Aplicação do algoritmo Random Forest para determinação da geração frustrada em parques eólicos em função de restrições operacionais do ONS

Abstract

This work presents a methodology for building a wind power generation prediction model during operational restrictions established by the National System Operator (ONS). The adopted approach involves the application of Advanced Analytics techniques and stands out for the use of the Random Forest algorithm, providing the predictive capability needed to estimate the generated energy in one-hour intervals based on known variables of the wind farm. For model validation and training, 8,670 real samples were collected, covering data on the availability of generating units, power and generated energy, and anemometric data. A data pre-processing procedure was applied to carefully select the variables to be used in the model training, ensuring the representativeness and relevance of the inputs. The entire computational process was carried out using the KNIME Analytics computational platform. The results achieved with the methodology indicated that the predicted energy generation has a Root Mean Square Error (RMSE) of 0.354 MWh – for a variable whose average value is 11.5 MWh – and a Mean Absolute Percentage Error (MAPE) of 3.8%, demonstrating a good accuracy of the developed model in predicting the target variable. This reinforces confidence in its use for predicting wind generation during operational restrictions. With the model, it was possible to determine that for the analyzed wind farm, the frustrated generation from Jan/2022 to Dec/2022 was 557.94 MWh, representing a financial loss of R$ 48.000 for the agent. Thus, the proposed methodology can be used by wind generation agents to estimate the amount of frustrated generation in each period and use this information for managerial decisions or to request compensation for the energy lost