基于K近邻互信息估计的原油管道电耗预测

Study on Power Consumption Prediction Method based on K-neighbor Estimating Mutual Information

为使用最少的特征实现管道电耗的精准预测,建立基于K近邻互信息估计的BPNN模型来预测管道电耗。利用管道输送理论公式扩充原始数据集,利用K近邻互信息估计提取强相关性特征,将提取出的特征喂入BPNN来建立原油管道电耗预测模型,最后对比利用不同输入特征建立的模型的预测精度。研究结果表明:利用K近邻互信息估计能够选出多个与电耗相关的重要特征;利用相关性最强的前5个特征建立的BPNN预测模型时,模型的平均绝对百分比误差比利用单个特征建模时降低了39.28%,达到5.79%;该模型平均训练时间也比利用全部特征建模时缩短22.49%。证明K近邻互信息估计能够提取管道电耗的相关特征,与BPNN结合后能够实现管道电耗的准确预测。

In order to achieve accurate prediction of pipeline power consumption with the least characteristics,this study established a BPNN model based on K-neighbor estimating mutual information to predict pipeline power consumption.The theoretical formula of pipeline transportation was utilized to expand the original data set,and the strong correlation features were extracted by using the K-neighbor estimating mutual information,and the extracted features were fed into BPNN to establish the prediction model of crude oil pipeline power consumption.At last,the BPNN model was established to predict the power consumption of crude oil pipeline.By comparing the prediction accuracy of the models established with different input characteristics,the results show that the K-neighbor estimating mutual information can effectively identify important characteristics associated with power consumption.The BPNN prediction model was established by the first five characteristics with the strongest correlation.The mean absolute percentage error of this model,reaching 5.79%,was decreased by 39.28%,compared with that of the model established by a single characteristic.The average time for training this model was also 22.49%shorter than that of the model established by all characteristics.The K-neighbor estimating mutual information can extract relevant features of the pipeline power consumption.The K-neighbor estimating mutual information combined with BPNN can improve the prediction accuracy of pipeline power consumption.