RULE-BASED ENERGY MANAGEMENT SYSTEM

Download this paper in Adobe PDF format.

ABSTRACT

Real time pricing of electricity and rising fossil fuel costs are causing renewed interest in energy management systems (EMS). This paper describes a rule-based EMS which has been implemented at several large industrial powerhouses. It is a PC-based supervisory system which is interfaced to a DCS. The EMS includes boiler allocation, turbine load allocation, bark optimization, real-time pricing (RTP) tie-line control and coordinated header pressure control. It optimizes the powerhouse operations to meet the steam and electrical requirements of the mill at minimum cost subject to all of the operating constraints imposed on the generation equipment.

Steady state optimization methods, such as linear and non-linear programming, are not suited for online optimization of powerhouse operations since the process is rarely at steady state. Instead, the optimization strategy is reduced to a fairly small number of fuzzy rules. It has proven to be capable of optimizing the powerhouse operations while keeping the powerhouse within a safe operating envelope.

One of the unique features of the EMS is the safe operating envelope. It is defined by setting up a prioritized list of process, equipment and environmental constraints.

BACKGROUND

The purchased energy (fossil fuel and electricity) cost component for producing a ton of paper is significant and small incremental changes can make a big impact on the profitability of a mill. Minimizing this cost should be one of the major goals at every mill. There are two factors that are starting to have a significant effect on purchased energy bills (which accounts for nearly 10% of the overall manufacturing cost (1) in the U.S. pulp and paper industry). Many utilities are now offering industrial customers a real-time pricing (RTP) rate schedule where the marginal prices usually vary on an hourly basis. The schedule for next day's prices is available about eight hours before they go into effect. This RTP rate results in the "good news-bad news" scenario. The "good news" is that the mill obtains a lower cost supply of purchased power for a significant percentage of the time. Customers can usually purchase all the power needed without worrying about exceeding a previous demand limit and incurring high demand charges. This all allows more flexibility in the operation of the powerhouse and the entire mill. The "bad news" is that the RTP rate can cause large sudden swings in steam generation as the powerhouse maneuvers to respond to the RTP price. The price of power can change dramatically during the summer months. There may be days where the RTP price exceeds $1000/megawatt.hour during the midday hours. During most days in the summer the powerhouse can typically expect to shift from a maximum "buy" period to a maximum "make" period around 11: 00 a.m. Large economic penalties can occur in a short period of time if the powerhouse does not respond quickly to the swing in purchase prices.

The second factor starting to impact the purchased energy bills of mills is the cost of fossil fuels. Prices have recently started to rise after years of remaining fairly steady. The mills can respond by maximizing the amount of steam supplied by their self-generation fuel sources such as bark and black liquor. They can also work to optimize the boiler loading and boiler fuels to produce the mill's steam requirement from the power boilers at the lowest cost. An energy management system (EMS) is a tool that can be used for this task.

The power house has a number of environmental, equipment and process constraints that must be adhered to as the powerhouse equipment is maneuvered to meet the mill's energy demand at the lowest possible cost. Balancing the optimization function with all the constraints is a difficult task requiring a significant amount of operator intervention.

A properly designed EMS can be quite useful to the operation of the powerhouse. It can perform the optimization function while adhering to all constraints. Like the operator, the EMS sacrifices cost optimization whenever a constraint is reached. This results in robust process control. The control priorities are:

1. Meet all environmental constraints (stay out of jail).
2. Meet all equipment constraints.
3. Meet all process constraints.
4. Meet energy requirements at minimum costs ($).

STEAM AND ELECTRICAL NETWORK

The steam and electrical network for a typical pulp and paper mill is shown in Figure 1. There are usually multiple recovery and power boilers for steam generation. More than half of the steam that is generated comes from burning process byproducts (black liquor and bark). Most of the process steam demand is for low-pressure steam (in this case 160 psig and 60 psig steam). Steam is generated at a higher pressure (in this case 900 psig) and throttled to the desired value through turbine-generators. A significant amount of electrical power, termed "extraction power" or "cogeneration", is generated as a result of this throttling action. Pressure reducing valves (PRVs) offer an alternative way to throttle the steam to the lower pressure headers. However, since no power is generated, the steam flow through PRVs should be minimized. Turbine Generator 2 (TG2) has a large condensing stage which can be used to control the purchased power. There is a 60 psig vent to the atmosphere which can be used to control purchased power on a limited basis.

Figure 1 - Typical Boiler -Turbine Network Configuration

There is significant variability in the process steam and electrical power demand. Batch digester operation, wood yard log chippers, soot-blowers, paper machine disturbances and pulping process upsets all contribute to this variability. A "sheet break" on a large paper machine and subsequent rethreading of the sheet can result in large sudden steam demand swings of up to 200,000 lbs./hr. in a period of less than a minute. Sometimes the power boilers must go from maximum load to minimum load and back again to maximum load within several minutes. Power boilers seldom operate at steady state conditions unless they are base loaded (i.e. the boiler master is placed in "manual"). It is this variability that makes realtime optimization of the powerhouse operations so challenging. Steady state optimization methods simply do not provide the solution when the process is rarely at steady state.

Boiler Load Allocation

A schematic of the 900 psig header pressure control with the embedded boiler load allocator is shown in Figure 2. The plant master is implemented with a fuzzy matrix controller that offers some significant advantages over a PID version. Fuzzy logic controllers can exhibit superior control performance compared to a PID controller, especially for nonlinear, complex processes. But the tuning of fuzzy controllers is a trial and error procedure that involves adjusting many parameters. A simple method to help with the tuning of fuzzy controllers has been developed. By overlaying a phase-plane plot on the rule matrix and analyzing the phase-plane trajectories, it becomes relatively easy to adjust membership functions and modify the rules to obtain the desired trajectories.

Figure 2 - 900 psig Header Pressure Control with Embedded Boiler Allocator

The fuzzy controller executes once per second and sends a request to the boiler load allocator for an incremental steam change. A 900 psig header steam balance calculation is performed. This allows changes in the header pressure to be anticipated so that corrective action can be taken before the actual header pressure changes significantly. This feed-forward is in the form of an incremental steam request which is sent to the boiler allocator. Both requests are summed to obtain a total steam request for the allocator.

The boiler load allocator design involves integration of three distinct functions. A safe operating envelope representing prioritized environmental, equipment and process constraints is defined which the allocator must respect. An optimization method is used which adjusts multiple boilers and fuels to obtain the most economical operating solution. The issue of header pressure control stability is addressed so power boilers with widely varying response capabilities can work in concert. Balancing these three functions is key to a successful design.

The boiler load allocator observes all predefined constraints before adjusting boiler fuel flows. These constraints create a safe operating envelope. Observing constraints prevents boiler damage and keeps the process out of undesirable operating regions. Constraints are prioritized in order of importance. Typical constraints for a boiler are (listed in order of priority):

1. Maintain opacity (6-minute average) below maximum.
2. Keep ID fan speed within control range.
3. Prevent furnace draft from going positive.
4. Maintain drum level in safe range.
5. Prevent excess oxygen from going too low.
6. Keep boiler steam generation within limits.

An incremental steam request intended for a boiler that is constrained is shifted to unconstrained boilers. In some cases, the constraint changes the boiler load so that the process moves back inside the safe operating envelope.

The boiler load allocation problem is analogous to the economic dispatching problem faced by an electric utility company whenever transmission losses can be ignored. For optimal allocation, the utilities must operate the units at equal incremental generating costs. Often the boiler load allocation problem has been posed as a static optimization problem (3). But in reality, the allocation function is embedded in the header pressure control loop that transforms it into a dynamic control problem. Since there are continuous disturbances to header pressure (caused by variations in steam demand), boiler load allocation also takes place on a continuous basis. Direct application of steady state optimization methods does not work for a process that is never at steady state. Instead, a dynamic boiler allocation method is used. In this solution, the optimization method has been converted to an optimization rule set that is integrated into the overall rule set.

An incremental steam generation cost ($/MLbs of steam) is continuously calculated for each boiler (fuel) based on the fuel cost ($/MMBtu), the selected "swing" fuel and incremental boiler efficiency for the selected fuel. This efficiency number is entered based on historical data. An on-line efficiency calculation is not required.

For incremental steam increase requests, boilers and fuels with lower incremental steam costs are favored more than boilers and fuels that have higher costs. All of the boilers move in concert to prevent one boiler from taking all of the load swings. For incremental steam decrease requests boilers and fuels with higher incremental steam costs are favored. In the long run, the most economical boilers and fuels take most of the steam load. The more expensive steam producers are kept at the minimum value. In the short term, if more expensive steam is required for good header pressure control, it is used. If the control is properly tuned the penalty for better control is usually not significant.

The variations in the header pressure error provide the driving function for boiler load allocation. Optimal boiler allocation will be achieved, regardless of the initial allocation. This is demonstrated in Figure 3. All three power boilers start out with each boiler master at 50% load (an arbitrary allocation). The boiler load allocator is turned "on" after 10 minutes. There is no significant change in the process steam demand during the period shown, but the natural noise of the header pressure error causes the boiler to reallocate to obtain the most economical operating condition (coal is the cheapest fuel, then gas and lastly, oil).

Figure 3 - Convergence to an Optimal Allocation from an Arbitrary Allocation

Bark Optimization

Bark optimization is incorporated in the boiler load allocation function. The operator enters a minimum and maximum bark rate limit for PB1 and PB2. It is desired to keep the bark rate for each boiler at its maximum value as much as possible.

In a multi-fuel boiler, each fuel is treated as if it were a separate boiler by the boiler load allocator. The cost of bark ($/MMBtu) is entered as a very low value (significantly less than coal which is the cheapest fossil fuel). This maximizes the bark rates for both boilers.

There is a significant lag time (several minutes) associated with the transport of bark from the bark bin to the boilers. This lag time prevents bark from being an "effective" swing fuel. However, an operator adjustable "aggressiveness" factor is used to allow bark to be treated as a pseudo "swing" fuel and maintain stable header pressure control. The "aggressiveness" factor is discussed in more detail in the next section of this paper.

Normally, bark flow will remain at the maximum limit (entered by the operator) and header pressure control is accomplished by adjusting coal, gas and oil flows. However, there are periods of low steam demand when bark flow must be reduced to prevent excess venting of steam to the atmosphere. When all three fossil fuels are at minimum limits and further steam generation reduction is needed, the boiler allocator will reduce the bark flows of PB1 and PB2. When the process demand increases, bark starts to increase. Bark is considered "somewhat" base loaded, since it always works its way back to the maximum limit.

Header Pressure Control Stability

One of the major challenges of implementing boiler load allocation is to maintain stable header pressure control for all combinations of boilers, fuels and equipment conditions. Boilers have different response times. Variable fuel quality and moisture content can effect the boiler's response time. Mechanical problems can limit the rate of load changes for a boiler.

Multi-fuel boilers, where bark is burned on a traveling grate, seem to create problems. Wet bark, long lag times in the bark feed system and bark "piling" on the grates can make using the boiler for header pressure control quite challenging.

An "aggressiveness" factor is assigned to each boiler fuel. It determines how much a boiler fuel is asked to participate in header pressure control. The factor varies from 0 to 1. When set to zero (0), the fuel does not participate in header pressure control. It becomes base loaded. When set to one (1), the boiler fuel has full participation in header pressure control. For any value in between, there is partial participation. Matching the "aggressiveness" factor to the responsiveness of each boiler is important for achieving stable header pressure control. Reducing the participation of boiler fuels that have poor steaming response is essential. However, there must be at least one boiler fuel (preferably two) that has a fast steam response if satisfactory header pressure control is to be obtained.

Sometimes boiler constraints reduce header pressure control effectiveness. Each boiler's constraints are checked once per second to insure process limits are not being violated. As a boiler approaches a limit, its participation in header pressure control is reduced to zero. When some limits are exceeded, such as boiler steam generation, constraint controllers may make counter control moves to place the boiler back inside the safe operating envelope. Counter control moves are usually to the detriment of good header pressure control. This means that the header pressure is not the highest control priority. In fact, it is the lowest priority.

Turbine Load Allocation

This subsystem provides supervisory control for 160-psig and 60-psig extraction flows of TG1 and TG2 to maximize the total power that is generated. This excludes the power generated by TG2 from condensing flow above minimum condensing. TG2 is assigned primary responsibility for control of the 160 psig header pressure, 60 psig header pressure and purchased power. RTP tie-line control adjusts the load (condensing flow) of TG2. The TG2 controls are operated independently of turbine load allocation. TG1 is assigned primary responsibility for control of 400-psig header pressure and turbine load allocation with secondary responsibility for control of 160 psig header pressure and 60 psig header pressure.

A safe operating envelope for turbine load allocation has been defined that will:

1. Maintain all TG1 parameters (V1, V2, V3, MWs ... etc.) within the minimum & maximum limits.
2. Provide override control for 900 psig header pressure (outside of minimum & maximum limits).
3. Provide override control for 160 psig header pressure (outside of minimum & maximum limits).
4. Provide override control for 60 psig header pressure (outside of minimum & maximum limits).
5. Maintain extraction flows on TG2 in control range for extraction pressure control.
6. Maintain sufficient swing range for TG2's condensing flow to accommodate RTP tie-line control.
7. Give preference to 60 psig exhaust flow over 160 psig extraction flow on TG1.

The turbine load allocator adjusts the 160 psig and 60 psig flow setpoints in TG1's Mark V controller to achieve turbine load allocation. TG1 is slightly more efficient than TG2 for 160 psig and 60 psig extraction flows. Achieving optimal load allocation means maximizing TG1's 160 psig extraction and 60 psig exhaust flows. Both flow setpoints continue to be "bumped" up until a constraint is reached.

To understand how turbine load allocation functions, let us assume that all of the mill's processes are at steady state. Suddenly one of the paper machines increases it's demand for 60 psig steam. TG2's 60 psig extraction pressure control initially provides this additional steam. The turbine load allocator, which is tuned much slower than TG2's pressure controls, then increases the 60 psig exhaust flow on TG1 (constraints permitting). As TG1's 60 psig exhaust flow picks up, the 60 psig extraction flow on TG2 backs off an equivalent amount. Unlike boiler load allocation, turbine load allocation is not embedded within a pressure control loop. This design takes advantage of fast pressure control provided by the turbine controller and has proven to be robust.

ELECTRIC UTLILITY RATE SCHEDULES

Most industrial customers purchase power from an electric utility company on a 15 or 30 minute interval. This type of rate schedule has a demand component and fixed energy charges for on and off-peak periods. The demand charge is usually based on the highest (peak) interval demand in the last 11 or 12 months. Interval demand is the average purchased power over an interval. Exceeding a previously set peak demand may cost hundreds of thousands of dollars since this new peak demand is usually ratcheted in as the minimum demand charge for the following 12 months.

Real Time Pricing (RTP) is a new type of rate schedule offered to industrial customers by some electric utilities. Under the rate schedule there is no demand charge or demand interval. Instead, the price of electricity varies on an hourly basis. Customers can purchase all the power they need without worrying about setting a new peak demand. During summer periods, when the power demand becomes high, the midday hourly price is usually quite expensive. On some days it may even exceed $1000/megawatt.hour. The customer obviously doesn't want to buy any more of this expensive electricity than is absolutely necessary.

RTP TIE-LINE CONTROL

Tie-line control has three modes: 1) RTP, 2) Demand MW and 3) Constant Purchase MW
However, it is the RTP mode that is primarily used in this application. The primary control objective is to adjust TG2's condensing flow to minimize the cost of providing the mill electrical deficit while staying within a predefined safe operating envelope.

Mill Electrical Deficit

The mill electrical deficit is an important concept to grasp when implementing RTP tie-line control. Figure 4 shows the mill electrical deficit components. The electrical deficit is defined as the mill's total electrical power demand minus the power being generated due to the turbine's extraction flows and minimum condensing. There are 3 sources of power that can be used to meet the deficit:

1. Purchased power
2. Forced condensing power
3. Venting (60 psig steam) power

Figure 4 - Mill Electrical Deficit Components

RTP Control Algorithm

A schematic of the RTP tie-line control algorithm is shown in Figure 5. The prioritized constraints are shown in the top part. They define the safe operating envelope. Using this constraint boundary, the turbine is "herded" to stay within the envelope while the tie-line control function is performed. It does not allow condensing to increase when:

1. Condensing flow is high.
2. Condenser vacuum is low.
3. Purchase power is at low limit.
4. TG2 generated MWs is high.
5. TG2 160 psig extraction flow is high.
6. TG2 throttle flow is high.
7. Total power boiler steam generation is very high.
8. 900 psig header pressure is very low.

Figure 5 - RTP Tie-line Control Algorithm

The number in front of the constraint is its priority.

It does not allow condensing to decrease when:

1. Condensing flow is at minimum.
2. Purchased power is at a high limit.
3. TG2 generated MWs is low.
4. TG2 160 psig extraction flow is low.
5. TG2 throttle flow is low.
6. Swing PB steam generation is low.
7. 900 psig header pressure is very high.

Each day the utility provides tomorrow's hourly prices by electronic mail or Internet to each RTP customer. Around 5:00 PM each day, tie-line control automatically downloads these prices (see Figure 6). At midnight the prices are automatically transferred into a buffer for use.

Figure 6 - RTP Hourly Prices

The control system continuously calculates the incremental cost to generate the next megawatt. hour by forced condensing. It is based on the incremental cost of steam generation and the amount steam required to generate a megawatt from forced condensing. The price of condensing power is compared to the cost of purchased power. When it is less expensive to buy power, the control decreases turbine condensing until it encounters a process constraint. When it is less expensive to make power, the control increases condensing until it encounters a process constraint. When the cost to buy versus make is nearly the same, condensing is controlled to mid-range the loads on the power boilers for maximum operating flexibility. The control adjusts the turbine's load to minimize electrical costs only when all variables are within the safe operating envelope. The control sacrifices minimum cost for safe process performance.

The performance of the RTP tie-line control is demonstrated in Figure 7. The mill electrical deficit is slightly over 14 megawatts during the period of these trends. One trace shows the cost ($/hour) to provide the entire deficit by forced condensing. Another trace shows the cost to provide the entire deficit by purchasing it. The third trace shows the actual cost to meet the deficit. The actual cost is usually a combination of forced condensing and purchased power.

For the first 17 minutes of the trend, the cost of purchased power is $25.00/megawatt.hour. The price of forced condensing power is $39.00/megawatt.hour for the entire 60 minutes. During the first 10 minutes, condensing is reduced to minimum and then the utility provides 100% of the mill electrical deficit. At 17 minutes the price of purchased power increases to $60.00/megawatt.hour. It becomes less expensive to make power so condensing starts to increase. A process constraint is reached which prevents the condensing from increasing any further. The absolute minimum cost cannot be safely achieved.

CONCLUSIONS

The rule-based EMS described in this paper has been designed and implemented in several powerhouses. The savings attributed to these implementations are substantial.

The design is based on fuzzy logic control. A new inference engine and defuzzification method is employed. It is the heart of this new supervisory software package. This methodology integrates online optimization and a set of prioritized constraints. A list of process, equipment and environmental constraints are converted to a set of linguistic variables (fuzzy variables) which are used to define a safe operating envelope. When the process is operating inside the envelope, the EMS optimizes the powerhouse to provide process steam and electrical power at the lowest cost possible. The EMS usually operates the process on the boundary of the constraints.

Figure 7 - Typical Performance of RTP Tie-line Control

This new control technology is applicable for many other, online process optimizations in pulp and paper mills and other industrial facilities.

ACKNOWLEDGMENTS

The authors wish to express gratitude to J. Vail, M. Blackburn and R. Stewart, P.E. for editorial assistance.

REFERENCES

1. Pulp and Paper 1995 North American Factbook - North America: Energy Use and Conservation, Miller Freeman, San Francisco.
2. Magnabosco, P.W. and O'Sheasy, M.T. , Tappi J. , 79(4): 107(1996)
3. Thomasson, F.Y. , 1979 Proceedings of the American Power Conference, AMERICAN POWER CONFERENCE, Chicago. P. 853.
4. Snow, C.A. and Realff, M.J. , Tappi J. , 81(12): 142(1998).