Research On Photovoltaic Power Prediction Method Based On Chaotic Time Series

With the increasing depletion of fossil energy and people's increasing attention to environmental pollution,the development and utilization of photovoltaic(PV)power generation has become a common solution to the problem of energy and environmental constraints in countries around the world.PV power generation systems have become an important part of power systems.However,PV power is strongly random,fluctuating,and intermittent due to various factors such as weather,clouds,and temperature,which will cause serious impact on the safe and stable operation of large power grid.An effective PV power prediction method can reduce the operating cost of the power grid and protect the safe and stable operation of the power grid.Power workers can also formulate a reasonable economic dispatch plan based on this.In view of the human subjectivity of traditional PV power prediction methods based on meteorological data,this thesis starts from digging deeply into the dynamic behavior of PV power time series,and gives the research direction of PV power prediction methods based on chaotic time series,namely PV power prediction based on univariate time series and PV power prediction based on multivariate time series.Firstly,the definition and basic concepts of chaos are reviewed,and the phase space reconstruction theory,the specific solution methods of the two key parameters of delay time and embedding dimension are clarified.The correlation dimension method,the largest Lyapunov exponent method,and the recurrent plot method for determining chaotic time series are introduced.The data of actual PV power plants is used to verify that the PV power time series has chaotic characteristics.Secondly,in the chaos theory,it is often difficult to obtain the best embedding dimension of the time series with the existing embedding dimension calculation methods.The embedding dimensions obtained by different calculation methods are slightly different,and the prediction results of the corresponding PV power are also different.In order to reduce the impact of the embedding dimension on the prediction results,a PV power adaptive prediction method based on multi-embedded dimension Volterra filters is studied.Taking the actual PV power time series as the research object,the mutual information method and Cao's method are used to determine the delay time and the embedding dimension respectively.A Volterra combination prediction model based on multiple embedding dimensions is constructed.The combined model adopts neural network to combine the Volterra single models under each embedding dimension.Simulation results verify the feasibility and effectiveness of the proposed method.Then,in order to avoid the neural network falling into local optimization effectively,a global PV power prediction method based on beetle swarm optimization(BSO)-Elman neural network is studied.Based on analyzing the characteristics of particle swarm optimization(PSO)and beetle antennae search(BAS)algorithms,the BSO algorithm combining PSO and BAS is applied to the prediction of Elman neural network.After reconstructing the phase space of the PV power time series,BSO performs the first iteration to find the global optimal solution and uses it as the optimal initial weight of the Elman neural network.Based on this,the Elman neural network performs a second iteration to complete the training and predicts the PV power.Taking the actual PV power plant data as an example,the prediction results and error comparison analysis verify that the proposed method is more adaptable.robust and stable.Finally,in view of the shortcomings of the univariate prediction method of PV power time series,a global PV power prediction method based on multivariable phase space reconstruction and radial basis function(RBF)neural network is studied.Based on the correlation analysis,the historical PV power and meteorological factor time series of the actual PV power plant are selected to form a multivariate time series.Then,the C-C method and the false nearest neighbors(FNN)method are used to reconstruct the multivariable phase space of the PV power prediction,and its chaotic characteristics is identified by small data method.Based on this,combined with the strong nonlinear fitting ability of neural network,a global PV power prediction model based on multivariate phase space reconstruction and RBF neural network is established.The analysis of example shows that compared with the univariate prediction method,the proposed multivariable phase space reconstruction prediction method has better performance.