Estimation of Solar Power by Combining Ground Measured and Satellite Data Using Artificial Neural Networks

The inherent intermittency of solar energy requires better understanding of its characteristics for the planning and operation of power systems. Solar energy measurements are normally collected at the ground level or using satellite data. Ground measurements are accurate but with limited coverage, in terms of location and covering period. Satellite data is usually available for any location and with decades of records but lacking the accuracy needed. To merge and align these two sources of solar energy measurements is rapidly becoming part of the chores for system planners and operators. To tackle this issue, this paper proposes a series of multi-layered backpropagation neural network (MLBPNN) to estimate historic or future ground measured data from overlapping existing satellite and ground measurements. The study successfully demonstrates that the proposed method can indeed achieve accurate projection. We also demonstrate the sensitivity of the mismatch between satellite data and ground measurements is highly depending on geographical and seasonal factor. Finally, we present a summary of analysing the mismatch comparison amongst different locations in Asia Pacific and discuss various applications using such method in system operation and planning.