基于改进的Chebyshev神经网络的用水量预测

FORECAST OF WATER USING IMPROVED CHEBYSHEV NEURAL NETWORK

针对目前Chebyshev神经网络存在的不足,从算法和网络结构方面进行了综合改进。改进后的Chebyshev神经网络不仅符合生物神经网络的基本特征,算法简单,收敛速度快,而且网络输入可以是任意值,是一种多输入的多层前向神经网络模型,因而扩大了网络辨识模型的能力与学习适应性,并有逼近任意线性和非线性映射的优异特性。用改进的Chebyshev神经网络对城市的家庭用水需求量进行建模和预测。仿真结果表明,改进的Chebyshev神经网络为预测家庭用水需求量提供了一种有效的方法,它不仅具有优良的预测能力,而且在相同精度的前提下,其收敛速度也优于一般的BP网络。

A kind of Chebyshev neural network which was improved from algorithm and network structure was presented according to existing insufficiency of Chebyshev neural network. Improved Chebyshev neural network does not only accord with the basic characteristics of biology neural network, but also has a simple algorithm, a high speed convergence of learning process, and input value of this network can be random. It is a multi-input network structure, so it expands the identification ability and learning adaptation of the network, which has excellent characteristics in the linear and nonlinear accurate approximation. This relatively new technique of improved Chebyshev neural network was proposed to modeling and forecasting the water demand in urban areas. The simulation results indicate that the improved Chebyshev neural network offers an effective way to forecast domestic water demand. It has a good estimate ability and high convergent speed with the same precision compared to the general network of BP.