|Title:||Quality control methods in sequential data assimilation system and applications in short-term traffic flow forecasting|
|Advisors:||Shi, Wenzhong (LSGI)|
|Subject:||Traffic flow -- Measurement|
Hong Kong Polytechnic University -- Dissertations
|Department:||Department of Land Surveying and Geo-Informatics|
|Pages:||182 pages : illustrations|
|Abstract:||Data assimilation (DA) is an important methodology that can integrate physical model information and measurements. It plays a significant role in many areas. The basic purpose of data assimilation is to estimate state vectors more accurate through fusing the advantages of model and measurements. However, there are a lot of uncertainties in model and measurements which cause serious negative effects on the reliability of data assimilation. Research on quality control methods in sequential data assimilation system means to study how to control these uncertainties within the scope of application requirements, so as to improve the reliability of sequential data assimilation system. The present dissertation mainly does the following researches: 1. From both model and measurements aspects, uncertainties factors affecting the quality of data assimilation system were analyzed, the quality control methods were reviewed. Then, existing problems in review were analyzed. The objective of this research is the sequential data assimilation system quality control methods from both model and measurements aspects. The basic concepts of data assimilation were summarized, such as its elements, assimilation methods, etc. Also, the characteristics of sequential and continuous data assimilation methods were compared. Then, the theory of uncertainties in sequential data assimilation system was introduced, which helped to establish theoretical foundation for research on quality control methods. 2. A systematic study was conducted on filtering divergence, which affected the reliability of models in the sequential data assimilation system. To restrain the filtering divergence, a new method based on L1 -norm constraints was proposed. The method can adjust weights of the model and measurements depending on new measurements when filtering divergence is about to occur. The L1 -norm constraint method was compared with two other existing methods, i.e., covariance weighted and adaptive methods, in the filtering divergence numerical experiment. Compared with the results from the covariance weighted and adaptive methods, the relative root mean square error values acquired from the L1 -norm constraint method were reduced by 9.71% and 3.10%, respectively. Meanwhile, the L1 -norm constraint method proposed was applied to the short-term traffic flow prediction of part of the highway network in England. To restrain the filtering divergence, the best filter divergence suppression performances were produced from L1-norm-constraint-based method. The average root mean square error value from the L1-norm-constraint-based method decreased by 88.01% on Monday compared to the results from the Kalman filter method with the incorrect model and reduced by 90.87% on Saturday. Furthermore, the average mean absolute percentage error values from the L1-norm-constraint-based method decreased by 89.05% on Monday and 89.64% on Saturday compared to those obtained using the Kalman filter method with an incorrect model. This proved that the L1 -norm constraint method could deal with the filtering divergence efficiently.|
3. To control the issues caused by colored noise in the model, a modified adaptive method based on state residual information was proposed. The method can directly estimate the colored noise based on state residuals, which reduced the effects of colored noise on the assimilation results. The modified adaptive method was then compared with the traditional Kalman filter method, vector amplification method, and covariance matrix adaptive method in target tracking experiments. Under three conditions, the relative root mean square error values of the modified adaptive method were reduced by 60.07%, 18.61%, and 27.30%, respectively, compared to the traditional Kalman filter method. Compared with the results from the covariance matrix adaptive method, which is a common method for dealing with colored noise, the relative root mean square error values acquired using the state residual adaptive method were reduced by 47.41%, 7.86%, and 10.31% under the three conditions, respectively. Meanwhile, a modified adaptive method was applied to the short-term traffic flow prediction of part of the highway network in England. The root mean square error values and mean absolute percentage error values of the prediction results based on the modified adaptive method were 10.47% and 9.66% lower than traditional Kalman filter method, respectively. Conclusions can be drawn that the modified adaptive method proposed in this paper could control the effects of colored noise on the assimilation prediction results and improve its accuracy. 4. Research on the quality control in measurements of sequential data assimialtion system mainly focused on how to reduce the impacts of local noises in the measurements on the assimilation results. A multi-scale sequential data assimilation method was proposed. The method was the combination of traditional Kalman filter method and three noise separation methods, i.e., the wavelet transform method, empirical mode decomposition method, and fast Fourier transform method, respectively. The method proposed can reduce influences of noisy information in measurements on accuracies of assimilation models construction and results prediction. In addition, an adaptive noise separation method was proposed for noise separation in the FFT method. This method could separate noisy information based on the observation characteristics. The average root mean square error and mean absolute percentage error values of the prediction results based on this method were 17.09% and 17.61% lower than the results acquired from the traditional Kalman filter method, respectively. This proved the effectiveness of the adaptive noise separation method proposed in this dissertation. 5. The multi-scale sequential data assimilation system for short-term traffic flow predictions was built based on the methods proposed in the model and measurements for improving the reliability of sequential data assimilation system. It was then applied into the short-term traffic flow prediction of the highway network in England. The influence of the noise separation scales on the assimilation prediction results was discussed. With the optimal noise separation scales, the root mean square error and mean absolute percentage error values of the prediction results obtained from multi-scale sequential data assimilation system were lower than ones from traditional sequential data assimilation system. This proved that the multi-scale data assimilation methods could suppress the influence of local noise in measurements on the assimilation system and achieve better precision in traffic flow prediction results.
|Rights:||All rights reserved|
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item: