构建高速公路服务区交通自洽能源系统是实现交通与能源融合的关键技术,而对其进行系统状态估计是近年来的研究重点.考虑到高速公路服务区交通自洽能源系统的复杂性和多样性,单一的数据或模型驱动方法难以全面、准确地估计系统实时状态.因此,研究了 1种数据与模型驱动的复合方法,旨在实现更高效的系统状态估计.在数据驱动方面,尽管基于深度学习的光伏功率预测模型性能优越,但通常忽视输入特征间的互相依赖机制.为此,建立了基于自注意力机制(self-attention,SA)的时间卷积-双向长短期记忆网络(time convolu-tion-bidirectional long short-term memory network,TCN-BiLSTM-SA),用于预测系统光伏出力情况.SA重新分配TCN-BiLSTM输入特征的权重,从而提升时空信息提取的有效性.在模型驱动方面,考虑高速公路路网车流量分布,建立了高速公路电动汽车出行轨迹概率模型;基于蒙特卡洛模拟法抽取初始和充电时电池容量,综合考虑车主用车习惯、环境温度等多种不确定性因素,来预测电动汽车充电负荷时空分布.通过利用新疆某高速公路服务区交通自洽能源系统的实际数据进行仿真验证,结果表明:在光伏预测上,所提模型在平均绝对误差、均方根误差和决定系数这3个指标上,相较于最佳模型分别提高了 25.3％、16.7％、0.7％;在负荷预测上,所提模型有效预测高速公路电动汽车的充电负荷时空分布;在系统状态估计上,所提方法的精度达到了 89.1％.
A Method for Data and Model Driven Estimation of Traffic Self-Consistent Energy System States in Highway Service Areas
The development construction of a self-consistent energy system for highway service areas is essential technology for integrating transportation and energy.A key focus for recent research is system state estimation.Giv-en the complexity and diversity of these energy systems,relying soley on either data-driven or model-driven ap-proaches often leads to challenges in accurately and comprehensively estimating real-time system states.Therefore,this study explores a hybrid method combining data-driven and model-driven approaches for more efficient system state estimation.In terms of data-driven methods,although deep learning-based photovoltaic power forecasting mod-els demonstrate superior performance,they fail to account for the interdependencies among input features.To ad-dress this issue,we developed a Time Convolution-Bidirectional Long Short-Term Memory network(TCN-BiL-STM-SA)based on a Self-Attention mechanism(SA)to predict the photovoltaic output of the system.The SA mod-ule adjusts redistributes the weights of the TCN-BiLSTM input features,improving the extraction of spatiotemporal information.On the model-driven side,considering the traffic flow distribution of highway networks,we estab-lished a probability model for electric vehicle travel trajectories.By utilizing Monte Carlo simulations,the initial and charging battery capacities are extracted while accounting for various uncertainties such as driver behavior and environmental temperature,thereby approximating accurate results to predict the spatiotemporal distribution of elec-tric vehicle charging loads.Simulation validation using actual data from a self-consistent energy system at a high-way service area in Xinjiang indicated that,in terms of photovoltaic forecasting,the proposed model improved the mean absolute error,root mean square error,and coefficient of determination by 25.3％,16.7％,and 0.7％,respec-tively,compared to the best model.Furthermore,the proposed model effectively predicted the spatiotemporal distri-bution of electric vehicle charging loads on highways,achieving a system state estimation accuracy of 89.1％.
integration of transportation and energysystem state estimationTCN-BiLSTM-SA modeldata-and model-drivenMonte Carlo
万方数据
维普
