Short Interval Solar Power Prediction for Energy Harvesting with Low Computation Cost on Edge Computation Network

Solar energy forecasting methods including artificial neural network (ANN) are mostly for planning large or medium scale grid connected photovoltaic (PV) systems. The small prediction step will cause expensive computational cost with ANN. Using the concept of edge computing (EC), the paper proposes an effective short time step PV energy predicting model I based on ANN and wavelet transformation to reduce the computation costs. Sparse coefficients, which are tried by both different wavelet types and similar day classification, and a few forecast meteorological data are used as inputs to reduces the scale of ANN sharply. The simulation results demonstrate that, when the proposed method processes the whole year data, the proposed ANN model I has low computation complexity and superior precision.