Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar1047

Solar Power Prediction Using LSTM & RNN

R. Indhuja1; Sivakami G.2; Syamala Devi S.3; Sowndarya A. K.4
1,2,3,4
Department of Artificial Intelligence & Data Science Kamaraj College of Engineering and Technology
(An Autonomous Institution - Affiliated to Anna University, Chennai)
K. Vellakulam, Virdhunagar – 626 001

Publication Date: 2025/03/31

Abstract: Solar power prediction using Long Short-Term Memory (LSTM) and Recurrent Neural Networks (RNN) in
Python is a crucial application of deep learning for renewable energy optimization. This study focuses on leveraging time-
series forecasting capabilities of LSTM and RNN to predict solar power generation based on historical data, including
temperature, sunlight intensity, humidity, and other meteorological factors. By preprocessing data, normalizing inputs,
and training models using TensorFlow and Keras, the study enhances prediction accuracy. The comparative analysis of
LSTM and standard RNN highlights the superior performance of LSTM in capturing long-term dependencies and
mitigating vanishing gradient issues. The results demonstrate that deep learning models can effectively forecast solar
energy output, aiding energy grid management and sustainable resource planning.Solar power is one of the most
promising renewable energy sources, playing a crucial role in sustainable energy solutions. However, its efficiency depends
on various meteorological factors, such as sunlight intensity, temperature, humidity, and cloud cover, making accurate
prediction a challenging task.

This study explores the application of LSTM and RNN models for predicting solar power generation using Python-based machine
learning frameworks such as TensorFlow and Keras. By leveraging historical meteorological data, the proposed models aim to i mprove
forecasting accuracy, aiding energy management systems in optimizing solar energy utilization and grid stability. The research also
includes a comparative analysis of RNN and LSTM to assess their effectiveness in predicting solar power generation.

Keywords: Solar Power Prediction, Time Series Forecasting, Recurrent Neural Network (RNN), Long Short-Term Memory
(LSTM), Renewable Energy, Deep Learning, Energy Management, AI in Forecasting, Flask Web Application.

How to Cite: R. Indhuja; Sivakami G.; Syamala Devi S.; Sowndarya A. K. (2025) Solar Power Prediction Using LSTM & RNN.
International Journal of Innovative Science and Research Technology,
10(3), 1545-1548. https://doi.org/10.38124/ijisrt/25mar1047.

I. INTRODUCTION benefits energy providers, researchers, and policymakers by

improving decision-making in renewable energy
Solar power prediction is essential for efficient energy management.
management, grid stability, and optimizing the use of
renewable resources. Solar energy generation depends on A. Overview
various environmental factors like temperature, humidity, Solar power prediction is crucial for optimizing
and cloud cover, making accurate forecasting a challenge. energy management and ensuring grid stability, especially
Traditional statistical models often fail to capture the as renewable energy adoption increases. Our AI-driven
nonlinear patterns in solar power generation. Advanced forecasting system enhances energy planning by:
deep learning techniques, particularly Recurrent Neural
Networks (RNN) and Long Short-Term Memory (LSTM)  Improving Prediction Accuracy using deep learning
networks, are highly effective in handling time-series data, models like RNN and LSTM.
learning temporal dependencies, and improving prediction  Aiding energy providers in balancing supply and
accuracy. demand efficiently.
 Offers a user friendly flask Web-Application for real
This project utilizes RNN and LSTM models to time solar power forecasting.
forecast solar power generation based on historical weather
and solar energy data. By analyzing past trends, the model The integration of AI in renewable energy forecasting
predicts future energy output, aiding in efficient energy helps maximize efficiency, reduce reliance on non-
distribution and planning. A Flask web application is renewable sources, and support sustainable energy
integrated to provide an interactive user interface for management for a greener future.
inputting data and visualizing predictions. This system

IJISRT25MAR1047 www.ijisrt.com 1545

Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar1047

B. Objective CNNs for detecting skin cancer from images, achieving

accuracy comparable to dermatologists. While focused on
 To analyze historical solar power generation data along skin conditions, the methodology of leveraging deep
with meteorological factors such as sunlight intensity, learning for image-based medical diagnosis is closely
temperature, humidity, and cloud cover. related to pneumonia detection from chest X-rays.
 To develop and implement RNN and LSTM-based
models for time-series forecasting of solar power "Sequence to Sequence Deep Learning Models for
generation using Python frameworks such as Solar Irradiation Forecasting",Bhaskar PratimMukhoty,
TensorFlow and Keras. Vikas Maurya, Sandeep Kumar Shukla,Year 2023.This
 To compare the performance of RNN and LSTM paper explores the application of sequence-to-sequence
models based on evaluation metrics such as Mean deep learning models, particularly LSTM networks, for
Squared Error (MSE), Root Mean Squared Error short-term forecasting of Global Horizontal Irradiance
(RMSE), and Mean Absolute Error (MAE). (GHI).
 To optimize the model parameters to improve
prediction accuracy and efficiency. The study demonstrates that LSTM models effectively
capture temporal dependencies in solar irradiation data,
C. Social Impact leading to improved forecasting accuracy compared to
traditional methods.
 Sustainable Energy Management: Enhances the
efficiency of solar power utilization, reducing "An Integrated Multi-Time-Scale Modeling for Solar
dependence on fossil fuels. Irradiance Forecasting Using Deep Learning”,Sakshi
 Optimized Grid Stability: Helps balance energy Mishra, Praveen Palanisamy,Year 2022.This research
supply and demand, preventing power shortages or proposes a unified architecture combining RNN and LSTM
wastage. for multi-time-scale solar irradiance forecasting. The study
 Support for Remote Areas: Enables better energy demonstrates that the integrated model achieves lower root-
mean-square prediction error compared to existing
planning in regions with limited access to reliable
methods, highlighting its effectiveness for intra-day and
electricity sources.
intra-hour forecasting horizons.
 Cost-Effective Energy Solution: AI-driven forecasting
reduces energy costs by improving resource allocation
III. DESIGN ARCHITECTURE
and minimizing inefficiencies.
A. Design Architecture
II. REALTED WORK
This section illustrates a Flask-based web application
that integrates deep learning for solar power prediction. The
"A Novel Approach Based Deep RNN Using Hybrid
system architecture comprises four main components:
NARX-LSTM Model For Solar Power
Forecasting",Mohamed Massaoudi, Ines Chihi, Lilia
 User Input & Flask Server: Users upload historical
Sidhom, Mohamed Trabelsi, Shady S. Refaat, Fakhreddine
S. Oueslati,Year 2022,This paper proposes a hybrid model solar power data in CSV format through a web-based
interface. The Flask server receives the request and
combining Nonlinear Autoregressive Exogenous (NARX)
temporarily stores the file in a predefined folder
models with LSTM networks to address the challenges of
solar power forecasting. The study demonstrates that (e.g., /uploads). The user's input parameters, such as
integrating LSTM with NARX enhances prediction location and weather conditions, are recorded in the
accuracy by effectively capturing long-range dependencies Flask server for further processing.
in time-series data.  Preprocessing & Model Inference: The system
retrieves the uploaded dataset and applies preprocessing
"Short-Term Power Prediction for Renewable Energy steps, including missing value handling, feature scaling,
Using Hybrid Graph Convolutional Network and Long and time-series formatting. The preprocessed data is
Short-Term Memory Approach”,Wenlong Liao, then passed into the pre-trained deep learning models
BirgitteBak-Jensen, Jayakrishnan Radhakrishna Pillai, Zhe (RNN and LSTM, saved in .h5 format). The model
Yang, Kuangpu Liu,Year 2023.This study introduces a predicts future solar power generation based on
hybrid model that combines Graph Convolutional historical trends.
Networks (GCN) with LSTM networks to capture both  Database Storage (SQLite): The prediction results,
spatial and temporal correlations in renewable energy data. along with the uploaded dataset’s metadata, are stored
The proposed approach outperforms traditional models in in an SQLite database. This allows users to retrieve past
short-term power prediction, demonstrating the predictions and analyze long-term energy trends for
effectiveness of integrating spatial and temporal features. better planning.
 Output & User Interface: The forecasted solar power
Esteva, A., et al. [3] (2017): “Dermatologist-level output is sent back to the Flask server, which processes
classification of skin cancer with deep neural networks.” and displays the results on the web interface. Users can
Published in Nature, this study explored the application of view visualized graphs of predicted power generation

IJISRT25MAR1047 www.ijisrt.com 1546

Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar1047

trends, helping them make informed energy with user login, where users authenticate themselves to
management decisions. access the platform securely. Once logged in, they enter
the city name, which acts as an input for retrieving real-
time weather data. The system then fetches weather data
through an API (such as Open Weather Map or NASA’s
POWER API), gathering key meteorological parameters
like solar radiation, temperature, humidity, wind speed, and
cloud cover. This data is essential for predicting solar
energy output.

After fetching the weather data, the system

performs data preprocessing, handling missing values,
normalizing inputs, and structuring them for time-series
analysis. The preprocessed data is then fed into a deep
learning model based on Recurrent Neural Networks
(RNN) and Long Short-Term Memory (LSTM) algorithms.
These models are trained to learn complex temporal
dependencies and patterns in historical solar power data,
improving forecasting accuracy.

The prediction results are stored in an SQLite

database, allowing users to track past forecasts and
compare them with actual energy generation data.

Finally, the predicted solar power output is

displayed on the web interface, where users can visualize
trends through interactive graphs and numerical values.
This AI-driven forecasting system helps energy providers,
Fig 1: Flow Chart researchers, and policymakers make informed decisions
about solar energy utilization, ultimately contributing to
B. Model Mechanism sustainable energy management and optimized grid
The model mechanism for solar power prediction is stability.
designed to provide accurate and real-time forecasting
through a Flask-based web application. The process starts

IV. RESULTS

Fig 2: The Solar Energy Prediction Output Provides Forecasted Power Generation Based on Historical and Real-Time
Weather Data, Enabling Users to Optimize Energy Planning, Manage Grid Stability, and Enhance Renewable
Energy Utilization Efficiently

IJISRT25MAR1047 www.ijisrt.com 1547

Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar1047

Fig 3: The Weekly Solar Power Prediction Provides Forecasted Energy Generation Trends using RNN and LSTM Models,
Helping Users Analyze Patterns, Optimize Energy Usage, and Plan Efficient Resource Allocation for Sustainable Power
Management

V. CONCLUSION & FUTURE WORK deployment on low-power devices. Ensemble learning

approaches, such as stacking or boosting multiple deep
In summary, this study explored the use of Long learning models, may also improve forecast stability and
Short-Term Memory (LSTM) and Recurrent Neural reliability.
Networks (RNN) for predicting solar power generation.
The results demonstrated that LSTM models effectively REFERENCES
capture long-term dependencies in time-series data,
outperforming traditional forecasting techniques in [1]. U. K. Das, K. S. Tey, M. Seyed mahmoudian, S.
accuracy and reliability. Mekhilef, M. Y. I. Idris, W. V. Deventer, B. Horan,
and A. Stojcevski, “Forecasting of photovoltaic
The implementation of deep learning-based models power generation and model optimization: A
provided significant improvements in short-term solar review,” Renewable and Sustainable Energy
power prediction, reducing errors and enhancing decision- Reviews, vol. 81, pp. 912 – 928, 2023.
making for energy management. Despite the promising [2]. M. Abuella and B. Chowdhury, “Solar power
outcomes, challenges such as model complexity, data probabilistic forecasting by using multiple linear
quality, and computational requirements were identified. regression analysis,” in Southeast Con 2021.
Addressing these limitations through feature selection, IEEE,2022, pp. 1–5.
hybrid models, and improved training techniques can [3]. A. Gensler, J. Henze, B. Sick, and N. Raabe, “Deep
further enhance prediction accuracy.In conclusion, LSTM learning for solarpowerforecastingâ ˘AˇTan
and RNN models present a powerful approach for solar approach using autoencoder and
power forecasting, contributing to efficient renewable lstmneuralnetworks,” in Systems, Man, and
energy integration. Future research could focus on Cybernetics (SMC), 2023 IEEE Inter-national
combining deep learning with external factors like satellite Conference on. IEEE, 2023, pp. 002 858–002 865.
data, weather forecasts, and hybrid AI techniques to refine [4]. J. Zeng and W. Qiao, “Short-term solar power
predictions and support sustainable energy solutions. prediction using a support vector machine,”
Renewable Energy, vol. 52, pp. 118–127, 2023.
In future work, transfer learning can be explored to [5]. S. Ferrari, M. Lazzaroni, V. Piuri, L. Cristaldi, and
develop models that generalize well across different M. Faifer, “Statistical models approach for solar
geographic locations. To optimize computational radiation prediction,” in Instrumentation and
efficiency, techniques such as model pruning, quantization, Measurement Technology Conference (I2MTC),
and lightweight architectures can be implemented, enabling 2013 IEEE International. IEEE, 2023, pp. 1734–
faster and more efficient predictions suitable for 1739.

IJISRT25MAR1047 www.ijisrt.com 1548