Case Introduction
This case will demonstrate how to use Recurrent Neural Networks (RNNs) for time series prediction. Specifically, we will use RNNs to predict the future values of a variable that depends on its own historical values.
In this case, we will use a temperature dataset. We will provide the temperature values from the past few days and then use the RNN to predict the temperature for a future day.
Algorithm Principle
Recurrent Neural Networks (RNN) are a special type of neural network designed to handle sequential data, such as time series. Unlike traditional neural networks, RNNs introduce recurrent connections that allow the network to retain internal states when processing sequential data.
RNNs pass information through time steps in a loop. At each time step, the RNN receives an input vector and the state vector from the previous time step, producing an output vector and the current time step’s state vector. This allows the recurrent network to capture temporal dependencies in the sequence.
Below is the formula derivation for RNN:
Dataset
We will use a dataset named “temperature_data.csv”. This dataset contains daily temperature data collected from a certain location.
The format of the dataset is as follows:
date,temperature
2023-11-01,25.74
2023-11-02,27.37
2023-11-03,26.93
2023-11-04,25.34
2023-11-05,25.37
2023-11-06,22.73
2023-11-07,24.24
2023-11-08,27.37
2023-11-09,26.93
2023-11-10,24.34
2023-11-11,26.77
2023-11-12,25.43...Calculation Steps
Data Preprocessing: Read the dataset and convert the dates to datetime objects, normalizing the temperature values.
Data Preparation: Create training and testing sets, using the past few days’ temperature values as input and the temperature value three days ahead as output.
Build RNN Model: Use the Keras library to build a simple RNN model.
Model Training: Train the RNN model using the training set.
Model Evaluation: Evaluate the trained model using the testing set.
Temperature Prediction: Use the trained model to predict the temperature three days ahead.
Python Coding Process
First, we import the necessary libraries, including numpy, pandas, scikit-learn, and Keras.
We use the pandas library to read and process the temperature dataset. We convert the date column to datetime objects and normalize the temperature values using MinMaxScaler.
Next, we create input features X and output labels y. For each time step t, the temperature values from the past three days are used as input X and the temperature two days ahead (t+2) as output y.
We use the Sequential model to build the RNN model. In this simple RNN model, we use one SimpleRNN layer and one fully connected layer.
We split the dataset into training and testing sets and train the model using the Adam optimizer and mean squared error loss function.
After training, we evaluate the model’s performance using the testing set and calculate the loss on the testing set.
Finally, we predict the temperature three days ahead by passing the last sample’s input from the testing set to the model.
This case demonstrates how to use Recurrent Neural Networks (RNNs) for time series prediction. By using the past few days’ temperature values as input, the RNN model can learn the temporal dependencies in the temperature sequence and accurately predict the temperature for a future day.
Python code (available for paid reading if needed; otherwise, you can try it yourself based on the above content!)