Professional Documents
Culture Documents
Boiler Und Turbine
Boiler Und Turbine
Keywords
Ancillary services, advanced control, fossil plant.
Abstract
This article presents the new advanced coordinated controller implemented at Cordemais Fossil Power
Plant – France.
Cordemais Power Plant – Unit 5 is a 600 MW once-through coal-fired plant. The boiler, which is very
sensitive to load transients, is mainly operated in low ramp rate load-following mode or at base load. In
particular, non linear increases in waterwall temperature are frequently observed at the bottom of the
boiler even during slow load drops. This is the main reason why the existing controller that coordinates
the boiler and the turbine aims at protecting the boiler to the detriment of turbine performance (this
control strategy is generally called turbine-following strategy). As a consequence, the plant does not
provide ancillary services to the grid.
From a fleet optimization point of view, this inability strongly penalizes EDF. Therefore, EDF launched
a project in order to allow Cordemais 5 to provide ancillary services, in spite of its boiler sensitivity.
The first stage of the project consisted in retuning the existing controller. However, the results of this
optimization were unsatisfactory. Indeed, the boiler-turbine system is a very complex non-linear and
multivariable process that is hard to control with PID-based controllers. As a result, EDF decided to
develop and implement a new advanced controller (NAC).
The NAC could not be implemented in the existing DCS (MicroZ – YOKOGAWA), for it is essentially
matrix-based. As a consequence, it was implemented in an industrial PC dedicated to the NAC. The
operators can switch from the existing controller to the NAC thanks to a bumpless transfer system that
was developed.
This document describes the stages that led to the implementation of the NAC at Cordemais 5. The first
part deals with the principles of the NAC and presents simulation results :
Introduction
Cordemais is an EDF power plant made up of three operating units:
- Unit 2 (700 MW oil-fired unit) started operation in 1976.
- Unit 4 (600 MW coal-fired unit) started operation in 1983.
- Unit 5 (600 MW coal-fired unit) started operation in 1984.
In October 2005, EDF and the French independent system operator (ISO) known as RTE signed a new
protocol regarding the ancillary services supply. EDF undertook the task of reporting each of its units
performance (in regard to ancillary services) to RTE before June 2006. This declared performance will
be officially registered as part of an additional clause to the protocol.
Cordemais – Unit 5 is part of RTE’s grid and as such, falls under the new protocol. However, Cordemais
5 does not provide ancillary services, which particularly penalizes EDF’s fleet optimization. This poor
performance is due to a very conservative control strategy that was designed in order to protect
Cordemais 5 highly sensitive boiler. To improve the load ramp rate, tests performed in 2004 at
Cordemais 5 highlighted the necessity for modifying the existing controller that coordinates the boiler-
turbine system.
Plant staff first tried to optimize the existing controller. However, retuning the parameters in the DCS
(MicroZ, YOKOGAWA) did not provide any improvement. EDF’s experience since 1998 in advanced
control applications [1] led plant personnel to decide to change the control strategy used at Cordemais 5.
A new project started up in September 2005 with the aim of developing and implementing the NAC in a
dedicated industrial PC.
This paper begins by going over the principles of the NAC and details the results obtained in simulation
before implementation. The project is particularly innovative since a new multivariable anticipation
technique was used in order to better Cordemais 5 load-following performance while taking into account
boiler sensitivity. Next, the article describes the interconnection between the MicroZ DCS and the
industrial PC. Finally, it shows the technical improvements and economic benefits obtained from the
implementation of the NAC.
General glossary
Peb : Electrical load.
Ps : Superheat pressure.
Hsep : Steam enthalpy at separator.
Psep : Pressure at separator.
Tsep : Temperature at separator.
Copyright 2006 ISA. All rights reserved. www.isa.org
Presented at the 16th Annual Joint POWID/EPRI Controls and Instrumentation Conference
49th Annual ISA POWID Symposium, 4-9 June 2006, San Jose, California
TSBT : Low temperature superheater output temperature.
Qv : Steam flow.
Ts : Superheat temperature.
Tr : Reheat temperature.
Qis : Desuperheater water flow.
Qis/Qea : Desuperheater water flow ratio.
P0 : Base load demand.
N.Pr : Secondary reserve demand.
K.∆f : Primary reserve demand.
ref Peb : Electrical load setpoint.
ref Ps : Superheat pressure setpoint.
ref Hsep : Setpoint of steam enthalpy at separator.
ref Qc : Coal flow setpoint.
ref Qea : Feedwater flow setpoint.
ref OHP : High pressure throttle valves opening setpoint.
ref Qis : Desuperheater water flow setpoint.
ref Ib : Burner tilting.
NAC : New advanced controller.
MBA : Model-based anticipation.
DCS : Distributed control system.
PID : Proportional integral derivative.
GF : Function generator.
Process modeling
In order to develop the NAC, a dynamic model of Cordemais 5 boiler and turbine was necessary. The
model to be built is a process-only model and as such, does not include any controller. Open-loop tests
were performed on-site to identify the process behavior.
The tests performed in open loop (Peb, Ps, Hsep, Ts and Tr existing controllers disconnected) consisted
in applying steps to ref Qc, ref OHP, ref Qea, ref Ib and ref Qis (see Figure 3).
Figure 5 : Process outputs and model outputs Figure 6 : Model step responses
Comparison between low load and high load model step responses
Figure 9 presents a comparison between low load and high load model step responses (data centered
around zero). This figure allows to assess process behavior differences according to the load. The effect
of ref Qea on TSBT as well as the effect of ref Qc on all outputs seem to be particularly sensitive to the
electric load.
P.S. : A low load model in humid separator configuration was build from the low load model in dry
separator configuration. It was used in simulation in order to test NAC’s robustness at low load (the
plant never operates in humid separator configuration at high load).
Innovative principles of the NAC
Overall structure of the NAC
The analysis of the existing controller as well as the strong couplings observed between the boiler and
the turbine at Cordemais 5 led EDF to propose a new multivariable controller that :
- takes explicitly into account physical coupling of the process.
- uses multivariable control methods in order to efficiently coordinate the main actuators.
- elaborates appropriate setpoints in order to “accompany” the actuators during load transients.
The main feature of the NAC is that it is fundamentally multivariable. Three actuators – ref Qea, ref Qc
and ref OHP – simultaneously control three process variables – Peb, Ps and TSBT. In the new controller
structure, the Hsep loop is replaced by the TSBT loop. This change allows to indirectly control Tsep and
secures a margin regarding saturation temperature. In other words, it explicitly guarantees that Tsep will
always be above saturation temperature, and consequently avoids transition to humid separator.
The other important characteristic of the NAC lies in the use of anticipation signals (ref Ib and ref Qis)
in order to better control the electric load.
The three setpoints of the NAC are :
- ref TSBT that replaces the enthalpy at separator setpoint (ref Hsep) and varies according to P0
+ N.Pr. The TSBT load-varying setpoint is deduced from the existing Hsep load-varying
Copyright 2006 ISA. All rights reserved. www.isa.org
Presented at the 16th Annual Joint POWID/EPRI Controls and Instrumentation Conference
49th Annual ISA POWID Symposium, 4-9 June 2006, San Jose, California
setpoint. The margin between Tsep and saturation temperature is approximately 10°C and
theoretically guarantees operation in dry separator configuration.
- ref Peb = P0 + N.Pr – K.∆f where :
P0 is the base load demand.
N is the secondary reserve signal (see below “ISO criteria for load response”).
Pr is the secondary reserve of the plant (in MW) declared to the ISO.
K is the frequency characteristic of the plant (in MW/Hz).
∆f = F – F0 is the instantaneous deviation from 50 Hz, or frequency error (in Hz).
- ref Ps that varies according to P0 + N.Pr.
With the NAC, the control of Qis/Qea is no more ensured. Indeed, the interest of this control loop is
rather limited considering that it only aims at guaranteeing a non-zero Qis. Since it unnecessarily
disturbs the boiler, it was simply removed. In fact, on-site tests show that the process naturally
guarantees – on average – a non-zero Qis.
As far as the existing Ts and Tr control loops are concerned, they were not replaced since :
- they provide satisfactory dynamic behavior to Ts and Tr.
- ref Ib and ref Qis (actuators of Ts and Tr control loops) are measured as anticipation signals in
the NAC.
To conclude with the overall structure of the multivariable controller, Psep is measured by the NAC (but
not controlled) as a tendency signal in order to better control Ps variations.
Internal structure of the NAC
The model used in order to synthesize the controller is called “synthesis model”. In this advanced control
application, the synthesis model is a ponderated mean model obtained from the low load and high load
identified models.
A model-based advanced controller was built from the synthesis model using Easy H∞ software
(commercial product developed by EDF and IPSIS). The Easy H∞ user can specify dynamic performance
for the controller, and the software calculates the parameters of controller matrixes A, B, C and D.
Moreover, the H∞ technique can desensitize the system to changes in operating conditions (to a certain
extent) and guarantees stability and performance in face of uncertainties such as unmodeled dynamics
[2]-[4]. For instance, in this application, uncertainties on ref Qc were introduced so as to model process
dynamics variations caused by coal quality variations.
The boiler sensitivity was implicitly taken into account in the controller synthesis. Indeed, one of the
controller synthesis objectives is based on a combination of ref Qc and ref Qe. To be more accurate, the
objective was defined so that ref Qc/(1 + Tc.p) and ref Qea/(1 + Tea.p) cancel each other most of the
time. Tc = 150 seconds represents the theoretical time effect of ref Qc on waterwall temperature, whereas
Tea = 15 seconds stands for the approximate time effect of ref Qea on waterwall temperature. By tending
to cancel [ref Qc/(1 + Tc.p) – ref Qea/(1 + Tea.p)], the controller implicitly tends to lower stress on
waterwall.
The most innovative part of the project lies in a model-based anticipation (MBA) that was developed
specifically for this application using H∞-like methods. The MBA elaborates an additive control signal
from the primary and secondary reserve demand. As a consequence, the MBA is activated on ancillary
services demand only. The additive control signal generated by the MBA is distributed in a coordinated
way on P0, ref Ps, ref TSBT, ref Qc, ref Qea and ref OHP according to the model dynamics. In other
words, the MBA “helps” the H∞ controller by anticipating the ancillary services demand directly on
actuators : the MBA is at the heart of the performance provided to the process.
Figure 10 shows a schematic representation of the NAC including the H∞ controller and the MBA.
Closed-loop simulator
A closed-loop simulator was developed in Matlab-Simulink simulation environment. It consists of a full
range model [300 MW-600 MW] interconnected with either the existing controller or the NAC.
The full range model was obtained from the low load and high load identified models. In fact, it is a
continuous family of interpolated models that covers a large scope of operating conditions. For example,
in a simulation of a typical load ramp from 350 MW to 550 MW, the full range model continuously
switches from the low load identified model to the high load identified model.
Furthermore, the existing controller and the NAC were programmed in the simulator and interconnected
to the full range model, which allows the user to simulate scenarios with either controller.
P.S. : In order to interconnect the existing controller, the dynamic behavior of Hsep was identified and
added as an output of the full range model.
To validate the simulator with the existing controller, various transients were tested in closed-loop. The
most validating transient is an on-site frequency step applied to Cordemais 5 in 2004. Figure 11 shows a
process output and a simulator output (electric load) for this transient.
The simulator is rather representative of the closed-loop behavior of the process operated with the
existing controller.
ISO criteria for load response
Primary reserve
• Primary reserve : The portion of a generating unit’s unloaded capability which can be “immediately”
loaded, or loaded capability which can be “immediately” unloaded in response to K.∆f where :
- K is the frequency characteristic of the plant (in MW/Hz).
- ∆f is an increase or drop in grid frequency.
The criteria that precisely define the notion of “immediately” are detailed below.
• 1st criterion for primary reserve release : For a 50 mHz grid frequency ramp in 10 seconds, the plant
is in accordance with ISO rules if :
- 50% of the expected load variation is released in 20 seconds (see Figure 12).
- 90% of the expected load variation is released in 60 seconds (see Figure 12).
10 s
90%
K.50mHz
50%
t
tini tini + 20 tini + 60
• 2nd criterion for primary reserve release : For a 50 mHz grid frequency decreasing ramp in 10
seconds, the plant is in accordance with ISO rules if plant load response is above [K.(F(t)-
F(tini))/(1+Tf.p) – q] for at least 75% of observation window (see Figure 13) with :
- F = grid frequency in Hz.
- tini = beginning instant of the ramp.
- tini < t < tini + 900.
- Tf = 20 seconds.
- Observation window : [tini ; tini + 900].
- q is the frequency measurement quantification.
10 s
q
K.50mHz
K.(F(t)-F(tini))/(1+Tf.p)
t
tini t1 t2 t3 t4 tini + 900
MW
P0 - Pr
t
tini tini + 120 t1 t2 tini + 920 tf
Figure 19 : Simulation of criterion for secondary reserve release – Other process variables
Industrial PC solution
Since the NAC cannot be implemented in the existing DCS (MicroZ – YOKOGAWA), EDF decided to
integrate it in an industrial PC connected to the MicroZ DCS through a local network. The existing
operation modes were conserved (“Classical automatic mode” and “Manual mode”). The activation of
the NAC constitutes a 3rd operation mode (“NAC automatic mode” or “Ancillary services mode”). The
NAC automatic mode can only be activated when plant staff already operates the process in Classical
automatic mode.
The operators can switch either from the Classical automatic mode to the NAC automatic mode or from
the NAC automatic mode to the Classical automatic mode thanks to a bumpless transfer system
implemented in the existing DCS and the PC.
Principles
The interconnection between MicroZ DCS and industrial PC was done by using standard modules.
These modules allow to concentrate and convert each MicroZ card TTY physical protocol toward only
one Ethernet port. A driver integrated in the industrial PC platform interprets MicroZ data and updates
cyclically a real time database. This data feeds the NAC integrated in this same platform.
The implementation of the whole system is easy and reliable. The C code corresponding to the NAC
block diagrams developed in Matlab-Simulink is generated automatically using a specific toolbox. The C
code is then compiled using the “control module” of the industrial PC.
Description
The industrial PC platform that accommodates the NAC is made of three modules :
- A real time server.
Copyright 2006 ISA. All rights reserved. www.isa.org
Presented at the 16th Annual Joint POWID/EPRI Controls and Instrumentation Conference
49th Annual ISA POWID Symposium, 4-9 June 2006, San Jose, California
- A control module in which NAC algorithms are embedded.
- A communication driver that interprets MicroZ data through an Ethernet interface.
Every MicroZ card is connected to the Ethernet interface through two standard conversion modules :
- Module 1 : Current loop to RS232.
- Module 2 : RS232 to Ethernet.
The driver simultaneously collects data from many MicroZ cards. The NAC cyclically processes this
data in order to calculate control signals (such as ref Qc, ref Qea and ref OHP). The driver then
simultaneously feeds the appropriate MicroZ cards with the signals generated.
PC platform material
The PC platform must be compatible with an industrial environment. As a consequence, there are no
integrated moving parts in the PC :
- Cooling is performed without ventilation fan.
- Hard disks are Compact Flash or Flash IDE
disks.
The operating system is Windows 2000 Pro or
Windows XP Embedded.
The PC that was used in order to implement the NAC is
a Wincomm WPC-651-P1G (see Figure 23). It is high
pressure and high temperature resistant. The main
technical features of the PC are :
- Processor : VIA Eden 1 GHz.
- SDRAM : 256 MB. Figure 23 : Industrial PC
- Graphic Card : VIA VT8606 AGP 4x.
- Size : L150 mm – W177 mm – H66 mm.
- Power supply : 85~264V AC, 5V/12V DC
P.S. : The overshoot is due to an uneven distribution of ref OHP signal on high pressure throttle valves.
This non-linear phenomenon (that was not observed during the open-loop tests) will be removed by plant
staff.
Figures 24 and 25 show that the NAC allows the plant to roughly respect the ISO criteria.
Figures 26 through 28 present other process variables behavior in response to two ISO criteria transients,
and a classical N.Pr and K.∆f transient. Figures features are as follows :
- The green dotted line represents the electrical load setpoint.
- The red solid line represents the load response with the NAC.
Figure 26 : 2nd criterion for primary reserve release – Other process variables
Copyright 2006 ISA. All rights reserved. www.isa.org
Presented at the 16th Annual Joint POWID/EPRI Controls and Instrumentation Conference
49th Annual ISA POWID Symposium, 4-9 June 2006, San Jose, California
Figure 27 : Criterion for secondary reserve release – Other process variables
No increase in waterwall temperatures was observed during the tests. However, a long term statistical
study will be carried out in order to assess waterwall stress when the plant is operated with the NAC.
Copyright 2006 ISA. All rights reserved. www.isa.org
Presented at the 16th Annual Joint POWID/EPRI Controls and Instrumentation Conference
49th Annual ISA POWID Symposium, 4-9 June 2006, San Jose, California
Conclusion
This paper briefly presents an advanced control application in a fossil plant. The results of this project
show that advanced control techniques noticeably improved Cordemais 5 performance. The new MBA
designed using H∞-like techniques is at the heart of this performance.
The immediate benefits for the plant were :
- A new Ancillary services mode, which is compatible with a safe operation of the process (in
particular regarding boiler sensitivity). This new mode is a valuable asset for EDF from a fleet
optimization point of view.
- A lower occurrence of waterwall temperature increases in normal operation.
The net present value (difference between the sum of the discounted cash flows which are expected from
the investment and the amount which is initially invested) of the project, estimated from 2006 through
2008, approximately comes at 20 M$.
The following step will consist in applying this innovative solution to EDF’s Q600 fleet.
References
[1] L. Deprugney, J.B. Liters, « New excess air controller at Le Havre Power Plant », 14th Annual Joint
ISA/POWID/EPRI Controls and Instrumentation Conference, 2003.
[2] J. Doyle, K. Glover, P. Khargonekar and B. Francis, « State-space solutions to standard H2 and H∞
control problems », IEEE Trans. Aut. Control, vol. 34, pp. 831-847, Aug. 1989.
[3] J. Doyle and G. Stein, « Multivariable Feedback Design : Concepts for a Classical/Modern
Synthesis », IEEE Trans. Autom. Contr., vol. 26.1, pp. 1-16, 1981.
[4] G.J. Balas, J.C. Doyle, K. Glover, A. Packard and R. Smith, « µ-analysis and Synthesis Toolbox »,
The Math Works Inc, 1993.