Table 3.
Alternative recurrent neural network models for the three cities
| City | Model | Learning rate | Dimensions of hidden layer | Number of epochs | MAPE (%)a | MAPE (%)b | MAPE (%)c |
|---|---|---|---|---|---|---|---|
| Xuzhou | RNN1 | 0.05 | 3 | 500 | 16.14 | 15.99 | 16.46 |
| RNN2 | 0.05 | 3 | 500 | 13.42 | 13.30 | 14.41 | |
| RNN3 | 0.2 | 3 | 150 | 13.08 | 11.95 | 12.07 | |
| RNN4 | 0.05 | 3 | 600 | 10.33 | 10.33 | 10.40 | |
| RNN5 | 0.05 | 5 | 600 | 8.45 | 8.25 | 8.54 | |
| RNN6 (RNN5 + MAS1) | 0.05 | 3 | 1000 | 7.36 | 7.33 | 7.33 | |
| RNN7 (RNN5 + MAS2 + MST2) | 0.05 | 3 | 800 | 6.38 | 6.31 | 6.42 | |
| RNN8 (RNN5 + MAT3 + MAS3 + MP3 + MST3) | 0.05 | 5 | 600 | 4.78 | 4.89 | 4.97 | |
| RNN9 (RNN5 + MAS1 + MAS2 + MST2 + MAT3 + MAS3 + MP3 + MST3) | 0.05 | 10 | 600 | 5.75 | 5.40 | 5.90 | |
| Nantong | RNN1 | 0.05 | 3 | 500 | 21.91 | 21.99 | 21.78 |
| RNN2 | 0.2 | 5 | 80 | 16.92 | 17.81 | 16.31 | |
| RNN3 | 0.2 | 3 | 150 | 13.82 | 14.26 | 13.86 | |
| RNN4 | 0.2 | 3 | 150 | 12.78 | 12.84 | 12.80 | |
| RNN5 | 0.2 | 5 | 100 | 11.38 | 11.44 | 11.24 | |
| RNN6 (RNN5 + MAS1 + MAH1) | 0.05 | 5 | 1000 | 9.19 | 8.82 | 8.84 | |
| RNN7 (RNN5 + MAS2 + MAH2) | 0.05 | 5 | 1000 | 8.58 | 8.26 | 8.52 | |
| RNN8 (RNN5 + MAS3 + MAH3) | 0.05 | 10 | 800 | 8.87 | 8.79 | 8.69 | |
| RNN9 (RNN5 + MAS1 + MAH1 + MAS2 + MAH2 + MAS3 + MAH3) | 0.05 | 5 | 800 | 8.79 | 9.21 | 9.19 | |
| Wuxi | RNN1 | 0.1 | 10 | 150 | 23.76 | 23.81 | 23.77 |
| RNN2 | 0.05 | 5 | 400 | 19.93 | 19.54 | 20.17 | |
| RNN3 | 0.05 | 10 | 250 | 18.23 | 17.84 | 18.59 | |
| RNN4 | 0.05 | 10 | 400 | 17.15 | 17.40 | 17.31 | |
| RNN5 | 0.05 | 5 | 600 | 14.10 | 13.93 | 13.95 | |
| RNN6 (RNN5 + MAT1 + MAP1 + MAS1 + MAH1 + MST1) | 0.05 | 3 | 1500 | 13.01 | 13.39 | 13.04 | |
| RNN7 (RNN5 + MAS2) | 0.1 | 5 | 800 | 12.62 | 12.36 | 12.80 | |
| RNN8 (RNN5 + MAT3 + MAS3 + MAH3) | 0.05 | 10 | 1000 | 12.71 | 13.06 | 12.94 | |
| RNN9 (RNN5 + MAT1 + MAP1 + MAS1 + MAH1 + MST1 + MAS2 + MAT3 + MAS3 + MAH3) | 0.1 | 3 | 1000 | 12.81 | 12.80 | 13.46 |
RNN recurrent neural network, MAPE mean absolute percentage error, MAT monthly average temperature, MAP monthly average atmospheric pressure, MAS monthly average wind speed, MAH monthly average relative humidity, MP monthly precipitation, MST monthly sunshine time, 1 1 month prior, 2 2 months prior, 3 3 months prior
a MAPE of the model with the testing set after the first training
b MAPE of the model with the testing set after the second training
c MAPE of the model with the testing set after the third training