Short-term Load Forecasting Considering Long-term Fluctuations

The power industry is closely related to national security and social and economic development,and is the basis for ensuring the safe and stable operation of human life.Short-term load forecasting of power systems is one of the core tasks of modern power management systems.It is of great significance for maintaining safe and stable operation of power systems,optimizing power dispatching,and improving social and economic benefits.Therefore,improving the accuracy of power load forecasting is currently being studied by many researchers worldwide.With the continuous development of modern science and technology,a variety of load forecasting technologies have emerged one after another.BP(Back Propagation)neural network forecasting technology has been w idely used due to its powerful nonlinear mapping ability and self-learning ability.The main work of this thesis is to propose some modified algorithms based on traditional BP neural network to improve the accuracy of load forecasting.By researching a lot of relevant literatures,the author firstly analyses the current development status of power load forecasting technology at home and abroad,and gives a comparative analysis of different load forecasting methods;secondly,the basic concepts,characteristics and classification of power load are re-visited.The influencing factors and basic steps of power load forecasting,the causes of forecast errors and the calculation methods of errors are introduced.Then the basic idea,structure,principles and related optimization algorithms of BP neural network are detailed.The advantages and limitations of BP neural networks are explained.This forms the foundation for the establishment of the enhanced BP neural network model later.In this thesis,based on the traditional BP neural network,using gradient descent algorithm and L-M algorithm as the optimization algorithm of network weight and threshold,two different BP neural network load prediction models are established.The electric load and related climate data of New York Independent System Operator(NYISO)in the United States were used as samples for training and testing.The load prediction results were compared and analyzed.It was found that the L-M algorithm had higher prediction effect than the gradient descent algorithm.By analyzing the load prediction results of the traditional BP neural network,it is found that the prediction errors of the two models have a certain directionality,that is,the negative deviation of the relative error is more than the positive deviation.Therefore,in the process of BP neural network training,this thesis considers the prediction deviation as a correction factor to form a new objective function,and proposes an improved gradient descent algorithm and an improved L-M algorithm for optimization.This leads to the establishing of a new neural network-based load forecasting model with deviation feedback.Similarly,the load data of NYISO and local climate data were used to study the performance of these two improved BP neural network models.It is found that the prediction accuracy of the improved model is higher;this confirms the superiority of the method.