为确保电力系统的安全与稳定,需对未来一天内各时段的电力需求进行精准预测。然而,随着可再生能源的增加,电荷预测变得更加复杂和不可预测。因此,本文介绍了一个基于GCformer构建的新模型GGSAGCformer,通过图卷积神经网络和门控循环单...为确保电力系统的安全与稳定,需对未来一天内各时段的电力需求进行精准预测。然而,随着可再生能源的增加,电荷预测变得更加复杂和不可预测。因此,本文介绍了一个基于GCformer构建的新模型GGSAGCformer,通过图卷积神经网络和门控循环单元提取数据中的空间和时序特征,再引入多头注意力机制(Multi-head Attention Mechanism Layer)。在此层中加入了地理相似空间自注意力模块(Geographically Similar Spatial Self-Attention, GSSA)和气候自注意力模块(Climate Self-Attention, CSA),旨在深入探索数据中的潜在关联,输出层使用GCformer来处理预测结果,以提升预测的准确性。实验结果显示,本文模型在输出步长为192的情况下,与传统模型GCformer、Informer和Reformer相比,MSE分别降低了15.3%、25%和29.3%。In order to ensure the safety and stability of the power system, it is necessary to accurately predict the power demand at each time of the day in the future. However, with the increase in renewable energy, charge prediction has become more complex and unpredictable. Therefore, this paper introduces a new model based on GCformer, GGSAGCformer, which extracts spatial and temporal features from the data through graph convolutional neural network and gated recurrent unit, and then introduces the multi-head attention mechanism layer. The Geographically Similar Spatial Self-Attention (GSSA) and Climate Self-Attention (CSA) modules are added to this layer to deeply explore the potential associations in the data, and the output layer uses GCformer to process the prediction results to improve the accuracy of the prediction. The experimental results show that the MSE of the proposed model is reduced by 15.3%, 25% and 29.3% compared with the traditional models GCformer, Informer and Reformer when the output step size is 192, respectively.展开更多
区域交通流量预测是智慧交通系统的一项重要功能。联邦学习可以支持多位置服务提供商(Location Service Provider,LSP)的联合训练,使得训练数据集可以更加全面地覆盖整个区域的交通流量,提高预测准确率。但是,当前基于联邦学习的区域交...区域交通流量预测是智慧交通系统的一项重要功能。联邦学习可以支持多位置服务提供商(Location Service Provider,LSP)的联合训练,使得训练数据集可以更加全面地覆盖整个区域的交通流量,提高预测准确率。但是,当前基于联邦学习的区域交通流量预测方案存在车辆数据去重、训练节点背叛以及隐私泄露等问题。为此,构建了基于联邦学习的隐私保护区域交通流量预测(Privacy-Preserving Regional Traffic Flow Prediction based on Federated Learning,PPRTFP-FL)模型。模型采用中心部署架构,由联邦中央服务器协调各个LSP联合完成模型的训练,并对全局模型进行梯度聚合与模型更新;采用交叉评价加权聚合的策略来防御部分不可信节点对全局模型的恶意攻击,提升了全局模型的鲁棒性;预测阶段使用同态加密聚合算法,各LSP在不泄露自身运营数据的情况下实现了更准确的流量预测。利用相关数据集进行测试,测试结果表明当训练数据集覆盖区域流量充分的情况下,本模型相比本地模型的预测准确率有明显的提升。对模型进行不同比例的恶意节点攻击实验,由实验结果可知,系统在存在恶意节点情况(当恶意节点数量小于总节点数量50%时)下仍具备较好的防御效果。展开更多
文摘为确保电力系统的安全与稳定,需对未来一天内各时段的电力需求进行精准预测。然而,随着可再生能源的增加,电荷预测变得更加复杂和不可预测。因此,本文介绍了一个基于GCformer构建的新模型GGSAGCformer,通过图卷积神经网络和门控循环单元提取数据中的空间和时序特征,再引入多头注意力机制(Multi-head Attention Mechanism Layer)。在此层中加入了地理相似空间自注意力模块(Geographically Similar Spatial Self-Attention, GSSA)和气候自注意力模块(Climate Self-Attention, CSA),旨在深入探索数据中的潜在关联,输出层使用GCformer来处理预测结果,以提升预测的准确性。实验结果显示,本文模型在输出步长为192的情况下,与传统模型GCformer、Informer和Reformer相比,MSE分别降低了15.3%、25%和29.3%。In order to ensure the safety and stability of the power system, it is necessary to accurately predict the power demand at each time of the day in the future. However, with the increase in renewable energy, charge prediction has become more complex and unpredictable. Therefore, this paper introduces a new model based on GCformer, GGSAGCformer, which extracts spatial and temporal features from the data through graph convolutional neural network and gated recurrent unit, and then introduces the multi-head attention mechanism layer. The Geographically Similar Spatial Self-Attention (GSSA) and Climate Self-Attention (CSA) modules are added to this layer to deeply explore the potential associations in the data, and the output layer uses GCformer to process the prediction results to improve the accuracy of the prediction. The experimental results show that the MSE of the proposed model is reduced by 15.3%, 25% and 29.3% compared with the traditional models GCformer, Informer and Reformer when the output step size is 192, respectively.
文摘区域交通流量预测是智慧交通系统的一项重要功能。联邦学习可以支持多位置服务提供商(Location Service Provider,LSP)的联合训练,使得训练数据集可以更加全面地覆盖整个区域的交通流量,提高预测准确率。但是,当前基于联邦学习的区域交通流量预测方案存在车辆数据去重、训练节点背叛以及隐私泄露等问题。为此,构建了基于联邦学习的隐私保护区域交通流量预测(Privacy-Preserving Regional Traffic Flow Prediction based on Federated Learning,PPRTFP-FL)模型。模型采用中心部署架构,由联邦中央服务器协调各个LSP联合完成模型的训练,并对全局模型进行梯度聚合与模型更新;采用交叉评价加权聚合的策略来防御部分不可信节点对全局模型的恶意攻击,提升了全局模型的鲁棒性;预测阶段使用同态加密聚合算法,各LSP在不泄露自身运营数据的情况下实现了更准确的流量预测。利用相关数据集进行测试,测试结果表明当训练数据集覆盖区域流量充分的情况下,本模型相比本地模型的预测准确率有明显的提升。对模型进行不同比例的恶意节点攻击实验,由实验结果可知,系统在存在恶意节点情况(当恶意节点数量小于总节点数量50%时)下仍具备较好的防御效果。
