Advancing Solar Forecasting with LSTM Algorithms

Solar energy is rapidly emerging as one of the most important renewable energy sources for reducing the current dependence on fossil fuel-based electricity generation. The recent improvements in renewable energy integration, specifically solar energy, into the power grid have significantly increased the importance of reliable solar energy forecasting techniques. This is because solar power generation is primarily influenced by meteorological and climatic conditions.

Image Credit: behindlens/Shutterstock.com
Image Credit: behindlens/Shutterstock.com

This article deliberates on the role of long short-term memory (LSTM) algorithms in solar energy forecasting.

Reliable Forecasting and LSTMs

In the last few decades, the penetration of renewable energies like solar energy has increased as they have become a promising solution to the global energy challenges. Photovoltaic technology is the most technologically attractive and convenient way to convert solar radiation into electricity.

Photovoltaic systems are preferred owing to their numerous advantages, including being environmentally friendly, having simple structures and easy applications, and storing the surplus of generated energy in batteries. Additionally, industrial electric power generation using photovoltaic systems requires fewer machines and labor and has reduced carbon emissions.

Yet, the photovoltaic power output is impacted by various factors like clouds and location. Solar irradiance's variation impacts the solar systems’ stability and the grid’s safety. Thus, accurate solar energy forecasting has become crucial with the recent advancements in energy conversion technology and reducing the cost of solar panels.

Such forecasting is necessary for reliable and cost-effective grid operation and control due to their rising contribution to grid electricity production. This increased the importance of developing forecasting techniques for reliable future prediction to ensure a comprehensive and balanced operation.

LSTMs are a type of recurrent neural network (RNN) that can remember information for much longer periods. LSTMs are one of the commonly utilized RNN models for time series data predictions, making them suitable for photovoltaic solar power production forecasting problems. Recently, LSTMs have also been widely adopted for short-term wind speed predictions.

Typically, capturing long-term time dependencies in time series while using RNNs is not easy. Thus, LSTM models, which are an extended version of RNNs, were designed to address this limitation.

LSTM models are efficient and flexible, and can effectively handle time dependency in data. Compared to other models like the autoregressive integrated moving averages model, the LSTM model provides a more accurate prediction in the presence of noisy data.

Long-term Solar Power Forecasting

A paper recently published in Energy Science and Engineering proposed a novel long-term (1 month–1 year) solar photovoltaic power forecasting approach using the LSTM model with the Nadam optimizer. This new and improved model could substantially enhance forecasting accuracy through the effective utilization of optimizers.

Moreover, the proposed approach can be implemented for a large number of step counts. The initial study comprised meteorological parameters like wind speed, solar insolation, relative humidity, and atmospheric temperature. These parameters significantly affect the solar photovoltaic panels’ output power.

The data used for this study contained 366 days of operational data obtained from a 250.25 kW grid-connected solar photovoltaic plant in Bhopal, India. None of the studies before this work considered such a large photovoltaic plant for long-term forecasting, which is essential for the efficient placement of renewable plants, reserve planning, and operation management.

The LSTM model showed superior performance with the time-series data as it contains information on more time steps. The experimental models were realized on the 250.25 kW solar photovoltaic power system in Bhopal. Researchers compared the proposed model’s performance with two time-series models and eight neural network models using LSTM with various optimizers.

The LSTM with Nadam optimizer significantly improved forecasting accuracy. For instance, 30.56% and 47.48% improvements in accuracy were realized using the LSTM with Nadam optimizer over the autoregressive integrated moving average and seasonal autoregressive integrated moving average, respectively.

Additionally, 58.29%, 50.69%, 11.84%, 4.88%, 3.51%, 1.43%, and 1.35% improvements were achieved over models using Ftrl, Adadelta, Adagrad, SGD, Adamax, Adam, and RMSprop, respectively. These results confirmed the feasibility of using the proposed method for solar photovoltaic power forecasting and improved system management and planning.

One Day-ahead Solar Power Forecasting

A study published in IEEE Access introduced a simplified LSTM algorithm built over the machine learning architecture/architecture of machine learning methodologies based on LSTM to forecast one-day-ahead solar power generation. Raw data were obtained from weather data acquisition and photovoltaic energy production equipment like datalogger, energy meter, and photovoltaic inverter, and preprocessed.

Then, the preprocessed data were sampled and split into two datasets for validation and training. To fit LSTM input requirements, these datasets were preprocessed to select and scale their features for dimensionality reduction. Different size train datasets contained data from two photovoltaic sites located in MFU (Thailand) and KHH (Kaohsiung).

Through the machine learning processes of data processing, cross-validation, model fitting, hyperparameters tuning, and metrics evaluation, the results showed that the proposed simplified LSTM model outperforms the multilayer perceptron (MLP) model for one day ahead of solar power forecasting. The root mean squared errors (RMSEs) of the LSTM model based on the one-month train dataset of MFU and KHH were 0.828 and 1.653, respectively.

Moreover, the forecast of the LSTM model could successfully capture intra-hour ramping on various weather scenarios, including rainy-sunny, sunny-cloudy, low light, rainy, cloudy, and sunny. The average RMSE was 0.512, which indicated that the proposed architecture and methodology can be effective for short-term solar power forecasting applications.

New Developments

A paper recently published in the journal Energies proposed a modified multi-step convolutional neural network (CNN) stacked LSTM network with dropout to predict two solar energy variables, including plane of array (POA) irradiance and solar irradiance/global horizontal irradiance (GHI).

Initially, a modified multi-step CNN was developed with several layers by improving the conventional CNN, which extracts features from the observed data independently. The sequenced CNN output data was used as input to a stacked LSTM network with dropout architecture to obtain the target variable without overfitting.

Solar data collection was performed at one-hour intervals from the Sweihan photovoltaic independent power project for July 2019. The input dataset was transformed into a supervised frame and split into testing and training datasets. During the testing period, the data were fed as input to assess the performance of the models using statistical measures, including RMSE, mean absolute percentage error (MAPE), mean squared error (MAE), R2, and normalized RMSE.

Although the LSTM outperformed the CNN in predicting POA and solar irradiance, the hybrid multi-step CNN-stacked LSTM model displayed better performance than both standalone LSTM and CNN models. The proposed model realized the best RMSE and R2 values of 61.24 and 0.96 for POA prediction, and RMSE and R2 values of 0.36 and 0.98 for solar irradiance prediction.

The proposed model's performance can be further refined through additional research and careful optimization of its key parameters. Thus, additional datasets will be utilized for in-depth investigation of the model, and the model must be evaluated independently in several locations.

Conclusion

Despite the effectiveness of LSTMs in solar energy forecasting, several challenges still exist regarding their adoption. For instance, efficiently comparing the accuracy between various LSTM prediction models is challenging due to several factors, including the use of different evaluation metrics, size of input parameters, forecasting horizons, and weather conditions of selected regions.

Hybrid models typically outperform standalone models while predicting solar irradiance. Specifically, the evaluation measures of hybrid models are significantly lower compared to standalone models. Hybrid models like CNN–LSTM require complex input data like images as it includes a CNN layer.

Lastly, the prediction accuracy for models running a large batch size of data is lower compared to other prediction models using small data batch sizes. This is because when more data are extracted, the process of producing the most accurate prediction becomes more complicated.

References and Further Reading

Jailani, N. L. et al. (2023). Investigating the Power of LSTM-Based Models in Solar Energy Forecasting. Processes, 11(5), 1382. DOI: 10.3390/pr11051382, https://www.mdpi.com/2227-9717/11/5/1382

Sharma, J. et al. (2022). A novel long-term solar photovoltaic power forecasting approach using LSTM with Nadam optimizer: A case study of India. Energy Science & Engineering, 10(8), 2909-2929. DOI: 10.1002/ese3.1178, https://scijournals.onlinelibrary.wiley.com/doi/full/10.1002/ese3.1178

Liu, C. H., Gu, J. C., Yang, M. T. (2021). A simplified LSTM neural networks for one day-ahead solar power forecasting. IEEE Access, 9, 17174-17195. DOI: 10.1109/ACCESS.2021.3053638, https://ieeexplore.ieee.org/abstract/document/9333638

Elsaraiti, M., Merabet, A. (2022). Solar power forecasting using deep learning techniques. IEEE Access, 10, 31692-31698. DOI: 10.1109/ACCESS.2022.3160484, https://ieeexplore.ieee.org/abstract/document/9737470

Konstantinou, M., Peratikou, S., Charalambides, A. G. (2021). Solar Photovoltaic Forecasting of Power Output Using LSTM Networks. Atmosphere, 12(1), 124. DOI: 10.3390/atmos12010124, https://www.mdpi.com/2073-4433/12/1/124

Elizabeth Michael, N., Mishra, M., Hasan, S. (2022). Short-Term Solar Power Predicting Model Based on Multi-Step CNN Stacked LSTM Technique. Energies, 15(6), 2150. DOi: 10.3390/en15062150, https://www.mdpi.com/1996-1073/15/6/2150

Last Updated: Sep 3, 2024

Samudrapom Dam

Written by

Samudrapom Dam

Samudrapom Dam is a freelance scientific and business writer based in Kolkata, India. He has been writing articles related to business and scientific topics for more than one and a half years. He has extensive experience in writing about advanced technologies, information technology, machinery, metals and metal products, clean technologies, finance and banking, automotive, household products, and the aerospace industry. He is passionate about the latest developments in advanced technologies, the ways these developments can be implemented in a real-world situation, and how these developments can positively impact common people.

Citations

Please use one of the following formats to cite this article in your essay, paper or report:

  • APA

    Dam, Samudrapom. (2024, September 03). Advancing Solar Forecasting with LSTM Algorithms. AZoAi. Retrieved on December 10, 2024 from https://www.azoai.com/article/Advancing-Solar-Forecasting-with-LSTM-Algorithms.aspx.

  • MLA

    Dam, Samudrapom. "Advancing Solar Forecasting with LSTM Algorithms". AZoAi. 10 December 2024. <https://www.azoai.com/article/Advancing-Solar-Forecasting-with-LSTM-Algorithms.aspx>.

  • Chicago

    Dam, Samudrapom. "Advancing Solar Forecasting with LSTM Algorithms". AZoAi. https://www.azoai.com/article/Advancing-Solar-Forecasting-with-LSTM-Algorithms.aspx. (accessed December 10, 2024).

  • Harvard

    Dam, Samudrapom. 2024. Advancing Solar Forecasting with LSTM Algorithms. AZoAi, viewed 10 December 2024, https://www.azoai.com/article/Advancing-Solar-Forecasting-with-LSTM-Algorithms.aspx.

Comments

The opinions expressed here are the views of the writer and do not necessarily reflect the views and opinions of AZoAi.
Post a new comment
Post

While we only use edited and approved content for Azthena answers, it may on occasions provide incorrect responses. Please confirm any data provided with the related suppliers or authors. We do not provide medical advice, if you search for medical information you must always consult a medical professional before acting on any information provided.

Your questions, but not your email details will be shared with OpenAI and retained for 30 days in accordance with their privacy principles.

Please do not ask questions that use sensitive or confidential information.

Read the full Terms & Conditions.