Development and maintenance of an online expert system as a power system dispatchers' aid.

摘要

Expert system (ES) technologies have received wide attention in the power engineering community as potential tools for handling planning and operation problems. This research addresses two issues which need to be resolved before the feasibility of ESs for on-line operations can be established.; The first issue deals with implementation of an ES in an on-line environment of a power system control center. This dissertation reports the design and development of an OPS83 based ES, CRAFT (Customer Restoration and Fault Testing), which is capable of identifying the faulted section of a transmission line based on the status of automatic devices. CRAFT incorporates another rule module, AUTOSTATUS, which finds the status of unsupervised automatic devices following a fault. CRAFT's problem solving strategy is based on logic embedded in the automatic devices on a power line, experience acquired from power dispatchers, and operating priorities and policies of the power company. An interface is designed to transfer alarms generated due to faults from the energy management system computer to an appended ES computer. CRAFT started on-line operation at Puget Sound Power and Light control center in July 1988. Overall, accuracy and speed of CRAFT has been outstanding.; The second issue deals with maintenance of rule-based systems. For development of maintenance tools, the concepts of Attribute Space (AS, domains of all attributes) and Pseudo Working Memory Elements (PWME) are introduced. Individual rules are represented by regions in the AS, where each rule is satisfied. The AS representation is used to define six relations between rules: cause-effect, mutual exclusion, conflict, subsumption, redundancy and implication. These relations are defined in a scenario- independent manner. A relation between two rules is either complete or partial; a complete relation holds for all valid data values in the AS while a partial relation holds only for some of them.; An algorithm is developed to identify complete relations between a new rule and existing rules in a rule base. The algorithm is implemented using the Rete net representation of a rule base. PWMEs enable chaining of set relations among regions in the AS. This allows use of the same algorithm for both rule-chaining and relation checking of a rule-based system. Application of the relation-checking algorithm to CRAFT is illustrated. Modifications to the algorithm for detection of partial relations are also discussed.