The Application Study Of The Principle Of Curve Fitting Is Load Forecasting

With the rapid development of technology and economy, the electric power production department can use accurate load forecasting, to economic unit start-stop reasonably, and to make more reasonable maintenance plan, thereby reducing the load spare capacity, reduce cost, achieve the goal of improve the economic benefit. Research has high precision and strong practicability for electric power load forecasting method of marketization and the development of smart grid is very necessary.In order to further improve the prediction effect of medium and long-term load forecasting, the paper mainly studied the following contents:(1) this paper introduces the three kinds of polynomial fitting model. Algebraic polynomial are introduced respectively, chebyshev polynomial and ray let DE polynomial. Three kinds of polynomials are widely used in curve fitting and numerical analysis. Which belongs to the orthogonal polynomial algebraic polynomials, and chebyshev polynomial and ray let DE polynomial are orthogonal polynomials, but also because of excellent properties of the two polynomials, compared to a polynomial is more suitable for curve fitting and other applications.(2) the currently the most widely used three kinds of algorithm of gradient descent method and conjugate gradient method and gradient least squares method, has carried on the comparison and analysis. The nature of the analysis of the three algorithms, the gradient descent method take the negative gradient direction for the search direction, the procedure is simple, each iteration of the less amount of calculation and save the storage space, for less good initial point can be convergence. But its linear convergence speed, convergence speed is limited by a lot. Conjugate gradient method is often used to solve large-scale nonlinear optimization problem. This method only USES objective function value and gradient function, avoids the problem of slow convergence speed gradient descent method. In addition, it also has no matrix storage and quadratic termination, etc. But the convergence of conjugate gradient method depends on the step length factor alpha has a good value, if the value of alpha step length factor is undeserved, convergence may not that don't work or convergence to a solution of point. The method training weight coefficient by the recursive least squares method, and the historical data are fitted in order to realize the forecast model parameter estimation. And recursive least squares method possesses the characteristics of algorithm is simple and easy to implement and therefore is widely used.(3) put forward a prediction method based on Chebyshev polynomials. The method using the recursive least squares method training the neural network weight coefficient, in order to obtain high accuracy parameter estimation, so as to realize optimum fitting of Chebyshev polynomial model of load, reuse trained Chebyshev polynomial model to realize the medium and long-term load forecasting.Simulation results show that this method can better simulate the load change law of predicted results improve an order of magnitude, have certain ability of medium and long-term load forecasting. In addition, also can see through simulation examples, using Laguerre polynomial neural network model of prediction method also has the same accuracy.