Implementation Of ANN And Random Forest To Short-Term Solar Power Generation Forecasting

Solar power is the most promising resource of electrical power generation.However,the electrical power generated is intermittent and it needs to be foretasted for its proper utilization.Even though human predictions has been playing major role in the solar forecasting up until now things need to be spice up with the technology.The electrical power generation can be forecasted by using different machine learning algorithms.In this research,I have used the Random Forest and Artificial Neural Network for the forecasting of the electrical power generation from the solar power system.Both of these algorithms can effectively predict the solar power generated at a specific place during its daytime.I have operated on the past solar data of Lahore,Pakistan.The main factors in the data used are Temperature and Irradiance.All of this implementation has been done using MATLAB which provides real-time simulation of the PV system.For the accuracy purposes three different solar panels are used and compared for our work namely TATA solar panel,Telesun Solar Cell and Taienergy Solar Power.To train and optimize ANN,models like Levenberg-Marquardt and Backpropagation(BP)are tested.After implementing these both and comparing their errors with the predefined number of epochs,I have concluded that LM produces less training and test error as compared to BM.Therefore,using LM I have done the final forecasting.The Random Forest algorithm applied is a form of decision tree-based search.After clearing the data,the final performance evaluation was classified into three RF models based on the weather condition.RF model 1,RF model 2 and RF model 3 for clear,cloudy and rainy/snowy days respectively.The performance evaluation parameters applied for the RF are Mean Absolute Error(MAE)and Mean Absolute Percentage Error(MAPE).However,the random Forest is less computationally intensive and does not require a GPU to complete the training.The random forest area can produce a special explanation for the selection tree,but the yield is higher.The ANN need much more training time than the ordinary algorithm i.e.,random forest algorithm.The neural community may reduce the interpretability of its function to the point that it no longer makes sense for performance reasons.Thus,the soft computing techniques i.e.,Random forest is used for forecasting data for a day but the neural networks are used for the forecasting of solar power generation for long periods i.e.,months or years.After carefully executing and comparing the solar forecast results of my research I was able to know the runtime of ANN was 7 times more than that of RF.