Hybrid Approach for Short-term Wind Power Prediction to Increase Flexibility of Renewable Integration in Power System

Renewable energy (RE) is rapidly increasing as an alternative to conventional power systems. One of the examples is the power system of Krabi province in the south of Thailand, whose 2026 projected energy consumption, is expected to be met almost entirely by RE sources. Generation from these volatile renewable sources largely depends on the weather, which causes problems on the transmission, e.g., the excess power flow reversing or backfeeding, and the 1ack of production or shortage. These problems cause particular difficulty in the wind power prediction of the power operation system in planning the future schedule. Consequently, inaccuracy of generating future scheduling diminishes the flexibility of the power system. In this paper, we describe a system to mitigate the problems by proposing a hybrid model of wind power prediction combining long short-term memory (LSTM) and Markov chain (MC) with quantization (QTZ) and denoising (DN) techniques. The experimental results show that the hybrid model increases the accuracy of the wind power prediction.