基于优化小波BP神经网络的燃气短期负荷预测

Short-Term Gas Load Forecasting Based on Optimized Wavelet BP Neural Network

在燃气短期负荷预测问题的研究中,燃气负荷由于受天气、人为活动等因素的影响,呈现出一种非线性特性,单个神经网络的局限性限制了其预测精度。为了有效的预测天然气短期负荷,提出了一种混沌遗传算法优化的小波BP神经网络预测模型。小波网络结合小波变换良好的时频局部特性和神经网络的自学习能力,加强了网络的函数逼近能力。利用混沌遗传算法的全局优化搜素能力对网络连接权值、阈值和伸缩平移尺度的优化求解,加快了网络的收敛的速度,建立最优的燃气负荷预测模型。将组合模型应用于上海燃气短期负荷预测,结果表明改进检测模型具有更好的非线性拟合能力和更高的预测精度。

As affected by weather, human activities and other factors, the gas load presents non-linear characteristics. The shortcomings of single neural network limit its prediction accuracy. In order to effectively forecast short-term gas load, a combinational model based on wavelet BP neural network optimized by genetic algorithm was proposed. Combining the well time-frequency local feature of wavelet transform with the self-learning ability of neural network, the wavelet network strengthened the function approximation capacity of the network. And the global search ability of chaos genetic algorithm was used to optimize connection weights, thresholds and telescopic pan scale of net- work, which accelerates the convergence speed of the network. Then the optimal gas load forecasting model was established. This proposed model was applied to short-term gas load forecasting for Shanghai and the result of simulation indicates that this algorithm has better non-linear fitting ability and higher prediction accuracy.