1. Deep learning method
Compared with traditional time series forecasting models, neural networks have the following advantages:
1. Automatically learn how to incorporate series features such as trend, seasonality and autocorrelation into forecasts.
2. Capable of capturing very complex patterns.
3. Many related series can be modeled simultaneously instead of processing each series individually.
But neural networks have some disadvantages:
1. Model building can be complex.
2. Model construction is computationally expensive.
3. It is more difficult to explain the basis for the predictions made by the model.
4. If you want the model to perform well, you need a large data set.
2. What is RNN?
Recurrent Neural Network (RNN) is a neural network for processing sequence data. Compared with the general neural network, it can handle the data of sequence change.
A recurrent neural network maps an input sequence to a predicted output.
1. The most common format is many-to-one, which maps an input sequence to an output value
2. The input at each time step is used to sequentially update the hidden state or memory of the RNN cell.
3. After processing the input sequence, use the hidden state information to predict the output.
Application of RNN in Time Series Forecasting