Testing Artificial Intelligence Models for Solar Cook Stove Performance Prediction: Case Study of Agadez, Niger

Abstract

Solar cook stoves are becoming an increasingly popular and sustainable option for cooking food, especially in areas where access to traditional cooking fuels is limited or where deforestation is a concern. However, the performance of solar cook stoves can be affected by various factors such as weather conditions, time of day, and geographic location. The developed stove was tested in Agadez, Niger Republic on 9th December 2022, in the cold season, and obtained a temperature of 76.4 ◦C with radiation at noon of 727 W/m2 and a total daily radiation of 5044 W/m2. To improve the performance of solar cook stoves, researchers have turned to artificial intelligence (AI) based models, specifically artificial neural networks (ANN), for prediction. In this research, ANN was adopted to predict the performance of solar cook stoves. The ANN model employed the use of feed-forward back propagation as network type, trainlm as a training function, learngdm as learning function, and Mean Square Error (MSE) as performance function, best result was obtained with 2 layers, 10 neurons in the hidden layer and tansig as a transfer function. For the algorithms, Levenberg-Marrquadt was used in training. The experimental data were divided by 70% for training and 30% for testing. Based on the various interpretations provided, it appears that the AI based models for the prediction of Solar Cook Stove performance have shown promising results. In the sensitivity analyses, M11 was the function for the Test Temperature, and M12 was for Test Temperature and Solar Radiation. Model M11: Training Phase: MSE = 0.1, RMSE = 0.2, Testing Phase: MSE = 0.05, RMSE = 0.15, Model M12, Training Phase: MSE = 0.05, Testing Phase: MSE = 0.05, RMSE= 0.1, RMSE = 0.45, Model M12 has a higher RMSE during testing compared to Model M11, indicating lower predictive accuracy. The scatter plots of observed versus predicted values for both M11 and M12 indicate a fairly strong positive correlation. Furthermore, the radar plots of goodness of fit (measured by R-squared values) suggest that both models perform relatively well, with M12 showing a higher degree of fit compared to M11. The error of fit graph, presented as a bar plot, also shows that the testing phase for M12 is significantly more accurate, with a much lower mean squared error (MSE) and root mean squared error (RMSE) compared to M11.