A Probabilistic Density Prediction Method for Power Plant Production Data Based on QR-GRU

The article proposes a probability density prediction method for forecasting time series production data in power plants, which is based on Gated Recurrent Neural Network Quantile Regression (QR-GRU). The method utilizes information such as the peak load, power generation, plant power consumption, and grid-connected power, and constructs a sliding time window as input to QR-GRU. It predicts future results at different quantiles and obtains the probability density distribution of each data variable per month using kernel density estimation. Experimental results demonstrate that the QR-GRU combined with kernel density estimation can effectively address the problem of probability density prediction for various types of generated data in power plants.