Energy demand forecasting in a food processing industry

In food processing companies that produce seasonal products, production planning is carried out depending on more than one variable, such as the variation in demand for the product due to the variations occurring in seasonal normals, and the inability to estimate production quotas issued by the government. A successful energy demand forecasting is needed in order to optimize the production of energy in the plants. In this study, a part of the required electrical energy is produced by the cogeneration method in the company. This competence requires production planning as well as estimating the energy that will be needed. If the energy demand forecasting can be made more effective, the enterprise will be able to make savings by producing it within its own generators instead of supplying the energy need from the distributor companies. While the more effective and successful energy demand forecasting can increase operating gains, inefficient demand forecasting leads to additional costs for the enterprise. These incomes and expenses balance shows how much the energy demand forecasting is important on the plant. Also, energy demand forecasting is made with daily production and energy consumption data in the mentioned food establishment which produces different foods according to seasons. The parameters that affecting the amount of energy consumption were determined and estimated by quantile regression, which is a statistical method, to determine the relationship between these parameters and the amount of energy consumption. It is observed that the quantile regression model is not a good predictor on the entire data set. Therefore, the new prediction model was developed and tested using artificial neural networks, which is one of the most popular methods that frequently used in the estimator algorithms of data with high variability. It is observed and reported that the developed artificial neural networks-based energy demand prediction model works with quite low error compared to quantile regression model.