|
Power system short-term load forecasting is an important sub-system of energy management system (EMS), it is the indispensable part of electric power dispatching system operation. Accurate short-term load forecasting can help managers to formulate reasonable generating plan, appropriate measures to move peak and valley, ensure grid security, economic operation, reduce power cost and improve the economic benefit and social benefit. Along with the development of electric power system, the traditional load forecasting technology has been difficult to satisfy the power sector increasingly high load forecasting accuracy requirement, how to get fast, accurate and stable load forecasting has always been the focus of research in recent years.Firstly, this paper review to power system load forecasting of traditional methods and modern methods, key research the application of artificial neural network in short-term load forecasting, established the neural network short-term load forecasting model. Aiming at the faults of most commonly used BP algorithm in neural network slow convergence speed, easy into the local minimum, this paper adopts Levenberg - Marquardt learning algorithm to training the network, at the same time, introduce particle swarm optimization (PSO) algorithm to neural network, aiming at the shortcomings of the particle swarm optimization algorithm gives some suggestions to improve the speed and position optimizing formula, using the improved particle swarm algorithm to optimize the network initial weights threshold, further improve the load forecast speed and precision. Through simulation examples, compare basic BP algorithm neural network predictive model, L-M algorithm improving neural network predictive model and particle swarm optimization L-M algorithm improved neural network predictive results, and finally proves particle swarm optimization improved neural network hybrid algorithm whether in prediction accuracy or speed are great degree rise, a further development of practical application value.
|