IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 26, NO. 1, FEBRUARY 2011 411 Decentralized Demand-Side Contribution to Primary Frequency Control Angel Molina-García, Member, IEEE, François Bouffard, Member, IEEE, and Daniel S. Kirschen, Fellow, IEEE Abstract—Frequency in large power systems is usually con- trolled by adjusting the production of generating units in response to changes in the load. As the amount of intermittent renewable generation increases and the proportion of flexible conventional generating units decreases, a contribution from the demand side to primary frequency control becomes technically and economically desirable. One of the reasons why this has not been done was the perceived difficulties in dealing with many small loads rather than a limited number of generating units. In particular, the cost and complexity associated with two-way communications between many loads and the control center appeared to be insurmountable obstacles. This paper argues that this two-way communication is not essential and that the demand can respond to the frequency error in a manner similar to the generators. Simulation results show that, using this approach, the demand side can make a sig- nificant and reliable contribution to primary frequency response while preserving the benefits that consumers derive from their supply of electric energy. Index Terms—Decentralized control, demand-side response, load frequency control, primary frequency control. I. INTRODUCTION I MBALANCES between load and generation must be cor- rected within seconds to avoid frequency deviations that might threaten the stability and security of the power system. Routine deviations from this balance are usually corrected by adjustments in the output of conventional generating units driven by their governor in what is called primary frequency response [1]. The load is used explicitly to restore this balance only when the imbalance is severe and cannot be remedied quickly enough using fast acting generation. In such cases, blocks of loads are interrupted following the action of underfre- quency relays. This control philosophy may need to be revised in the coming years as the demand side may take a more active role in the control of the system. As their relative size increases, intermittent and variable output renewable energy sources such as wind farms will contribute larger random fluctuations to the load/generation balance [2]. At the same time, the number of Manuscript received July 16, 2009; revised November 06, 2009. First pub- lished May 24, 2010; current version published January 21, 2011. This work was supported in part by the Spanish Fundación Séneca, 08747/PI/08 Refer- ence. Paper no. TPWRS-00538-2009. A. Molina-García is with the Department of Electrical Engineering, Technical University of Cartagena, Cartagena, Spain (e-mail: angel.molina@upct.es). F. Bouffard and D. S. Kirschen are with the School of Electrical and Elec- tronic Engineering, The University of Manchester, Manchester M60 1QD, U.K. (e-mail: francois.bouffard@manchester.ac.uk, daniel.kirschen@manchester.ac. uk). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TPWRS.2010.2048223 conventional generating plants that have the flexibility required to take part in primary frequency control is likely to decrease as coal-fired plants are decommissioned. One possible scenario would see the bulk of the electrical energy being produced by a combination of renewable sources and nuclear power plants [3]. Under such conditions, performing primary frequency control using only supply-side resources may become not only prohibitively expensive but also technically difficult; see, for example, [4]. It is therefore important to explore how a sufficient proportion of the loads could assume a routine role in primary frequency control to maintain the stability of the system at an acceptable cost, considering this load participation as an example of the contribution that consumers could make to ancillary services [5], [6]. The obvious challenge in including loads in frequency control is the large increase in the number of potential participants. Even in the largest control areas, at most a few hundred large genera- tors contribute to frequency control. On the other hand, partici- pation from the demand side might involve tens of thousands if not millions of consumers. Though this may appear technically daunting and economically unrealistic, it has to keep in mind that conventional primary frequency control is a distributed con- trol system that relies on the availability of the frequency as a measure of imbalance between load and generation. Indeed, the response of each generating unit is determined by its droop char- acteristic and a local frequency measurement, not by a signal sent from a control center. Communication to and from the con- trol center is used only in the slower secondary and tertiary con- trol loops for better economic optimization and network secu- rity. A load or consumer thus does not have to be plugged into a communication network to take part in primary frequency con- trol. Schweppe et al. originally proposed this idea in 1980 and patented this concept as the Frequency Adaptive Power Energy Rescheduler (FAPER) [7]. In the last few years, research effort on the design and appli- cation of FAPER-like controllers applied to primary frequency control gained significant momentum. The Grid Friendly Appli- ance controller [8] developed by the Pacific Northwest National Laboratory has shown great promise as a means to modulate load in response to certain trends in the system frequency. This controller is to be fitted into individual appliances which are es- sentially energy users rather than power users. Energy users are appliances which can modulate their power consumption over time as long as the final energy consumption is sensibly the same. These include primarily heating, ventilation and air con- ditioning equipment, tumble dryers, immersion water heaters, etc. Lu and Hammerstrom in [9] discuss, simulate, and test in a laboratory environment the effect of the triggering frequency 0885-8950/$26.00 © 2010 IEEE