Optimizing the Hydro System

First published in Hydro Review, volume 18, no. 5, August, 1999.

by
Alan Livingstone (Project Manager, FPL Energy - Maine, Lewiston, ME)
Douglas I. Smith (Operations Software Engineer, BC Hydro, Vancouver, BC. Canada)
Tung Van Do (President, Powel Technologies Inc, Victoria, BC, Canada)
Charles D. D. Howard (Senior Advisor, Charles Howard & Associates, Ltd., Victoria, BC, Canada)

Abstract

FPL Energy - Maine owns and operates a hydroelectric system on three principle river systems in southern Maine. The former owner of the hydroelectric system, Central Maine Power Co., embarked in 1995 on a project to develop a computerized system for hydroelectric operations scheduling. The purpose of the system was to maximize the value of the limited water resource. Since its inception, the scope has expanded to include all aspects of river management, including use of the operations scheduling software for bidding into the wholesale electricity market, and for re-licensing studies.

The FPLE Hydroelectric System

The FPLE hydro system has 31 conventional hydroelectric stations containing 92 units scattered throughout its service territory. Nearly 57% of FPLE's hydro capacity is located on the Kennebec River with the balance located on the Androscoggin River, Saco River and five smaller tributaries. All but five of the stations are fully or semi-automated, thus requiring a limited operations staff. Additional hydro stations on the Kennebec River, owned by other entities, are affected by operation of the FPLE owned plants. Each of the three river systems currently has it own River Control Center (RCC), manned 24 hours a day, that operates or oversees the operation of all FPLE owned stations on that river.

The FPLE stations range in capacity from less than one MW to over 88 MW, and in head from less than 20 feet to over 150 feet. Unit types include horizontal and vertical Francis, horizontal and vertical Kaplan, horizontal and vertical propeller, and horizontal multiple runner Francis. The 31 stations have a total of capacity of 373 MW.

Harris and Wyman on the Kennebec River, and Gulf Island on the Androscoggin River have ponds of sufficient size to allow weekly cycling. Shawmut and Williams on the Kennebec, and Bonny Eagle and Skelton on the Saco are capable of some cycling on a daily basis. All other stations are classified as run-of-river stations, some with pondage that can be used to regulate flows over a few hours..

The Kennebec and Androscoggin rivers have substantial storage capacity. Reservoirs near their headwaters operate on an annual cycle to store snow melt in the spring and fall rains and to augment flows during periods of low natural inflow in summer and winter.

Overview of the Hydro Modeling System

The Real-time Hydro Operations Model (RHOM) is FPLE's implementation of Charles Howard & Associates Ltd.'s HYDROPS system. It was developed over a period of approximately three years starting from existing HYDROPS software developed for other hydroelectric systems. The modeling system envisioned in 1995 covered all aspects of water and hydro operations management on time scales ranging from near real-time to annual. The time steps used by the computer system range from minutes, to hours, to weeks. Functions included in the operations support system include:

• weekly scheduling of reservoir operations,
  
• hydro maintenance planning,
  
• hourly and seasonal inflow forecasting,
  
• on-line unit loading optimization within each station, and
• system wide hourly scheduling of all generators.

Hourly scheduling of the generating units in the system has developed into the central part of the RHOM project. With the advent of retail competition, there are clear advantages to having an operations model that can quickly provide an optimum dispatch schedule based on projections of market prices, future inflow forecasts and the current watershed state. The following is a brief description of the hardware and the software system including the components and capabilities of each module.

The system is set of client/server applications running on PC-based Windows NT workstations with a MS SQL Server database running on a Windows NT Server. The SQL database supports various workstations over FPLE's corporate wide area network (WAN). The HYDROPS database includes a comprehensive system of usage tracking and data set version control to facilitate ease of use, system security and data ownership.

The software consists of a tightly integrated system of individual modules. There are three time horizons:

• near real-time,
  
• one week ahead scheduling in hourly time steps, and
  
• one year scheduling in weekly time steps.

The modules schedule the operation of the generating units, the stations, the use of storage, and in total effect they determine the operation of the river. They also pinpoint the schedules for annual maintenance.

There are nine main software modules:

1. Satellite Down-link, Hydrometric Data-viewer, and Editing Module
   
2. Inflow Short-term Forecast Module
3. Inflow Stochastic Forecast Module
   
4. Dispatch Decision Support System Module
   
5. Annual Storage Module
   
6. Maintenance Module
   
7. Station Optimization Module
   
8. Engineering Module
   
9. Re-licensing Module

Satellite Down-link, Hydrometric Data-viewer, & Editing

Real-time hydrometric data is picked up from the three river basin areas via direct reception of satellite transmissions. Temperature and rainfall gages located at USGS river gage sites store and transmit the meteorological data along with river stage to the GOES-8 satellite. The rebroadcast from a commercial satellite is received by a dish and decoder system at FPLE where a communication server formats and stores the data in the central database. Other meteorological data from the internet are automatically collected and loaded into the database. The Satellite Data-viewer converts river stage to discharge and provides convenient system-wide views of the river flows and the meteorological data. These data are required by the hydrologic inflow forecast system.

Inflow Short-Term Forecast Module

Twice each day the communication server automatically downloads a 7-day weather forecast from a commercial weather forecasting service. The weather forecast along with the hydrometric data obtained from the satellite and the Internet are loaded into the forecast module. Based on the current watershed situation, this module provides forecasts of future hourly inflows to the rivers and reservoirs over the next 7 days. The module is operated at least once each day for each river basin. The results are automatically loaded into the database for use by other modules.

The inflow-forecasting module is a custom version of HFAM hydrologic model developed by Hydrocomp, Inc. HFAM is a derivative of USEPA's HSPF and the Stanford Model, which is the precursor of all physically based hydrologic digital models.

Selection of the proper watershed model is an important consideration in the early stages of the project. The inflow-forecasting module is a physically based model which incorporates the physics of all hydrological processes. This type of model was chosen because of the scarcity of gages in the river basins and the fact that river gages are frozen and useless during much of the winter. Calibration of this model was an expensive and time-consuming process. Operation of the model is also time consuming. Frequent adjustments are required to keep the model on track with the observed hourly flows in the unregulated indicator streams. New interfaces to expedite this process are currently being tested.

Inflow Stochastic Forecast Module

This module provides probabilistic forecasts of inflows to the major storage reservoirs for the next 52 weeks. The inflow calculation is based on 50 years of historical daily meteorological data. The calculations begin from the current initial soil moisture and snow conditions determined by the short-term forecast model.

For each year in the weather record the model determines the daily inflow for one year ahead from the current date. This process is repeated until inflow forecasts have been made for all 65 years of weather record. The result is 65 forecast sequences of inflows for the coming year, all starting from the current date. This ensemble of forecasts is used to construct probability distributions of cumulative inflows to the reservoirs. These are used in optimizing the reservoir operations and the maintenance schedule. The operator's level of risk aversion is reflected by selecting 52 week inflow sequences from up to seven probability levels. The entire process is repeated when initial conditions of soil moisture or snow change.

Dispatch (Short-term) Decision Support System

The Dispatch Decision Support System (DDSS) module is the heart of the system. It optimizes the schedules of hourly generation from hydro units for the next seven days (168 hours). The objective is to maximize revenue within the limits of the available water resources, contract commitments, and the license constraints. The DDSS module produces the optimum hourly unit operations schedule based on the following information:

• scheduled releases from the Annual Storage Module,
  
• the inflow forecasts from the Short-term Forecast module, and
  
• the forecasts of hourly market energy prices.

To initialize to current conditions in the river the DDSS uses current pond levels and current river flows collected by the SCADA system. These are automatically fed into the database by a communications server. Other information used by the DDSS includes:

• operating restrictions, such as rough zones
  
• unit maintenance schedules,
  
• licensing restrictions such as pond levels and minimum flows,
  
• rafting releases, and
  
• river system set flows

This information provides hard constraints in the optimization problem formulation.

The module consists of a customized user interface and a commercial mathematical programming package (CPLEX). The CPLEX package is claimed by the vendor to be the world's fastest optimizer, and it is very fast, dealing with over 15,000 variables and 10,000 constraints in a few minutes on a standard PC. Hourly generation schedules for the current week for all units in the hydro system are re-optimized daily, five times a week at a minimum, on a rolling 7 day time horizon.

In the new New England Power Pool (NEPOOL) market system the bid schedule for tomorrow's generation must be posted to NEPOOL by noon. After the unit operations for the next day are acceptable to the scheduler at FPLE's energy trading & marketing, and by NEPOOL, they are posted to the database where they can be accessed by the River Control Centers for implementation.

There is a separate sub-module for each river basin and each is customized to handle the special requirements of that particular river. For instance, on the Kennebec River a required constant discharge at Madison (the set-flow) is established frequently by the Kennebec Water Power Committee, a committee of all owners of generation on the river system. The set-flow presents a hard constraint that affects upstream operating decisions at previous times.

The Kennebec module optimizes the operation of individual units on the river to compensate for river routing and the short-term local inflow forecast. If the specified set-flow exceeds the capability of the river system at any hour during the one week schedule the model will alert the operator to the infeasibility of the set-flow. A message will be displayed describing the situation and the hour and location that is the problem. When this occurs the program operator must adjust the set-flow to a value that is feasible. Conversely , the model can be used to automatically calculate feasible set flows from reservoir releases and forecast unregulated inflows. Infeasibilities diagnostics also indicates violations of other hard constraints which must be corrected manually. Resolving infeasibilities requires judgment and experience as infeasibilities diagnostics indicates the infeasible constraints, not necessarily the direct cause. One constraint can make many constraints infeasible.

Annual Storage (Long-term) Module

Inflow forecasts produced by the Stochastic Forecast module are used by the Annual Storage Module, which incorporates the Maintenance Scheduling routine. This module is operated at least once each week to re-optimize the storage operation schedules for the Kennebec and Androscoggin basins.

Releases from storage reservoirs are scheduled by this module for the next 52 weeks and updated each week. Input to the Annual Storage Module includes:

• output from the Inflow Stochastic Forecast module,
  
• current storage levels,
  
• available generating capacity in each station for each weekly period, and
  
• the projected average weekly market value of power

The output is the generation flow schedule from each reservoir that will maximize the expected value of power and meet all of the license constraints. The operator can select up to seven levels of inflow probability to analyze simultaneously. The module will produce the optimum release schedule for the next week, considering all of the selected levels of probability for future inflows.

The Annual Storage Module includes customized spreadsheet-type worksheets that are used to fine tune the storage releases during the week and to allow for special conditions such as rafting releases.

Maintenance Module

The Annual Storage module contains a sub-module called the Maintenance Module which provides an optimized maintenance schedule. This module optimizes the schedule for unit outages throughout the year with the objective of minimizing lost revenue. Maintenance is incorporated into the optimization routine of the Annual Storage module because outages directly impact the operation of storage. The optimization is constrained within an operator specified window and outage duration for each unit. Outages can also be entered with fixed schedules. The maintenance routine determines the scheduled outages which are posted to the database. From there they are used by the Dispatch Decision Support System and the Annual Storage module.

The appropriate river engineer runs the Maintenance Module. Outage schedules must be coordinated with the Independent System Operator (ISO/Maine Satellite).

Station Optimization (Near Real Time) Module

The current output of each unit and water level data is provided by a SCADA system to the Energy Management System (EMS) at FPLE. The EMS downloads these station data every five minutes to the Communications Server, which posts it to the database. The Station Optimization module loads the station data and displays two alternative set-ups for the loading of the units in the plant:

1. Maximize the current power output with the current rate of plant discharge.
   
2. Minimize the plant discharge at the current plant power output.

The station operators at the River Control Centers use the Station Optimization module. Through the wide area network, managers can monitor the current operating status and efficiency of all stations in the system. Off-line, the Station Optimization module provides a tool to study station operations under various conditions.

Engineering Module

The Engineering module provides a convenient interface for editing portions of the database that deal with physical characteristics of the projects. These data include time dependent licensing restrictions; maximum and minimum pond levels, minimum flows, and bypass flows. Station level and unit level engineering data that are controlled through this module include; turbine maximum and minimum limits, generator limits, turbine rough zones, unit efficiencies, stage-storage curves, and tailwater curves. The module records the dates of time dependent licensing restrictions, which are used by the Dispatch Decision Support Module to alert the operator to license violations that might be inherent in the problem setup.

Re-licensing Module

The Re-licensing module provides the Annual Storage and Dispatch Decision Support modules with a database of historical flows and energy prices. With this information a study can accurately determine the effect on operations and revenues of changes in licensing or other operating restrictions. The Re-licensing module will also be a valuable engineering tool to study changes or additions to generating units.

In the past, re-licensing and engineering studies have used either HEC-5, which is a simulation model, or been based on manual calculations with flow duration curves. Limitations of simulation models for studies include having to fix operating parameters ahead of time and the use of fixed rule curves. Simulation models do not automatically redevelop the operating rules for the changes being examined in the studies. The Re-licensing module uses the automatic optimization routines in the Dispatch Decision Support System and the Annual Storage Model.

Re-licensing studies typically examine operating conditions for many combinations of flow conditions by simulating the operating environment over many years. The Re-licensing module accurately simulates the operating environment by including all of the details of actual hourly operations. For a study encompassing a year or more, this can require several hours of computer time.

Operating a Hydro System in the New NEPOOL

Beginning in May 1999 the New England Power Pool began operation as a "residual wholesale electricity market". This means that electric generation companies may sell electricity into a wholesale market, subject to the rigid bidding and operating rules of the Pool. The rules are described on the ISO web page at www.iso-ne.com under the section on Market Rules and Procedures. In brief, the rules require that by noon of the current day hydro generation resources must be scheduled for each hour of the next day.

Depending on the resource's characteristics, there are three methods of bidding a hydro generation resource:

1. as a self-schedule (SS), which is essentially a run-of-river unit;
   
2. as a specific output self schedule (SOSS), which is a run-of-river unit with reserve capability;
   
3. as limited energy optimization unit(LEO); which allows the ISO to schedule the unit output within limits provided by the participant.

A self-schedule is a listing of the MW hours and associated bid price for each hour for the following day for each ISO hydro unit.

To further complicate the issue 'ISO Hydro Unit' under the ISO definition may consist of a single hydro unit, multiple units in a single station, or multiple stations closely connected in parallel or series and operated in combination. Each ISO Hydro Unit must be bid separately and operated to the hourly schedule to within tolerance of +- 1 MW per hour.

Within the bid structure, scheduling is particularly problematic for run-of river hydro stations. For these stations, the range of allowable operations is very narrow because little or no pond is available to re-regulate flows from upstream. If the operator finds it impossible to meet the schedule which was bid the previous day, the ISO must be notified through a process know as redeclaration. NEPOOL is currently reviewing bidding rules hydro stations due to the difficult many run-of-river stations have in meeting NEPOOL's bidding criteria.

Frozen Schedule

The software described above can reschedule the entire hydro system quickly to take advantage of changes in hydrology or market conditions. But the software was designed before the new system of bidding had been developed. A number of software changes were required to work within the complex bidding requirements of the new NEPOOL. Now, when updating the schedule for the coming week the schedule already committed to NEPOOL for the remainder of the current day can be "frozen". The system operator now may select the fixed schedule from the previous day's optimization run, edit it and then use it to "freeze" the output schedule for each ISO hydro unit for the remainder of the current day. Thus, the current day's operating schedule, which was produced the previous day, is not modified by the program.

Each morning, with today's schedule frozen, the Dispatch Decision Support System module generates the schedule for the next day. All continuity and licensing requirements must apply to the frozen schedule, but since the schedule was determined, conditions such as the hydrologic forecast may have changed. Under the old NEPOOL system this was not a problem because the software would automatically re-optimize the schedule. Under the new system, the program will determine whether the schedule remains hydrologically feasible for the remainder of the current day. If the schedule is no longer feasible it must be re-declared to the ISO.

Available Reserve Analysis

The new bidding system initiated a new requirement for a schedule of High Operating Limit. This is required for bidding reserve generating capability. In order to qualify for reserve capability, an ISO hydro unit must have a reserve of 1 MW or more available and have the pondage capability to sustain the High Operating Limit for one hour. Reserve bids must be submitted for specific units for each hour of the bid period. The difference between the specific output self schedule (SOSS) and the High Operating Limit for a unit in a given hour is the available reserve. A routine was added to the software to deal with reserves by determining the additional energy that can be generated for one hour from each unit with the available water.

Wholesale Market Operation

The NEPOOL wholesale electricity market began on May 1, 1999. The modeling system was tested in a "mock" market operation conducted by NEPOOOL in November of 1998 and performed well. Since the market's opening day the RHOM model has functioned as the scheduling tool for the river system and the interface between energy marketing and station operations. The process of coordination with river operations personnel and the marketing group continues to evolve and personnel training continues, but it is already apparent the model functions as intended and its use will continue to evolve and grow.

Conclusions

FPL Energy - Maine has a large number of dams and generating units which are accurately modeled by a desk top computer system on a wide area network. The comprehensive system of computer models schedules reservoir operations over the seasons and optimizes the hourly operation of all generating units to maximize revenue within license limits. The time steps used by the computer system range from minutes, to hours, to weeks. Functions included in the operations and re-licensing support system include:

• acquisition, display, and data quality control of hydrometric and plant operating data
  
• weekly scheduling of reservoir operations,
  
• hydro maintenance planning,
  
• hourly and seasonal local inflow forecasting,
  
• on-line unit loading optimization within each station, and
  
• system wide hourly scheduling of all generators.

The RHOMS system operates from a relational database controlled by a dedicated computer which acts as a communication server. Convenient user interfaces and data set version tracking minimize manual data entries and opportunities for errors in the input data.

The on-line unit-loading module has been in use for over one year. The operations scheduling software has successfully been successfully operated in the New England wholesale electricity market. The module for re-licensing studies is still being tested.

The NT computer environment is adequate for this application. The fast pace of Windows software upgrades has created problems due version incompatibilities of components. Management and tracking of software and system configuration changes has been an important function during the development.

The CPLEX optimization system is reliable, fast, and flexible. Its speed for large problems has been key to making these detailed operating models viable for near real time operations planning.

The development of this software was a complex interwoven process of thoughtful design, re-direction, engineering judgment and re-judgment, and previous related software development experience. There were low points of frustration and high points of satisfaction. During the development period there were rapid major advances in proprietary software and new hardware that enhanced RHOM/HYDROPS system. These advances are likely to continue.

Acknowledgments

The software was developed by a closely coordinated team led by Alan Livingstone of FPL Energy in Lewiston, ME and Douglas I. Smith at Charles Howard & Associates, Ltd., in Victoria, B.C., Canada. Hydrocomp of Palo Alto, California provide the calibrated HFAM watershed models for the three river basins under the direction of Norm Crawford. Our sincere appreciation is extended to the many people at FPL Energy, Charles Howard & Associates and Hydrocomp who have worked diligently to ensure the success of this project.