【作者】 杨欣；

【导师】 严洪森；

【作者基本信息】 东南大学 ， 控制理论与控制工程， 2015， 硕士

【摘要】 随着科学技术的不断发展,高新技术武器大量运用于海上军事作战领域,同时海洋环境要素也极大地影响着武器效能的发挥。研究在特定气象水文要素环境下武器作战效能的评估问题,对于提高武器作战水平,更好地辅助军事行动决策具有重要意义。另外数据库建立初期,武器作战信息匮乏,解决这种情况下的武器效能评估问题,同时提高模型的评估精度和参考性,是很多学者的共同研究目标。为此,本文对这些问题展开了一系列研究。支持向量回归机(SVR)能较好地解决武器效能评估中出现的非线性、高维度和小样本等问题,但是武器效能受作战环境要素影响显著且不同的要素影响程度不尽相同。为了充分利用环境要素重要性差异,提高评估精度,本文提出了一种特征加权支持向量回归机(WSVR)评估模型。该模型采用基于G1法的主观法与灰色关联分析的客观法的组合权重方法来获取环境要素权重并据此改进了SVR模型,给出了算法实现步骤。最后基于实例验证并与其他模型的对比分析,来验证模型的性能。为进一步提高模型评估精度,有必要对模型实施参数优化。为了克服标准粒子群算法(PSO)易陷入局部极值、后期收敛速度慢、精度差等缺点,提高PSO算法的全局收敛能力,本文综合模拟退火(SA)算法和混沌(C)思想,提出了一种改进的混沌模拟退火粒子群(CSA-PSO)算法,并给出了算法的流程图及实现步骤。然后,基于实例分析来测试验证算法的性能,并对WSVR评估模型的参数进行了优化。最后,针对数据信息缺乏的武器效能评估问题,本文引入衍生推理机制。利用与待评估实例具有较高相似度的一种或多种武器的样本数据,从通用模型库中自适应地选择最合适的模型,通过样本训练得到该武器专用评估模型。为了提高模型的评估精度和可靠性,本文引入了评估模型的自学习机制,按照一定规则进行模型的二次训练,并更新专用模型以完成模型自学习。本文最后,基于UML建模和.NET三层架构模式,完成了武器效能衍生推理和模型自学习功能模块的设计与实现。更多还原

【Abstract】 With the development of science and technology, a large number of high-tech weapons used in naval warfare field, marine environment elements, at the same time also greatly affect the weapon effectiveness. Research on operational effectiveness of naval weapons under the specific meteorological and hydrographic elements of the marine environment, is of great significance to raise the operative level of weapons and to better support military operations decisions. In addition, at the early of database established, weapon operation is short of information. It is a common research goal of many scholars to solve the weapon effectiveness evaluation problem under this kind of circumstance, at the same time, to improve the evaluation precision of the model and the reference. So the thesis conducts a series of research on these problems.Support vector regression machine can well deal with weapons in the effectiveness evaluation of small samples, nonlinear and high dimension problems, but operational environment elements significantly affect the weapon effectiveness and the influence degree of different elements are not nearly the same. In order to take full advantage of the importance of differences of the environmental elements and to improve the evaluation accuracy, the thesis proposes an elements weighted support vector regression machine (WSVR) evaluation model. The model uses a combination weighting method based on G1 subjective method and grey relation analysis objective method to get the weights of environmental elements and improves the SVR model, and the algorithm implementation steps are given. At the end, the thesis utilizes example analysis and comparing with other models to verify the performance of the model.In order to improve the evaluation precision of the model, it is necessary to execute model parameters optimization. To overcome deficiencies of standard particle swarm optimization (PSO) algorithm, such as easily being lost in local optimum, the slow evolutionary convergence speed and poor search accuracy and so on, and to improve the global convergence of the PSO algorithm, In the thesis, combining simulated annealing (SA) algorithm and the chaotic (C) theory, an improved chaotic simulated annealing particle swarm algorithm(CSA-PSO) is proposed, and the flowchart and realization steps of the algorithm is provided. Then, the thesis test and validate the performance of algorithms based on the example analysis and uses the algorithm to optimize the parameters of WSVR model.At last, in view of effective evaluation problem of weapons which are short of data information, the thesis introduces derivative reasoning mechanism. Using the sample data of one or more other weapons having a high similarity with the weapon to be evaluated, and adaptively choosing the most suitable model from the general model library, the special evaluation model of weapon is obtained by training the model with the sample. In order to improve the evaluation precision and reliability, the thesis introduces self-learning mechanism of the evaluation model, executing secondary training of the model according to certain rules and updating the special model to complete the self-learning. At the final of the thesis, based on the UML modeling and.Net three-layer architecture model, the function module design and implementation of the derivative reasoning of weapon efficiency and self-learning of the model have been completed.更多还原