Data-Driven-Based Maximum Power Point Tracking And Power Forecasting Methods Of Photovoltaic Power Generation Systems

Energy issue is related to the national economy and people’s livelihood,and is the cornerstone of social production development.With the rapid development of social economy,the energy demand is increasing.At the same time,the defects of traditional fossil energy,such as oil and coal,have gradually appeared.Developing renewable energy has become an important energy strategy in China.In recent years,the installed capacity of renewable energy systems,especially photovoltaic power generation systems,has increased explosively.PV module is the key element in photovoltaic power generation system.Modeling PV module is one of the important technologies to evaluate the output performance of photovoltaic power generation system under complex operating conditions.However,the attenuation rate of PV modules is consistent under actual operating conditions.The inconsistency of attenuation rate puts forward higher requirements for PV module modeling technology.Modeling the attenuated PV modules with high precision is an urgent problem to be solved.At the same time,the output of the photovoltaic system has obvious nonlinear characteristics.When the PV array is shaded,its maximum power point will shift significantly.In order to improve the economy of photovoltaic power generation system,the maximum power point tracking technology needs to be optimized.In addition,compared with the traditional power generation system,the output characteristics of photovoltaic power generation system is highly dependent on meteorological parameters,such as irradiance,temperature and relative humidity,etc.When meteorological parameters suddenly change,the photovoltaic output power will fluctuate in a large degree.Therefore,the photovoltaic output has the problems of intermittency,randomness and fluctuation.It is urgent to carry out small-timeresolution photovoltaic power forecasting research to provide accurate reference information for dispatching department of power grid.The output characteristics of photovoltaic power generation is highly nonlinear with irradiance.Based on data-driven method,this paper carried out the researches on optimal operation and power forecasting methods of photovoltaic power generation systems,including PV module modeling,maximum power point tracking,power forecasting of photovoltaic power plant and regional PV output power forecasting.The specific research contents are as follows:（1）Parameter extraction technique of five-parameter PV model based on characteristics of Ⅰ-Ⅴ curve.Firstly,a data set containing multiple Ⅰ-Ⅴ characteristic curves is established by assigning random values to the five unknown parameters.The characteristic parameters of Ⅰ-Ⅴ characteristic curves can then be obtained.The training set of artificial neural network（ANN）is established by taking the characteristic parameters of Ⅰ-Ⅴ curves as the known input and five parameters as the unknown output.The mapping relationship between the characteristic parameters of Ⅰ-Ⅴ curves and five parameters can be established.In order to improve the prediction accuracy of ANN as well as decrease computation burden,Pearson correlation coefficients between five parameters and characteristic parameters drawn from Ⅰ-Ⅴ curves are calculated respectively.The characteristic parameters with higher Pearson correlation coefficient will be implemented as input parameters of ANN to extract the unknown parameters.Finally,the accuracy of the proposed method is verified by simulations and experiments.The proposed method can realize off-line analysis of the output characteristics and data collection of PV modules as well as PV array under different operating conditions,which could support the data set establishment to realize data-driven based maximum power point tacking and power forecasting method.（2）Global maximum power point tracking（GMPPT）using characteristics mapping method.Firstly,the shape of P-V curve is described based on the characteristic quantities under several characteristic points.Through off-line modeling,the P-V curves of various PV arrays under different shading conditions are collected and analyzed.The selection rules of characteristic points as well as characteristic quantities are determined based on statistical analysis.Then,the mapping relationship between the shape of P-V curve and the global maximum power point is established based on ANN,which could realize the fast locating of the global maximum power point by scanning several characteristic points in P-V curve.Finally,the high-precision global maximum power point tracking is realized by implementing the adaptive perturbation and observation（P＆O）algorithm.The high tracking speed and accuracy of the proposed GMPPT method are verified by simulations and experiments.（3）Inverter-data-driven second-level power forecasting of PV plant.Firstly,the structure and initial parameters of artificial neural network are optimized based on particle swarm optimization（PSO）algorithm and statistical method,which aims to improve the convergence speed and prediction accuracy of the model.Then,the shading condition of PV array is expressed by the irradiance matrix.By reducing the dimension of the irradiance matrix,the problem of low converging speed caused by highdimension input parameters of ANN is solved.In addition,the mapping relationship between the shading condition of PV array and its output power is established based on ANN.The shading condition of PV array is reversely deduced by using the output power of inverters.The virtual cloud image is introduced to accurately describe the shape,thickness,moving direction and moving speed of cloud based on the output power of inverters as well as the position of PV arrays.Finally,the second-level power forecasting of PV plant is realized according to the layout structure of PV plant,and the forecasting accuracy of the proposed method is verified by simulations and experiments.（4）Data-driven minute-level regional PV output power forecasting.Firstly,data cleaning of the collected historical data of PV plants is carried out based on Lagrange interpolation formula,which could solve the problem of the lack and inaccuracy of historical output data.Secondly,the correlation between the output power of each PV plant and the regional PV output power is analyzed.The statistical scaling up method is implemented to carry out regional PV power forecasting,which could significantly reduce the required amount of data.Then,the selection rules of reference PV plant in statistical scaling up method are optimized based on ANN,which could effectively improve the regional power forecasting accuracy and solve the problem of forecasting accuracy decreasing caused by the output fluctuation of reference PV plant.In addition,by integrating the forecasting results of ANN models in different time resolutions,accurate regional output power rolling forecasting is realized.Finally,the forecasting accuracy of the proposed method is verified based on the measured historical data.Aiming at the shortcomings of the existing researches,this thesis has carried out relevant research works in PV module modeling,maximum power point tracking,power forecasting of photovoltaic power plant and regional PV output power forecasting,which has laid a solid technological foundation for high-proportion photovoltaic system accessing to the power grid and could guide the large-scale application of grid-connected photovoltaic power generation system to a certain extent.