Prevent overbuilding of negatively-priced generators

[grgmiller]

As noted in commit 2020.09.22 (Version 0.3.0) (https://github.com/grgmiller/SWITCH247/commit/2454e18af657a88ae796e04ae5808a614327578a), there is an issue when a generator's weighted average Pnode revenue is greater than its PPA cost, because it means that the generator will have a net negative cost, and the model will want to build as much of it as possible (leading to an unbounded problem if the generator capacity is not constrained). At the time, we had dealt with this by removing the Pnode revenue from excess generation from the objective function, in order to penalize building more than was needed to meet load. We added this excess Pnode revenue back into the cost in the summary report.

However, there are concerns that excluding this from the objective function could lead to sub-optimal outcomes, or that this puts a very large penalty on excess generation. It could also lead to strange behavior when nodal prices are negative, becuase it incentivizes artificially pushing DispatchGen to ExcessGen to avoid paying to inject the power into the grid, which then causes uncecessary system power consumption.

One alternative approach could be to identify the generators that are at risk of negative pricing and put a hard constraint on the amount of excess generation that is allowed from these generators.

One challenge with this approach is identifying the generators that are at risk of negative pricing. It is simple enough to calculate the weighted average Pnode revenue and PPA cost based on the variable capacity factor, but if anything causes the plant's VCF to change (for example by dispatching economic curtailment, or if it is a hybrid paired with storage), then the price can go negative. One way to screen for this would be to pre-simulate curtailment and storage dispatch.

Curtailment could be pre-modeled by assuming that 2% of generation from the lowest cost hours would be curtailed. Screening the storage would likely require running a simple linear program. But this may not be necessary

We also currently only check variable generators, but it is possible that storage could be priced low enough that it earns more arbitrage revenue than its capacity cost. That has not yet been an issue, but something to look for.

[grgmiller]

Exclude Pnode revenue approach

Advantages

• Guarantees that excess generation will have a non-negative cost
• Reflects extreme risk if all excess generation were curtailed but we still had to pay the PPA cost
• Doesn't require setting a static/artificial limit

Cons

• Because we are essentially setting a penalty equal to the Pnode revenue for each gen at each time, this means that generators that produce excess during times when prices are low or negative will have less of a penalty than generators that have excess when market prices are high. This is the opposite of what the market would want - from the system perspective, they want more generation during times when prices are high
• Can cause an incentive to artificially push DispatchGen to ExcessGen when prices are negative, just to avoid paying the Pnode cost, and use system power instead, even if it is not needed

What if we just discount the Pnode price: that would be like only allowing excess generation if it were a really good deal

[grgmiller]

Why penalize excess gen at all?

• We don't want to prevent cheap generators from being built, but we only want them to be built if their shape matches our need.

Setting a flat penalty

Advantages

• Penalizes excess generation equally from all generators
• Generators with excess gen during high cost times are incentivized, rather than generators who have excess during low-value times (however this may just attenuate the existing preference, because generators with high Pnode revenues are already the most likely to be overbuilt.
• There is no incentive to push generation from another project to be excess because it is all penalized the same.

Disadvantages

• Has to be set high enough to overcome negative pricing for all generators
• Uncertainty about what is the right penalty

Potential options for penalty magnitude:

• Average nodal price for the entire year (for all nodes?) - similar to pnode, but affects all generators equally
• PPA price - creates a differential penalty for different generators
• The magnitude of the uncertainty on nodal price forecasts (if nodal prices were lower than forecasted

[grgmiller]

Congestion cost approach: calculate the delivery cost of delivering energy to the DLAP. In reality we only pay for dispatched rather than excess, but pricing this way in the model would create an incentive to have excess generation, since it would be cheaper than regular generation.

What if we charge delivery cost on all generation? Makes all generation more expensive.

[grgmiller]

Implemented a flat penalty on excessgen