[50] - Solar Array Data Prediction and Analysis

• 

  RL-TD3 model on daily data

• 

  RL-TD3 model on daily data

• 

  Transformer on hourly data

• 

  Transformer of daily test data

Predicting Daily and Hourly Energy Production of the CAMBUS Solar Array Using Weather Data from NSRDB

ENGIE Solar Performance

Utilizing energy system data spanning 11 years, I predict both daily and hourly energy production of the CAMBUS solar array. I have employed transformer models, MLP, and RL-TD3 models for this prediction task. The transformer model has shown the best predictive performance, achieving a Mean Squared Error (MSE) loss of 4000-7000 on daily test data and approximately 80 on hourly test data.

Inspiration

Acknowledging the impact of technology on the environment, I delve into analyzing and predicting solar energy production using weather data from NSRDB.

What it does

Leveraging over a decade of solar irradiance data and information about the CAMBUS solar panel array, my program calculates, compares, and visualizes each array's energy production based on weather data. Understanding and harnessing weather data, specifically tailored to solar energy generation, allows us to develop efficient and sustainable energy solutions. The goal is to optimize the performance of solar arrays by leveraging weather data and its correlation with energy production. By doing so, we contribute to a cleaner, greener future while advancing our understanding of renewable energy technologies.

How I built it

To construct this predictive model for daily and hourly energy production of the CAMBUS solar array, I began by leveraging the rich dataset from the National Solar Radiation Database (NSRDB). The dataset contained vital weather-related features such as 'day_of_year,' 'year,' 'air_temperature,' 'alpha,' 'aod,' 'asymmetry,' 'cld_opd_dcomp,' 'cld_reff_dcomp,' 'clearsky_dhi,' 'clearsky_dni,' 'clearsky_ghi,' 'cloud_press_acha,' 'cloud_type,' 'dew_point,' 'dhi,' 'dni,' 'ghi,' 'ozone,' 'relative_humidity,' 'solar_zenith_angle,' 'ssa,' 'surface_albedo,' 'surface_pressure,' 'total_precipitable_water,' 'wind_direction,' and 'wind_speed.'

To predict energy production, I opted for various machine learning models, including transformer models, Multi-Layer Perceptrons (MLP), and RL-TD3 models. Through rigorous experimentation and testing, the transformer model emerged as the most effective, displaying the lowest Mean Squared Error (MSE) loss. On daily test data, the transformer model showcased a remarkable MSE loss range of 4000-7000, while achieving an impressive MSE loss of around 80 on hourly test data.

Throughout this development process, I faced initial challenges, particularly in understanding the intricate technical and non-technical aspects associated with the project. Overcoming these challenges required quick learning, effective organization, and adapting to new knowledge and skills. The dedication to sustainability and the determination to create a positive impact through efficient energy solutions were the driving forces that fueled my progress.

Challenges I ran into

Initiating the project posed a significant challenge, given my initial lack of technical and non-technical knowledge related to this domain. Additionally, organizational and planning shortcomings were obstacles. However, I've learned to adapt and progress, gaining valuable insights into project management and the technical aspects of my work.

Accomplishments that I'm proud of

I take pride in dedicating my efforts to sustainable solutions, aligning with the university's commitment to efficient sustainability. This project allows me to refine my skills and contribute to a sustainable future.

What I learned

Participating in this project taught me coding best practices, effective feature selection, and rapid learning of new skills and technologies. Engaging with APIs and tackling a multifaceted project broadened my understanding and capabilities.

What's next for ENGIE Solar Performance

In the future, ENGIE Solar Performance can aim to forecast peak energy production by utilizing precise weather reports. This invaluable insight will enable optimal utilization of solar panels, enhancing energy output. Leveraging Artificial Intelligence, they may envision an automated system to monitor and regulate solar panels, driving increased productivity seamlessly and efficiently.

CREATED BY

A K M MUHITUL ISLAM

Built With

• python