Research On Early Warning Of Wind Power Ramp Events Based On Deep Learning Technology

The increased capacity of wind power brings great economic and environmental benefits,and also has a great impact on the security and stability of power system.The large-scale centralized distribution of wind farms makes the wind power output show a strong correlation within a period of time,and there may be a large change in the output of wind power within a short period of time.If the reserve and regulating capacity of the power grid is insufficient to balance this change,the frequency of the power grid may reduce due to the large power shortage,and even a large amount of load is lost.Wind power output has strong fluctuations and uncertainties.Errors in wind power prediction will further increase the difficulty of control.The system needs more reserve capacity to deal with the adverse effects of wind power changes,otherwise,it may cause system frequency out of bound and large load loss.Therefore,in-depth research on the rapid and accurate prediction,assessment and early warning of wind power ramp events is of great significance to ensure the security and stability of power system.From the perspective of ensuring the security and stability of the power system,this thesis used artificial intelligence technology such as deep learning to predict,evaluate and warn wind power ramp events.Firstly,a multi-level feature extraction method for wind power ramp events prediction based on stacked denoising autoencoder(SDAE)was proposed.The K-means clustering algorithm was used to divide the samples into different categories,which facilitated the classification of samples with higher similarity into one category.For different categories,SDAE was used for feature extraction,and the features of each hidden layer were used as the input of the prediction model to achieve fast and accurate prediction of wind power ramp events.The case analysis results of Shandong power grid showed that the use of SDAE could improve the accuracy of wind power ramp events prediction to a certain extent,and reduced the rate of missed alarms and false alarms.Secondly,a wind power ramp events prediction method based on support vector machine(SVM)was proposed,which comprehensively considered the regulation capacity of conventional generating units and pumped storage power stations,and used the hidden layer features extracted by SDAE as inputs to train a classification model SVM.Use wind power,load historical data,real-time data,and forecast data as the input of the SVM and whether a wind power ramp events occured as an output.The SVM predicted whether a wind power ramp event occurs quickly.Based on the operation data of Shandong Power Grid,the comparison of different methods showed that the proposed method had higher accuracy,lower false alarm rate and missing alarm rate.Finally,taking into account the uncertainty of wind power output and certain errors in load prediction,further analysis was made on the scenario where the wind power ramp event may occur.Use Monte Carlo simulation to generate multiple operating scenarios and calculate the power imbalance of each scenario.Based on the utility theory,a severity function was established to classify the severity of a wind power ramp event,which visually showed its severity grade.Analyze the effectiveness of severity classification and assessment.The simulation results showed that the proposed method could accurately classify and warn the severity of wind power ramp events.