There's no difference. Pick the best model, apply it to the whole data set, and produce the forecasts.
fc_tbl %>% accuracy(
iniciativa,
by = ".model",
list(rmse = RMSE, mae = MAE, mape = MAPE, mase = MASE, crps = CRPS, winkler = winkler_score)
) %>%
arrange(rmse)
#> # A tibble: 4 x 8
#> .model .type rmse mae mape mase crps winkler
#> <chr> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 tslm1 Test 28.9 25.0 225. 0.741 16.8 117.
#> 2 tslm2 Test 29.6 25.5 234. 0.756 17.2 120.
#> 3 tslm3 Test 30.7 26.0 233. 0.771 17.8 127.
#> 4 tslm4 Test 30.7 26.0 233. 0.771 17.8 127.
# Best model (lowest rmse) is tslm1 (it might differ due to random generation of data)
# Re-fit to whole data set and produce forecasts
fc <- iniciativa %>%
model(tslm1 = TSLM(n ~ trend() + fourier(K = 1))) %>%
forecast(h=20)
fc %>% autoplot(iniciativa)

fc
#> # A fable: 20 x 4 [1D]
#> # Key: .model [1]
#> .model data_planejada n .mean
#> <chr> <date> <dist> <dbl>
#> 1 tslm1 2020-07-19 N(51, 822) 50.8
#> 2 tslm1 2020-07-20 N(49, 822) 48.9
#> 3 tslm1 2020-07-21 N(47, 822) 46.7
#> 4 tslm1 2020-07-22 N(46, 823) 45.8
#> 5 tslm1 2020-07-23 N(47, 823) 46.9
#> 6 tslm1 2020-07-24 N(49, 823) 49.2
#> 7 tslm1 2020-07-25 N(51, 823) 50.9
#> 8 tslm1 2020-07-26 N(51, 824) 50.7
#> 9 tslm1 2020-07-27 N(49, 824) 48.8
#> 10 tslm1 2020-07-28 N(47, 824) 46.6
#> 11 tslm1 2020-07-29 N(46, 824) 45.7
#> 12 tslm1 2020-07-30 N(47, 824) 46.8
#> 13 tslm1 2020-07-31 N(49, 825) 49.0
#> 14 tslm1 2020-08-01 N(51, 825) 50.7
#> 15 tslm1 2020-08-02 N(51, 825) 50.6
#> 16 tslm1 2020-08-03 N(49, 826) 48.7
#> 17 tslm1 2020-08-04 N(46, 826) 46.4
#> 18 tslm1 2020-08-05 N(46, 826) 45.5
#> 19 tslm1 2020-08-06 N(47, 826) 46.6
#> 20 tslm1 2020-08-07 N(49, 827) 48.9
Created on 2020-12-25 by the reprex package (v0.3.0)