Short-term power load forecasting method based on improved generalised regression neural network

In this paper, a short-term power load forecasting method based on improved generalised regression neural network is proposed. The autocorrelation, timing, and periodicity characteristics of short-term power loads based on time series characteristics are determined, and load data change rules based on changing factors are obtained. The trend of daily and weekly load changes through load data scatter charts is determined, load data at different stages is extracted, and the Spearman correlation coefficient to collect power load data is introduced. A generalised regression neural network architecture is constructed, that is, the number of neurons is determined, input sample short-term power load data, and the number of neurons is kept consistent with the training load. Weights to optimise neuron attributes are introduced, the latest short-term power load data output layer is then constructed, and finally, the timely power load forecasting is achieved. Experimental results show that the proposed method can reduce prediction errors and is feasible. © 2023 Inderscience Enterprises Ltd.