What is the error cost in solar power forecasting?- Berkeley Lab

2022-07-09 0 By

What is the error cost in solar power forecasting?- Berkeley Lab Solar Modules.Characteristic pictures: Jackiso/Shutterstock.com Jan. 24 (Renewables Now) – accident may lead to a gloomy solar power lower than expected, as a result, other generator will need to use additional output to compensate.One possibility is that fast-acting gas turbines will fill this generation gap, and while this is a suitable solution from a reliability point of view, more fuel-efficient combined-cycle generators may be able to compensate for reduced solar output with sufficient advance warning.This is an example of the cost of solar forecasting error, where a group of generators fails to meet optimal demand.But given the complexity of the grid, how to determine the dollar cost of solar forecasting errors?One solution is to use the market price of the previous day and the real-time market to determine the cost of forecast errors.This approach does not indicate which generators are required to accommodate prediction errors, but it does provide an indication of the system cost of prediction errors.A new Berkeley Lab study, published in the journal Solar Energy, looked at the cost of solar forecasting errors at more than 600 plants in five major U.S. electricity markets from 2012 to 2019.The study looked at two types of predictions, a simple “persistence” prediction approach, in which today’s solar map is expected to repeat tomorrow, and a publicly available numerical weather forecast forecast (the North American Mesoscale Model, or NAM).The study used local hourly prices for each plant location, as well as an hourly profile of actual generating capacity developed independently of the NAM weather model, and was de-biased based on recorded generating capacity for each plant and each region.The average cost of published forecasts is lower (most years the study found that the average cost of forecast errors using the NAM method was $1 or less per MWH in all years except 2016, of which $1.50 per MWH (see Figure 1).In contrast, in most years, persistence based projections cost more, approaching $1.50 / MWH throughout the year.This shows that even using a simple, open forecasting technique like NAM can provide value by reducing the cost of forecasting error.Still, the forecast error cost is low, about $20 to $40 per megawatt hour (depending on the year), relative to typical overall electricity prices.One concern about solar forecasting errors is that, as a larger share of total electricity generation comes from solar generators, their costs can increase.This concern is simply based on the idea that a portion of the solar error is regionally related, so additional solar deployments could result in larger absolute forecast errors that could be more costly to resolve.We find mixed evidence that error costs are affected by regional solar penetration levels.On the one hand, NAM forecasts an average error cost of nearly $1 / MWH in California and New England for areas with high solar penetration from 2017 to 2019, compared to an average error cost of nearly $0 / MWH in areas with low penetration (such as regional power markets SPP, PJM and ERCOT).This finding suggests that, on average, NAM has a very low cost of forecasting errors in low permeability areas and a moderate cost in high permeability areas.The story is complicated, however, because Solar penetration in California is much higher than in New England, but the error cost is similar.The paper discusses these complexities in further detail.However, within the broad regional trends described above, error costs vary widely from plant to plant and year to year.For more details on regional, factory-level, and time variations in error costs, see the full article.What is the value of solar farms participating in the market one day ahead of time?We further examine the value of solar power plant participation in the day-ahead market, even taking into account the cost of forecast errors.Although most factories are not really “commercial” factory (that is, most of the factories to sign a form of long-term purchase agreement), but we can use the wholesale price to determine solar power plants to participate in one day in advance and real-time market (and not just the real-time market participation) provided by the system value, method is to use public NAM predict bidding to drive one day in advance.The study found that, on average, the value of participating in the previous day’s market was modest, even after accounting for NAM prediction errors.This value varies by year and ranges from -$0.5 / MWH to $5.2 / MWH.What are the costs of future mistakes?This study provides a basis that can be used to set future expectations and examine prospective models.Against this backdrop, the study found that solar error costs remain modest through 2019, despite a dramatic increase in solar penetration in California.This finding can serve as a starting point for prospective studies.Unanswered questions remain, however.In particular, how will the cost of forecast error change as solar deployment increases to previously unseen levels, or as storage is massively integrated into the grid?In addition, the study did not examine the cost of forecasting errors for solar + storage hybrid plants, of which more and more are expected to be deployed in the coming years.Solar magazine article on “the us one day in advance the cost of solar energy prediction error” is “open access”, everyone can use: https://doi.org/10.1016/j.solener.2021.12.012