Single-Step and Multi-Step Forecasting of Global Horizontal Irradiance Using Long Short-Term Memory Network /

Abstract

Renewable resources are very effective to meet energy needs due to growing cost of fuel and decline in fossil fuel holds. This shifts the focus on providing affordable and authentic sources of electricity. For incorporating solar energy into power system, there are two challenges: First is accurate forecasting of solar irradiance and second is solving economic dispatch issue. This thesis proposes a novel technique which is from RNN class such as LSTM for optimum day ahead scheduling of three conventional generators and solar photovoltaic power generation as a cost minimization problem. Solar irradiance forecasting is a non-linear time series problem and LSTM has given fine results in solving non-linear time series problems. Multivariate single-step and multi-step GHI forecasting models have been established in this work depending upon meteorological variables and historical data of GHI. Hyper-parameter tuning is being employed for achieving optimal parameters and in this way best model is evolved for solar irradiance forecasting. The simulations for GHI prediction are being performed on dataset gathered in Meteorological Station of NUST Islamabad, Pakistan. ANN has the ability of rapid convergence near the solution so it is very good AI technique for solving power system optimization problems. Economic dispatch problem for three conventional generators is solved by lambda iteration method and LSTM technique. Training and testing patterns of powers, total costs and power transmission losses for LSTM are obtained by using lambda iteration method, dataset of load for this research is taken from NUST grid. Univariate multi-step LSTM models by doing hyper-parameter tuning are developed for doing optimal scheduling of generators with and without solar energy incorporation into the power system. By comparing results for one day, it is seen that solar energy integration results in saving of operating cost and reduction of transmission power loss. Simulations shows the effectiveness of LSTM in prediction of GHI and solving economic dispatch problem by comparing results on the basis of different evaluation metrics.