Estimate nonlinear vector autoregression models (also known as the next generation reservoir computing) for nonlinear dynamic systems. The algorithm was described by Gauthier et al. (2021) doi:10.1038/s41467-021-25801-2.
You can install the development version of NVAR from GitHub with:
# install.packages("devtools")
# devtools::install_github("Sciurus365/NVAR")
install.packages("NVAR", repos = c('https://sciurus365.r-universe.dev', 'https://cloud.r-project.org'))
This is an example for the Lorenz model.
library(NVAR)
testdata <- nonlinearTseries::lorenz()
testdata <- tibble::as_tibble(testdata)
t1 <- NVAR(data = testdata, vars = c("x", "y", "z"), s = 2, k = 2, p = 2, alpha = 1e-3)
t1_sim <- sim_NVAR(t1, length = 5000)
realdata <- nonlinearTseries::lorenz(time = seq(0, 100, by = .01)) %>% tibble::as_tibble()
library(ggplot2)
ggplot(realdata) +
geom_line(aes(x = 1:10001, y = x), color = "red", alpha = 0.4) +
geom_line(aes(x = 1:10001, y = x), data = t1_sim, color = "blue", alpha = 0.4) +
geom_vline(xintercept = 5000) +
theme_bw() +
xlim(c(4900, 8000)) +
labs(x = "time", y = "x")
# Red line: real data.
# Blue line: simulated data with the NVAR.
# Black vertical line: when the simulation starts.