279 A BEST-FIRST STRATEGY FOR VOLTAQE AND REACTIVE POWER CONTROL E. Carpaneto, G. Chicco, R. Napoli, F. Piglione Politecnico di Torino, Italy ABSTRACT The paper presents an expert system based on a modified best-first search strategy, which assists the power system planning operator in voltage control and reactive power dispatching. An experimental PROLOG program, based on the proposed strategy, has been developed. Tests on several study systems have shown promising results. I" In the past years the voltage/reactive power control problem was often considered as a secondary issue which could be solved on a local basis using locally available compensation devices. However, the continuous increase of the load demand has brought the transmission systems to operate near to their own security limits. Moreover, the growing complexity and interconnection of the networks has made the traditional local voltage regulation a very difficult task. As consequence, in the planning studies the voltage/reactive power control problem is now a global problem, which must be tackled and solved on the overall system. The problem approach is traditionally heuris- tic: after an execution of the load-flow program on the active power balanced system, the operator detects voltage and reactive power violations, and then employes his own physical problem and specific81 network know- ledge to eliminate the detected violations, acting on the existing control devices. How- ever, the increasing complexity of the actual transmission systems makes this task very cumbersome, therefore it may be faced only by experienced network planners. Applications of Artificial Intelligence (AI) techniques to the power system problems have attracted large interest in recent years. Indeed the heuristic human approach employed on many power system problems to find fast approximate local solutions, therefore avoid- ing full algorithmic solution, is well suited to automatic implementation by Expert Systems (ES). Unlike numerical algorithms, ES can explain their results to the user in terms of heuristic choices. Therefore they make the human supervision of the whole decisional process easier. In the last years many ES and knowledge-based methods have been developed to assist the operator in voltage/reactive control task. Some authors employ general purpose ES shells to build production systems based on rules written in natural-like language, Liu et al. (l), Tweed and Weatherwax (2). Other authors use instead appropriate heuristic search strategies, directly coded in an AI language (LISP, PROLOG), Cheng et 81. (3), Pessi et al. (4). Although the former approach seems very direct and simple, the latter is often more effective in terms of coding task and operating performances. This paper presents an expert system for the voltage/reaotive power control which uses a modified best-first search strategy to explo- re the state space of a system linear model, searching feasible solutions. The search is guided by an index, representing the voltage- /reactive violations amount for each state. The developed program is written in PROLOG, language highly oriented to state space sear- ching, and interacts with a FORTRAN load-flow program and some C language routines for numerical calculations. PROBLEM DESCRI PT ION The physical and mathematical background of load-flow (LF) method is well known. The decoupling between active (P/8) and reactive channel (Q/V) greatly simplifies the voltage/reactive power control problem. From this viewpoint, it is possible to distinguish the involved variables in dependent variables (load bus voltages, generator bus reactive power) and control variables (generator voltages, shunt capacitor values and transformer tap changers set points). Both dependent and control variables are subject to physical constraints. Therefore the reactive planning problem lies in adjusting the control variables values inside their operating limits so that the constraint violations of dependent variables are eliminated or at least reduced. PERT SYSTEM APPROACH The expert system emulates the operator's behaviour, using a solution search strategy which takes the place of the empirical compensation rules and a linear network model which represents the specific operator know- ledge on the network. The linear model allows the expert system to quickly formulate and verify nany compensation hypotheses, so taking into account mutual interaction which the operator cannot analyse simultaneously. The overall scheme of the expert system approach is depicted in Fig.1. The ES module is composed of the solution search strategy and the linear model. It receives the LF output data and the system linear model built at the operating point. The solution search strategy tries to eliminate the existing violations verifying every control action on the linear model. If a solution is found, the control values are returned to the LF module to check on the full system model the effectiveness of the reached solution. The User Interface (UI module) allows the operator to supervise the whole process, accepting or correcting the expert system suggestions. The two main parts of ES module, the linear model and the search strategy, will be now briefly described. The linear m o m The system linear model is used to quantify control actions and their secondary effects on the other buses, and to calculate the new system state resulting from a control action.