1、The development of an effective methodology for the design of intelligent control systems undoubtedly requires the synthesis of many concepts from artificial intelligence, realtime computing, and control engineering. Three approaches that have the potential for intelligent control are expert systems
2、 as adaptive elements in a control system; fuzzy calculations as decision-producing elements in a control system; and neural networks as compensation elements in control systems. Expert systems, or rule-based systems, embody a rule-based solving paradigm built around if-then rules. When the procedur
3、e works forward from a sequence of if conditions to a sequence of then actions, it is called forward chaining. While the approach may be applicable to the decision-making needs of a control system, it is generally too slow for high-speed systems and it has limited learning capability. Backward chain
4、ing starts with a desired sequence of then actions and works backward to determine whether the appropriate if conditions are met. It may be more appropriate for control system applications because a feedback loop is generated when an error detection and correction capability is added to a backward c
5、haining system.In fuzzy set theory, the membership function assigned to the elements is continuous and lies between zero and unity, which is different with the standard set theory. If the membership function takes on only the extreme values of zero and unity, then fuzzy sets reduce to standard set t
6、heory. Fuzzy set theory is useful in those situations in which data and relationships cannot be written in crisp mathematical terms. For example, an aircraft pilot who states that the rudder is not working correctly is providing significant fuzzy information that should be integrated into the decisi
7、on-making process of the outer control loop even though it can not be characterized by standard mathematical models.3Neural networks provide a powerful approach for developing empirical nonlinear models for a wide variety of physical phenomena. With neural networks, the problem of control can be con
8、sidered as a pattern recognition problem, where the patterns to be recognized are change signals that map into action signal for specified system performance. The intelligent controller should recognize and isolate patterns of change in real time and learn from experience to recognize change more qu
9、ickly, even with incomplete data.We are convinced that these techniques are certainly useful and an important step toward intelligent control, but are not the answer to intelligent control of complex and large-scale systems. In recent years mechanical, electrical, and computer engineering have fused
10、 to the field of mechatronics. Intelligent devices including sensor, actuator, and software are built into intelligent mechatronic components integrating all three aspects naturally in minimum space. The complexity of a mechatronic module is mainly in the control software often referred to as the in
11、telligence of the system.To summarize the proposition for an approach of intelligent control: system and technology integration including mechatronics, computer science, system design optimization, communication, and human interaction may yield what we expect from Intelligent Control. The view of In
12、telligent Control is illustrated in Fig. 14-A-1.Fig. 14-A-1. Key technologies for Intelligent Control systemOne reason to include human interaction and treat it as an essential part to intelligent control system is unfulfilled high expectations of artificial intelligence to create goal-oriented inte
13、lligent behavior.4System design optimization and generation of optimal reference trajectories for complex intelligent control systems is an important issue. There are various powerful numerical optimization tools available and a recent development in semidefinite programming, linear matrix inequalit
14、ies （LMI） and interior point methods has created increasingly efficient algorithms. One important key issue remaining is to improve performance of intelligent control systems iteratively without impairing already learned knowledge. One may call this careful exploration to improve performance from tr
15、ial to trial. It seems still difficult to achieve this with present methodology from adaptive control or newer techniques like artificial neural network or fuzzy approaches.Ongoing research using the intelligent control scheme is pointing in a direction where the continuous control schemes have to b
16、e augmented with a discrete event model and supervisory controller.5New Wordsadaptive dptiv adj. 适应的biology baildi n. 生物学dimensionality di,mennlti n. 维度distributed distribjutid adj. 分布式的，分散式的criteria kraitiri n. 标准，条件（criterion的复数）qualitatively kwliteitivli adv. 定性地，从品质上讲discipline disiplin n. 学科，纪律
17、imitate imiteit vt. 模仿，仿效，仿造，仿制paradigm prdim n. 范例empirical empirikl adj. 根据经验的，经验主义的mechatronic ,mektrnik adj. 机电一体化的mechatronics ,mektrniks n. 机械电子学methodology ,medldi n. 方法学，方法论perceive psi:v vt. 感知，察觉，感觉，理解，认知；vi. 感到，感知proposition ,prpzin n. 命题，建议psychology saikldi n. 心理学recognition ,rekgnin n.
18、 识别spectrum spektrm n. 光谱，频谱，范围stringent strindnt adj. 严厉的，迫切的PhrasesBayesian probability 贝叶斯统计fuzzy logic 模糊逻辑machine learning 机器学习evolutionary computation 进化计算genetic algorithms 遗传算法decision maker 决策者decision space 决策空间noise level 噪声电平cope with 处理，应付neural network 神经元网络a sequence of 一系列feedback lo
19、op 反馈回路membership function 隶属度函数supervisory controller 监督控制AbbreviationsAI （Artificial Intelligence） 人工智能ANN （Artificial Neural Networks） 人工神经网络LMI （linear matrix inequalities） 线性矩阵不等式Notes 1. As a result, over the past four to five decades numerous terms borrowed from psychology and biology, such a
20、s pattern recognition, adaptation, learning and self organization, have been introduced into the systems literature.因此，在过去的四、五十年内，很多心理学、生物学等学科的概念，如模式识别、适应性、学习和自组织，被引入系统论中。2. A good definition for intelligence control is the discipline in which control algorithms are developed by emulating certain ch
21、aracteristics of intelligent biological systems.智能控制的一个恰当的定义是一门学科，控制算法的发展是通过模拟智能生物系统的某些特点来实现的。 3. For example, an aircraft pilot who states that the rudder is not working correctly is providing significant fuzzy information that should be integrated into the decision-making process of the outer cont
22、rol loop even though it can not be characterized by standard mathematical models. 例如，飞机员指出方向舵工作错误是给出的一个重要模糊信息，即使这一信息不能用经典数学模型来描述，也必须被加入外部控制回路的决策过程中。 4. One reason to include human interaction and treat it as an essential part to intelligent control system is unfulfilled high expectations of artifici
23、al intelligence to create goal-oriented intelligent behavior.将与人的交互作用包括在智能控制系统中，并将其看作一个重要的组成部分的一个原因是，人工智能难以实现目标导向的智能行为。5. Ongoing research using the intelligent control scheme is pointing in a direction where the continuous control schemes have to be augmented with a discrete event model and supervisory controller.正在进行中的研究利用智能控制在一个方向上计划，在这个方向上，连续控制计划必须与离散事件模型和监控器相结合。B：Artificial Neural NetworkArtificial neural networks, also called neural networks, are information proces
