Research On Short-term Load Forecast Of Power System With Distributed Energy

Load forecasting has been known as one of the key issues in the field of power grid since published in the literature.With the advent of the era of intelligent power grid,and the increasing seriousness of the problem of energy and environment,new energy power generation has developed rapidly,and the distributed energy has become the mainstream.Because of the randomness and intermittent nature of the distributed energy generation,the grid connected will bring some harm to the power grid.The research on short-term load forecast of power system with distributed energy can help the relevant departments to develop or adjust the grid ahead generation scheduling,scheduling and conventional units,thereby reducing the influence of grid power system distributed energy brought to the power system safe and stable operation.To forecast the short-term load of electric power system,characteristics of the load,including daily,weekly,monthly,and the influence of the factors on the load need to be studied firstly.After that,the load forecasting model is established based on the deep belief nets,and the influence of different parameters on the prediction results is analyzed and the final parameters are determined.Finally,forecast the load,and calculate the mean absolute percentage error of forecast results is 1.68%.Meanwhile,the mean absolute percentage error of the prediction results obtained by BP neural network is 3.43%.What can conclude is that the forecast results obtained by deep belief nets are better than that obtained by BP neural network,so deep learning algorithm can be used to forecast the short-term load of power system.Short term output power prediction of photovoltaic power generation system needs to study the factors that affect the photovoltaic power generation firstly,to obtain the correlation between each factor and the output power of photovoltaic power generation.These factors include solar radiation intensity,air temperature,relative humidity,wind speed and weather type.Then,the model inputs are solar radiation intensity,atmospheric temperature,relative humidity and the historical data of photovoltaic power generation output of the 5 days before the forecast day.Finally,forecast the photovoltaic power generation output of 2016.6.15-6.17 after data preprocessing.The average absolute percentage errors of the three days' forecast results were 9.25%,11.17%,13.30%,respectively,which have reached the requirement of forecast accuracy.So,deep learning algorithm can be used to forecast short-term output power of photovoltaic power generation system too.To understand the concept of equivalent load,and the equivalent load forecasting method is put forward,and the flow chart of equivalent load forecasting is drawn according to the previous research results.To sum up,the achievements have some practical value for the research on short-term load forecast of power system with distributed energy.