Add new metrics to capture the financial impact of the improved PV forecast. This isn't what is called a "proper scoring metric", in that a worse forecast can actually help your financial metric if you happen to be lucky. But its actually very valuable to be able to quantify the financial impact of our forecasts for end users.
The financial loss metric is relevant for day ahead forecasts (as it is comparing the day ahead price with the balancing price). It could in theory be applied to forecats a few hours ahead, but it's relevance is primarily for day-ahead forecasts. It can be used for site-level forecasts, but also for national forecasts.
The theory is in the screenshot attached:
The figures s_a and s_f are the target / forecast. The figures p_f, p_rt are the EPEX and Balancing (SSP / SBP) prices in the csv attached.
P_f is day ahead price and is column "price" in file epex_prices.csv
p_rt is real-time price (balancing market, or system sell price) and is column "ssp" in file ssphourly.csv
The figures are calcualted for each hour (take sum of forecast and actual generation figures across 2 half-hours to convert to hours), and then the expectation of the hourly figures calcualted across the period being tested - this coul be a year, by month, or by time of day, etc.
dantravers
commented
1 year ago
Add new metrics to capture the financial impact of the improved PV forecast. This isn't what is called a "proper scoring metric", in that a worse forecast can actually help your financial metric if you happen to be lucky. But its actually very valuable to be able to quantify the financial impact of our forecasts for end users.
The financial loss metric is relevant for day ahead forecasts (as it is comparing the day ahead price with the balancing price). It could in theory be applied to forecats a few hours ahead, but it's relevance is primarily for day-ahead forecasts. It can be used for site-level forecasts, but also for national forecasts.
The theory is in the screenshot attached:
The figures s_a and s_f are the target / forecast. The figures p_f, p_rt are the EPEX and Balancing (SSP / SBP) prices in the csv attached.
The figures are calcualted for each hour (take sum of forecast and actual generation figures across 2 half-hours to convert to hours), and then the expectation of the hourly figures calcualted across the period being tested - this coul be a year, by month, or by time of day, etc.