Ayun Hapsari, Budi Prasetiyo
Air temperature is an important element because it describes the weather conditions in a place. Temperature forecasts are made by collecting quantitative data about the state of the atmosphere. This research performs the process of forecasting air temperature using data mining methods using the backpropagation method. The artificial Neural Network Algorithm, commonly called the ANN algorithm, has a high success rate. In the forecasting process, the backpropagation method is often stuck in the local minimum problem. This is because initial weight initialization is done randomly and it is difficult to set network architecture parameter values such as learning rate, momentum, and training cycle. If the initialization weight is too low then the weight will drop to a very low value and eventually become useless. However, if the initialized weight is too large, the sigmoid function tends to be flat. The initial weight initialization problem can be solved by using the Particle Swarm Optimization Algorithm which can be called PSO, namely by representing the weights into the position matrix of the PSO particles. While the problem of determining parameter values is solved using a Genetic Algorithm which can be called GA, by generating a population that contains the training cycle chromosome, learning rate, and momentum. The results showed that PSO and GA had succeeded in reducing the RMSE value to an average of 0.006727 from ten trials after being applied to weather forecasting using the Artificial Neural Network algorithm. © 2023 Author(s).
Department of Computer Sciences, Universitas Negeri Semarang, Semarang, Indonesia
