| With the development of the economy,China’s demand for energy is increasing,while fossil energy is increasingly depleted,making the issue of energy supply and demand increasingly serious.Therefore,it is imperative to vigorously develop renewable energy.Currently,wind power has become one of the most mature renewable energy power generation methods.However,due to various factors such as wind speed and atmospheric pressure,wind power output has randomness,which poses challenges to grid operation and scheduling after grid connection.Therefore,studying wind power prediction technology is of great significance.Accurate wind power prediction results are conducive to formulating reliable scheduling strategies.Interval prediction of wind power can provide uncertainty information about the prediction results,and help to assess the uncertainty and risk factors of wind power grid connection.In this context,this paper focuses on the research of interval prediction of wind power and economic dispatch of power systems based on interval prediction of wind power.The main contents are as follows:Firstly,the output characteristics of wind power and the causes of bad data are analyzed,and three criteria are used to identify and eliminate abnormal data.Different interpolation methods are used based on the degree of missing data,so as to obtain reliable and complete wind power data;Improved complete ensemble empirical mode decomposition with adaptive noise and variational mode decomposition methods are used to decompose the wind power time series twice to alleviate the impact of strong non-stationary characteristics of the wind power time series on prediction accuracy.Considering that there are too many sub components obtained after decomposition,their data processing is cumbersome and requires too long training time,sample entropy is used to analyze their complexity and reconstruct them into trend components,oscillatory components,and random components of wind power as input to the prediction model.Then,in the point prediction stage of wind power,bidirectional long short-term memory neural network that can fully exploit the temporal characteristics of time series in deep learning is used for prediction,and bayesian optimization is used to optimize the network’s parameters to obtain the point prediction values of trend components,oscillation components,and random components;In the stage of constructing prediction intervals for wind power,The mixed kernel density estimation method is used to estimate the error distribution of prediction errors for oscillatory and random components,perform probability interval prediction,and then superimpose the point prediction values to obtain the overall wind power prediction interval results.An example is analyzed to verify the effectiveness of the method.Finally,in view of the uncertainty of wind power forecasting,an economic dispatch model of power system based on wind power interval forecasting is established.The interval forecasting information is included in the dispatch model.In the dispatch model,besides the generation cost of generating units,the cost of pollution treatment and the cost of system reserve capacity are also considered.The artificial bee colony algorithm is improved by formulating the optimal guiding strategy of the bee colony and introducing dynamic probability,and the scheduling model is solved using the algorithm.The feasibility of this method is verified by an example. |
