Characterizing and Mitigating the Impact of Solar Forecast Errors on Grid Planning and Operations
In recent years, the contribution of photovoltaic (PV) power production to the electric grid has been increasing. Still, a number of challenges remain for a reliable and efficient integration of solar energy. While conventional electric power generated by gas turbines can be adjusted to follow the grid load, the stochastic nature of solar radiation makes it difficult to control the PV output, which hinders its integration in the grid. Accurate solar forecasts help grid operators integrate solar energy by enhancing power quality and reducing grid operation costs. Following the development of sky imager hardware and algorithms at UC San Diego, we present a variety of models and methodologies to reduce sky imager forecasts errors by improving the accuracy of meteorological parameters, compensating the power mismatch caused by solar forecasts errors, and mitigating the impact of solar forecast errors on real world grid planning and operations.
First, a low-cost instrument for measuring local cloud motion vectors (CMVs) was developed. Three algorithms for estimating local cloud base height (CBH) using a single sky imager paired with either distributed ground irradiance sensors or measured CMVs were then designed and tested. Since sky imager forecasts are often used in conjunction with other instruments for measuring CBH, cloud velocity, and/or solar irradiance measurements, our approaches decrease instrumentation costs and logistical complexity. More importantly, through these algorithms, local measurements improve sky imager forecasts by adding information that is unobservable from a single sky imager.
Second, integrating battery systems into a PV plant can compensate the power imbalance caused by solar forecast errors. Battery system size can be optimized by determining the energy reserve required to offset the possible maximum power ramp. Because passing cloud shadows are the main cause of the power ramps, a simple model based on physics variables that are available globally can determine the worst power ramp rates. Local CMV measurements enable even more accurate maximum ramp rate estimates. The key merit of the method is that it is universally applicable in the absence of high frequency measurements.
Finally, issues when integrating imperfect solar forecasts in grid operations are evaluated. Both physics-based forecasts and Numerical Weather Prediction (NWP) based machine learning forecasts that are commonly utilized in the grid operations exhibit autocorrelated forecast errors. First, a deterministic valley-filling problem through EV charging is formulated to investigate how the autocorrelated forecast errors increase peak demand and cause grid net load variability. Then a corrective optimization framework is proposed to minimize the deviation of the realistic valley filling solutions from the ideal solutions. In addition, with the goal of operational deployment, stochastic programming incorporating real time updates of solar forecast and EV charge events to address real-world uncertainty is employed. The optimal valley filling problem is solved in an innovative way and executed under a predictive control scheme in the presence of autocorrelated forecast errors. The proposed corrective stochastic optimization framework successfully mitigates the impact of autocorrelated forecasts errors on grid operations.