search

openclimatefix / open-source-quartz-solar-forecast

Open Source Solar Site Level Forecast
MIT License
60 stars 51 forks source link

Quartz-Solar-Forecast_demo #87

Open weiyang22 opened 6 months ago

weiyang22 commented 6 months ago

Detailed Description

Given the absence of specific data for the model, I proceeded to predict photovoltaic power generation by leveraging preprocessing and feature engineering techniques using data retrieved from https://dkasolarcentre.com.au/download?location=alice-springs. This demonstration encompasses four distinct models: LSTM, LSTM+CNN, LSTM+CNN+Attention, and RNN.

Context

In this demo, extracting strongly correlated features from cleaned data after data preprocessing will yield a better prediction performance. In community projects, I believe that data preprocessing and feature engineering are crucial aspects. How to clean and merge features from different datasets through data preprocessing operations in PSP will be a key point for the "Add other model architectures to Open Quartz Solar" project.

Possible Implementation

https://github.com/weiyang22/Open-Source-Quartz-Solar-Forecast/blob/main/Quartz-Solar-Forecast_demo.ipynb

peterdudfield commented 6 months ago

Hi @weiyang22 thats great to use that free data. And that looks really cool whats your done with various models. Some quick questions if thats ok?

  1. Do you have a summary table of which model was best?
  2. Could you also run quartz-solar-forecast on this datestamos and see how it does? To compare?

We've got some specific data that we've used, but admitely its not in an easy to use csv. Perhaps that could be a task for someone. The pv data is here and the NWP data is here.

weiyang22 commented 6 months ago

Hi @weiyang22 thats great to use that free data. And that looks really cool whats your done with various models. Some quick questions if thats ok?

  1. Do you have a summary table of which model was best?
  2. Could you also run quartz-solar-forecast on this datestamos and see how it does? To compare?

We've got some specific data that we've used, but admitely its not in an easy to use csv. Perhaps that could be a task for someone. The pv data is here and the NWP data is here.

Thank you very much for your compliment.

1、In this dataset, the best models are likely those utilizing LSTM_CNN and LSTM_CNN_Attention architectures, as they demonstrate the best fit between predicted and actual results. 2、Initially, I intended to modify the data provided by the community. However, Jacob mentioned that some data couldn't be disclosed publicly, so I opted for alternative data sources.

In my opinion, factors affecting solar energy generation in real life include:

1、The efficiency of solar panels (which decreases with installation). 2、The angle of solar panels. 3、Altitude, solar radiation, temperature (which affects wiring and efficiency). 4、Weather and other local conditions at the solar installation site. For instance, for location A, relevant features would be altitude, weather, temperature, sunrise/sunset times, installation angle of solar panels, installation status, time at location A, solar radiation, and efficiency of the solar panels installed at location A.

The predictive code in the community utilizes metrics like clear sky irradiance. Therefore, effective preprocessing and feature engineering of the raw data will significantly enhance prediction accuracy. Additionally, factors like latitude and longitude might not be crucial. I'm eager to contribute to the community's efforts and assist in optimizing this work.

peterdudfield commented 6 months ago

Hi @weiyang22 thats great to use that free data. And that looks really cool whats your done with various models. Some quick questions if thats ok?

  1. Do you have a summary table of which model was best?
  2. Could you also run quartz-solar-forecast on this datestamos and see how it does? To compare?

We've got some specific data that we've used, but admitely its not in an easy to use csv. Perhaps that could be a task for someone. The pv data is here and the NWP data is here.

Thank you very much for your compliment.

1、In this dataset, the best models are likely those utilizing LSTM_CNN and LSTM_CNN_Attention architectures, as they demonstrate the best fit between predicted and actual results. 2、Initially, I intended to modify the data provided by the community. However, Jacob mentioned that some data couldn't be disclosed publicly, so I opted for alternative data sources.

In my opinion, factors affecting solar energy generation in real life include:

1、The efficiency of solar panels (which decreases with installation). 2、The angle of solar panels. 3、Altitude, solar radiation, temperature (which affects wiring and efficiency). 4、Weather and other local conditions at the solar installation site. For instance, for location A, relevant features would be altitude, weather, temperature, sunrise/sunset times, installation angle of solar panels, installation status, time at location A, solar radiation, and efficiency of the solar panels installed at location A.

The predictive code in the community utilizes metrics like clear sky irradiance. Therefore, effective preprocessing and feature engineering of the raw data will significantly enhance prediction accuracy. Additionally, factors like latitude and longitude might not be crucial. I'm eager to contribute to the community's efforts and assist in optimizing this work.

Sounds great!

Let us know any questions you have to get started