Probabilistic forecasting of ultra-short-term PV output using the improved GWO and TCN-QRF

Abstract

As the share of photovoltaic （PV） power generation grows increasingly within electric power systems， the accurate probabilistic forecasting of PV output can be help for grid regulation and operation. To enhance forecasting precision， a probabilistic forecasting method for ultra-short-term photovoltaic output using the improved grey wolf optimization （GWO）， temporal convolutional neural networks and quantile random forests （TCN-QRF） is proposed. Firstly， the preprocessed time series dataset for PV output is converted into a supervised learning dataset. Then， the TCN is used to extract the temporal features of PV output as the input to the QRF， constructing the TCN-QRF model. Finally， the GWO is improved using the nonlinear convergence factor and Gaussian mutation strategy. The improved GWO efficiently selects hyperparameters for TCN-QRF， enabling a more precise probabilistic forecasting of PV output.

Keywords

• improved gwo
• probabilistic forecasting
• pv output
• tcn-qrf