine the submarine impose wire-guided torpedo() attack. Assessment conclusion is that the ship be taken to avoid wire-guided torpedo () maneuver. At t=1 time slice, submarine weapon types increased. IFF spacing(r) and the azimuth of submarine (A) meet firing conditions of submarine-to-ship missile (SSM), to determine the submarine impose wire-guided torpedo() and submarine-to-ship missile (SSM) attack. The threat is increases. Assessment conclusion is that the ship be taken to avoid the wire-guided torpedo () and submarine-to-ship missile (SSM) maneuver. At t=2 time slice, target detection means richen. The depth of submarine (H) does not meet firing conditions of submarine weapon systems, to determine the submarine can not attack. The threat is eliminate. Assessment conclusion is that the ship maintenance normal sailing.

7 Applied Mechanics and Materials Vol For sample 5. At t=0 time slice, IFF spacing(r) and the azimuth of submarine(a) does not meet firing conditions of submarine weapon systems, to determine the submarine can no attack. There is no direct threat. Assessment conclusion is that the ship maintenance normal sailing. At t=1 time slice, submarine weapon types increased. IFF spacing(r) and the azimuth of submarine (A) meet firing conditions of submarine-to-ship missile (SSM), to determine the submarine impose submarine-to-ship missile (SSM) attack. The threat is increases. Assessment conclusion is that the ship be taken to avoid the submarine-to-ship missile (SSM) maneuver. At t=2 time slice, target detection means richen. The depth of submarine (H) meet firing conditions of submarine-to-ship missile (SSM), to determine the submarine impose submarine-to-ship missile (SSM) attack. Threat remains unchanged. Assessment conclusion reaffirm the ship be taken to avoid submarine-to-ship missile (SSM) maneuver. Thus situation assessment conclusions of sea-battlefield assessment derived decision tree model basically credible. Meanwhile decision tree algorithm based on derived attribute can well parallel processing dynamic changes in sea-battlefield, such as the type of object increase and situation information richen, reducing the ambiguity of the sea-battlefield situation assessment. Summary In the sea-battlefield situation assessment can be applied to improve decision tree method. First, the use multi-attribute classification ability of decision tree algorithm can be to solve complex nonlinear input/output relationship problems in sea-battlefield. Then based on the derived attributes immediate introduction of new classification rules to solve parallel processing problems of varying-dimensions data sets, which generated by sea-battlefield objects randomly generated disappearance. In the next step of research will continue to explore the pruning rules of derivative branches set, to improve the efficiency of established the sea-battlefield situation assessment decision tree model. References [1] Marvin S Cohen, Bryan B Thompson, Leonard Adelman., et al. training critical thinking for the battlefield volume 1: basis in cognitive theory and research [M]. Now York: Cognitive Technologies, [2] HINMAN M L.Some computational approaches for situation assessment and impact assessment[c]//proceedings of the 5th International Conference on Information Fusion. New York, USA, 2002: [3] Zkim Le. Fuzzy relation compositions and pattern recognitions. Inf Sci, 1996, 89: [4] Saaty T L. The analysis hierarchy process [J].Management Science, 1986, 32(7), [5] Fried M A, Brodeley C E. Decision Tree Classification of Land Cover from Remotely Sensed Data [J].Remote Sens. Environ, 1997, 61: [6] Wei Fan, E. Greengrass, J. McCloskey, et al. Effective estimation of posterior probabilities: Explaining the accuracy of randomized decision tree approaches[c], In Fifth IEEE International Conference on Data Mining, IEEE, pages [7] Quinlan J R Inductìón of decision trees [J]. Machine Learning.1986, l (l): [8] Murthy S K. Kasif S, Salzberg S. A system for induction of oblique decision trees [1]. Journal of Artificial Intelligence Research, :1-32.