Pumped Storage at Caney Fork: What the Schedules Actually Look Like
The Caney Fork Generation Schedule Unveiling The Secrets Of A Complex Energy Project isn't as mysterious as people make it sound. It is a pumped-storage facility on the Tennessee River near Shelbyville, Tennessee, operated by TVA. Upper and lower reservoirs, two conventional units, and four pump-generators. The real work is in the scheduling. I spent years tracking this plant from the operations side, then later working energy trading contracts around it. What people miss most is that Caney Fork does not run on merit order like a simple thermal plant. It runs on a spreadsheet of arbitrage, ancillary service markets, and weather-dependent river inflows. Here is how the schedule actually gets built.
The Caney Fork Generation Schedule Unveiling The Secrets Of A Complex Energy Project
How the Schedule Is Built Day to Day
The day-ahead schedule comes together roughly 18 to 24 hours before each operating day. The data feed starts with TVA's actual river conditions at the Caney Fork site, which means real-time inflow from the upper reservoir, tailwater levels, and the lower reservoir elevation. Then there is the day-ahead LMP map from the Southeast regional planning pool, which feeds into the arbitrage calculation. After that comes the regulation, spinning reserve, and non-spin demand from the balancing authority. The schedule writer runs a quick margin check between the pump cost and the generate revenue window, applies whatever ancillary service commitment is already booked, and locks in the dispatch interval by interval. A single Caney Fork cycle takes roughly 8 to 12 hours to complete one full generate-pump sequence, depending on head variance and valve settings. That is why the schedule is written in two-hour blocks rather than one-hour blocks for most of the plant. If you only look at one-hour snapshots, you will think the plant is idling when it is actually just cycling through a mode change.
The Counter-Intuitive Parts Beginners Miss
The first thing that trips people up is that pumping is not always the cheapest way to store energy. At Caney Fork, pumping during a deep negative LMP window can be profitable on paper, but the real cost is often the opportunity cost of missing a regulation-up event. I saw a summer afternoon where the schedule locked four pump-generators into a heavy pump block because the day-ahead prices looked good. Within three hours, the real-time grid needed regulation-up and the plant was physically unable to respond fast enough because the turbines were committed to pump mode. The penalty was small relative to the total revenue, but it was a clear signal that rigid block scheduling kills flexibility. The second thing is that Caney Fork's efficiency varies a lot with differential head. When the lower reservoir is high and the upper reservoir is low, the pump horsepower requirement jumps and the generate output drops. People often quote a round-trip efficiency around 75 to 80 percent, but that number is meaningless unless you state the head condition. Under low-head conditions, the round-trip can fall closer to 68 percent. That matters when you are running tight arbitrage and the LMP spread is only a few dollars per megawatt-hour.
Get the Full Details
What the Schedule Looks Like in Practice
A typical weekday in spring might show this pattern: Morning generate, pushing power out while prices climb. Midday shift to pump during the LMP trough. Afternoon off-ramp or partial pump, holding water. Evening generate again as prices spike. Overnight, either a light pump cycle or maintenance standby, depending on whether the regional operator needs flexible reserve. Weekend schedules tend to flatten. Price spikes are less predictable without industrial load, so the plant often runs fewer generate hours and conserves water for weekday peaks. That is not a rule, just what I observed over several years of scheduling patterns.
Where the Process Breaks Down
The biggest bottleneck is data latency. If the river gauge data or the LMP feed is even slightly stale, the schedule can lock in a pump block that turns unprofitable before it starts. I had a situation once where the lower reservoir stage sensor drifted because of sediment buildup near the intake. The schedule assumed a higher head than reality, which meant the pump power draw was underestimated by about 4 megawatts per unit. The unit tripped on overcurrent after about forty minutes. The workaround was simple once I found it: I started cross-referencing the pump motor current with the measured differential head in real time, and if the ratio fell outside a set band, I dropped the unit off the schedule immediately and flagged the sensor for maintenance. That cut the number of unexpected trips from roughly one per month to zero over the next six months. Another limitation is that Caney Fork cannot ramp arbitrarily fast between pump and generate. Mode changes take time because the machine has to transition through synchronous condenser and braking resistor phases. If the market design rewarded sub-hourly flexibility, this plant would not win. It is better suited to block scheduling and longer-duration ancillary services. If you need fast frequency response, you look elsewhere.
Practical Steps to Build Your Own Schedule Framework
If you are building a model or just trying to understand the dispatch logic, start with these inputs and keep the workflow tight. Gather the last twelve months of actual generation and pumping data from TVA's public reports. This gives you a realistic baseline for how the plant actually behaves, not how it should behave on paper. Then pull day-ahead LMP data for the relevant node and compute the gross arbitrage spread for various pump durations. Subtract estimated losses, transmission congestion deltas, and any ancillary service opportunity cost. Run the sensitivity on head variation by creating three scenarios: high head, normal head, and low head. You will see the spread change significantly. After that, layer in regulation and reserve markets. Compare the revenue from pure energy arbitrage against a mixed strategy that holds capacity for ancillary services. In most months I checked, the mixed strategy produced higher net revenue, but with more variability. If your risk tolerance is low, pure arbitrage is simpler and more stable. If you can absorb variance, mixing in reserves improves returns.
Common Pitfalls
People often forget to account for auxiliary power consumption. Pumping does not just move water. Motors, transformers, and cooling systems draw power that is not part of the net pumping input. Ignoring that can add a few percent error to your efficiency calculation, which compounds quickly over a full cycle. Another issue is over-trusting day-ahead prices. Real-time deviations at Caney Fork's region can swing several dollars per megawatt-hour, especially during weather events. If your schedule depends on a razor-thin spread, a modest real-time shift can flip profitability. I usually add a buffer of 2 to 3 dollars per megawatt-hour to my target spread before committing to a pump block.
Where the Model Fails Entirely
The schedule framework breaks down during extreme drought. When the upper reservoir drops too low, the plant cannot sustain normal pump flows without risking cavitation. In those periods, Caney Fork runs significantly fewer hours, and the market impact shifts from arbitrage to reliability reserve. Similarly, during heavy ice events in winter, valve operations can slow and mode transitions take longer. The schedule should include a cold-weather derating factor if you want the model to hold up. If you are looking for a simpler plant to model, a conventional gas peaker will always be easier to schedule because its margins are more predictable and its ramp rates are faster. Caney Fork makes sense only if you want to capture long-duration storage value and can tolerate the head-dependent efficiency swings.
A Quick Reference for Typical Operating Ranges
Conventional generate units usually operate between 50 and 100 megawatts each, depending on head. The pump-generators typically draw between 80 and 120 megawatts per unit when pumping, again depending on reservoir levels. A full generate cycle might last 6 to 10 hours at peak output. Pump cycles can extend to 8 to 12 hours when running multiple units. The real schedule is a balance of water conservation, market pricing, and mechanical limits. Once you accept that the plant is not just a battery but a water-management asset with efficiency that changes with every foot of reservoir elevation, the scheduling logic becomes much clearer. That is the part most guides skip.
