Short-term Forecast Of Photovoltaic Output Based On Similar Days And Improved PSO-DBN

With the continuous consumption of fossil energy and the environmental pollution problems it brings,governments of all countries are paying more and more attention to the development and utilization of clean energy.Photovoltaic power generation has now become a research hotspot in the field of renewable energy.However,the photovoltaic power generation system is affected by the external environment,and the output power sequence is obviously non-linear and intermittent.Accurately predicting photovoltaic output and planning well in advance can effectively reduce the impact of large-scale photovoltaic grid-connection on the large power grid.However,the current photovoltaic output power prediction algorithm usually has problems such as easy to fall into local extreme value and insufficient prediction accuracy.Based on the above problems,this paper combined the deep belief network technology to study the short-term photovoltaic power generation forecast,the main research results are as follows:(1)By analyzing the equivalent circuit and working principle of photovoltaic cells,a simulation model is established,and the output characteristics of photovoltaic cells are analyzed in MATLAB.On the basis of historical power generation data and meteorological data,in-depth analysis of the relevant factors that affect the output power of the photovoltaic power generation system.(2)Aiming at the problem that the existing shallow network prediction model cannot fully extract the deep features in the photovoltaic output power sequence and the prediction error is large,an improved PSO-DBN network prediction model is proposed.The strategy of linearly decreasing weights and the idea of adaptive learning factors are introduced into the PSO algorithm to improve the PSO algorithm.The improved PSO algorithm is used to optimize the initial weights of the DBN network to improve the problem that the initial weights of the DBN network are random and easy to fall into the local optimal solution.(3)Photovoltaic output power is related to many meteorological factors,and the traditional forecasting method of classification and modeling according to weather type is difficult to ensure accurate forecasting results.In order to improve the prediction performance of the model and shorten the model training time,this paper proposes a prediction method based on similar days and an improved PSO-DBN network.First,the original data is divided according to the generalized weather type,and then a variety of factors other than the weather type are integrated Select a set of historical power generation days with high similarity to the predicted day to form a model training sample.Since the output power sequence of a similar day is similar to the prediction day,using its data as a training sample can improve the accuracy of the prediction model.(4)According to the prediction method proposed in this paper,the actual calculation examples are simulated and verified,and compared with the improved PSO-BP model,traditional DBN model,BP neural network model,fully verify the feasibility and effectiveness of the prediction method in this paper.