针对动力电池荷电状态(state of charge,SOC)的估算问题,利用长短期记忆(LSTM)循环神经网络建立SOC估算模型,以实验室恒流放电数据训练模型并测试,测试最大绝对误差为2.7%。进一步以FSEC赛车电池实测数据验证,最大测试误差为3.9%。但在...针对动力电池荷电状态(state of charge,SOC)的估算问题,利用长短期记忆(LSTM)循环神经网络建立SOC估算模型,以实验室恒流放电数据训练模型并测试,测试最大绝对误差为2.7%。进一步以FSEC赛车电池实测数据验证,最大测试误差为3.9%。但在工程应用时,考虑到实际运行过程中的环境复杂性以及不同驾驶习惯对动力电池造成的不一致性,需要根据车辆实际行驶工况数据对其进行训练与测试,但是由于该数据中的SOC直接由BMS报文解析而来,无法确定BMS内的SOC算法是否准确,故不能直接用作训练模型时的标签,此时需计算出正确的训练标签或借助已有标签的模型,在其基础上根据实际运行数据对其模型参数进行动态调整。为解决无标签数据的训练问题,本文采取第二种方法,首次提出将迁移学习中的领域自适应网络(DaNN)与LSTM组合形成LSTM-DaNN的SOC估算算法,利用有标签数据预先训练好LSTM模型,再将其模型参数迁移至LSTM-DaNN,最后综合有标签与无标签数据一起对LSTM-DaNN模型进行训练。测试结果表明LSTM-DaNN可以在没有实际行驶工况标签(SOC)的情况下完成训练,最大测试误差为4.8%,相比模型自适应调整前误差下降了14.1%,且保证绝对误差<5%,满足实际需求。展开更多
The domain adversarial neural network(DANN)methods have been successfully proposed and attracted much attention recently.In DANNs,a discriminator is trained to discriminate the domain labels of features generated by a...The domain adversarial neural network(DANN)methods have been successfully proposed and attracted much attention recently.In DANNs,a discriminator is trained to discriminate the domain labels of features generated by a generator,whereas the generator attempts to confuse it such that the distributions between domains are aligned.As a result,it actually encourages the whole alignment or transfer between domains,while the inter-class discriminative information across domains is not considered.In this paper,we present a Discrimination-Aware Domain Adversarial Neural Network(DA2NN)method to introduce the discriminative information or the discrepancy of inter-class instances across domains into deep domain adaptation.DA2NN considers both the alignment within the same class and the separation among different classes across domains in knowledge transfer via multiple discriminators.Empirical results show that DA2NN can achieve better classification performance compared with the DANN methods.展开更多
Deep learning based channel state information(CSI)fingerprint indoor localization schemes need to collect massive labeled data samples for training,and the parameters of the deep neural network are used as the fingerp...Deep learning based channel state information(CSI)fingerprint indoor localization schemes need to collect massive labeled data samples for training,and the parameters of the deep neural network are used as the fingerprints.However,the indoor environment may change,and the previously constructed fingerprint may not be valid for the changed environment.In order to adapt to the changed environment,it requires to recollect massive amount of labeled data samples and perform the training again,which is labor-intensive and time-consuming.In order to overcome this drawback,in this paper,we propose one novel domain adversarial neural network(DANN)based CSI Fingerprint Indoor Localization(D-Fi)scheme,which only needs the unlabeled data samples from the changed environment to update the fingerprint to adapt to the changed environment.Specifically,the previous environment and changed environment are treated as the source domain and the target domain,respectively.The DANN consists of the classification path and the domain-adversarial path,which share the same feature extractor.In the offline phase,the labeled CSI samples are collected as source domain samples to train the neural network of the classification path,while in the online phase,for the changed environment,only the unlabeled CSI samples are collected as target domain samples to train the neural network of the domainadversarial path to update parameters of the feature extractor.In this case,the feature extractor extracts the common features from both the source domain samples corresponding to the previous environment and the target domain samples corresponding to the changed environment.Experiment results show that for the changed localization environment,the proposed D-Fi scheme significantly outperforms the existing convolutional neural network(CNN)based scheme.展开更多
文摘针对动力电池荷电状态(state of charge,SOC)的估算问题,利用长短期记忆(LSTM)循环神经网络建立SOC估算模型,以实验室恒流放电数据训练模型并测试,测试最大绝对误差为2.7%。进一步以FSEC赛车电池实测数据验证,最大测试误差为3.9%。但在工程应用时,考虑到实际运行过程中的环境复杂性以及不同驾驶习惯对动力电池造成的不一致性,需要根据车辆实际行驶工况数据对其进行训练与测试,但是由于该数据中的SOC直接由BMS报文解析而来,无法确定BMS内的SOC算法是否准确,故不能直接用作训练模型时的标签,此时需计算出正确的训练标签或借助已有标签的模型,在其基础上根据实际运行数据对其模型参数进行动态调整。为解决无标签数据的训练问题,本文采取第二种方法,首次提出将迁移学习中的领域自适应网络(DaNN)与LSTM组合形成LSTM-DaNN的SOC估算算法,利用有标签数据预先训练好LSTM模型,再将其模型参数迁移至LSTM-DaNN,最后综合有标签与无标签数据一起对LSTM-DaNN模型进行训练。测试结果表明LSTM-DaNN可以在没有实际行驶工况标签(SOC)的情况下完成训练,最大测试误差为4.8%,相比模型自适应调整前误差下降了14.1%,且保证绝对误差<5%,满足实际需求。
基金The work was supported by the National Natural Science Foundation of China under Grant Nos.61876091 and 61772284the China Postdoctoral Science Foundation under Grant No.2019M651918the Open Foundation of Key Laboratory of Pattern Analysis and Machine Intelligence of Ministry of Industry and Information Technology of China.
文摘The domain adversarial neural network(DANN)methods have been successfully proposed and attracted much attention recently.In DANNs,a discriminator is trained to discriminate the domain labels of features generated by a generator,whereas the generator attempts to confuse it such that the distributions between domains are aligned.As a result,it actually encourages the whole alignment or transfer between domains,while the inter-class discriminative information across domains is not considered.In this paper,we present a Discrimination-Aware Domain Adversarial Neural Network(DA2NN)method to introduce the discriminative information or the discrepancy of inter-class instances across domains into deep domain adaptation.DA2NN considers both the alignment within the same class and the separation among different classes across domains in knowledge transfer via multiple discriminators.Empirical results show that DA2NN can achieve better classification performance compared with the DANN methods.
基金supported in part by the China National Key R&D Program under Grant(YFA1000500)in part by the Key Research and Developement Program of Shaanxi under Grant(2017DCXL-GY-04-02).
文摘Deep learning based channel state information(CSI)fingerprint indoor localization schemes need to collect massive labeled data samples for training,and the parameters of the deep neural network are used as the fingerprints.However,the indoor environment may change,and the previously constructed fingerprint may not be valid for the changed environment.In order to adapt to the changed environment,it requires to recollect massive amount of labeled data samples and perform the training again,which is labor-intensive and time-consuming.In order to overcome this drawback,in this paper,we propose one novel domain adversarial neural network(DANN)based CSI Fingerprint Indoor Localization(D-Fi)scheme,which only needs the unlabeled data samples from the changed environment to update the fingerprint to adapt to the changed environment.Specifically,the previous environment and changed environment are treated as the source domain and the target domain,respectively.The DANN consists of the classification path and the domain-adversarial path,which share the same feature extractor.In the offline phase,the labeled CSI samples are collected as source domain samples to train the neural network of the classification path,while in the online phase,for the changed environment,only the unlabeled CSI samples are collected as target domain samples to train the neural network of the domainadversarial path to update parameters of the feature extractor.In this case,the feature extractor extracts the common features from both the source domain samples corresponding to the previous environment and the target domain samples corresponding to the changed environment.Experiment results show that for the changed localization environment,the proposed D-Fi scheme significantly outperforms the existing convolutional neural network(CNN)based scheme.
