Research Article
A Novel Hybrid Deep Learning Model for Complex Systems: A Case of Train Delay Prediction
Table 1
Disadvantages of state prediction in some areas of real-world complex systems.
| Prediction target | Model | Disadvantage | Reference number |
| Building thermal load | LSTM | Weather data need to be processed separately | [22] | Heating demand | RNN | Difficult to process long-term thermal load | [24] | Thermal load of district heating networks | DNN | The attribute of characteristic should be considered | [25] | Building energy consumption | GRU | Only classify the building energy data into time categories | [26] | Numerical weather prediction | DNN | The spatial characteristic of the grid point should be mined by the model | [27] | Storm surges | CNN_LSTM | No model is built for spatial characteristics | [28] | PM2.5 concentration | CNN_LSTM | PM2.5 influencing factors need to be distinguished | [29] | Temperature and wind speed | 3D CNN_FNN | The time characteristics of dynamic systems should be considered separately | [30] | Useful life of the complex system | LSTM | The efficiency of the model could be improved | [31] | Thermal and cooling consumption of building | 1D CNN | The model’s ability to process time series data should be improved | [4] | Train delay | DNN | External influence should be further analyzed | [32] | Train delay | LSTM | Spatial relationships among different stations should be considered | [33] | Train delay | LSTM | Time sequence characteristics mining of train influencing factors should be improved | [34] |
|
|