1 Introduction
2 Materials and Methods
2.1 Case Studies
2.2 Traditional Hybrid Models Based on Two Linear and Nonlinear Models
2.2.1 Modeling the Linear Component of Time Series
2.2.2 Individual Models for Modeling Nonlinear Components of Time Series
2.3 Improved Hybrid Model Based on One Linear Model and an Ensemble of Nonlinear Models
-
The first configuration is based on residual subseries with different antecedent time steps (t-1), (t-2), (t-3), as the following:$$e_{t} = f(e_{t - 1} ,e_{t - 2} ,e_{t - 3} ) + \varepsilon_{t}$$
-
The second configuration is based on the linear component (\({\widehat{L}}_{t}\)), the original observed time series with antecedent time steps (yt-1 or yt-12), and the residual sub series with different antecedent time intervals from step 1 as follows:$$y_{t - an} = f(y_{t - 1} ,y_{t - 2} ,\hat{L}_{t}^{{}} ,e_{t - 1} ,e_{t} ) + \varepsilon_{t}$$$$y_{t - mon} = f(y_{t - 1} ,y_{t - 12} ,\hat{L}_{t}^{{}} ,e_{t - 1} ,e_{t} ) + \varepsilon_{t}$$
2.4 Evaluation of the Model Performance
3 Results
3.1 Modeling Linear Component of Precipitation Time Series
Station | Tabriz | Rasht | |
---|---|---|---|
Normality test | Kolmogorov–Smirnov | 0.157 | 0.000 |
Shapiro–Wilk | 0.399 | 0.000 | |
Stationarity test | ADF | -1.87 | -0.34 |
Model identification | Model | ARIMA(0,1,2) | SARIMA (0,1,1) × (0,1,1)12 |
SBC | 347.31 | 456.88 | |
BIC | 8.47 | 2.55 | |
Parameters | e1 = 0.64, e2 = 0.5 | e1 = 0.96, E12 = 0.9 | |
Diagnostics | P-value | 0.002, 0.017 | 0.0001,0.0001 |
P-value | 0.497 | 0.6 |
3.2 Modeling Nonlinear Component of Precipitation Time Series
model | Parameters | Tabriz | Rasht | ||
---|---|---|---|---|---|
Configuration | |||||
First | Second | First | Second | ||
GEP | Number of generations | 1000 | 100 | 100 | 80 |
Chromosome length | 50 | 50 | 50 | 50 | |
Head size | 8 | 7 | 7 | 7 | |
Number of genes | 4 | 3 | 3 | 3 |
SVR | Kernel function | radial basis function | sigmoid function | linear function | radial basis function |
---|---|---|---|---|---|
Penalty parameter | 0.5 | 1 | 0.5 | 2 | |
GMDH | Maximum number of neurons in a layer | 5 | 10 | 10 | 10 |
Maximum number of layers | 5 | 7 | 5 | 5 | |
Selection pressure | 0.6 | 0.5 | 0.6 | 0.1 |
3.3 Improved Hybrid Models
Model type | Precipitation time series | Structure of hybrid model | RMSE | RRMSE | MSE | RPD | APB | dm | U2 | AMPE | NSE |
---|---|---|---|---|---|---|---|---|---|---|---|
Hybrid | Tabriz | ARIMA + SVR | 11.88 | 0.041 | 141.21 | 4.5 | 0.034 | 0.86 | 0.04 | 40.83 | 0.94 |
ARIMA + GEP | 17.59 | 0.061 | 309.6 | 3.05 | 0.052 | 0.77 | 0.06 | 40.85 | 0.87 | ||
ARIMA + GMDH | 9.87 | 0.034 | 97.49 | 5.45 | 0.028 | 0.89 | 0.034 | 40.83 | 0.95 | ||
Improved hybrid: Combination of multiple nonlinear models | Tabriz | Iv + SSE | 8.81 | 0.03 | 77.67 | 6.1 | 0.023 | 0.9 | 0.03 | 40.82 | 0.96 |
MSEI + SSE | 7.36 | 0.025 | 54.21 | 7.3 | 0.02 | 0.91 | 0.025 | 40.82 | 0.97 | ||
Iv + SMAPE | 7.73 | 0.027 | 59.9 | 6.95 | 0.023 | 0.9 | 0.026 | 40.82 | 0.97 | ||
MSEI + SMAPE | 8.91 | 0.031 | 79.46 | 6.03 | 0.024 | 0.89 | 0.03 | 40.82 | 0.96 | ||
SVR | 5.57 | 0.019 | 31.03 | 9.6 | 0.017 | 0.93 | 0.019 | 40.81 | 0.98 | ||
GA | 9.39 | 0.032 | 88.31 | 5.7 | 0.025 | 0.89 | 0.032 | 40.82 | 0.96 | ||
LSR | 9.01 | 0.031 | 81.2 | 5.97 | 0.025 | 0.89 | 0.031 | 40.82 | 0.96 | ||
ANN | 7.64 | 0.026 | 58.43 | 7.04 | 0.02 | 0.91 | 0.026 | 40.82 | 0.97 | ||
Hybrid | Rasht | SARIMA + SVR | 20.02 | 0.187 | 400.9 | 3.77 | 0.142 | 0.97 | 0.15 | 4.58 | 0.92 |
SARIMA + GEP | 30.51 | 0.28 | 931.3 | 2.47 | 0.19 | 0.94 | 0.23 | 4.64 | 0.82 | ||
SARIMA + GMDH | 24.22 | 0.22 | 586.7 | 3.11 | 0.18 | 0.96 | 0.18 | 4.63 | 0.89 | ||
Improved hybrid: Combination of multiple nonlinear models | Rasht | Iv + SSE | 15.92 | 0.149 | 253.4 | 4.74 | 0.11 | 0.98 | 0.12 | 4.56 | 0.95 |
MSEI + SSE | 17.2 | 0.16 | 296.17 | 4.39 | 0.12 | 0.98 | 0.13 | 4.56 | 0.94 | ||
Iv + SMAPE | 16.52 | 0.154 | 272.9 | 4.57 | 0.11 | 0.98 | 0.12 | 4.55 | 0.95 | ||
MSEI + SMAPE | 18.08 | 0.169 | 326.9 | 4.17 | 0.13 | 0.98 | 0.13 | 4.57 | 0.94 | ||
SVR | 13.03 | 0.12 | 177.56 | 5.66 | 0.091 | 0.99 | 0.1 | 4.53 | 0.96 | ||
GA | 11.48 | 0.1 | 131.84 | 6.57 | 0.078 | 0.99 | 0.088 | 4.52 | 0.97 | ||
LSR | 14.37 | 0.13 | 206.55 | 5.26 | 0.11 | 0.98 | 0.11 | 4.55 | 0.96 | ||
ANN | 17.7 | 0.16 | 294.99 | 4.39 | 0.1 | 0.98 | 0.13 | 4.54 | 0.94 |