Research On Short-term Wind Power Forecasting Method Based On Similar Days And Feature Reduction

Today,wind power generation has become an important measure for countries to promote the adjustment of energy structure,and the wind power itself is random,volatile and intermittent due to the influence of natural wind.When large-scale wind power is connected to the grid,the balance capacity of the traditional power system is destroyed,which brings great challenges to the regulation strategy of the power sector.To ensure the safe and stable operation of the power system,the use of wind power prediction technology is one of the effective measures.Shortterm wind power prediction refers to predicting the power generation of wind turbines within the next 72 hours.Accurate prediction is very important for the stable operation of the power system.With the development of the wind power industry,the data volume of the wind farm is large and the quality is poor,which makes it difficult for the prediction model to be effectively trained.Therefore,this paper has conducted in-depth research on short-term wind power forecasting methods from the perspective of training samples,input features,and forecasting models,and has mainly done the following work:First,correct erroneous data according to empirical rules and low-order polynomial fitting methods.To reduce the redundant information of training data and enhance the relevance of training samples,this paper draws on the principle of "similar days" in power load forecasting to refine the selection of training samples,and establishes a similar day-SVM prediction model.Experimental results show that this method can effectively reduce the training time of the model and improve the accuracy of short-term wind power prediction,which has practical significance.Then,through a holistic analysis of the error sources generated by the similar day-SVM prediction model,it is found that wind power is not only related to common meteorological factors such as wind speed,wind direction,and air temperature,but changes in wind time series cause power prediction The main reason for the large local error,and the SVM method is difficult to effectively deal with time-series information,so the wind change is characterized as the wind speed change rate and wind direction change rate are reflected in the sample.On this basis,the fuzzy rough set method is introduced to reduce the characteristics of various factors affecting wind power,simplifying the input of the SVM model.Finally,for the traditional machine learning modeling method,the coupling relationship between the input data is split,and the continuous effect of the wind speed change on the fan is not reflected.The neural network is easy to improve due to the diversity of connection methods and excitation functions.In this paper,the excitation function of the hidden layer of the BP neural network is changed to a dual Sigmoid function with hysteresis to enhance the neural network's ability to remember historical information,and The improved bird swarm algorithm(BSA)is used to optimize the hysteresis parameters.The experimental results show that the hysteresis feature is introduced into the BP neural network,which makes the response of the neuron correlate with the historical input,effectively reflects the physical process of wind turbine power generation,and the prediction accuracy has been correspondingly improved.