Abstract
To address the uncertainty in photovoltaic (PV) outputs for day-ahead home energy scheduling in hourly timescale, a novel stochastic optimization strategy based data-driven method is proposed. Based on available historical PV outputs, the Gaussian mixture model (GMM) algorithm combined with improved prediction strength of clustering method is applied to establish the forecasted probabilistic PV outputs model. Based on the seven-step approximation model of Gaussian distribution, only PV outputs with larger probability level at each hour are used to generate scenarios. Then the typical scenario set can be constructed by scenario reduction method. By finding the solution in typical scenario set using mixed-integer nonlinear programming (MINLP), the scheduling strategy will be closer to real cases. Test results verify the effectiveness of proposed probabilistic PV output model and solution method.
Highlights
The day-ahead home energy scheduling in hourly timescale makes great contribution to reducing residential energy cost [1], [2]
A variety of optimization methods have been proposed. They can be roughly classified into three types: deterministic optimization methods, stochastic optimization methods, and robust optimization methods
Since the stochastic optimization methods [12]–[14], [16]–[21] can address uncertainty in PV outputs rather than conservation as in robust optimization methods, they have been widely applied to the day-ahead home energy scheduling
Summary
The day-ahead home energy scheduling in hourly timescale makes great contribution to reducing residential energy cost [1], [2]. When the number of initial scenarios is enormous, the reserved scenarios can follow specified PV probability distribution How many it is cannot be determined for different cases. [21] employs a two-point estimate method to model the uncertainty in PV outputs, which only generate larger probability level scenarios. To construct approximately PDFs of PV outputs more precisely, the multi-model probability distribution is constructed by employing GMM algorithm combined with improved prediction strength of clustering method. 2. Based on the seven-step approximation model of Gaussian distribution, only PV outputs with larger probability level at each hour are used to generate scenarios. Too many smaller probability level scenarios may be generated in MCS when the number of scenarios is not larger enough This make the scheduling scheme not meet the real cases.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
