عنوان مقاله [English]
One of the major issues for design and operation of power systems is load forecasting for the same hour in the next few days, known as a Short-Term Load Forecasting (STLF). Forecasts are required for proper scheduling activities, such as generation scheduling, fuel purchasing activities, maintenance scheduling, investment scheduling, and for security analysis. Accurate forecasting of electrical load leads to energy saving and careful planning. The aim of this study is to predict short-term consumption of electrical energy in one of the states of Iran (i.e., Mazandaran). This study used several techniques and tools of data mining to predict electrical energy consumption and demand in short-term time. Several methods, such as Neural Network, Support Vector Machine were used for forecasting and their results were examined. The first phase of this research is to identify the parameters that affect electrical energy consumption. Then, among these factors, those with the greatest effect will be selected. In the next step, data analysis and different behaviors of electrical energy consumption are discussed and classified based on their similarity. Afterwards, the required inputs will be identified and pre-processing will be performed. In the next step, Pervious electricity load values with related data of each category are presented for the Multilayer Perceptron Neural Network and support Vector Machine recursively. In this model, the support Vector Machine could supply a better result. Then Principle Component Analysis (PCA) is used to reduce the dimension of input variables. New data will be tested once again with proposed systems to observe the effects of principle component analysis on each method. Finally, the results of all methods are compared with each other. The result will be compared with two measures: including coefficient of determination (R2) and root mean square error (RMSE). The result shows the improvement in Neural Network and Support Vector Machine with the use of principal component analysis, which provide better results compared to classical predictions.