Forecasting multiple waste collecting sites for the agro-food industry

The agro-food industry wastes tons of oil and grease not suitable for immediate consumption. Their collection mostly relies on the experience of managers and this results in inaccurate visits by truck drivers and operations teams. Indeed, the measurement of by-products waste is complex and thus information is imprecise, making the collecting operations inefficient. In this paper, we propose a model that forecasts the daily input of thousands of industrial and commercial sites of the agro-food industry based on historical data. The algorithm rejects errors and mistakes in the routing-collection-measuring process. In our model, the site container capacity is known and remains constant. The main contribution of this study is to propose a model based on the Theil-Sen constrained regression (Theil-Sen CR) that rejects errors and outliers to simplify the forecast of future collections. We apply this method to a real case study and compare its performance at different collecting sites. The forecasting error is significant compared to Linear Regression (LR). We have calculated, for our industrial partner, based on 12.2 km between sites and a fleet of 200 trucks, a potential reduction of 940 tCO2 equivalent per year.