Advanced process control in electric utilities
Model-based predictive control (MPC) is a valuable, proven advanced process control (APC) approach that has been used for some time, especially in the chemical and hydrocarbon processing industries. It has been underutilized until recently in the power industry, and in that segment, MPC is now also showing dramatic results. The strength of MPC is in its ability to model and predict dynamic process behavior in an extremely precise manner.
Putting MPC to work
Since it only takes one or two failures to make operators never turn on the MPC again, it is important for MPC to be robust under all circumstances, lest it suffers the fate of yet one more solution that was abandoned in-place. MPC is typically employed to reduce process variability, resulting in operation closer to target specification limits, as shown in Figure 1. Once it is applied, the smaller deviations allow the process to run closer to either the upper or lower limit, whichever is desired, in order to enhance the process. Whether the enhancement goal is to increase throughput, improve product quality through reduced variability, increase profit, or any combination of those benefits, MPC can be a significant aid in enhancing the operational excellence and profitability of a process.
Figure 1. Reduction in variance after MPC is applied.
APC typically uses a mathematical model. A popular choice is auto regressive with exogenous input (ARx), to approximate the dynamic response of the process. The ARx model approach is proven to show superior performance with unmeasured dynamic load changes and load dispatch, which is a typical disturbance in an operating plant. This model is representative of both the steady state and dynamic gain of the process. Tracking how various outputs from the model change as various inputs change, the associated process responses can be modeled and used as a mathematical approximation for the control action for various variables like:
• Dead time
• Multiple input/output interactions
• “Shape” of the transient response, and
• Steady-state gain.
One way to make APC more robust is to enhance the ARx model with a finite impulse response (FIR) model. A FIR model increases the accuracy of predictions by expanding the number of variable terms within the control model. This combination, a key part of Connoisseur MPC advanced process control software from Invensys Operations Management, helps provide the greatest accuracy in describing real-world processes. Either portion of the combined method would not be sufficient alone in modeling real-world processes; this would result in substantial approximations, large errors, and poor control. However, by combining both ARx and FIR into a single controller matrix, the accuracy and dynamic performance improves dramatically. This robust solution handles real-world cases as well as special cases, because of the rigor attained from including complex logic with the MPC model.
Building better accuracy
There are a few ways to enhance the robustness of an MPC model. Most processes typically consist of multiple control loops, which under favorable circumstances can work together to provide adequate control. However, there is the possibility that different control loops can interfere with one another. In order to create a more robust MPC model, a technique called decoupling is employed. Decoupling compensates the responses from multiple control loops so that the isolated loops can be addressed separately. Decoupling the control loops allows the MPC model to consider all control variables together, allowing for the best control for the process as a whole.
Once those independent models have been appropriately modeled, the models can be combined to yield a more robust model of the overall process. Another component, linear programming (LP) optimization, locates an operating point that maximizes the profit within the operating constraints of the control matrix, based on the MPC model. Since this is included as part of the MPC model, the LP profit function quantifies all relevant process factors. LP optimization assures the most profitable operation based on the robust MPC model.
A third way to improve the robustness of an MPC model is by enabling adaptive control to manage changing process conditions in real time. Most processes change their behavior over time. This means that the MPC model accuracy changes (degrades) over time from the original design. One way to combat that is to enable the MPC model to adapt to the process, and update the control approach accordingly. The result is a model that continues to remain robust even as the process changes.
Feedwater heater example
The power industry provides two recent examples where MPC can really make an impact. The first is feedwater heater control, where the application of MPC allows for improved Rankine cycle efficiency in a power plant. The Rankine cycle (Figure 2) is used in conventional fossil-fired steam power plants to maximize power plant efficiency and power generation.
Figure 2. Basic Rankine cycle: the control challenges stem from the fact that all elements of the system are interrelated. You can’t change one without affecting others, which is why it is such a good example of a process that benefits from MPC.
The design of the feedwater heaters is based on information from the turbine, boiler, and various subsystems, while still in its design phases prior to construction of the plant. The turbine extraction points determine the final feedwater heater temperature as a function of turbine steam flow. The boiler then heats the feedwater to generate superheated steam, targeting the desired operating pressure and temperature. Drum-style steam power plants have fixed surface areas for each functional area of the steam process; economizers for water preheating, evaporators for steam generation, superheaters for superheating steam, and reheaters for reheating steam. These surface areas are designed to achieve specific design steam temperatures for design steam flows and fuels.
However, in actual operation, modifications to a boiler section surface area, or changes in fuel quality (coal energy value can vary greatly, depending on the source), can result in an energy distribution imbalance and an inability to attain steam temperatures as originally designed. Controlling the high-pressure feedwater heater outlet temperature provides an important additional element of control to attain design steam temperatures. The potential improvement in cycle efficiency through achieving design steam temperatures more than offsets any efficiency loss due to cooler feedwater. So, in this case, feedwater heater outlet temperature control becomes a vital component of an MPC steam temperature control system. An additional benefit is increased generation output and the potential for reduced NOx emissions.
The objective of a feedwater boiler is to heat the feedwater to generate superheated steam. However, as noted, sometimes the original design is not adequate to provide the desired superheated steam temperatures because of deviations in the quality of the fuel, design changes in the boiler, or emission regulation constraints. The objective of applying MPC to a feedwater heater is to increase the short and long-term load generation capacity, while reducing emissions and increasing efficiency.
Applying MPC makes the control approach more effective by adding a key manipulated variable to the already large list of manipulated variables that influence boiler steam temperatures, load generation, and emissions. This can be done because the combination of ARx and FIR in the control approach allows additional variables to be considered. Since boiler feedwater temperature directly affects the superheat steam temperature, the high-pressure feedwater heater outlet temperature can be controlled to attain and sustain design steam temperatures. The improvement from MPC on the feedwater heater outlets increases overall generation levels and reduces the potential emissions.
Burner optimization example
A common issue in coal-fired burners for utility boilers is that the distribution of coal is frequently uneven, resulting in inefficient operation. Some areas within the burner have more coal than others, causing oxygen imbalances. This can cause problems since the burner was designed for equal distribution of fuel and oxygen. The day-to-day reality is that coal distribution differs depending on which coal pulverizers are operational, or simply due to a phenomenon called roping. Typically, burners have numerous air registers to provide a steady stream of oxygen for combustion. However, the air registers are often calibrated during system start-up and rarely recalibrated for ongoing changes in operating conditions.
An MPC model is also built and calibrated during start-up, including an online adaptive feature that allows the model to recalibrate the dampers controlling the air registers to account for the changing conditions in real-time. The MPC model tests the parameters and updates itself, automatically, without operator intervention. Since this is a continuous process, operators are better able to deal with potential mismatches between the MPC model and the process. Automation of the damper calibration on the air registers allows the system to adapt to changing conditions and gives the system a chance to attain additional benefits, such as lower NOx and CO emissions with associated reduction in ammonia consumption.
Payoffs for plants
Model-based predictive control provides a sound basis for applying advanced process control that can be used to improve process profitability by enhancing quality, increasing throughput, and reducing energy consumption. Combining the ARx method with the FIR method yields a flexible mathematical model that is capable of controlling real-world problems. This can also support improved quality and increased profit, typically providing a return on investment in as little as three to six months, plus the ongoing financial benefits that continue to accrue over time, as shown in Figure 3.
Figure 3. Typical financial benefit over time for an MPC application.
A variety of industries, and electric utilities in particular, can benefit greatly from the sophisticated capabilities of modern MPC platforms. The ability to incorporate a wider range of variables makes the control solution more robust, allowing, for instance, vastly improved steam temperature control. The tighter control behavior of MPC also helps power generation processes meet regulatory requirements and cope with changing load conditions by achieving reduced variability and compliance with multiple targets and varying load demands.
Joseph McMullen is product marketing manager of advanced applications for Invensys Operations Management.
For more information, visit: