| The high penetration of wind power and photovoltaic power(PV)has become a prominent feature of new power system in China.Intermittency and volatility of wind and PV power bring various uncertainties to the safe,stability,and economic operation of power system.Accurate forecasting of wind/ PV power and uncertainty quantification is an important technique to solve the above issues.The coupling characteristics between power generation and load is variability and complexity in new power system,which make it difficult for traditional forecasting methods to improve substantially in short-term power forecasting for wind and PV stations.Based on the latest state-of-art machine learning and artificial intelligence algorithms,a study on wind,PV power forecasting,and optimal operation strategic for multi-energy complementary system is carried out,focusing on power forecast of a separate wind farm/PV power station---power forecast of regional wind farms/PV power stations---uncertainty quantification for power forecast---optimization operation strategy for multi-energy complementary power system.The main study works are as follows:(1)Based on data-driven and machine learning theory,an adaptive Stacking ensemble learning framework for short-term forecast of wind/PV power is proposed,which can achiever optimal performance self-adaptively with the variation of data to improve forecast accuracy and stability.For a separated wind farm or PV station with complete meteorological data monitoring,considering multi-source heterogeneous characteristics of historical power data and Numerical Weather Prediction(NWP)data,the better base models are adaptively selected according to the fitting evaluation score of the training data based on Stacking ensemble learning theory;For performance differences of various base models on different data sets,linear model is selected as an metamodel to integrate multiple based models,and the weight of each base model is automatically obtained according to the principle of minimum cross-validation error.The proposed framework integrates the advantages of multiple models to adapt to varies in dataset,which is validated for a real-world wind farm and PV station.For all the data sets,the results demonstrate a certain increase in prediction accuracy.In addition,the stability and robust performance of proposed model are analyzed and proved.It is concluded that the proposed Stacking ensemble learning framework based on data driven has ability to improve the accuracy for wind and PV power short-term forecast.(2)A spatio-temporal dynamic graph neural network(AST-DGCN)with attention mechanism is constructed,which can capture the dynamic spatial-temporal correlation of wind /PV station output power in a region,and provide a new ideal for improving forecast accuracy with incomplete or missing meteorological monitoring data.Based on the idea of graph neural network,the topology of graph network represents the spatial dependence among multiple irregularly-arranged wind farms or PV power stations in a region,and an attention mechanism is introduced to explain the dynamic spatial dependence weight among them.The graph neural network module deeply learns and extracts the dynamic information of spatial dependencies.The Gated Recurrent Neural Network(GRU)module deeply extracts the time dependence information of output power for multi-wind farms or multi-PV power stations,and the attention mechanism is used to describe the dependence weight in different time dimensions.The results are generated for a real-world multi-wind farms and PV power stations,which demonstrate the better adaptability and forecast accuracy for the complex and high-dimensional spatiotemporal data sets.The validity of the proposed AST-DGCN model is verified.(3)A novel bidirectional long-short-term memory neural network(Bi LSTM)model with distribution-free is proposed for uncertainty quantification of wind power and PV power,which overcomes the weakness of data description using prior probability distribution,and makes the uncertainty quantification easier to understand and implement in engineering applications.The Bi LSTM model has ability to capture the forward temporal correlation and reverse temporal correlation,which can deeply extract the hidden dependence features from time-series wind or PV power data.Bootstrap error resampling technology improves the diversity and difference of error samples,which can more accurately describe the distribution characteristics of prediction errors.Then,the quantile method can directly output the predicted values of different quantiles,and after that,the upper and lower limits of the prediction interval with specified confidence level can be obtained.A comparative analysis is carried out to validate the effective and reliability.The overall results show that the proposed method can obtain high-quality prediction intervals with narrow average coverage width,which outperform the state-of-art baselines.(4)A novel low-carbon operation model of multi-energy complementary system with wind power,PV power,coal power,and pumped storage is proposed,which take pumped storage as flexible and schedulable resources,and pay respect to the uncertainty of wind and PV output power prediction.For a new power system with high penetration wind and PV renewable energy,how to ensure stable,low-carbon and economic operation is the key issue.Firstly,impacts of meteorological factors on the load,wind power and PV power are detailed analyzed using the Pearson correlation coefficient and mutual information entropy.Secondly,the complementary and consistent indicators are used to analyze the complementarity of wind-PV power and the change consistency of wind-PV-load with different weather scenarios.Then,the effects of conventional hydropower integrated into the power system are discussed.Finally,under the two scenarios of high and low output of wind-PV power,the superiority of the explored model in terms of three single objectives including net load fluctuation,carbon emission and wind-PV consumption,and the economic total objective is demonstrated for different weather conditions,which guides the safe,stability and low carbon economic operation of future new power system. |
