Research On Intelligent Forecasting Of PV Generation And Load And Optimization Of Energy Dispatch In Microgrid

Due to the high cost of the electricity network and the difficulty of maintenance,traditional centralized power generation technology is gradually unable to meet the needs of regional distributed and diversified development.As a distributed power supply technology,microgrid can efficiently integrate a variety of clean energy sources,including solar energy,fuel cells,et.al,and can be applied flexibly according to local conditions.However,affected by the external environment and weather factors,the uncertain changes of renewable energy such as wind,light and inelastic loads have brought huge challenges to the optimization of energy dispatching in microgrids.In the end,the thesis takes photovoltaic power generation units,solid oxide fuel cell systems and energy storage units as the research objects to study the high-precision intelligent prediction of photovoltaic power generation and load.In addition,the corresponding energy dispatch optimize the strategy to ensure the economy and stability of the microgrid system operation.For the prediction of photovoltaic power generation,the thesis first analyzes and extracts the key factors that affect the total solar irradiance and photovoltaic power generation characteristics.Then,the K-Means clustering algorithm based on the elbow method is used to preprocess the training set,and two prediction methods are proposed for the intermittent and periodic characteristics of photovoltaic power generation.The first method is a data-driven-mechanism model combination prediction method: predict the total solar irradiance based on the Gaussian Process Regression(GPR)algorithm,and then use the irradiance as input to construct a photovoltaic power physical model to predict photovoltaic power generation power;The second is a data-driven prediction method based on the online T-S fuzzy model.A cascade T-S fuzzy prediction model is established to predict the total solar irradiance,and then predict the photovoltaic power generation.Finally,actual measurement data are used to verify the effectiveness of the above two methods.After analysis and comparison,it is found that the GPR combined with physical model method has the highest accuracy,however,it is only suitable for short-term photovoltaic forecasting,while the online cascade T-S fuzzy model method is suitable for photovoltaic forecasting with a wide range of operating hours.For inelastic load forecasting,firstly,the key factors affecting load characteristics are analyzed and extracted,and then the offline long and short-term memory network direct forecasting method and the improved online Long Short-Term Memory(LSTM)rolling forecasting method are respectively proposed for the time series and periodicity of the load demand curve.The offline LSTM prediction method has the advantages of less training times,fast convergence speed and small amount of calculation,but it also has some shortcomings,such as a fixed training set,which cannot be updated in real time and cannot learn new data laws.Therefore,the thesis further adopts the improved online LSTM rolling prediction method to realize online rolling prediction by updating the training set in real time.Finally,the actual measurement data is used to verify the effectiveness of the two proposed methods.After analysis and comparison,it is found that the prediction accuracy of the online LSTM rolling prediction method is higher than that of the offline LSTM prediction method.Based on the previous photovoltaic and load forecasting results,the thesis establishes a microgrid energy day-ahead dispatching economic model,using particle swarm optimization(PSO)to optimize the dispatching strategy.The optimized result is a 19%reduction in operating costs compared to the conventional rule control strategy without a forecast link.In summary,this thesis aims at the uncertainty change of photovoltaic power generation and inelastic load in microgrid,and proposes corresponding prediction methods based on dynamic feature analysis,which achieves high-precision photovoltaic and load forecasting on different time scales.Based on the prediction results,the day-ahead dispatch strategy optimization of microgrid energy is carried out.The results show that,compared with the non-predictive dispatching strategy,the method proposed in this thesis significantly improves the economy and stability of microgrid operation.