| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 3.27 MB | Adobe PDF |
Authors
Abstract(s)
Optimizing indoor hydroponic lettuce production is important for enhancing its economic viability. This study examined machine learning approaches for its optimization. Among the options analyzed, an artificial neural network (ANN) combined with a brute-force algorithm was chosen. This choice proved compatible with the available time and materials.The gain in fresh matter mass was defined as the optimization goal. And, the irrigation time off, the electric conductivity, and the pH of the nutrient solution were defined as the optimization parameters. The employed method involved four steps: conducting exploratory plantings to generate a database; training an ANN to create a regression model; using the brute-force algorithm to find the optimal combination in the generated model; and conducting validation plantings to verify if the found combination produces the predicted value. The exploratory plantings generated results with variations. The variations
were normalized and additional lighting data were collected. The optimization result predicted that the found combinations would generate an average normalized production of 114.22g per individual. Two validation plantings were conducted: the first exceeded the prediction by 34.43%, while the second was 17.20% below the expected. The first result surpassed the values of the exploratory plantings, but the second did not. Due to this divergence, further studies will be necessary to reinforce the validation of the results.
Description
Mestrado de dupla diplomação com a UTFPR - Universidade Tecnológica Federal do Paraná
Keywords
Modified hydroponic shipping container Controlled-environment agriculture Artificial neural networks Brute-force algorithm