Predicting crop yield and field energy output for oil palm using genetic algorithm and neural network models

For many years, the Malaysian oil palm industry has been facing the challenge of reduced rate of palm oil yield due to sizeable difference between the crop’s actual yield and the crop’s genetic yield potential. This gap has grown wider over time and has been of great concerned since oil palm is a very important commodity that contributes significantly to the country’s GDP. Currently, Malaysia has devoted a high percentage of the land resource and material inputs to agriculture, whereby a large proportion of them are used for oil palm cultivation. However, the typical yields are only 50–60% of the potential, and artificial intelligence research on modelling of the crop yield and energy consumption is still at its infancy. Forecasting oil palm production and selecting significant variables that effects production are complex activities. Accurate prediction results are required for this type of analysis and can provide the basis for the decisions and plans for the management of agricultural crops in the local, regional, and global scale. In the field of agricultural engineering, artificial intelligence has helped to reduced operational periods and costs. There was not enough information available on the implementation of neural networks and genetic algorithm for the prediction and selecting input variables in oil palm yield and output energy. This research presents the development of a GA and SW as a variables selection method in ANN and NARX models for predicting oil palm yield and output energy. Data were collected from 11 districts in 11 states in Malaysia for FFB and PO models, which includes Kedah, Kelantan, Johor, Melaka, Penang, Pahang, Perak, Selangor, Terengganu, Sabah, and Sarawak. The study is based on monthly data from 2005 to 2015. In FFB and PO models, the data used 15 variables, namely: percentage of mature area and percentage of immature area, rainfall, rainy days, humidity, radiation, temperature, surface wind speed, evaporation and cloud cover, O3, CO, NO2, SO2, and PM10. The study used input energy data from 8 variables for developing energy models. These data included human power, electricity, fuel, water, fertilizers, and seed. Data were collected from Peninsular Malaysia, Sabah and Sarawak over a period of 11 years (annual data from 2005 to 2015). Results showed that GA was able to select the variables correctly, while also being an easy-to-use variable selection tool. It proved to be more effective than the Stepwise. The findings of this research, using 11 years of climate change and air pollution, have significantly affected the oil palm production. Surface wind speed and humidity were recorded at an impact ratio of up to 100%, which correlated negatively on the productivity of oil palm plantations. Surface wind speed and humidity reduced the productivity of oil palm FFB plantations for 5.12 and 4.61 ton/ha/11year in the Sabah and Sarawak respectively. Additionally, the surface wind speed is considered the most essential variable recorded with an impact ratio of up to 100% on FFB in Selangor, Terengganu, and Kelantan while the cloud cover, average NO2 in the air, average PM10 in the air, humidity, radiation, and O3 recorded the most significant impact up to100% on FFB in Perak, Melaka, Johor, Kedah, Penang, and Pahang respectively. Fuel consumption, water, and P-fertilizer consumption are considered the most important variables in oil palm plantation operations, its importance being the relative values of 45%, 34.3 %, and 23 %. These variables impacted oil palm operation during the 11 years at 67.764, 45.38, 16.24 GJ /ha for Peninsular Malaysia, Sabah and Sarawak, respectively. In this study, the performances of six models (namely, ANN, GA-ANN, SW- ANN, NARX, GA-NARX and SW-NARX) are compared with one another as well as with multiple linear models. The GA-NARX was chosen as the best yield model in 9 states (Perak, Sabah, Sarawak, Selangor, Terengganu, Pahang, Kedah, Kelantan and Penang), while the GA-ANN was considered the best yield model recorded in Melaka and Johor. Additionally the GA-NARX was chosen as the best energy model in Peninsular Malaysia, Sarawak and Sabah, with the average accuracy percentage simulation being 0.95.07, 95.55 and 87.43 % respectively. Finally, this research concluded that a genetic algorithm is useful for selecting input variables in oil palm production. It is a user-friendly variable selection tool with excellent results compared to Stepwise, especially in a large search space. The GA-ANN and GA-NARX models perform markedly better than the other models in the most training algorithms with different numbers of hidden layers.

Text FK 2019 36 ir.pdf Download (1MB)

Actions (login required)

View Item