The Research Of Interval Prediction For Wind Power Based On LSTM Network

The rational development of wind energy is of great significance in improving China's energy structure and alleviating energy crisis.As a major method of developing wind power,wind-power generation is affected by impact of the volatility and intermittence of wind speed,which causes a series of problems such as difficulty in connecting to the grid and accommodating wind power.Actively promoting research of wind-power prediction will give assistance to real-time adjustment of dispatch plans for power grid,and effectively reduce the impact on security and stability of power system economic operation,which can increase the overall economic benefits of wind power.However,the volatility and complexity of the wind power data increases the difficulties of wind-power prediction to some extent.This paper starts with the research of interval prediction for wind power,and builds a model for interval prediction based on Long Short Term Memory(LSTM)networks,optimization techniques of deep learning,variational modal decomposition(VMD)algorithm,and proposes two optimization methods based on the construction of loss function and prediction interval,to achieve the effective training of the LSTM prediction model;Then,in order to improve the effect of LSTM prediction model,and combined with the ability of the VMD algorithm to decompose and simplify the sequence data,an LSTMVMD combination interval prediction model is constructed to achieve the improvement of prediction effect.The main work of the paper can be divided into the following three parts:(1)The wind-power data shows strong randomness and volatility.In order to achieve the improvement of prediction results,this paper proposes a wind-power prediction model based on LSTM.Based on the structural advantage of LSTM to process temporal logic,the paper uses LSTM to extract the basic feature of wind-power data.Combined with deep learning related technologies,the deep network completes the further feature extraction and outputs the upper and lower bound.Finally the paper achieves a more efficient interval prediction model with stronger nonlinear mapping capability than traditional prediction models based on shallow machine learning.(2)The prediction interval needs to ensure the balance between reliability and clarity.Based on the deep learning optimization technology and the thought of stochastic gradient descent optimization,this paper designs two training methods for the interval prediction model: First,the double loss functions are constructed for synchronously optimizing both sides of coverage and width of prediction interval,then model use back-propagation algorithm to achieve the update of the weight.The second is to construct a interval by the historical data of wind power,and design an adaptive adjustment strategy for the width of interval.In essense,it is a way to directly build training tags needed by model.Finally,the experimental comparison shows that the prediction effect of the LSTM interval prediction model is much better than the traditional LUBE model.(3)In order to further improve the prediction effect of LSTM model,this paper proposes a combination interval prediction model LSTM-VMD from the perspective of simplifying complexity of data.The VMD algorithm is used to simplify and decompose wind-power data into several smoother sequence components.Based on this,the LSTM prediction model is constructed separately,then total prediction interval is obtained by the summation of prediction interval of each component.Finally,through the analysis of the experimental results,it is proved that the LSTM-VMD combination prediction model shows a more effective performance than the basic LSTM prediction model.