Unscented Trainable Kalman Filter Based on Deep Learning Method Considering Incomplete Information
Rapid changes of states and occurrence of data missing in power systems cause accurate state estimation very hard. In this paper, an unscented trainable Kalman filter (UTKF) with a deep learning prediction model is proposed to provide accurate state estimation under incomplete information. First, th...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2023-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10129202/ |
| _version_ | 1797813918002839552 |
|---|---|
| author | Yanjie Yu Qiang Li Houyi Zhang |
| author_facet | Yanjie Yu Qiang Li Houyi Zhang |
| author_sort | Yanjie Yu |
| collection | DOAJ |
| description | Rapid changes of states and occurrence of data missing in power systems cause accurate state estimation very hard. In this paper, an unscented trainable Kalman filter (UTKF) with a deep learning prediction model is proposed to provide accurate state estimation under incomplete information. First, the CNN-LSTM architecture, a typical deep learning model, is applied to form a trainable prediction model (TPM), which offers more accurate prediction of states. However, sometimes states are incomplete due to data losses in transmissions. To deal with incomplete information, historical time-series states are employed and fed to the TPM in order to develop a missing data filling method. In this way, the prediction errors can be lower through the online training and parameter adjustment of the TPM. Combining with the TPM and the missing data filling method, an unscented trainable Kalman filter (UTKF) is proposed to improve the state estimation of power systems when incomplete information is involved. Finally, three cases are designed, and the simulation results show that for the prediction of states, the root mean square error (RMSE), an indicator of accuracies, is reduced by about 3 multiples, if our missing data filling method is added. Furthermore, the accuracy of state estimation is improved about 5 multiples by the proposed UTKF method, even if incomplete information is involved. |
| first_indexed | 2024-03-13T07:58:51Z |
| format | Article |
| id | doaj.art-1529328015eb468daa0e4d6ff8ac172e |
| institution | Directory Open Access Journal |
| issn | 2169-3536 |
| language | English |
| last_indexed | 2024-03-13T07:58:51Z |
| publishDate | 2023-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Access |
| spelling | doaj.art-1529328015eb468daa0e4d6ff8ac172e2023-06-01T23:00:29ZengIEEEIEEE Access2169-35362023-01-0111507005070910.1109/ACCESS.2023.327761610129202Unscented Trainable Kalman Filter Based on Deep Learning Method Considering Incomplete InformationYanjie Yu0https://orcid.org/0000-0003-4359-8578Qiang Li1https://orcid.org/0000-0002-1899-2808Houyi Zhang2State Key Laboratory of Power Transmission Equipment and System Security and New Technology, School of Electrical Engineering, Chongqing University, Chongqing, ChinaState Key Laboratory of Power Transmission Equipment and System Security and New Technology, School of Electrical Engineering, Chongqing University, Chongqing, ChinaElectric Power Research Institute, Guizhou Power Grid Company Ltd., Guiyang, ChinaRapid changes of states and occurrence of data missing in power systems cause accurate state estimation very hard. In this paper, an unscented trainable Kalman filter (UTKF) with a deep learning prediction model is proposed to provide accurate state estimation under incomplete information. First, the CNN-LSTM architecture, a typical deep learning model, is applied to form a trainable prediction model (TPM), which offers more accurate prediction of states. However, sometimes states are incomplete due to data losses in transmissions. To deal with incomplete information, historical time-series states are employed and fed to the TPM in order to develop a missing data filling method. In this way, the prediction errors can be lower through the online training and parameter adjustment of the TPM. Combining with the TPM and the missing data filling method, an unscented trainable Kalman filter (UTKF) is proposed to improve the state estimation of power systems when incomplete information is involved. Finally, three cases are designed, and the simulation results show that for the prediction of states, the root mean square error (RMSE), an indicator of accuracies, is reduced by about 3 multiples, if our missing data filling method is added. Furthermore, the accuracy of state estimation is improved about 5 multiples by the proposed UTKF method, even if incomplete information is involved.https://ieeexplore.ieee.org/document/10129202/Dynamic state estimationdeep learningprediction modelmissing data fillingunscented Kalman filter |
| spellingShingle | Yanjie Yu Qiang Li Houyi Zhang Unscented Trainable Kalman Filter Based on Deep Learning Method Considering Incomplete Information IEEE Access Dynamic state estimation deep learning prediction model missing data filling unscented Kalman filter |
| title | Unscented Trainable Kalman Filter Based on Deep Learning Method Considering Incomplete Information |
| title_full | Unscented Trainable Kalman Filter Based on Deep Learning Method Considering Incomplete Information |
| title_fullStr | Unscented Trainable Kalman Filter Based on Deep Learning Method Considering Incomplete Information |
| title_full_unstemmed | Unscented Trainable Kalman Filter Based on Deep Learning Method Considering Incomplete Information |
| title_short | Unscented Trainable Kalman Filter Based on Deep Learning Method Considering Incomplete Information |
| title_sort | unscented trainable kalman filter based on deep learning method considering incomplete information |
| topic | Dynamic state estimation deep learning prediction model missing data filling unscented Kalman filter |
| url | https://ieeexplore.ieee.org/document/10129202/ |
| work_keys_str_mv | AT yanjieyu unscentedtrainablekalmanfilterbasedondeeplearningmethodconsideringincompleteinformation AT qiangli unscentedtrainablekalmanfilterbasedondeeplearningmethodconsideringincompleteinformation AT houyizhang unscentedtrainablekalmanfilterbasedondeeplearningmethodconsideringincompleteinformation |