基于TSNE-BiGRU模型短期电力负荷预测

Short-term Power Load Forecasting Based on TSNE-BiGRU

短期电力负荷预测是电力系统合理调度与安全稳定运行的基础。为提高电力负荷预测精度,提出一种基于t分布邻域嵌入(t-SNE)算法和双向门控循环单元(Bi-GRU)网络的短期电力负荷预测方法。该方法首先通过多标签处理将电力负荷时序数据转换成高维时间戳数据,进而在维持数据信息完整性的前提下通过t-SNE算法对其降维,并结合实时电价数据,基于Bi-GRU网络学习时间戳数据、实时电价数据及实时负荷数据之间的非线性特性,最后经全连接输出层聚合相关信息给出预测结果。基于新加坡地区电力基准数据集进行试验,对比分析所建模型TSNE-BiGRU与基准模型Bi-GRU及GRU的预测性能。试验结果表明所建模型TSNE-BiGRU具有良好的鲁棒性,能有效提高短期电力负荷的预测精度。其平均百分比误差值为0.49%,相较Bi-GRU与GRU,分别降低了23.44%与32.88%;其平均绝对误差值为30.58,相较两基准模型分别降低了22.19%与32.84%;其均方根误差值为39.40,相较两基准模型分别降低了17.16%与27.88%。

Short-term power load forecasting is the basis for the rational dispatch and safe and stable operation of the power system.In order to improve the accuracy of power load forecasting,a short-term power load forecasting method based on t-distributed neighborhood embedding(t-SNE)algorithm and bidirectional gated cyclic unit(Bi GRU)network is pro-posed.Firstly,the power load time series data is converted into hight dimensional timestamp data through multi-label processing,and then the dimensionality is reduced by the t SNE algorithm under the premise of maintaining the integrity of the data information.Combined with real-time electricity price data,based on the nonlinear characteristics between Bi-GRU network learning timestamp data,real-time electricity price data and real-time load data,the relevant information of layer aggregation is output through full connection to give prediction results.Based on the Singapore power benchmark:dataset,the prediction performance of the established model TSNE-BiGRU and the benchmark model Bir GRU and GRU are compared and analyzed.The experimental results show that the TSNE BiGRU model has good robustness and can effectively improve the prediction accuracy of short-term power load,and the average percentage error value is0.49%.The experimental results show that the proposed model TSNE-BiGRU has good robustness and can effectively improve the pre-diction accuracy of short-term power load,and its average percentage error value is 0.49%,which is 23.44%and 32.88%lower than Bi GRU and GRU,respectively.The average absolute error value of TSNE BiGRU is 30.58,which is 22.19%and 32.84%lower than the two benchmark models,respectively.The root mean square error value of TSNE-BiGRU is 39.40,which is 17.16%and 27.88%lower than the two benchmark models,respectively.