Research On Short-term Load Forecasting Of Electric Power System

Power system short-term load forecasting is an important means of power system dispatching operation, load forecasting accuracy directly affect the stability of power system operation, safety and economy. Nonlinear and stochastic characteristics of power load itself determine the difficulty of the prediction, but given its importance in the power industry, load forecasting has become a frontier and hot issue of academic research. With increasing the capacity and the complexity of power system, load variation regularity is more complex, the traditional prediction method of the adaptive ability is poorer, predict cannot get satisfactory results. Therefore, intelligent integrated prediction method of study to become one of the focuses on load forecasting.According to the characteristics of the power load, this paper established two working days and holidays load change model, the effect in the model such as climate, real-time electricity price child element, using intelligent optimization calculation method and the comprehensive prediction technology, accurate short-term load forecasting of power system changes, for power system operation management to provide scientific decision basis. The main research work and innovative results are as follows:(1)This paper introduces the Dongying economic situation as well as its power load with the characteristics and analyzing the load characteristics, highlighted several factors affect the accuracy of load forecasting Dongying region, while error analysis.(2)This paper analyzed the main influence factors in the load forecasting model and error factor, according to the interference data of the samples, is proposed based on the kernel function of the weighted fuzzy c-means clustering algorithm-WKFCM, the algorithm adopts a simple two nuclear induction distance and replaces the Euclidean distance as the clustering of complex target formula of similarity measure function, to reduce the computational complexity and error. After the data clustering, using the pattern classification ability, convergence speed of neural network identification interference data correction model, reducing the data on the model, the influence of improve the prediction accuracy.(3) For working load model, the article puts forward the excitation function based on the strategy of adaptive adjustable BP learning algorithm. In this paper, the numerical examples show that the excitation function based on the strategy of adaptive adjustable optimization BP learning algorithm is better than BP algorithm based on chaos optimization of excitation function is more accurate and reliable, has more practical value.(4) In this paper, we will be holiday model divided into two categories, the weekend day of rest, and major holidays. Based on immune particle swarm optimization are put forward on the weekend day off, the least squares support vector machine (IPSO-LS-SVM) prediction model. The antibody diversity maintaining mechanism of the immune system is introduced into the particle swarm optimization algorithm, while keeping high fitness particle at the same time, ensure the diversity of particles, so as to improve the convergence performance. Load forecasting for holiday long time span, refers to the historical data of quantity is little, affected by meteorological factors is more outstanding, the characteristics of gray markov chain model are proposed for load forecasting, and then use comprehensive factors affecting matching model to modify predicted results, improve the precision of prediction.