1576 IEEE Transactions on Power Systems, Vol. 12, No. 4, November 1997 z rtificial Neural Network ynamic Load Modeling Takashi Hiyama, Masatomo Tokieda, Walid Hubbi* Electrical and Computer Engineering, Department, Kumamoto University, Kumamoto 860, Japan *Electrical and Computer Engineering Department New Jersey Institute of Technology Newark, zyxw NJ 07102 Niroaki Andou Research Laboratory, Kyushu Electric Power Co. Fukuoka 815, Japan Abstract: The aim of this research is to investigate the use of Artificial Neural Networks zyxwvutsr (ANN) with feedback loops for modeling power system dynamic loads using field data. In addition to the power demand before the transient, only the phase voltages will be assumed available at the recall stage. It is found that the load dynamics can be identified using the ANN developed for the season and the location of the load being emulated. The frequency response of the tested load is also obtained using the achieved ANN model. Improvement in DLM is necessary, however, not only because the load is an important component of the system and it is dynamic in nature but also because an inaccurate Load Model (LM) will render the high accuracy in the models of the other components less valuable. Of course, the design of any control system acting on a PS, such as a PS stabillzer, is only as good as the model upon which it is based Instances of significant difference between the predicted and the actual behavior of the system have been reported [2,3]. In [4], studies were evaluated. The constant reactance method was found to give results having large errors rendering the method inadequate for long term simulations involving large deviations in system frequency. In [SI it is COncl~ded that in Some stability studies, the difference between results obtained using static LMs and those obtained using dynamic LMs can be quite pronounced. The effect of load dynamics on oscillation damping, is studied in [6] where it is found that using static LM~ gives oscillation dampingresults ln 1982, [71, found that a static LM is adequate to represent the voltage as well as to inter-area swing. For industrial load, ~sults Presented in this paper demonstrate that accurate dynamic load modeling even for residential load (at least for some) is necessary. Not all residential loads have the same characteristics. The question that naturally arises is, at what locations dynamic load zyxw (DL) representation is more important This is answered in [SI where techniques are presented to identify busses at w b c h load dynamics have the greatest effect Keywords: dynamic loads modeling, artificial neural different approachesto model the Ps load in long-term stability networks, power system dynamcs. 1. Introduction Accurate electric power dynamic load modeling (DLM) is becoming increasingly important because the Power System (PS) is being operated in a more and more stressed state. Opposition to nuclear generation, difficulties in obtaining right-of-way in some countries; the high rate of growth of electric power demand and lack of capital in other countries, are some of the reasons that contributed to forcmg higher utilization found in [I]. In this situation, the safety margins usually imbedded in the guidelines and limits obtained using off-line the increasing capabilities of power electronics switches, and computing devices and peripherals, are enabling the PS engineer to process system information in real-time and to take control actions at greater speeds. All this points out to the importance of fully understanding the static and dynamic responses of the various components of the PS Of the Ps mfrasmcture, growth data for a specific system can be residential/commercialload response to a sudden change in studies become increasingly unaffordable. On the other hand, howecler, the response is si@ficant. However, the Of ps is at a more advanced stage than so that other loads can be representedusing zyx their static models DLM. This is not surprising; the nature, the changing composition, and inaccessibility of ps loads make DLM exceedingly difficult in effect, the load is a nonlinear time varylng (almost inaccessible) System that affects the input to it through feedback interaction with the PS It can be concluded then that ignorlngload dynamics can lead to it is well-accepted that the problem of voltage instability is dynamic in nature with some aspects that cannot be predicted using static LMs. DLM is important also for damping low assessment On-line. OnceDLMat a certainbus is justified, a choice is made between an aggregate-measurement based approach or a component based approach. The latter approach is considered impractical for DLM except perhaps for the cases where the load composition is known to a high degree of confidence such as the results in transient and PE-439-PWRS-0-01-1997 A paper recommended and approved frequency OscillatiOm and to improve security by the IEEE Power System Engineering Committee of the IEEE Power Engineering Society for publication in the IEEE Transactions on Power Systems Manuscrlpt submitted JUIY 29 1996, made available for printing January 8 1997 0885-8950/97/$10.00 zyxwv 0 1997 IEEE