Short-Term Load Forecasting Based On Modified Fruit Fly Algorithm And Support Vector Machine

Accurate load forecasting is the foundation to ensure the economic, safe and reliable operation of power system, for power supply departments to formulate reasonable power plan, arrangement of the unit start-stop have very important effect.The key issue of load forecasting is the selection of predictive model and model parameters. In this paper, short-term load forecasting based on MFOA-SVM is proposed. SVM has better generalization ability and nonlinear processing capabilities, and can effectively solve the small sample size problem, however if choose the parameter is not good will seriously affect the learning and generalization ability of support vector machine. Therefore, in this paper a fruit fly algorithm is use to optimize the parameters of support vector machines. Fruit flies algorithm has good global optimization capability but in the process of optimization easily trapped in local optimal solution, so a modified fruit fly algorithm is proposed. With the modified algorithm the parameters of support vector machine (SVM) can be automatically optimized selection.When forecasting, take into account the influence factors such as daily maximum temperature, minimum temperature, rainfall, date type, and load trends, then a similar day method is used to select data sample set. And then in a certain area of Guangxi as example, predict96points load value in a day and compare with the prediction results of fruit flies algorithm optimization of support vector machine model and generalized regression neural network model. The result show that the proposed method has the best prediction accuracy, proved the feasibility of the proposed method. Finally, ordinary day and holidays are forecasted respectively, predict result further demonstrate the effectiveness of the proposed method.