基于BP神经网络与多元线性回归的短期燃气负荷预测

Short-term Gas Load Forecasting Based on BP Neural Network and Multivariable Linear Regression

影响燃气负荷变化的因素呈现非线性和随机性特征,单一数值算法很难进行精确预测。为了提高燃气负荷预测的准确度,使预测算法具备更好的适应性,提出了一种基于多元线性回归与BP神经网络的短期燃气负荷预测模型。该混合优化算法兼顾了多元线性回归算法的非线性特性和BP神经网络的泛化特性。以宁夏平罗县2011年城市居民燃气用气量为研究算例,应用灰色关联度对燃气负荷及影响因素进行相关性分析,并采用均方根误差及R2判定系数作为预测模型性能评价方法。通过仿真,验证了所建立模型是可行且有效的。相比单一的多元线性回归方法或BP算法,采用混合算法所建立预测模型具有更好的适应性,预测误差更小。

The factors affecting gas load change are nonlinear and random,so it is difficult to predict accurately for a single numerical algorithm.To improve the accuracy of gas load prediction and make the forecasting algorithms more adaptable,a short-term gas load forecasting model is proposed in this paper based on multivariable linear regression(MLR)and BP neural network.The hybrid optimization algorithm integrates the nonlinear feature of multivariate linear regression algorithm with the generalization property of BP.Firstly,the urban gas consumption in 2011 of Pingluo is taken as an example.Secondly,the root means square error and R2 coefficient are used as the evaluation methods.Meanwhile,the correlation between gas load and its influencing factors are analyzed by grey correlation analysis.Finally,the simulation results show that the model established is feasible and effective.Compared with the single MLR method or BP algorithm,the prediction model based on the hybrid algorithm has better adaptability and smaller prediction error.