Wind Power Prediction Based On Improved Vmd Pretreatment And Bidirectional Lstm

As a clean and renewable energy,wind energy itself does not produce pollution and is not limited by energy exhaustion,making it occupy an increasingly important proportion in the world energy field.However,wind power has high randomness and volatility,large-scale wind power grid connection will affect the stability of power system operation.In order to improve the influence of wind power non-stationarity on the stability of power system,the wind power prediction is carried out to assist the power department to formulate reliable scheduling plan,which can effectively improve the safety,economy and stability of power system operation,which has great significance to power system energy reform.Due to the non-stationary characteristics of wind power data,the instability of prediction model and the imperfect construction method of wind farm feature set,the current ultra-short-term wind power prediction accuracy is still difficult to meet the real-time scheduling requirements of wind farms and power grids.In order to improve the prediction accuracy of ultra-short term wind power,the prediction of ultra-short term wind power was improved from three aspects: data preprocessing of wind power,improvement of prediction model and construction of optimal feature set for wind power prediction.In view of the non-stationary influence model construction of wind power data in the process of wind power prediction,the time-frequency domain analysis method is adopted to preprocess the wind power data.Traditional time-frequency domain analysis method,such as empirical mode decomposition and wavelet decomposition method is easy to appear the modal aliasing problems,put forward using empirical mode decomposition and variational mode decomposition method which is hard to appear the modal aliasing problems to develop wind power data preprocessing,the combination of example verification show that the new method improves the pertinence and reliability of wind power pretreatment process.Aiming at the problems of instability,easy to be trapped in the local optimal and easy to forget the historical training state in the traditional prediction model,the Bi-directional Long Short-Term Memory network method optimized by Bayes Optimization was adopted,and the ultra-short-term wind power prediction model was built based on the pre-processing wind power data.The new method realizes the long-term memory of wind power sequence by means of memory module.By virtue of stable Bi-directional structure and strong feature learning method,the probability of falling into local optimal condition is effectively reduced,and the reliability and advancement of the new method are verified by experiment.Wind power is closely related to local meteorological factors.Improving meteorological feature set can enhance the pertinence of models and effectively improve the prediction accuracy.New method according to the formula of wind energy capture,based on the history of pretreatment of the wind power features,introduces the wind speed,temperature,humidity,air pressure and air density of 10 kinds of wind power related meteorological features,and the statistics of the extreme value of each region,poor,mean,variance and average 5 kinds of statistical features,constructed a relatively perfect wind power forecast feature set.Aiming at the problem of model accuracy and efficiency decline caused by information redundancy among meteorological features,a low-redundancy feature selection process was adopted to screen the input features,and the optimal prediction model was obtained on the premise of ensuring the accuracy of the prediction model.In the process of feature selection,the correlation between the input features and the pre-treatment wind power sequence is analyzed on a daily basis using the conditional mutual information method,and the feature importance ranking based on the conditional mutual information is obtained,and the candidate feature subset is constructed to reduce the redundancy between features.Finally,the new method is proved to be effective and advanced by using the optimal prediction model based on Bi-directional Long Short-Term Memory network to carry out experiments based on the measured wind power data.