Stochastic Unit Commitment in Various System Sizes under High Uncertainty Photovoltaic Forecast

Abstract

This paper proposes a stochastic unit commitment (SUC) approach to solve a day-ahead unit commitment (UC) problem in a system with high uncertainty net load which is caused by photovoltaic (PV) power plants. In contrast with robust unit commitment (RUC) which only considers the worst-case scenario, SUC considers every possible scenario with its probability. Multiple possible PV curves were obtained using k-means clustering on historical data. The proportion of cluster members was used as a weight factor representing the occurrence probability of PV curves. The test was separated into two-step tests, namely day-ahead and real-time markets, using IEEE 10 generating unit system and solved using CPLEX. The results showed that in a day-ahead UC, SUC ($539,896) had lower cost than RUC ($548,005). However, when the total energy generated was considered, the SUC (20.78 $/MWh) cost higher compared to RUC (20.75 $/MWh). It is because the solution proposed by SUC is as robust as the RUC, but the generation cost formulation also considers over-commitment. Thus, SUC produced a fairer price for the independent power producer and electric utility in the day-ahead calculation. The results also showed that in the test environment of the real-time market, SUC was able to produce a robust solution without going into over-commitment. It is clearly shown in a 30 units system test with 10 centroids, in which SUC had a cheaper solution (20.7253 $/MWh) compared to RUC (20.7285 $/MWh), without violating power balance or going to load shedding.

Keywords

• intermittency
• k-means
• mixed-integer linear programming
• stochastic unit commitment