Wind-Power Intra-day Statistical Predictions Using Sum PDE Models of Polynomial Networks Combining the PDE Decomposition with Operational Calculus Transforms

摘要

Chaotic processes in complex atmospheric circulation and fluctuation waves in local conditions cause difficulties in wind power prediction. Physical models of Numerical Weather Prediction (NWP) systems produce only coarse 24-48-h prognoses of wind speed, which are not entirely assimilated to local specifics and usually delayed to be produced every 6-h. Artificial Intelligence (AI) techniques can process daily forecasts or calculate independent statistical predictions using historical time-series in a few-hour horizon. The presented unconventional neuro-computing method elicits Polynomial Neural Network (PNN) structures to decompose the n-variable Partial Differential Equation (PDE), into a set of node-converted sub-PDEs. The inverse Laplace transformation is applied to the node produced rational terms, using Operational Calculus (OC), to obtain the originals of unknown node functions. The complete composite PDE model includes the sum of selected sub-PDE solutions, which allow detail representation of complex weather patterns. Self-adapting statistical models are developed using a specific increased inputs->-output time-shift to represent the current local near-ground conditions for predictions in the trained time-horizon of 1-12 h. The presented multi-step procedure forming statistical AI models allow more accurate intra-day wind power predictions than processed middle-scale numerical forecasts.