free counter
World

Predicting solar powered energy output with limited data sizes

Researchers in Greece are suffering from three different transfer-learning approaches for solar powered energy forecasting. The techniques derive from a stacked long short-term memory model, that is a sort of recurrent neural network that may learn order dependence in sequence prediction problems.

Scientists at the National Technical University of Athens purchased a machine-learning method referred to as transfer- learning (TL) to build up a new solar powered energy forecast modeling intended at helping developers with limited data size.

The TL method runs on the trained model using one task to repurpose in another, related task. The researchers used three TL strategies in conjunction with the stacked long short-term memory (LSTM) model, that is a sort of recurrent neural network with the capacity of learning order dependence in sequence prediction problems. The LSTM technique takes the relevant elements of a pre-trained machine learning model and applies it to a fresh but similar problem.

TL is exploited both for weight initialization of the LSTM model and for feature extraction, using different freezing approaches, they explained. LSTM depends upon weight updating between your neurons of the deep learning model, allowing the creation of pre-trained models. Thus, it facilitates pre-training the model on the baseline PV to be able to make use of the saved weights of the pre-trained model and apply TL on the prospective PV.

The stacked LSTM model considers temperature, humidity, solar irradiance, PV production, one-hot encoding representation of the month of the entire year, and sine/cosine transformation of the hour of day. The three strategies were predicated on three different approaches: keeping the weights of the layer fixed, fine-tuning the weights of the layer in line with the target domain data, and training the weights of the layer from scratch in line with the target domain data.

The TL Strategy 1 approach is reportedly in a position to extract features from the foundation domain and carry them to the prospective domain.

It is a trusted scheme when treating images, where in fact the first layers are employed as feature extraction layers and the final layers are accustomed to adjust to new data, the researchers explained.

With TL Strategy 2, weights of most layers of the TL model are initialized predicated on data from the foundation domain. They’re fine-tuned predicated on data from the prospective domain.

This process is extensively used in combination with problems where there’s a good amount of data in the foundation domain, but a scarcity of data in the mark domain, the group said.

In TL Strategy 3, the original layers of the TL model are frozen and the final layer is trained from scratch, popping the final layer of the bottom model and adding a fresh layer following the frozen layers.

This process is comparable to the initial one, nonetheless it differs in the truth that the weights of the final layer aren’t initialized predicated on data from the foundation domain, the academics said.

The researchers used the three ways of forecasting the hourly production of six solar plants located across several locations in Portugal. Their effectiveness was in comparison to that of conventional non-TL models.

The findings of the experimental application indicate that three TL strategies significantly outperform the non-TL approach with regards to forecasting accuracy, evaluated by several error indexes, the scientists said. Results indicate that TL models significantly outperform the traditional one, achieving 12.6% accuracy improvement with regards to root-mean-square error (RMSE) and 16.3% with regards to forecast skill index with twelve months of training data.

They introduced the model in Transfer learning approaches for solar powered energy forecasting under data scarcity, that was recently published in Scientific Reports.

This study may be the first rung on the ladder towards enhancing our knowledge of the impact of TL on solar plant power prediction, they concluded. Future work will focus on assessing the impact of the bottom models training data volume, investigating whether training base models with an increase of data or with data from different solar plants could further improve forecasting accuracy.

This article is protected by copyright and could not be reused. In order to cooperate around and wish to reuse a few of our content, please contact: editors@pv-magazine.com.

Read More

Related Articles

Leave a Reply

Your email address will not be published.

Back to top button

Adblock Detected

Please consider supporting us by disabling your ad blocker