基于神经网络的燃气日负荷智能预测方法

Research on combination forecasting method of gas daily load based on neural network

为解决城市燃气日负荷具有随机性和不确定性问题,首先分别采用广义回归神经网络(GRNN)、灰色?GRNN和梯度?GRNN对燃气日负荷数据进行预测,通过MATLAB仿真表明,得到的预测误差大部分都在20%以内,说明这3种预测模型总体上是可行的,但预测精度并不是很高;然后针对城市燃气日负荷可能发生的预测模型故障现象,将GRNN、灰色?GRNN和梯度?GRNN的预测数据作为组合预测模型的数据基础,采用广义动态模糊神经网络(GD-FNN)进行智能组合预测,仿真结果表明:GD-FNN的平均预测精度为93.639%,平均每组预测用时为7.668 s,从预测精度上看,组合预测模型的预测精度要明显高于单一预测模型的预测精度,尤其是在预测过程中发生故障现象时更显其优势。

In order to solve the problem of randomness and uncertainty of urban gas daily load,firstly,the data of urban gas daily load are forecasted by using generalized regression neural network(GRNN),grey-GRNN and gradient-GRNN respectively.The simulation results of MATLAB show that most of the forecasting errors are within 20%,which shows that the three forecasting models are generally feasible,but the forecasting accuracy is not very high,Then,aiming at the fault phenomenon of forecast model of urban gas daily gas load,which is likely to occur,the forecast data of GRNN,grey-GRNN and gradient-GRNN are taken as the data base of combination forecast model.Intelligent combination forecast is carried out by using the generalized dynamic fuzzy neural network(GD-FNN).The simulation results show that the average prediction accuracy of GD-FNN is 92.725%,and the average prediction time of each group is 7.668 seconds.The prediction accuracy of the combined forecasting model is much higher than that of the single forecasting model,especially when fault occurs in the forecasting process.