Check if timeseries has peaks with same frequency - r

Question

I have timeseries with several instances:
Some instances looks like this which is normal:
But some looks like that - they have peaks with same intervals (1 hour in this example):
I need to analyze data and find instances with that anomaly - peaks repeated with nearly same interval.
As a result I expect name of Instance with that anomaly (in my example data - only 'A'), it's period of peaks in seconds (3600 in my example data) and spread between let's say median and peaks of that instance.
How to do that?
Here is my example data:
library(dplyr)
library(lubridate)
library(ggplot2)
set.seed(900)
data1 <-
data.frame(
datetime = seq.POSIXt(as.POSIXct("2020-12-26 10:00:00"), as.POSIXct("2020-12-26 10:00:00") + 15*50001, "15 sec"),
Value = sample(1:10, 50002, replace = T),
Instance = "A"
)
data1.1 <- data.frame(
datetime= seq.POSIXt(as.POSIXct("2020-12-26 10:00:00"), as.POSIXct("2020-12-26 10:00:00") + 15*50001, "hour"),
Value = sample(10:100, 209, replace = T),
Instance = "A"
)
data1 <- rbind(data1, data1.1) %>% group_by(datetime, Instance) %>% summarise(Value = max(Value)) %>% ungroup()
data2 <- data.frame(
datetime = seq.POSIXt(as.POSIXct("2020-12-26 10:00:00"), as.POSIXct("2020-12-26 10:00:00") + 15*50001, "15 sec"),
Value = sample(1:100, 50002, replace = T),
Instance = "B"
)
data3 <-
data.frame(
datetime = seq.POSIXt(as.POSIXct("2020-12-26 10:00:00"), as.POSIXct("2020-12-26 10:00:00") + 15*50001, "15 sec"),
Value = sample(1:100, 50002, replace = T),
Instance = "C"
)
data4 <- data.frame(
datetime = seq.POSIXt(as.POSIXct("2020-12-26 10:00:00"), as.POSIXct("2020-12-26 10:00:00") + 15*50001, "15 sec"),
Value = sample(1:100, 50002, replace = T),
Instance = "D"
)
data <- do.call("rbind", list(data1, data2, data3, data4))

Answer 1

As I am still not quite sure how the answer is supposed to look like, this is only a preliminary attempt to help. this is probably not exactly what you need but maybe it can be step in the correct direction?
qt<-data[data$Value>=99,] # find peaks
qt_test<-qt
# length(qt$datetime)
# table(qt$Instance)
# create hourly difference between peak times (seconds were not useful, mins may be better)
x<-vector()
for(i in seq_along(qt_test[[1]])){
if (i == 1){
a<-0; x<-c(x, a)
}
if (i>1){
a<-difftime(qt_test[[1]][i], qt_test[[1]][i-1], units = "hours")
x<-c(x, a)
}
}
qt_test$difftime_in_hours<-x
# summary(qt_test$difftime_in_hours)
qt2<-qt_test[between(qt_test$difftime_in_hours, 0.04, 0.05),] # timeframe - +/- floor/ceiling mean (0.04647) - only for test purposes
# A tibble: 154 x 4
datetime Instance Value difftime_in_hours
<dttm> <chr> <int> <dbl>
1 2020-12-26 15:44:15 B 100 0.0417
2 2020-12-26 16:44:00 B 99 0.0417
3 2020-12-26 16:57:30 B 100 0.05
4 2020-12-26 17:58:00 B 99 0.0417
5 2020-12-26 19:15:30 B 100 0.05
6 2020-12-26 19:24:30 B 100 0.0417
7 2020-12-27 04:04:45 B 99 0.05
8 2020-12-27 09:37:00 B 99 0.0417
9 2020-12-27 11:16:00 B 100 0.0417
10 2020-12-27 11:55:30 B 99 0.05
# ... with 144 more rows
table(qt2$Instance)
B C D
69 47 38
results will differ due to random seed but this is an answer as far as i understand your question