Probabilistic Forecasting Of Bus Short-term Net Load Considering Renewable Energy Accesss

Bus load forecasting is the basis for realizing the refined management of smart grid dispatching.It is the prerequisite for energy-saving power generation dispatching,safety and stability analysis,dynamic state estimation,and maintenance plan preparation.The analysis and research on bus load forecasting will help to further improve the perception and control of power grid.At the same time,it is also conducive to the implementation and promotion of the power market.With the continuous expansion of the number and scale of renewable energy access in the grid,the traditional bus load has also changed from the original pure bus user load to the difference between bus user load and renewable energy output,that is,the net load of the bus.The inherent randomness and uncertainty of renewable energy which is contained in the bus net load poses a huge challenge to traditional forecasting methods.How to make full use of the historical load data bonus and the physical relationship between bus load and renewable energy output to reasonably modeling and effectively improve the performance of bus net load forecasting has become a significant technical problem to be studied.This dissertation conducts related research on the problem of bus net load forecasting under high renewable energy penetration rate.It mainly includes four parts: the identification and correction of historical abnormal load data,bus short-term load forecasting,renewable energy output short-term forecasting and bus net load hierarchical forecasting.The main work and conclusions of the research are as follows:(1)Aiming at the abnormal data problem of bus load,this thesis proposes an abnormal data correction method based on Affinity Propagation(AP)and Fuzzy C-means clustering(FCM),which flexibly clusters the historical data,identifies the abnormal values in the cluster by 3σ criteria,and corrects them by cluster mean.Two case studies are used to verify the feasibility and accuracy of the proposed algorithm.The time series forecasting model is also used to forecast the load data before and after the correction,which proves that the proposed identification and correction algorithm can effectively improve the performance of load forecasting.(2)Due to the lack of adaptive time series feature screening methods in traditional forecasting models,this thesis proposes a short-term bus load forecasting model based on phase space reconstruction and deep belief network(DBN).Firstly,the phase space reconstruction is used to project the time series into a moving point in the phase space,and then the DBN,which has excellent nonlinear fitting ability,is used to fit the trajectory to realize bus load forecasting.At the same time,cross-validation was applied to optimize the DBN’s structure.Finally,the real bus load data was used to verify the effectiveness and superiority of the proposed model.(3)In view of the inherent randomness of renewable energy and the difficulty of using deterministic forecasting to accurately describe renewable energy output,this thesis takes photovoltaic output forecast as an example,and puts forward a day-ahead probabilistic forecasting model of photovoltaic output based on Quantile Regression Neural Network(QRNN).First,a group of QRNN day-ahead photovoltaic quantile forecasting model with loss function changing according to quantile condition q is trained by using historical data of photovoltaic output and Numerical weather prediction(NWP)data.Then nonparametric probabilistic forecasting of photovoltaic output is formed by integrating the prediction results of each quantile.Finally,two case studies are used to verify the effectiveness,superiority and compatibility under small training samples of the proposed model.(4)Because the existing net load forecasting model fails to make full use of the information redundancy brought by the load big data to improve the forecasting effect.Based on the methods proposed in points(2)and(3),this thesis obtains the original forecast values of bus user load,net load and photovoltaic output.The minimum trace(Min T)hierarchical forecasting algorithm is used to modify the original forecast value of bus net load,and then quantile regression is used to summarize the prediction error distribution to form the final probabilistic forecasting of bus net load.In this thesis,measured data is used to verify the feasibility and superiority of the proposed method.