Exhibit 99.419
MEMORANDUM
From: Paul Gribik
To: File
Date: May 6, 1997
Subject: Gaming Issues and BEEP (Revised and Extended Version)
1.0 INTRODUCTION
The method of defining ex-post prices in the balancing energy market can lead to
unstable or wildly fluctuating prices. It can also provide the parties playing
in the markets with opportunities to "game" the system. The Trustee has had
consultants looking at gaming and efficiency issues in the forward energy
markets run by the PX and the forward transmission markets run by the ISO. To
date, I am not aware of anyone looking at such issues in the real-time balancing
energy market. However, it is likely that someone will look at this area.
In this note, I will briefly sketch one possibility for gaming the real-time
market. This is not meant to be a detailed description of an actual scenario. It
is only meant to illustrate a potential problem area. The problem I will
illustrate arises from a discontinuity between the sets of players and bids in
the forward energy markets and the set in the real-time balancing energy market.
To simplify, I will only consider the hour-ahead market and the real-time
market.
In the hour-ahead time frame, each SC essentially runs its own energy market
according to its own rules. We will assume that there is no transmission
congestion to simplify the example. Each SC schedules its generation according
to its rules in its own hour-ahead market. In the real-time balancing energy
market, the ISO runs a single market to balance the real-time energy in the
entire system. Many different parties converge in this market. The SCs can bid
adjustments to energy from sources that were scheduled in their individual
hour-ahead markets. They can also bid energy from resources that were selected
to provide ancillary services but which were not scheduled to generate in their
hour-ahead markets. The bids may come from:
- - supplemental energy bids (positive or negative adjustments) from
sources scheduled in the SCs' forward markets
- - supplemental energy bids (positive adjustments) from sources NOT
scheduled in the SCs' forward markets
- - energy from AGC sources (positive or negative adjustments) that were
scheduled to produce energy in the SCs'forward markets
- - energy from spinning reserves (positive adjustments only) that were NOT
scheduled to produce energy in the SCs'forward markets
- - energy from nonspinning reserves (positive adjustments only) that were
NOT scheduled to produce energy in the SCs'forward markets
- - energy from replacement reserves (positive adjustments only) that were
NOT scheduled to produce energy in the SCs'forward markets
1
The combined set of these real-time energy resources are not to be dispatched to
the point at which total energy costs are minimized in real-time. Instead, we
are to work from the SCs' schedules from their forward markets. If more energy
is needed in real-time than was scheduled in the forward markets, the cheapest
additional energy available from the combined set of all real-time sources is to
be used. If less energy is needed in real-time than was scheduled in the forward
markets, the most expensive energy scheduled from the combined set of all
real-time sources is to be backed down. Marginal costs are not clearly defined
in such cases. The approach that BEEP follows can lead to gaming.
2.0 GAMING EXAMPLE
Consider an example with the PX and one SC. Suppose that the PX schedules loads
of 10,000 MWh in its hour-ahead market and an equal amount of generation.
Suppose that the SC forecasts its load as 5,000 MWh in its hour ahead market. I
will look at two cases and the strategy that SC could follow in each to increase
its profits at the expense of the PX.
2.1 CASE 1
Let's assume that SC has detected a pattern in the PX's operations. The PX tends
to UNDER-FORECAST its loads in situations similar to the current conditions.
Assume that SC can forecast with a fair degree of accuracy that the PX will
under-forecast its load by 1,000 MWh.
In its hour-ahead market, SC schedules an additional 1,100 MWh of load. Suppose
that SC schedules:
Load = 6,100 MWh
Gen1_SC = 5,000 MWh
Gen2_SC = 1,100 MWh
The PX schedules:
Load = 10,000 MWh
Gen1_PX = 5,000 MWh
Gen2_PX = 5,000 MWh
In the real-time balancing energy market, SC submits the following strangely
priced supplemental energy bid:
0 (less than or equal to) Gen2_SC (less than or equal to) 1,100 MWh @
$1/kWh
In the real-time balancing energy market, the PX submits the following more
realistically priced supplemental energy bids:
0 (less than or equal to) Gen1_PX (less than or equal to) 5,000 MWh @
$0.05/kWh
0 (less than or equal to) Gen2_PX (less than or equal to) 10,000 MWh @
$0.06/kWh
2
In real-time:
PX's load is 11,000 MWh which exceeds its scheduled load in its
hour-ahead market by 1,000 MWh.
SC's load is 5,000 MWh which falls below its scheduled load in its
hour-ahead market by 1,100 MWh.
2.1.1 BEEP DISPATCH
The total system load in real-time is 100 MWh less than the load (and
generation) which was scheduled by PX and SC combined in their hour-ahead
markets. The ISO will call on the supplemental energy bids to reduce generation.
BEEP will call on the most expensive resource to back down. This is Gen2_SC.
Gen2_SC will be backed down by 100 MWh to 1000 MWh. It sets the price for
balancing energy equal to $1/kWh.
Settlements for Balancing Energy under BEEP Dispatch and Pricing:
PX Generation:
PX generates 10,000 MWh in real -time and had scheduled generation of
10,000 MWh in its hour-ahead market. Its scheduled generation equals
its real-time generation. PX's payments to the balancing market due to
mismatch between scheduled and real-time generation = $0.
PX Load:
PX load is 11,000 MWh in real-time and had scheduled load of 10,000 MWh
of load in its hour-ahead market . The PX consumes 1,000 MWh of
balancing energy for which it pays the ISO $1/kWh * 1,000 MWh * 1,000
kWh/MWh = $1,000,000.
PX Net:
PX pays $1,000,000 to ISO for net 1,000 MWh of balancing energy it
consumes.
SC Generation:
SC generates 6,000 MWh in real time and had scheduled generation of
6,100 MWh in its hour-ahead market. Its real-time generation falls
short of its scheduled generation by 100 MWh. It buys replacement power
to cover this shortfall from the ISO for $1/kWh * 100 MWh * 1,000
kWh/MWh = $100,000.
SC Load:
SC's load is 5,000 MWh in real time and it had scheduled load of 6,100
MWh in the hour-ahead market. Real-time load falls short of scheduled
load by 1,100 MWh. SC sells this excess through the ISO's real-time
balancing energy market. The ISO pays SC $1/kWh * 1,100 MWh * 1,000
kWh/MWh = $1,100,000 for this energy.
3
SC Net:
SC receives $1,000,000 from ISO for net 1,000 MWh of balancing energy
sold in the real time balancing market.
Effect of SC's Strategy if BEEP Dispatch and Pricing were Used by ISO:
SC was able to game the system to sell very high priced energy to the
PX (1,000 MWh at $1/kWh). The SC is able to set the price to an
arbitrarily high level.
2.1.2 REAL-TIME MARKET CLEARING DISPATCH
To see that SC was able to game the system to sell high priced power to the PX,
look at the prices and dispatch that would have occurred if the ISO were to
clear the real-time energy market. To clear the real-time balancing energy
market, the ISO would make the following changes to the dispatch:
Gen1_SC = 5,000 MWh
Gen2_SC = 0 MWh
Gen1_PX = 5,000 MWh
Gen2_PX = 6,000 MWh
Gen2_PX is now the marginal unit in the real-time dispatch. It sets the market
price to $0.06/kWh.
Settlements for Balancing Energy under Real-Time Market Clearing
Dispatch:
PX Generation:
PX generates 11,000 MWh in real -time and had scheduled generation of
10,000 MWh in its hour-ahead market. Its real-time generation exceeds
its scheduled generation by 1,000 MWh. The PX sells this energy to the
ISO's real-time balancing energy market. The ISO pays the PX $0.06/kWh
* 1,000 MWh * 1000 kWh/MWh = $60,000.
PX Load:
PX load is 11,000 MWh in real-time and had scheduled load of 10,000 MWh
of load in its hour-ahead market . The PX consumes 1,000 MWh of
balancing energy for which it pays the ISO $0.06/kWh * 1,000 MWh *
1,000 kWh/MWh = $60,000.
PX Net:
PX pays $0 to ISO for net 0 MWh of balancing energy.
SC Generation:
SC generates 5,000 MWh in real time and had scheduled generation of
6,100 MWh in its hour-ahead market. Its real-time generation falls
short of its
4
scheduled generation by 1,100 MWh. It buys replacement power to cover
this shortfall from the ISO for $0.06/kWh * 1,100 MWh * 1,000 kWh/MWh =
$66,000.
SC Load:
SC's load is 5,000 MWh in real time and it had scheduled load of 6,100
MWh in the hour-ahead market. Real-time load falls short of scheduled
load by 1,100 MWh. SC sells this excess through the ISO's real-time
balancing energy market. The ISO pays SC $0.06/kWh * 1,100 MWh * 1,000
kWh/MWh = $66,000 for this energy.
SC Net:
SC pays $0 to ISO for net 0 MWh of balancing energy.
2.2 CASE 2
Again, assume that SC has detected a pattern in the PX's operations. The PX
tends to OVER-FORECAST its loads in situations similar current conditions.
Assume that SC can forecast with a fair degree of accuracy that the PX is
over-forecasting its load by 1,000 MWh.
In its hour-ahead market, SC schedules 1,100 MWh of load LESS that it actually
expects to serve. Suppose that SC schedule in its hour-ahead market is:
3,900 MWh of load
Gen1_SC = 3,900 MWh
Gen2_SC = 0 MWh
The PX schedule in its hour-ahead market is:
10,000 MWh of load
Gen1_PX = 5,000 MWh
Gen2_PX = 5,000 MWh
In the real-time balancing energy market, the SC submits the following strangely
priced supplemental energy bid:
0 (less than or equal to) Gen2_SC (less than or equal to) 1,000 @
$0/kWh
In the real-time balancing energy market, the PX submits the following more
realistically priced supplemental energy bids:
0 (less than or equal to) Gen1_PX (less than or equal to) 5,000 @
$0.05/kWh
0 (less than or equal to) Gen2_PX (less than or equal to) 10,000 @
$0.06/kWh
In real-time:
PX's load is 9,000 MWh which falls below its scheduled load in its
hour-ahead market by 1,000 MWh.
SC's load is 5,000 MWh which exceeds its scheduled load in its
hour-ahead market by 1100 MWh.
5
2.2.1 BEEP DISPATCH
The total system load in real-time is 100 MWh more than the total load (and
generation) which was scheduled by PX and SC combined in their hour-ahead
markets. The ISO will call on the supplemental energy bids to increase
generation. BEEP will call on the cheapest resource to increase. This is
Gen2_SC. Gen2_SC will be increased by 100 MWh to 100 MWh. It sets the price for
balancing energy equal to $0/kWh.
Settlements for Balancing Energy under BEEP Dispatch and Pricing:
PX Generation:
PX generates 10,000 MWh in real -time and had scheduled generation of
10,000 MWh in the hour-ahead market. Its scheduled generation equals
its real-time generation. Payments to balancing market due to mismatch
between scheduled and real-time generation = $0.
PX Load:
PX's load in real-time is 9000 MWh and it had scheduled a load of
10,000 MW in its hour-ahead market. The PX's real-time load falls short
of its scheduled load by 1,000 MWh. The PX sells this energy through
the ISO's real-time balancing energy market. The ISO pays the PX $0/kWh
* 1,000 MWh * 1,000 kWh/MWh = $0 for this 1,000 MWh of energy.
PX Total:
The PX receives $0 from ISO for net 1000 MWh of balancing energy the PX
sells.
SC Generation:
SC generates 4,000 MWh in real time and it had scheduled generation of
3,900 MWh in its hour-ahead market. Its real-time generation exceeds
its scheduled generation by 100 MWh. It sells this excess 100 MWh on
the ISO's balancing market for $0/kWh * 100 MWh * 1,000 kWh/MWh = $0.
SC Load:
SC's real-time load is 5,000 MWh and it had scheduled a load of 3,900
MWh in its hour-ahead market. SC's real-time load exceeds its scheduled
load by 1,100 MWh. SC buys this 1,100 MWh through the ISO's real-time
balancing energy market. It pays the ISO $0/kWh * 1,100 MWh * 1,000
kWh/MWh = $0.
SC Total:
Pays $0 to ISO for net 1,000 MWh of balancing energy it consumes.
Effect of SC's Strategy if BEEP Dispatch and Pricing were Used by ISO:
SC is able to game the system to obtain 1,000 MWh of free energy from
the PX.
6
2.2.2 REAL-TIME MARKET CLEARING DISPATCH
To see that SC was able to game the system to obtain 1,000 MWh of free energy,
look at the prices and generation levels that would have occurred if the ISO
were to clear the real-time energy market. To clear the real-time balancing
energy market, the ISO would make the following changes to the dispatch:
Gen1_SC = 3,900 MWh
Gen2_SC = 1,000 MWh
Gen1_PX = 5,000 MWh
Gen2_PX = 4,100 MWh
Gen2_PX is now the marginal unit in the real-time dispatch. It sets the market
price to $0.06/kWh.
Settlements for Balancing Energy under Market Clearing Dispatch and
Pricing:
PX Generation:
PX generates 9,100 MWh in real -time and had scheduled generation of
10,000 MWh in the hour-ahead market. Its real-times generation falls
short of its scheduled generation by 900 MWh. The PX buys this as
replacement energy on the ISO's balancing energy market and pays the
ISO $0.06/kWh * 900 MWh * 1000 kWh/MWh = $54,000.
PX Load:
PX's load in real-time is 9000 MWh and it had scheduled a load of
10,000 MW in its hour-ahead market. The PX's real-time load falls short
of its scheduled load by 1,000 MWh. The PX sells this energy through
the ISO's real-time balancing energy market. The ISO pays the PX
$0.06/kWh * 1,000 MWh * 1,000 kWh/MWh = $60,000 for this 1,000 MWh of
energy.
PX Total:
The PX receives $6,000 from ISO for net 100 MWh of balancing energy the
PX sells on the ISO's balancing energy market.
SC Generation:
SC generates 4,900 MWh in real time and it had scheduled generation of
3,900 MWh in its hour-ahead market. Its real-time generation exceeds
its scheduled generation by 1,000 MWh. It sells this excess 1,000 MWh
on the ISO's balancing market for $0.06/kWh * 1,000 MWh * 1,000 kWh/MWh
= $60,000.
SC Load:
SC's real-time load is 5,000 MWh and it had scheduled a load of 3,900
MWh in its hour-ahead market. SC's real-time load exceeds its scheduled
load by 1,100
7
MWh. SC buys this 1,100 MWh through the ISO's real-time balancing
energy market. It pays the ISO $0.06/kWh * 1,100 MWh * 1,000 kWh/MWh =
$66,000.
SC Total:
Pays $6,000 to ISO for net 100 MWh of balancing energy it consumes.
2.3 POSTSCRIPT TO EXAMPLE
There may be ways for an SC to cause the PX to over or under schedule load in
its forward markets. There is nothing to prevent a large consumer/generator from
buying some of its energy from a UDC and self-providing the rest. Such a party
could act as an SC for part of its load. The UDC would have to forecast the
portion of that party's load that the UDC must serve and include it in its bid
to the PX. Such a consumer could easily induce forecasting errors in the UDC's
forecast by shifting demand in real-time between: - that which it self-provides,
and - that which the UDC provides.
It would need two meters and the ability to direct load to be served from its
"SC" meter to its UDC meter and vice versa.
3.0 POSSIBLE FIX
Fixing this problem may be a bit tricky. We will probably never be able to
eliminate gaming opportunities. After all, we are dealing with complex markets
and parties wish to maximize their profits. Developing strategies to maximize
profits may be viewed as positive. What we want to eliminate are incentives that
induce pathological behavior. In revising the protocols to eliminate one
problem, we must be sure that the fix does not open even worse gaming
opportunities.
The real time dispatch problem is not a static optimization problem. That is, we
are not looking at a single "snap-shot" or point in time. We actually have a
control problem in which time is an important variable. For example, we may
minimize costs over an hour period. Within the hour, want to find:
- - The dispatch of each resource in each five minute interval in the hour
- - Take into account dynamic characteristics that limit changes in
operating point from one five minute period to another (e.g. ramp
rates)
- - Meet changing loads over the hour
- - Take into account dispatch of each resource at the start of the hour
and the desired dispatch of each resource at the end of the hour
(targets).
Solving an optimization problem that treats coupled time periods (e.g. 5 minute
periods in an hour) is theoretically and practically possible. However, we most
likely could not have such a system in place in 1998. As such, we should look
for simpler problems that will give us reasonable performance but which are
closer to currently available software. Today's economic dispatch software only
looks at a single point in time. We will restrict ourselves to looking at
similar formulations.
8
Our goal is to develop rational market prices for energy purchased and sold on
the ISO's real-time balancing energy market. The goal is to use marginal cost
based pricing for this market. As in congestion management in the forward
markets, we can not define stable and appropriate marginal costs unless we
dispatch the ISO's balancing energy market to minimize costs subject to
appropriate constraints.
In the real-time balancing energy market, we cannot impose market separation
constraints between the different scheduling coordinators. The ISO will not have
the metering and telemetry that would be needed for such an approach.
Consequently, the ISO's real-time balancing energy market must operate as a
pool.
This is not to say that the entire California market structure is a pool. It
still has separate forward energy markets for the different SCs and the PX.
These parties schedule energy in their forward markets according to there own
rules. They compete for transmission in the ISO's transmission forward markets
to support these energy (and ancillary services) schedules. The SCs and the PX
may voluntarily bid some of their resources to the ISO's real-time balancing
energy market. Once these resources are bid to the ISO, the ISO will adjust
their dispatch as in a pool. Only the real-time balancing energy market is a
pool.
To start, we will ignore interzonal congestion.
We will look at twelve five-minute dispatch periods in an hour: t[0] to t[1],
t[1] to t[2], ..., t[11] to t[12], where t[0] is the start of the hour and t[12]
is the end of the hour. At the start of the s(th) period, we will know the
dispatch of the resources at t[s-1].
For the revised real-time dispatch, we must have forecasts of load plus losses
at the end of the hour. This forecast may be updated at the start of each five
minute period:
[FORMULA OMITTED]the forecast at t[s] of requirements (load plus losses) AT
THE END OF HOUR.
We must also have a measure of the load plus losses AT THE START OF A FIVE
MINUTE PERIOD:
[FORMULA OMITTED]the load plus losses at t[s-1].
The ISO will calculate a forecast of the load plus losses that it will be
serving by the end of a five minute period:
[FORMULA OMITTED] = the forecast made at t[s-1] of the requirements (load
plus that will exist AT THE END OF PERIOD S.
9
For scheduling coordinator k (SC k), we will have resources (generation and load
reductions) that were scheduled in the forward markets. These are assumed to
operate as scheduled in the forward markets:
[FORMULA OMITTED] is the scheduled effective real power generation from
resource i (i e SG[k]) at t[s]. This should be constant across all t[s] in
the hour.
[FORMULA OMITTED] is the scheduled reduction of demand for load j (j e SL[k])
at t[s]. This should be constant across all t[s] in the hour.
Scheduling coordinator k (SC k) may bid resources (generation and loads) that
the ISO can adjust to meet real-time balancing market needs. We will model these
as additional logical resources that are separate from the resources that SC k
scheduled in the forward markets:
[FORMULA OMITTED] is the effective real power generation from resource i (i e
AG[k]) at t[s].
[FORMULA OMITTED] for i e AG[k] is the adjustment range bid by SC k
[FORMULA OMITTED] is the ramp rate for real power generation from
resource i (i e AG[k]).
[FORMULA OMITTED] is the convex price function for effective generation
from resource i
[FORMULA OMITTED] is the reduction of demand for load j (j e AL[k]) at t[s].
[FORMULA OMITTED] for j e AL[k] is the adjustment range bid by SC k
[FORMULA OMITTED] is the ramp rate for load adjustment from load j (j e
AL[k]).
[FORMULA OMITTED] is the convex price function for reducing load j.
The ISO solves a small extension of the standard economic dispatch optimization
problem to dispatch and price real-time balancing energy:
[FORMULA OMITTED]
[FORMULA OMITTED]
[FORMULA OMITTED]
[FORMULA OMITTED]
[FORMULA OMITTED]
Since this is essentially the standard economic dispatch problem, this
formulation should greatly simplify the development of systems for 1/1/98.
10
To price balancing energy, we can simply find the resource that is on the margin
in the real-time balancing energy market.
This pool formulation of the real-time balancing energy market should simplify
the treatment of real-time congestion management. We could merge these
constraints into the DC OPF used for interzonal congestion management in the
forward markets. We would also remove the market separation constraints. The
result would be a transmission constrained economic dispatch problem. This
problem could be solved to dispatch the balancing energy market and develop
locational marginal costs. The same software that is being developed for
interzonal congestion management could be used to solve this problem.
The size of the DC-OPF formulation could possibly pose problems to solving it
every five minutes. In this case, we could shrink the formulation to treat zones
only. We would replace the DC power flow formulation by a network flow
formulation. This should be adequate (particularly if the interzonal network is
radial). We would then solve the much smaller problem using highly efficient
network flow software.
11