1

THE BLACKOUT OF 1987 IN TOKYO, JAPAN:

PURE VOLTAGE COLLAPSE
Chevalier, Samuel C.
College of Engineering and Mathematical Sciences
University of Vermont, Burlington, VT

Abstract—On July 23 of 1987, voltage collapse caused a large

scale blackout to occur in the Tokyo, Japan. Over 2.8 million
households serviced by the Tokyo Electric Power Company,
Inc. (TEPCO) lost electrical power for a period of over 3
hours. Ultimately, this blackout was caused by severe voltage
collapse, and the event has shaped TEPCO’s current operational
planning and reactive power control methods. This paper serves
to introduce and outline the phenomenon of voltage collapse in
the context of the Tokyo collapse. The causes of the blackout
and its cascading effects are subsequently explained. Finally, the
lessons learned by TEPCO through this disaster are analyzed.
Index Terms—Power System Stability, Voltage Collapse, Saddle
Node Bifurcation

I. I NTRODUCTION
A typical power system is modeled through a series of Dif-
ferential Algebraic Equations (DAEs), as given by reference Figure 1. An example of a saddle-node bifrucation is given. The bifurcation
[6]. These equations model the power flows, load dynamics, occurs once the parameter reaches a value of λ = 0. This figure is reproduced
and the dynamics of the voltage regulators, turbine governors, from reference [2].
and synchronous machines. Primarily, these equations are
functions of both algebraic and differential variables, but they
depend on two other important system values too. causes of the Tokyo collapse are explained. Finally, the lessons
learned by TEPCO are given.
   
ẋ f (x, y, λ, p)
= (1)
0 g(x, y, λ, p) A. Voltage Collapse: A Saddle Node Bifurcation
In (1), x is a differential variables and y is an algebraic As stated by [7], the 1987 blackout in Tokyo was caused by
variable. The variable p is not actually a variable but in fact a phenomenon called voltage collapse. Voltage collapse is a
a control setting which grid operators have total authority type of saddle node bifurcation. In Ian Dobson’s paper Voltage
over. This includes reactive support, regulator set points, and Collapse in Power Systems, a visualization of a saddle node
frequency settings. Finally, λ is a slowly varying system bifurcation is given. As indicated, the parameter λ is often
parameter, and this is typically chosen to be system load considered to be the aggregate system load. Once a system
P0 + jQ0 (and load noise ξ). In any given power system, load surpasses a certain value, the Power Flow Equations will
there is a point called the maximum loadability point, and it no longer converge (via the Newton Raphson Method).
corresponds to the maximum value of λ. Once this value of As shown in the figure 1, each value of λ has two solutions:
λ is reached, voltage collapse will occur. one is stable (xs ) and one is unstable (xu ). Once a power
True voltage collapse is a relatively uncommon event, as system reaches the bifurcation point, system load cannot
system dynamics often become unstable (in the form of a increase any more. Small perturbations will cause the system
Hopf bifurcation) long before voltage collapse occurs ([8],[9]). voltages to dynamically fall to very low levels, as the system
For this reason, the Tokyo collapse is a fascinating case will be operating in the highly unstable region on the lower
study on power system stability. As will be shown, there is side of the nose curve (the region with solutions corresponding
no major oscillatory instability, no operator error, no lighting to xu ). This is what the Tokyo system experienced in 1987.
strikes or faults, and no component failure. According to [7],
the blackout was due to pure voltage collapse.

II. PAPER B ODY: VOLTAGE C OLLAPSE OF THE TEPCO B. Causes of the Tokyo Collapse
N ETWORK The following details of the collapse are summarized from
In this section, the saddle node bifurcation is introduced, references [7] and [5]. Both of these references indicate that
as the blackout cannot be understood without it. Next, the on July 23, 1987, Tokyo experienced unusually hot weather. In
 2

1.4

1.2

1 PF = .97 lead
PF = 1
0.8
V2t

0.6 PF = .97 lag

0.4

0.2

0
0 0.5 1 1.5
System Load PD

Figure 3. This is the layout of the 1987 TEPCO network. This figure is
Figure 2. The “To” bus voltage of a small power system is plotted against reproduced from reference [7].
system load. Loads with three different power factors are plotted.

rate of increase, and the switched capacitor banks did not

some places of Japan, the temperature exceeded 39◦ Celsius respond quickly enough. Part of the reason for this sluggish
(102◦ Fahrenheit). This was the 9th hottest day on record in response is the fact that voltage collapse is a type of “long
the region, and so the load began increasing dramatically due term voltage stability”. This is explained in reference [4].
to increased air conditioner use. As the day progressed, the Such voltage stability exists on the time frame of 1 to 10
load forecast was increased from 38.5 GW to Version
Student 39 GW, and
of MATLAB
minutes. Therefore, responses often exist on this time frame
then around mid-morning it was again increased up to 40 GW. also. Switched capacitors may take several minutes to respond
The TEPCO system was nominally rated to handle up to 40 to voltage sags.
GW of load, but such operating conditions had never been In total, the TEPCO network had 10,570 MVar of switched
experienced. This peak forecast included 3.8% of operating capacitor reactive support. This is only a nominal rating
reserve so that the system could handle all potential increases. though. The reason for this can be seen in the equation which
Before the “lunch break” noon hour, the system demand computes the reactive power produced by a shunt capacitor:
2
declined dramatically from 39.1 GW to 36.5 GW. When Qinj = − VX . In this expression, V is the voltage differen-
this happened, a large block of shunt capacitors were kicked tial magnitude across the capacitor and X is the capacitive
offline. Shunt capacitors provide reactive support in the form reactance. As voltage sags, the reactive power produced by a
of reactive power. When this support is not needed, it must be capacitor bank falls with the square of the voltage. Therefore,
removed. The reason why is visualized in figure 2. if the voltage sags from 1 PU to .8 PU, the reactive power
In Figure 2, three nose curves are plotted for the same magnitude drops from 1 PU to .64 PU (assuming X = 1).
two-bus power system. The “To” (or load) bus voltage is
plotted against the load at that particular bus. The maximum Between 13:00 and 13:15, the 500KV backbone transmis-
loadability point for several types of loads can be seen clearly: sion lines were dropping by 4kV per minutes. As the system
this corresponds to the bifurcation point. parameter (load) increased, pushing the system closer to the
If this system is loaded down with say PD = .5 and a nose of the nose curve, the 500kV lines began dropping by
power factor of .97 lagging, the load voltage will be low 18kV per minute. This occurred from 13:15 to 13:19. When
(just over .8 PU). If, suddenly, reactive support kicks on, and these lines dropped below 370kV (.74 PU), lines around the
power factor correction occurs (to, say, .97 leading), the load system began to trip and a massive amount of load was shed.
voltage will jump well over 1 PU. Over-voltage spikes are very Figure 3 shows the physical layout of the system. Most notably
dangerous and damaging to a power system, so this would be are the two inter area tie lines. The Japanese power system has
a serious problem. For this reason, reactive support (such as two operating frequencies: one runs at 50Hz while one runs
shunt capacitors) must be removed when it is not necessary. at 60Hz.
Figure 4 shows a fascinating visualization of the saddle-
When TEPCO’s load dropped before lunch, the reactive node bifurcation experienced by the network. According to
support was removed. During the lunch hour, as the system [7], this plot was produced after the blackout using collected
demand increased again, this reactive support should have real time data as well as time domain simulations. The 500kV
kicked back online. The demand increase, though, occurred far backbone voltage is plotted against the increasing system load.
too rapidly, and the shunt support did not turn back online fast As can be seen, the time passage for this plot is not linear. The
enough. This was an issue caused by the automated response effects of the implemented reactive support can be seen very
of the Voltage Reactive Power Controller (VQC): the VQC clearly: at various times, when −Qinj increases, the voltage
did not respond to the load increase quickly enough. spikes back to a higher value (system nominal). At 13:03, the
During the peak load rise, the demand increase was ap- system is momentarily operating on the bottom side of the
proximately 400 MW per minute. This was an unprecedented nose curve, but then more reactive support is added, pushing
 3

C. How the 1987 Collapse has Influenced Operational Plan-

ning of TEPCO
As explained in [5], this failure was not due to network
faults, unexpected line trips, scheduled outages or mainte-
nance, or conscious operator error. There are many other root
causes behind this particular blackout, and the most apparent,
include insufficient system monitoring, lack of strategies to
guard against voltage collapse, inadequate control setting and
load characterization, and an uneven distribution of power
plants. If the reactive support had been turned online sooner (to
prevent reactive losses proportional toV2 ) or if the operators
wouldn’t have made assumptions about the speed at which
Figure 4. A visualization of TEPCO’s Saddle Node bifurcation is plotted. load can ramp up, the situation may have been avoided. The
This figure is reproduced from reference [7].
particular lessons learned by TEPCO are separated into seven
distinct categories. These are similar to the categories given
in [7].
the nose curve father out on the x axis. At 13:10, maximum 1) Elevating the Voltage Profile: In the Extra High Voltage
loadability is reached and a bifurcation occurs. At this point, (EHV) transmission lines, TEPCO has made it a priority to
the system cannot support any more load, and system voltages keep the voltage profile “high and flat”. This is mainly in order
begin diverging towards very low, unstable values. to leverage full usage of all shunt support and transmission
It is fascinating to note that the same amount of system capacity (the higher the voltage, the lower the current). Ef-
load can be served by multiple voltage values. At 13:03 for fectively, TEPCO raised all of its nominal set points, allowing
example, 38.9 GW of load is served with the transmission more room for error. For example, the nominal operating range
backbone at 520kV. Later, at 13:15, the same amount of load is for the 500 kV lines was 515~525 kV before the blackout.
served, but the transmission voltage has fallen to 460kV. This After the blackout, this range was increased to 525~550 kV.
is due to the nature of power consumed by a load: S = Ṽ I˜∗ . Of course, distribution voltages were not altered. When trans-
Clearly, the same amount of (complex) power can be served mission voltages increase, transformer winding ratios must be
with low current and high voltage, or with high current and altered (in some way) to ensure that consumers see no change
low voltage. in the electricity which serves their home.
Keeping a “flat” voltage profile mean operating far from the
bifurcation point (where bus voltage variance is very high for
One important part of the voltage collapse in Tokyo which given stochastic perturbations).
must be considered is the nature of the loads. According to 2) Increased Installation of Shunt Capacitors: At the time
analysis done by [5], there was a certain popular type of air of the blackout, the TEPCO network had 10.6 Gvar of shunt
conditioning unit which accounted for much of the increased capacitors in the system to support an upper bounded load
load. This unit can be approximately modeled as a constant of 40 GW. Many of these capacitors, though, had to remain
power load, meaning as the voltage sags, the current draw disconnected from the system during the collapse because of
increases. Typically, when the voltage sags in a circuit, the load upper bus voltage limits on the tertiary sides of the transform-
drops because the power (P = IV) decreases. For the Tokyo ers. Over voltage is typically a more serious constraint on the
system, the opposite situation occurred. As the voltage sagged, system than under voltage is, so these capacitors could not be
the power remained constant, so the current draw increased. used. This decreased the effective amount of capacitive support
Increased current draw corresponds to larger I2 Z losses in that the system had available. Over the five years after the
the transmission and distribution lines (where Z = R + jX). blackout, TEPCO has installed 11.9 Gvar of additional shunt
As seen by the network, transmission lines losses and load capacitors.
demand are virtually the same thing. 3) Increased Reactive Support through Dynamic Reactive
After 13:19, four 275 kV lines, four 275/66 kV trans- Power Reserves: Generators use Automatic Voltage Regula-
formers, and four 500 kV lines tripped. Under voltage, over tors (AVRs) in order to keep generator voltages at a proper
current, and impedance relays worked in conjunction to end level, but Power System Voltage Regulators (PSVR) have the
the collapse by providing the much needed load shedding of ability to keep the PV bus’ “sending end” voltage at a certain
8.168 GW of load (21% of the total load). The severe voltage prescribed level. TEPCO installed a number of PSVRs as well
drop on the lines caused the apparent impedance of the lines as SVCs in its network. While shunt capacitors provide long
to trigger the impedance relays to trip in some cases, while term voltage stability support, SVCs and PSVRs additionally
the voltage and current relays seem to be the causes of the provide short term (quick response) voltage stability support.
other trips. Interestingly, the impedance relays provided the Both types of stability must be maintained in order to ensure
most relief to the system, even though such devices are not stability of a power system network.
designed to prevent voltage collapse. Once the load was shed, 4) Reliance on UVLS in Extreme Situations: Under Voltage
the system rebounded and began operating at stable, nominal Load Shedding (UVLS) can be an extremely effective tool
voltage once again. when used properly, and it can be used in such a way that
 4

will prevent a large scale blackout. It represents a triage (or power configuration. This is because even though transmission
utilitarianistic) way of dealing with a looming collapse. voltages are falling, distribution voltages are not, and therefore
On the Western (weaker) side of the grid, TEPCO has four the loads are consuming the same amount of power. If the load
central units capable of implementing UVLS. When three of voltages are allowed to sag, power demand will begin to drop
the four units detect voltage sags at a specified (alarmingly accordingly (in most cases).
fast) rate, a signal will be sent to operators at local stations In order to combat this problem, TEPCO implemented an
recommending load shedding. Such an action is regrettable and interesting protocol after the 1987 blackout. When the con-
certainly a last resort, but the 1987 blackout itself proved that troller sees that all shunt capacitors are in service, and primary
UVLS is highly effective at stalling and reversing the effects side voltage falls lower than a certain value, the tap controller
of voltage collapse. locks itself. This prevents the tap-changing transformers from
5) Increased System Monitoring: Soon after the collapse, elevating secondary voltages. In effect, the aggregate load shift
TEPCO installed enhanced on-line voltage security monitoring from constant power to constant impedance (at a very high
systems. These system, along with the state estimation tools, level).
give operators the ability to draw PV and QV curves in real
time. By doing so, bifurcation proximity estimation can be III. C ONCLUSION
calculated by tracking the parameter λ. By having access to This paper outlines the causes, the consequences, and the
such curves, operators can make real time control decisions outcomes of the voltage collapse in Tokyo in 1987. The results
about how power should be dispatched, when and how reactive are primarily pulled from two interesting case studies done on
support should be implemented, and how much more loading this blackout ([5], [7]). As part of the conclusion, six insightful
the system can take. questions concerning the failure shall be addressed explicitly.
Such data can also allow operators to see long term trends.
Operators might see, for instance, that the system is often
operating close to a collapse, or that the system is often A. Was the cause of failure known a priori or was it an
having over-voltage spikes. Increasing the system monitoring unknown unknown?
will make long term systemic issues much more visible. Certainly, the cause of the blackout was known a priori, as
For instance, reference [1] outlines the causes of the 2011 voltage collapse is a well known and thoroughly understood
US Southwest blackout. In the official report, the authors sight phenomenon. Extremely high load levels coupled with limited
poor situational awareness (along with poor operational plan- reactive support will clearly lead to extreme sagging, if not
ning) as one of the primary causes of the failure. Situational total collapse. The situation was unique and unprecedented,
awareness will only be enhanced through elevated system but given all system information, a modeling platform could
monitoring. have predicted the collapse.
6) Enhanced Simulation Tools: In response to the blackout,
TECPO developed a software package entitled VQC Sim- B. Was the failure due to a random act or human error?
ulation. This simulator models the dynamic reactive power
Primarily, this collapse was due to poor long term opera-
units in the system: these include SVC, PSVR, and VQC
tional planning. Reactive support was not high enough, the
technology. This software serves to help grid operators and
maximum loading point was over estimated, VQC controls
planners study “slow” grid stability. More specifically, the
were not properly tuned, and the operators were not thoroughly
software helps the operators understand how the different
trained on how voltage collapse should be responded to. In [5],
control settings of these reactive devices influence the stability
the Tokyo region on July 23 is said to have experienced “high
of the overall grid. In this way, the control settings can be
temperatures caused by a stagnant atmosphere and strong
chosen in an optimized fashion without having to rely on
daylight due to a Pacific anticyclone.” This can certainly be
intuition or the guess and check methods.
seen as a random event, but it is one which is not outside the
TEPCO also developed a slimilar program called the
scope of imagination.
Voltage Stability Contained Optimal Power Flow (VSCOPF)
program. In this program, users can study voltage stability
margins and voltage sensitivity in the context of OPF using C. Could it have been prevented by some kind of monitoring
the gradient of the PV Curve. system?
7) Altered Control Settings for Tap-Changing Transform- Indeed, the disaster could have certainly been prevented
ers: When a system is tending towards voltage collapse, by an enhanced monitoring system. In 1987, TEPCO had
shunt reactors will be switched off and shunt capacitors will three power system control centers. These centers knew most
be switched on. Additionally, tap-changing transformers will system voltages, but the 500 kV lines voltage outside of their
begin to affect the system also. An on-load tap-changing respective territories were not known. This greatly inhibited
transformer has the ability to change its winding ratio (while operator situational awareness, and it prevented them from
the load is in service) in order to raise or lower the secondary taking early mitigating control actions (such as minor load
side voltage. Such devices are incredibly useful at keeping shedding or blocking the actions of the on-load tap changing
consumer voltages constant when system voltages are sagging. transformers). Today, Phasor Measurement Units (PMUs) are
When voltage collapse is looming, though, a tap changing becoming extremely widespread. These units measure instan-
transformer is effectively keeping the entire load in a constant taneous voltage phase and magnitude and can provide this
 5

for three to four hours. The indirect consequences are, in fact,

far more reaching than the direct ones, and society can learn
some very important lessons from this disaster. In order to
optimize constrained grid infrastructure, power systems are
often operated close to their limits (maximum loadability).
When unprecedented load levels are reached, though, this
paradigm is challenged, and cost of upgrading infrastructure
versus the cost of an unstable grid (i.e. a brownout or a
blackout) must be weighed.
Even when investments are made to upgrade a system,
though, reference [3] makes an important observation: large
investments do not always lead to a more stable grid. This is
primarily due to how the power is dispatched. Grid operators
(or in the US, Independent System Operators (ISOs)) will
dispatch the power flows in the most economical ways possible
which still maintain N-1 security. If new transmission lines
are added, and power dispatch does not drastically change,
Figure 5. A visualization of load during several days. This figure is
reproduced from reference [5]. then stability will certainly be enhanced. If the power flows
are totally redirected in order to re-optimize the system, then
in terms of stability, the system is more or less right where it
data to grid operators in real time. This data can also be used started: right on the brink of collapse. This insight is reflective
to compute current flows and power flows very quickly. If of human nature: constant dissatisfaction.
the TEPCO operators had access to such data, these control
actions could have been taken.
R EFERENCES
D. Did this failure have an effect on the future design of [1] Arizona-Southern Califronia Outages on September 8, 2011. Carto-
graphic Perspectives, 2012.
similar systems? [2] Ian Dobson. Observations on the geometry of saddle node bifurcation and
Yes indeed, this failure has been a well studied example of voltage collapse in electrical power systems. IEEE Transactions on Cir-
cuits and Systems I: Fundamental Theory and Applications, 39(3):240–
voltage collapse, and much work has been done to understand 243, 1992.
how systems might be updated to avoid such a collapse. [3] Daniel Kirschen and Goran Strbac. Why investments do not prevent
Because of the collapse, TEPCO began to plan out a 1000 kV blackouts. Electricity Journal, 17(2):29–36, 2004.
[4] Prabha Kundur, John Paserba, Venkat Ajjarapu, et al. Definition and
transmission system to be constructed in the 1990’s. TEPCO Classification of Power System Stability. IEEE Transactions on Power
also made many operational changes, as outlined in Section Systems, 21(3):1387–1401, 2004.
2.3. [5] Atsushi Kurita and T. Sakurai. The power system failure on july 23, 1987
in tokyo. In Decision and Control, 1988., Proceedings of the 27th IEEE
Conference on, pages 2093–2097 vol.3, Dec 1988.
E. Can the general public learn something from this? [6] A P Lerm, Claudio a Ca, and a B Lemos. Multi-parameter Bifurcation
In general, the general public did not learn very much from Analysis of Power Systems. Electrical Engineering, 18(October):1–7,
1998.
this disaster. Although costly, blackouts typically do not cause [7] T. Ohno and S. Imai. The 1987 tokyo blackout. In Power Systems
the sort of death and destruction associated with a bridge Conference and Exposition, 2006. PSCE ’06. 2006 IEEE PES, pages
collapse or a transportation system which cataclysmicly fails. 314–318, Oct 2006.
[8] William D. Rosehart and Claudio A. Cañizares. Bifurcation analysis of
The general takeaway is this: an overloaded system, when not various power system models. International Journal of Electrical Power
properly supported, will fail. The public does not have very & Energy Systems, 21(3):171–182, 1999.
much control over such things. TEPCO, on the other hand, [9] Peter W. Sauer and M. A. Pai. Power System Steady-State Stability and
the Load Flow Jacobian, 1990.
learned several very valuable lessons from this disaster.
Figure 5 shows a plot of the aggregate load seen by gen-
erators in the TEPCO network. Clearly, the event on 7/23/87
experiences a load level and load increase which, cumulatively,
cause the system to collapse. The massive load shedding which
occurs at 13:19 is interesting to note. Of course, the general Samuel C. Chevalier received a B.S. in Electrical
public cannot be held accountable for such a disaster, and Engineering from the University of Vermont in 2015.
their usage habits probably did not change after this blackout. He is currently pursuing an M.S. degree in Electrical
Engineering from UVM, and his research interests
TEPCO learned, though, that its system must be robust enough include stochastic power system stability, large scale
to handle such system demands. renewable energy penetration and Smart Grid.

F. What were the consequences for society and the engineer-

ing profession?
The direct consequences of the failure are obvious: several
million people in an urban area of Japan lost electrical power