于军琪,王佳丽,赵安军,等.基于AP相似日选取与FISOA-RBF的短期负荷预测[J].深圳大学学报理工版,2021,38(3):315-323.[doi:10.3724/SP.J.1249.2021.03315]
YU Junqi,WANG Jiali,ZHAO Anjun,et al.Short-term load forecasting based on AP similar days and FISOA-RBF[J].Journal of Shenzhen University Science and Engineering,2021,38(3):315-323.[doi:10.3724/SP.J.1249.2021.03315]
点击复制
基于AP相似日选取与FISOA-RBF的短期负荷预测
于军琪,王佳丽,赵安军,解云飞,冉 彤,赵泽华

西安建筑科技大学建筑设备科学与工程学院,陕西西安 710055

计算机神经网络; 吸引子传播; 相似日选取; 搜索者优化算法; 径向基; 建筑用电; 短期负荷预测

Short-term load forecasting based on AP similar days and FISOA-RBF
YU Junqi, WANG Jiali, ZHAO Anjun, XIE Yunfei, RAN Tong, and ZHAO Zehua

School of Building Services Science and Engineering, Xi'an University of Architecture and Technology, Xi'an 710055, Shaanxi Province, P.R.China

computer neural network; affinity propagation(AP); similar day selection; seeker optimization algorithm; radial basis function; electricity consumption of buildings; short-term load forecasting

DOI: 10.3724/SP.J.1249.2021.03315

备注

为提高建筑电力负荷的预测精度,在考虑天气信息和日期类型等影响因素的基础上,提出基于吸引子传播(affinity propagation, AP)相似日选取和改进搜索者优化算法-径向基(fusion improvement seeker optimization algorithm-radial basis function, FISOA-RBF)神经网络的建筑用电短期负荷预测模型.采用AP算法对短期电力负荷进行相似日选取,以克服外界环境对建筑电力负荷预测精度的影响; 以RBF神经网络的网络参数为优化对象,采用搜索者优化算法(seeker optimization algorithm, SOA)进行参数寻优,并引入融合改进策略提高传统人群算法的寻优性能,以进一步提高RBF神经网络的预测精度和学习速度; 根据FISOA 算法优化后的RBF神经网络对相似日数据进行训练,建立最优参数下的建筑短期电力负荷预测AP-FISOA-RBF模型.在相同数据集和气候特征条件下,与传统RBF、PSO-RBF和SOA-RBF预测模型相比,AP-FISOA-RBF模型平均预测绝对百分比误差分别降低了93.05%、83.60%和71.13%,平均预测速度分别提高了54.34%、 39.25%和23.96%,表明AP-FISOA-RBF模型在预测精度和预测速度上的表现更好.
In order to improve the accuracy of electrical load forecasting for large public buildings, we propose a forecasting model for the short-term load of large public buildings based on affinity propagation(AP)similar days selection and the fusion improvement seeker optimization algorithm-radial basis function(FISOA-RBF)neural network by considering the weather information, date type and other influencing factors. In order to overcome the influence of external environment on the accuracy of building electrical load forecasting, AP algorithm is used to select similar days of short-term electrical load. The structural parameters of RBF neural network are optimized by FISOA which uses the fusion improvement theory to further improve the prediction accuracy and the learning speed of RBF neural network. Finally, the similar daily load data are used to train an optimized FISOA-RBF to predict the short-term electrical load of buildings. In order to validate the effectiveness of the proposed model, the exhaustive experiments are conducted in comparison with RBF, PSO-RBF and SOA-RBF methods. The experimental results indicate that the proposed model outperforms the other models: the mean absolute percentage error(MAPE)is reduced by 93.05%, 83.60% and 71.13%, and the average prediction speed is increased by 54.34%, 39.25% and 23.96%, and thus demonstrate that AP-FISOA-RBF model in prediction accuracy and speed of prediction performance is better than other three RBF-based methods.
·