A Novel Fuzzy Time Series Markov Chain Model for Forecasting Demand Flexibility Services (DFS) Power Requirements Based on KMeans Clustering

Modern energy systems necessitated precise short-term forecasting of power requirements to ensure the efficient operation of Demand Flexibility Services (DFS). This work introduced an innovative hybrid forecasting model that combined KMeans clustering with Fuzzy Time Series (FTS) and Markov Chain methodologies. The suggested technique used KMeans to create adaptive fuzzy intervals from normalised historical data, in contrast to classic fuzzy models that relied on fixed partitions. The intervals facilitated the construction of fuzzy logical connections (FLRs) and a state transition matrix, so allowing dynamic forecasting of DFS power demand. The model was assessed with actual DFS data collected at 30-minute intervals. Forecasts were produced using a defuzzification technique informed by Markov transition probabilities. Experimental findings demonstrated a significant correlation between anticipated and actual values, yielding a Mean Absolute Error (MAE) of 74.60 MW and a Root Mean Square Error (RMSE) of 86.04 MW. The results demonstrated that the model accurately represented temporal demand patterns and maintained robustness across different load levels. The technique shown considerable promise for practical use in smart grid forecasting systems. Its simplicity, interpretability, and flexibility made it an invaluable instrument for real-time energy management and decision-making. Keywords: Demand Flexibility Services (DFS), Fuzzy Time Series (FTS), KMeans Clustering, Markov Chain Forecasting, Short-Term Load Predictions