Cloud movement makes short-term forecasting of solar photovoltaic (PV) panel output challenging. A better PV forecast can realize value for both grid operators and commercial or industrial customers with solar assets. In this study, we build convolutional neural network (CNN) based models to forecast power output from PV panels 15 min into the future. Model inputs are the PV power output history and ground-based sky images for the past 15 min. The key challenge is ensuring that due importance is given to each type of input. We systematically explore 28 methods of “fusing” these heterogeneous inputs in our CNN. These methods of fusion (MoF) belong to 4 families. We also systematically explore the many hyperparameters related to model training and tuning. Limited resources preclude an exhaustive search. We apply a three-stage “funnel” approach instead, wherein we narrow our search to the most promising one of these 28 MoF. We find that a two-step autoregression-CNN MoF has the best performance followed closely by a “mix-in” MoF that performs feature expansion and reduction to give appropriate importance to the two types of inputs. The two-step autoregression-CNN model has a forecast skill (FS) of 17.1% relative to smart persistence on the test set comprising 20 complete days (9 sunny, FS = 22%; 11 cloudy, FS = 16.9%). This optimization results in the improvement of FS from 14.1% for a previously published nonoptimized “baseline” model, a CNN wherein the PV history was simply concatenated to the end of the image-sourced vector obtained after convolution, pooling, and flattening operations.
Accurately forecasting wind and solar power output poses challenges for deeply decarbonized electricity systems. Grid operators must commit resources to provide reserves to ensure reliable operations in the face of forecast errors, a process which can increase fuel consumption and emissions. We apply neural network-based machine learning to expand the usefulness of median point forecast data by creating probabilistic distributions of short-term uncertainty in demand, wind, and solar forecasts that adapt to prevailing grid conditions. Machine learning derived estimates of forecast errors compare favorably to estimates based on incumbent methods. Reserves derived from machine learning are usually smaller than values derived using incumbent methods, which enables fuel savings during most hours. Machine learning reserves are generally larger than incumbent reserves during times of higher forecast error, potentially improving system reliability. Performance is tested using multi-stage production simulation modeling of the California Independent System Operator (CAISO) system. Machine learning reserves provide production cost and greenhouse gas (GHG) emission reductions of approximately 0.3% relative to historical 2019 requirements. Savings in the 2030 timeframe are highly dependent on battery storage capacity. At lower levels of battery capacity, savings of 0.4% from machine learning reserves are shown. Significant quantities of battery storage are expected to be added to meet California's resource adequacy needs and GHG reduction targets. Addition of these batteries saturate reserve needs and results in minimal within-hour balancing costs in 2030.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.