Artificial intelligence modeling for power system planning

The increasing complexity of modern power systems due to the integration of prosumers, renewable energy sources, and energy storage, has significantly complicated system organization and planning. Traditional centralized power plants are being replaced by decentralized structures, making the power flow more complex to predict. As a result, alternative methodologies for power system planning are imminent. This paper introduces a novel approach using a combination of artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) models for forecasting system states. Here, ANN model predicts energy consumption, while the ANFIS model forecasts thermal and hydro power plant production as well as CO₂ emissions. The accuracy of these models results from leveraging the collective expertise of power system planning professionals, utilizing extensive databases containing hourly data from measurements in Serbian power systems. These datasets encompass hourly production data from various energy sources, energy consumption patterns, and relevant environmental parameters (such as temperature, wind speed, and solar irradiation). To underscore the effectiveness of the proposed ANN model, predictions of power consumption from ANN are compared with predictions from ARIMA (autoregressive integrated moving average) model. The developed forecasting models are employed to predict annual and daily energy consumption, seasonal variations in thermal and hydro production, and annual CO₂ emissions. The dependencies between power consumption/production and ambient parameters are visually depicted in three-dimensional representations. Model accuracy is evaluated through graphical, numerical, and error-based analyses across four distinct error metrics. By utilizing historical data and expert insights from previous production scheduling, these models enhance the precision of future production scheduling decisions. This approach minimizes human error, maximizes the utilization of human expertise, and establishes a framework for effectively planning large-scale power systems. The primary contribution of this research lies in the integration of artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) methodologies. This combined approach minimizes the errors inherent in each individual methodology while leveraging their respective advantages. Specifically, the consumption prediction error achieved is 5.64%. When ANFIS is utilized with a training database based on ANN consumption prediction, the prediction error for CO₂ emissions is 1.27%.