Based on the BiLSTM regression forecast method

0, Preface

Use the BiLSTM algorithm to achieve regression prediction, and optimize the BiLSTM learning rate, number of iterations, batch size, and number of nodes in each layer.

1. Principle of BiLSTM

Please refer to the model explanation of the big guy, this article will not repeat it, please click principle

2. Performance affecting parameters

The learning performance of the neural network is affected by the learning rate, the number of iterations, the batch size, and the number of nodes in each layer.Reasonable setting of parameters can achieve ideal results.An optimization algorithm can be used for parameter optimization to replace the limitations of empirical settings.

The following figure is the prediction result after setting parameters directly based on experience

Performance Metrics:

Root Mean Square Error (RMSE): 7.8549
Mean Absolute Error (MAE): 6.1429
Mean Relative Percent Error (MAPE): 1.3486%
R-squared Coefficient of Determination (R2): 0.79258
> Nash coefficient (NSE): 0.78745

Using the sparrow optimization algorithm to optimize the BiLSTM learning rate, number of iterations, batch size, and number of nodes in each layer, the results are as follows:

SSA-BiLSTM
Root Mean Square Error (RMSE): 5.6401
Mean Absolute Error (MAE): 4.4521
Mean Relative Percent Error (MAPE): 0.97865%
R-squared Coefficient of Determination (R2): 0.89432
Nash coefficient (NSE): 0.89041