Scribd
Upload
13 views14 pages

Presentation 5

Uploaded by

ujjwal yadav
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
13 views14 pages

Presentation 5

Uploaded by

ujjwal yadav
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 14

Temperature Forecasting Leveraging Deep Learning with a Bi-

LSTM Model

Department of Computer Science and Engineering


Netaji Subhas University of Technology (NSUT)

NAME OF THE STUDENT ROLL NUMBER


Urvashi 2021UCM2365
Ujjwal 202IUCM2351
Kartik Yadav 2021UCS1719
Under the supervision
OF
Dr. Veenu
Problem Statement
The objective of this project is to develop a Bi-LSTM
(Bidirectional Long Short-Term Memory) model for weather
prediction. The model is tasked with predicting weather
parameters such as temperature, air pressure, wind
speed, wind direction, and relative humidity for the 121st
hour based on historical data from the past 120 hours.

It aims to leverage the Bi-LSTM architecture to effectively


capture both past and future dependencies in the
sequential weather data, enabling more accurate
predictions.
Motivation

Traditional weather prediction methods rely on solving complex equations, which can be computationally intensive
and struggle to capture non-linear weather patterns. Deep learning offers a powerful alternative by learning temporal
relationships directly from historical data.

This study explores the potential of Bi-LSTM models to improve the accuracy and efficiency of short-term weather
forecasting, which is crucial for decision-making in agriculture, transportation, and disaster management. By
leveraging deep learning, we aim to address the limitations of traditional methods and provide reliable, real-time
predictions.
Challenges in Existing Work

Complexity of Traditional Models: Computationally intensive and struggle with non-linear weather
patterns.

Limited Context Utilization: Traditional methods often fail to leverage long temporal relationships
effectively.

Real-Time Constraints: Many models lack efficiency for real-time forecasting applications.

Seasonal Variations: Capturing and predicting seasonal trends accurately remains a challenge.

Data Availability: Gaps or inconsistencies in historical weather data can impact model performance.
1-Hour Temperature
Forecasting Using Bi-
LSTM
• Background:
• Predicting temperatures is critical for applications like
weather monitoring, agriculture, and energy
management.

• Objective:
• Develop a deep learning model to predict the
temperature for the 121st hour using historical
weather data from the past 120 hours.

• Key Features:
• Leverage Bi-LSTM to capture temporal patterns.
• Provide hourly temperature forecasts.
Data Preprocessing Pipeline

Input Data: Steps: Split Ratios:


Parameters: Temperature, Standardize the dataset using Training: 70%, Validation: 15%,
pressure, wind speed, wind StandardScaler. Testing: 15%.
direction, humidity. Create sliding window sequences
Dataset: A 7-year-old historical of 120 hours for model input.
weather dataset, scaled and Separate the target variable
preprocessed for consistency and (temperature) for prediction.
optimal model performance.
Co-relation
of each
parameter
with others
Bi-LSTM Model Architecture

Overview: Layers: Parameters:

• The Bi-LSTM model captures • Input: Accepts sequences of • Optimizer: Adam


forward and backward temporal shape (120 hours × 5 features). • Loss Function: Mean Squared
dependencies in weather data. • Bi-LSTM: Error (MSE)
• 512 units.
• tanh activation.
• Dropout:
• Reduces overfitting by
randomly deactivating 25% of
neurons.
• Dense Output Layer:
• 1 unit with linear activation to
predict temperature.
Model Training Process

Training Process: Observations:


Epochs: 2 Training and validation loss decreased over
Batch size: 32 epochs.

Training Split: 70% Sufficient data preprocessing improved model


convergence.
Validation Split: 15%
Testing Split: 15%
Evaluation Metrics

Indicates the average


Root Mean Squared deviation between
RMSE: R-squared: Key Insight:
Error: 0.5616. actual and predicted
temperatures.

The model provides


Measures variance
reliable predictions
explained by the
for hourly
model: 0.993.
temperatures.
Results
Plot For 1 year
Conclusion

Successfully Provides a foundation


Achieved a reliable Model effectively
implemented a Bi- for future work in
RMSE and R-squared captures temperature
LSTM model for 24- advanced forecasting
value, demonstrating trends and provides
hour temperature models and real-time
the model's accuracy. meaningful forecasts.
prediction. applications.
References

• https://www.researchgate.net/publication/372826491_A_flexible_and_lightweight_de
ep_learning_weather_forecasting_model
• https://www.mdpi.com/2079-9292/13/17/3472
• https://www.sciencedirect.com/science/article/pii/S1385110124000054
• https://journal.esrgroups.org/jes/article/view/3557/2767

You might also like