Rule-based strategies

The mles used may be deterministic or non-deterministic. For instance, for an FC system, we know that an open-circuit operation is to be avoided if we wish to increase the durability of the system. Thus, one rale might be to prohibit this open-circuit operation. 

With the same spirit, we can also perform frequency filtering. This involves aibitrarily defining one or more hmit frequencies and, on the basis of the frequency of the power load demanded by the charge in comparison to these limit frequencies, adjusting the power demand to one storage device or another (typically high 

We can also use fuzzy rules (based on fuzzy logic theory) to produce excellent results [TEK07 SOL 11], or indeed neural networks, adaptive or otherwise [MOR 06], 

An example of a definition for some of these fuzzy rules might be as follows, in the context of a vehicle supplied by a fuel cell and supercapacitors (where 

Rule 2 IF 0AND IF dP, /dt dP )THEN (provide the required power using the fuel-cell system). 

---

An alternative starting network is one without stream spHts. The networks from the TI method maximize energy recovery and may introduce heat-load loops. Stream spHts ate not made in the initial steps of network invention. The ED method is proposed to be one in which heuristic rules and strategies would be used to improve the networks developed by the TI method. The importance of a thermodynamic base for evolutionary rules is stressed in this proposal, but there is no expHcit guidance for the evolutionary process. 

Section 9.2 will review traditional statistical process control/statistical quality control (SPC/SQC) techniques used in quality control. Section 9.3 will follow this review with a discussion of techniques based primarily on an experiential rule base and expert system technology. Section 9.4 will discuss control strategies that use an on-line process model a variety of models can be used in such model predictive control. Section 9.5 will discuss this variety of models. Section 9.6 will summarize this chapter and discusses future trends in the field. 

Whereas individual rules are relatively simple, their logical relations and interactions within a large set of rules are usually not straightforward. Complex rule-based systems make it difficult to understand how individual rules contribute to the overall problem solution strategy. 

ARC is not an expert system in the classical sense it does not provide ques-tion-and-answer strategies or a generic rule base. It includes nonlinear optimization algorithms and guides the user through the necessary steps for creation and evaluation of descriptors. 

A separate inference engine should be used to process the rule base, allowing for multiple inference strategies using the same rule base and providing easy rule modification. Additionally, the inference engine itself may be reused for other projects. 

---