Research On Short-term Power Combination Forecasting Of Wind Farm Based On Improved Deep Belief Network

With the gradual exhaustion of primary energy and the increasingly serious environmental pollution caused by traditional thermal power generation,the effective use of renewable energy has received more attention.As a clean energy with pollution-free,renewable and abundant resources,wind power generation plays an important role in adjusting energy structure and protecting the ecological environment.However,wind energy has the characteristics of randomness and volatility,which makes it difficult to grasp the output of wind farm.Therefore,it is necessary to study the key technology of wind farm power prediction to ensure the stable operation of power system.The main work in this thesis are as follows:Firstly,this paper takes a wind farm in Zhangbei as the research object,analyzes the characteristics of wind power generation,qualitatively discusses the relationship between wind power and wind speed,wind direction,temperature,humidity and air pressure,and determines the input vector of the prediction model.According to the wind farm data and meteorological data provided by the wind farm,the valuable data information is extracted.In order to ensure the validity of the data,the collected data need to be preprocessed.The normalization method is used to process the input vectors to eliminate the influence of dimension,and the weighting method is used to process the input vectors to improve the impact of important factors on wind power,so as to improve the prediction accuracy.Secondly,aiming at the problem that the traditional Deep Belief Network(DBN)takes long time and is easy to fall into local optimum in the training process,this paper analyzes the causes of the problem and finds out the problem.The simulated annealing algorithm is used to optimize the initial parameters and the input feature vectors of DBN structure in order to reduce the number of oscillations and improve the feature extraction ability of the data of the network model in the training process.The adaptive step size algorithm is used to reduce the time needed for the DBN training network model to reconstruct the error and further improve the network model.An improved deep confidence network wind speed prediction model is established and verified.Then,as the low stability and accuracy of the single prediction model,an improved short-term combined forecasting model of DBN wind speed is proposed.The combined forecasting method is an induced ordered weighted averaging algorithm.Before combined forecasting,the redundancy of single model participating in the forecasting is analyzed,and the optimal model is selected to eliminate the redundant model.The prediction accuracy of the optimal model is used as the inducing factor,and sorted in descending order,and the sum of squares function of prediction errors is constructed by predicting values corresponding to each time after sorting,thereby the weight vector is obtianed.Then combining the optimal model according to the weight vector and an improved DBN wind speed short-term combined forecasting model is established to obtain the predicted wind speed.The wind speed power curve is analyzed,the predicted power is obtained,and the error analysis is performed.The results prove that the prediction accuracy and efficiency are improved.Finally,based on the established prediction model,a software for wind farm power prediction is developed by using MATLAB and C#,which has certain practicability.