R Notebook

We used a automation script to extract most recent 15 days data from 13 Capital One assets. However for various reasons, none of them has that much data. The following table details the range of the available data for each asset.

data %>%
  group_by(id) %>%
  summarise(range=round(max(ts) - min(ts), 1), count=n()/2, change.count=sum((u_p - u != 0))/2) %>%
  select(id, range, count, change.count)

As shown above in the change.count column, there is not a lot cooling stage changes during this periods of time. For the purpose of slaving algorithm testing, the most 6 active RTUs in cooling are selected and listed as following:

index = data %>%
  group_by(id) %>%
  summarise(change.count = sum((u_p - u != 0))) %>%
  arrange(desc(change.count)) %>%
  filter(row_number()<=6)
index

Interpolation VS Forward Filling

data1 = data %>% filter(id %in% index$id)

As shown below, slaving with time interpolation is generally a little better than any counter parts of forward filling. This may due to time interpolated space temperature is closer to thermostat’s measure during the time of thermostat making decision.

data1 %>%
  mutate(method=recode(method, ffill='ffill.rate', interpolate='interpolate.rate')) %>%
  group_by(method, d1) %>%
  summarise(rate = mean(u == uhat)) %>%
  ungroup() %>%
  spread(method, rate)

Success Rate Scenario by Scenario

d1 = .5 shines when only consider stage changing scenarios, as shown below.

data1.plot1 %>%
  ungroup() %>%
  mutate(scenario=paste(u_p, 'to', u), method=recode(method, ffill='ffill.rate', interpolate='interpolate.rate')) %>%
  arrange(scenario, d1) %>% 
  select(scenario, method, d1, rate) %>%
  spread(method, rate)

Success Rate by method and asset

Success Rate consider each scenario as 1

0 to 0 scenario, for every asset d1 = 2 provide a higher chance is because d1 = 2 offers a wider range. 0 to 1 scenario, d1 = .5 is dominating. 1 to 0 scenario, d1 is not presented in the calculation, therefore every d1 perform about the same. 1 to 1