“…Almeida et al [29] used the meteorological data obtained by NWP as input data, and a probability prediction model based on a quantile regression prediction algorithm was established to study the probability prediction of photovoltaic power generation. Mohammad et al [30] combined the probability distribution theory with the Gaussian mixture method, and the prediction results are consistent with the actual probability distribution of photovoltaic power under different weather conditions.…”
Due to solar radiation and other meteorological factors, photovoltaic (PV) output is intermittent and random. Accurate and reliable photovoltaic power prediction can improve the stability and safety of grid operation. Compared to solar power point prediction, probabilistic prediction methods can provide more information about potential uncertainty. Therefore, this paper first proposes two kinds of photovoltaic output probability prediction models, which are improved sparse Gaussian process regression model (IMSPGP), and improved least squares support vector machine error prediction model (IMLSSVM). In order to make full use of the advantages of the different models, this paper proposes a combined forecasting method with divided-interval and variable weights, which divides one day into four intervals. The models are combined by the optimal combination method in each interval. The simulation results show that IMSPGP and IMLSSVM have better prediction accuracy than the original models, and the combination model obtained by the combination method proposed in this paper further improves the prediction performance.
“…Almeida et al [29] used the meteorological data obtained by NWP as input data, and a probability prediction model based on a quantile regression prediction algorithm was established to study the probability prediction of photovoltaic power generation. Mohammad et al [30] combined the probability distribution theory with the Gaussian mixture method, and the prediction results are consistent with the actual probability distribution of photovoltaic power under different weather conditions.…”
Due to solar radiation and other meteorological factors, photovoltaic (PV) output is intermittent and random. Accurate and reliable photovoltaic power prediction can improve the stability and safety of grid operation. Compared to solar power point prediction, probabilistic prediction methods can provide more information about potential uncertainty. Therefore, this paper first proposes two kinds of photovoltaic output probability prediction models, which are improved sparse Gaussian process regression model (IMSPGP), and improved least squares support vector machine error prediction model (IMLSSVM). In order to make full use of the advantages of the different models, this paper proposes a combined forecasting method with divided-interval and variable weights, which divides one day into four intervals. The models are combined by the optimal combination method in each interval. The simulation results show that IMSPGP and IMLSSVM have better prediction accuracy than the original models, and the combination model obtained by the combination method proposed in this paper further improves the prediction performance.
“…[2][3][4] Different from the traditional thermal power and nuclear power stations, the output power of PV power generations is random and uncontrollable. 5 Deeply mining the historical data is one of the intuition and effective ways to extract the characteristics of the system, 6 and the scientific establishment of data samples is the premise of statistical analysis related to the accuracy and validity of the system. 7 The acquisition granularity (it means the sampling time interval) and sampling span (it means the overall capacity of the sample) of the data are the feature factors that need to be identified first in the data collection or statistical process as the feature data of the evaluation data sample.…”
Section: Discussionmentioning
confidence: 99%
“…The characteristics of the reactive power change of renewable energy power stations have a great influence on the planning and capacity configuration of energy storage power stations . Different from the traditional thermal power and nuclear power stations, the output power of PV power generations is random and uncontrollable . Deeply mining the historical data is one of the intuition and effective ways to extract the characteristics of the system, and the scientific establishment of data samples is the premise of statistical analysis related to the accuracy and validity of the system .…”
Summary
The acquisition granularity (time feature quantity) and sampling span (spatial feature quantity) of the data are the feature factors to analyze the active power of renewable energy power stations. According to the time and space characteristics of photovoltaic (PV) power stations, the acquisition granularity and sampling span calibration methods of PV output power based on data mining technology are proposed this paper. The initial range of the acquisition granularity is determined by analyzing the maximum‐order difference components of the PV output power. Through deeply mining the continuous change state of the PV output power, an acquisition granularity calibration method of the PV output power based on the multiobjective optimization model is proposed from time characteristics. The particle swarm optimization algorithm is used to solve the model to obtain the optimal the acquisition granularity of the PV power station. Through the analysis of the sample information entropy change trend of the PV output power, a sampling span calibration method of the PV output power based on the information entropy theory is proposed from space characteristics. The sensitivity analysis of the acquisition granularity and sampling span of the data to the capacity of energy storage systems is realized by the smooth control of the PV output power using first‐order low filters. The simulation tests of the annual history operating data at a PV power station with the installed capacity of 40 MW in China verify the validity of the provided methods. The simulation results show when the acquisition granularity takes 60 seconds and the sampling span takes 33 days, it can satisfy the accuracy of the required data of energy storage systems to realize the smooth control of the PV output power.
“…The Extreme Learning Machine (ELM) is a fairly popular variant of ANN [11], [12], [13]. Gaussian Process Regression (GPR) [14] and Markov Chain (MC) [15] models are also becoming more frequent in the literature. The global forecasting competition GEFCOM 2014 [16] showed that the most efficient algorithms were often non-parametric, such as Quantile Regression Forests (QRF) [17] and Gradient Boosting (GB) [18].…”
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.