Research On The Day-ahead Wind Power Prediction And The Detection Of Power Ramp Event Considering Transitional Weather

Wind power plays an important role in realizing the "carbon peak and neutrality targets".As a kind of efficient and green renewable energy,wind power energy has been developing rapidly in an unprecedented way under the background of global Internet.However,wind power has strong randomness and volatility,and the integration of large-scale wind power into the power grid will affect the safety and stability of the power grid operation.Especially in extreme or transitional weather conditions,the output of wind farm stations will change dramatically in a short time,which makes it difficult for the power system to maintain the real-time balance of power supply and load,and makes the power grid face great challenges.Therefore,it is of great significance to carry out research on high-precision wind power prediction and power ramp event detection technology for improving wind power consumption capacity and safe,stable and economic operation of power system.In this context,researches on the accuracy improvement of wind power prediction models and the detection of power ramp events detection are constantly emerging.At present,there are mainly the following problems in wind power prediction and power ramp events detection under transitional weather conditions.First,the research on wind power prediction considering transitional weather mainly focuses on model optimization and combination.At this time,the input characteristic variables of the model are fixed,for example,the whole forecasting process only includes wind speed or wind direction,etc.However,when the weather conditions change,the leading sensitive meteorological factors will also change,so the fixed input characteristic variables will limit the improvement of prediction accuracy.Second,the power and meteorological data sets under a transitional weather process belong to small samples.However,the deep learning algorithm widely used in wind power forecasting relies too much on historical data and cannot give full play to its advantages through a large number of training,resulting in a large prediction deviation.Third,the existing wind power ramp event detection algorithm fails to consider the merging of adjacent trend segments and the accuracy of end points at the same time.The adjacent trend segments in the same direction,bump events and unreasonable ramp points destroy the integrity of climbing events,leading to the reduction of detection accuracy.Aiming at the above problems,in order to weaken the influence of transitional weather on wind power prediction and power system,and further improve the accuracy of wind power prediction and power ramp event detection,this paper proposes a day-ahead wind power forecasting based on multi-scene sensitive meteorological factor optimization and time-series generative adversarial network and a detection algorithm based on improved spinning door transformation:(1)In terms of model input feature selection,the generation mechanism of wind power prediction error under two kinds of transitional weather conditions is firstly analyzed,and the correlation between meteorological elements and wind power output under three different scenarios of two types of transitional weather is analyzed based on Pearson correlation coefficient method,and the input characteristics of the prediction model in each scenario are determined.(2)In the aspect of day-ahead wind power prediction considering transitional weather,a power forecasting method based on multi-scene sensitive meteorological factor optimization and time-series generative adversarial network are proposed.This method makes full use of the advantages of time-series generative adversarial network considering sequence autocorrelation,and realizes the expansion of various scenarios sensitive meteorological factors and measured power sample sets.In the model building stage,the expanded sample is directly taken as the training data,and the sensitive meteorological factors under various scenarios are taken as the input.Under each scenario,the combined prediction model combining the advantages of three different prediction algorithms,namely,long short-term memory,extreme gradient lifting and back-propagation neural network,is constructed respectively.The overall prediction accuracy of the model is improved by fully integrating the advantages of each method.Then particle swarm optimization algorithm was used to optimize the model parameters and weights to further improve the prediction accuracy of the combined model.(3)In terms of power ramp event detection,a wind power ramp detection algorithm based on the improved spinning door transformation is proposed.Firstly,the traditional spinning door transformation algorithm is used to extract the trend of the original power data,then the adjacent ramp sections in the same direction are merged,and the trend marking method is used to solve the fracture problem caused by bump events between the two ramp sections.Finally,the endpoint of ramp events is corrected by finding local extreme value points,so as to further improve the detection effect of the traditional revolving gate algorithm.Based on the data of a wind farm in Jilin Province,the paper carries out an example analysis,and selects different evaluation indexes to evaluate the prediction and detection results of the proposed model and the benchmark model.The experimental results verify the effectiveness of the proposed method,and can improve the prediction accuracy and detection accuracy to a certain extent.