Other attributes
Intelligent control is a computational procedure for directing a complex system with incomplete and inadequate representation and under incomplete specifications of how to do so in an uncertain environment toward a certain goal. Intelligent control often combines planning with online error compensation and requires the learning of a system and environment to be part of the control process. Overall, intelligent control systems seek to emulate important characteristics of human intelligence. These characteristics include adaptation and learning, planning under great uncertainty, and coping with large amounts of data. And as intelligent control has developed, it has begun to encompass everything that is not characterized as conventional control. As an interdisciplinary approach, intelligent control combines and extends theories from areas such as control, computer science, operations research, and mathematics, and takes inspiration from biological systems.
Classical control systems require an agent in the process and a designer to construct a mathematical model of the system and the dynamics of a plant that affects controlling it. In this system, the controller of the robotic system is the intelligence.
Whereas, an intelligent control system requires a designer to input system behavior and the intelligent control system abstractly models the system. This is also known as the Lazymans Approach, since the design does not need to know the internal dynamics to be controlled. Further, in an intelligent control system, the intelligence is shifted toward the software controlling the system. The designer must have some knowledge of the system but should not need to develop an accurate model of it for the intelligent control system to work.
The complexity of a controlled object that an intelligently controlled system may deal with includes model uncertainty, high nonlinearity, distributed sensors/actuators, dynamic mutations, multiple time scales, complex information patterns, big data processes, and strict characteristic indicators. To emulate human intelligence to solve these problems, various researchers continue to suggest intelligent control can include expert control, fuzzy control, neural network control, hierarchical intelligent control, anthropomorphic intelligent control, integrated intelligent control, combined intelligent control, chaos control, and wavelet theory, among others. Further, intelligent control systems can try to achieve adaptation and learning, planning under uncertainty, and decision making.
The basic model of the decision-making process of fuzzy control includes fuzzy concept, fuzzy judgment, and fuzzy reasoning. These are also known as the three basic forms of human fuzzy thinking. In a fuzzy controller, the fuzzy concept is a fuzzy linguistic variable represented by a fuzzy set. For example, the exact amount of error is converted to the fuzzy quantity on the discrete domain, also known as fuzzy quantization processing. A fuzzy control system can be summarized into several fuzzy control rules in language, which can be described by a fuzzy relation matrix, which is a general principle of the operating process also known as the language model of the controlled object.
System principle of fuzzy control.
The adaptive fuzzy controller works to make a control strategy described in language by observing and evaluating the performance of the controller. There are two types of adaptive control: direct adaptive control and indirect adaptive control. Direct adaptive control adds an adaptive mechanism to the basic feedback control, which allows the fuzzy control to enable it to adaptively modify the controller parameters to make the control. Indirect adaptive control uses online identification to identify the parameter of an object and use identified parameters to adjust the control parameters to continuously improve the parameter corrector and improve the control performance. The adaptive fuzzy controller is built on a similar structure as the basic fuzzy controller mechanism, with an adaptive mechanism added.
Structure of the adaptive fuzzy controller.
Artificial neural networks are circuits, computer algorithms, and mathematical representations inspired by massively connected sets of neurons that imitate biological neural networks. This offers an alternative computing network that has proven useful in pattern recognition, signal processing, estimate, and control problems. The neural network model offers distributed storage of information, information processing and reasoning with parallelism, information processing with the characteristic of self-organization and self-learning, and a strong non-linear mapping capability from input to output.
In the control system, the non-linear mapping capability can be used to model complex non-linear objects that are difficult to accurately describe, to act as controllers, to optimize calculations, to perform inference, for fault diagnosis, and to adapt to certain functions. Neural network control can be combined with fuzzy logic control, where fuzzy systems can directly express logic and knowledge and neural networks are better at learning to express knowledge implicitly through data. They are complementary and related systems, with fuzzy systems suitable for top-down expression and neural network systems suitable for bottom-up learning processes.
An expert control system uses an expert, or someone defined as an expert, based on their deep theoretical knowledge or practical experience in a given field and follows the expert's decision-making actions to solve difficult problems or achieve important results through the accumulation of theoretical knowledge and practical experience. The system then uses some kind of knowledge acquisition method to store the knowledge and experience and code those decisions into actions an intelligent control system can take. This means an expert system generally consists of a knowledge base, a database, an inference engine, and an interpretation part and knowledge acquisition system. These systems are also expected to have high reliability and long-term continuous operation, real-time nature online control, excellent control performance and anti-interference, flexibility, and easy maintenance.
Human-like intelligent control works to introduce some non-linear control methods, offering a more dynamic process and transient process, according to the needs of the system's characteristics, behavior, and control performance. This requires expert control experience and intuitive judgment and reasoning rules. These control systems are conducive to solving the contradiction between fastness, stability, and accuracy in the control system while also enhancing the system's adaptability and robustness of the system in regard to uncertain factors. In part, to achieve this, human-like intelligent control basically works to imitate human intelligent behavior for control and decision-making.
Often this provides a machine with the necessary operational training, and the artificially implemented control method can be close to optimal. The structure of the human-like intelligent control is similar to an expert controller, including four parts: acquisition and processing of characteristic information, feature pattern set, pattern recognition, and control rule set. The working process of the human-like intelligent controller can be summarized into three steps: the system judges the characteristic mode of the dynamic process, the inference mechanism searches for a matching control rule, and the controller executes the control rules to keep the object controlled.
Large systems often have the characteristics of high-level order of the system, a large number of subsystems and interrelationships, a large number of system evaluation goals, and conflicts between goals. In order to process these systems, people tend to break them down into hierarchical levels. For large-scale complex control systems, these systems aim to form a pyramid-like hierarchical control structure. The hierarchical intelligent control structure mimics the human central nervous system, which is organized according to a multilayer structure.
This structure is further divided into three basic processing methods and exchange methods: decentralized control, distributed control, and hierarchical control. The main structure of the system includes multiple descriptions and multi-level descriptions. Depending on the decision-making objectives, the system can be divided into single-stage single-objective systems, single-stage multi-objective systems, and multi-stage multi-objective systems. In this control structure, the configured controller receives information from an upper-level controller and is used to control the controller or subsystem at a lower level. To ensure possible conflicts between controllers are avoided, the system also uses coordinator systems, of which there are many methods but tend to be based on two basic principles of association prediction coordination and association balance coordination.
Hierarchical intelligent control systems are a branch of intelligent control that was first applied in industrial practice and played an important role in the formation of intelligent control systems. The intelligent control structure, according to the intelligence level, is divided into three levels: organization level, coordination level, and control level.
Bayesian probabilistic control should measure the confidence of an individual for an uncertain proposition and use this property to control it, making the system subjective in this sense. Using the probability theory proposed by Bayes, the sensitivity of the decision-making can be examined. Bayes further proposed the concept of prior and posterior probability, where the prior probability can be modified for new information to obtain the posterior probability; put another way, Bayesian probability control can be used to incorporate new information into the analysis. The model offers stable classification efficiency, can work well on small-scale data, can handle multi-class tasks, and is suitable for incremental training. Further, the model is not as sensitive to missing data as other models are, with a fairly simple algorithm.
The genetic algorithm control is a computational model offering global, parallel search and optimization methods developed on Darwinian principles. This model works with potential solutions to a problem, with each potential solution represented as a particular solution to the problem, generally expressed in some form of genetic code. The idea of these algorithms is based on the ideas from natural laws of biological genetics and is a search algorithm with an iterative process of "survival + detection." The genetic algorithm uses randomization techniques to guide an efficient search of an encoded parameter space. Among them, selection, crossover, and mutation constitute the genetic operation of the algorithm; the five elements of parameter coding, initial population setting, fitness function design, genetic operation design, and control parameter setting constitute the core of the genetic algorithm. The genetic algorithm control often produces interpretable results that are easy to apply, with a range of data types that can be processed, used for optimization, and integrated easily with the neural network.
